d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
3c502d3b75
freebsd-nq/module/zfs/zil.c

3689 lines
111 KiB
C
Raw Normal View History

Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
OpenZFS 9112 - Improve allocation performance on high-end systems Overview ======== We parallelize the allocation process by creating the concept of "allocators". There are a certain number of allocators per metaslab group, defined by the value of a tunable at pool open time. Each allocator for a given metaslab group has up to 2 active metaslabs; one "primary", and one "secondary". The primary and secondary weight mean the same thing they did in in the pre-allocator world; primary metaslabs are used for most allocations, secondary metaslabs are used for ditto blocks being allocated in the same metaslab group. There is also the CLAIM weight, which has been separated out from the other weights, but that is less important to understanding the patch. The active metaslabs for each allocator are moved from their normal place in the metaslab tree for the group to the back of the tree. This way, they will not be selected for use by other allocators searching for new metaslabs unless all the passive metaslabs are unsuitable for allocations. If that does happen, the allocators will "steal" from each other to ensure that IOs don't fail until there is truly no space left to perform allocations. In addition, the alloc queue for each metaslab group has been broken into a separate queue for each allocator. We don't want to dramatically increase the number of inflight IOs on low-end systems, because it can significantly increase txg times. On the other hand, we want to ensure that there are enough IOs for each allocator to allow for good coalescing before sending the IOs to the disk. As a result, we take a compromise path; each allocator's alloc queue max depth starts at a certain value for every txg. Every time an IO completes, we increase the max depth. This should hopefully provide a good balance between the two failure modes, while not dramatically increasing complexity. We also parallelize the spa_alloc_tree and spa_alloc_lock, which cause very similar contention when selecting IOs to allocate. This parallelization uses the same allocator scheme as metaslab selection. Performance Results =================== Performance improvements from this change can vary significantly based on the number of CPUs in the system, whether or not the system has a NUMA architecture, the speed of the drives, the values for the various tunables, and the workload being performed. For an fio async sequential write workload on a 24 core NUMA system with 256 GB of RAM and 8 128 GB SSDs, there is a roughly 25% performance improvement. Future Work =========== Analysis of the performance of the system with this patch applied shows that a significant new bottleneck is the vdev disk queues, which also need to be parallelized. Prototyping of this change has occurred, and there was a performance improvement, but more work needs to be done before its stability has been verified and it is ready to be upstreamed. Authored by: Paul Dagnelie <pcd@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Gordon Ross <gwr@nexenta.com> Ported-by: Paul Dagnelie <pcd@delphix.com> Signed-off-by: Paul Dagnelie <pcd@delphix.com> Porting Notes: * Fix reservation test failures by increasing tolerance. OpenZFS-issue: https://illumos.org/issues/9112 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/3f3cc3c3 Closes #7682
2018-02-12 20:56:06 +00:00
* Copyright (c) 2011, 2018 by Delphix. All rights reserved.
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
* Copyright (c) 2014 Integros [integros.com]
Wrong error message when removing log device In the case where the pool is loaded without the crypto keys necessary to playback the intent log, and log device removal is attempted, a generic busy message is received. Change the message to inform the user that the datasets must be mounted. Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Tom Caputi <tcaputi@datto.com> Signed-off-by: Paul Zuchowski <pzuchowski@datto.com> Closes #7518
2018-06-08 01:07:29 +00:00
* Copyright (c) 2018 Datto Inc.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
/* Portions Copyright 2010 Robert Milkowski */
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
#include <sys/zfs_context.h>
#include <sys/spa.h>
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
#include <sys/spa_impl.h>
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
#include <sys/dmu.h>
#include <sys/zap.h>
#include <sys/arc.h>
#include <sys/stat.h>
#include <sys/zil.h>
#include <sys/zil_impl.h>
#include <sys/dsl_dataset.h>
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
#include <sys/vdev_impl.h>
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
#include <sys/dmu_tx.h>
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
#include <sys/dsl_pool.h>
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
#include <sys/metaslab.h>
Enable use of DTRACE_PROBE* macros in "spl" module This change modifies some of the infrastructure for enabling the use of the DTRACE_PROBE* macros, such that we can use tehm in the "spl" module. Currently, when the DTRACE_PROBE* macros are used, they get expanded to create new functions, and these dynamically generated functions become part of the "zfs" module. Since the "spl" module does not depend on the "zfs" module, the use of DTRACE_PROBE* in the "spl" module would result in undefined symbols being used in the "spl" module. Specifically, DTRACE_PROBE* would turn into a function call, and the function being called would be a symbol only contained in the "zfs" module; which results in a linker and/or runtime error. Thus, this change adds the necessary logic to the "spl" module, to mirror the tracing functionality available to the "zfs" module. After this change, we'll have a "trace_zfs.h" header file which defines the probes available only to the "zfs" module, and a "trace_spl.h" header file which defines the probes available only to the "spl" module. Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Prakash Surya <prakash.surya@delphix.com> Closes #9525
2019-10-30 18:02:41 +00:00
#include <sys/trace_zfs.h>
DLPX-44812 integrate EP-220 large memory scalability
2016-07-22 15:52:49 +00:00
#include <sys/abd.h>
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* The ZFS Intent Log (ZIL) saves "transaction records" (itxs) of system
* calls that change the file system. Each itx has enough information to
* be able to replay them after a system crash, power loss, or
* equivalent failure mode. These are stored in memory until either:
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* 1. they are committed to the pool by the DMU transaction group
* (txg), at which point they can be discarded; or
* 2. they are committed to the on-disk ZIL for the dataset being
* modified (e.g. due to an fsync, O_DSYNC, or other synchronous
* requirement).
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* In the event of a crash or power loss, the itxs contained by each
* dataset's on-disk ZIL will be replayed when that dataset is first
Fix typos in module/zfs/ Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Ryan Moeller <ryan@ixsystems.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Andrea Gelmini <andrea.gelmini@gelma.net> Closes #9240
2019-09-03 00:56:41 +00:00
* instantiated (e.g. if the dataset is a normal filesystem, when it is
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* first mounted).
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* As hinted at above, there is one ZIL per dataset (both the in-memory
* representation, and the on-disk representation). The on-disk format
* consists of 3 parts:
*
* - a single, per-dataset, ZIL header; which points to a chain of
* - zero or more ZIL blocks; each of which contains
* - zero or more ZIL records
*
* A ZIL record holds the information necessary to replay a single
* system call transaction. A ZIL block can hold many ZIL records, and
* the blocks are chained together, similarly to a singly linked list.
*
* Each ZIL block contains a block pointer (blkptr_t) to the next ZIL
* block in the chain, and the ZIL header points to the first block in
* the chain.
*
* Note, there is not a fixed place in the pool to hold these ZIL
* blocks; they are dynamically allocated and freed as needed from the
* blocks available on the pool, though they can be preferentially
* allocated from a dedicated "log" vdev.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* This controls the amount of time that a ZIL block (lwb) will remain
* "open" when it isn't "full", and it has a thread waiting for it to be
* committed to stable storage. Please refer to the zil_commit_waiter()
* function (and the comments within it) for more details.
*/
int zfs_commit_timeout_pct = 5;
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
/*
* See zil.h for more information about these fields.
*/
zil_stats_t zil_stats = {
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
{ "zil_commit_count", KSTAT_DATA_UINT64 },
{ "zil_commit_writer_count", KSTAT_DATA_UINT64 },
{ "zil_itx_count", KSTAT_DATA_UINT64 },
{ "zil_itx_indirect_count", KSTAT_DATA_UINT64 },
{ "zil_itx_indirect_bytes", KSTAT_DATA_UINT64 },
{ "zil_itx_copied_count", KSTAT_DATA_UINT64 },
{ "zil_itx_copied_bytes", KSTAT_DATA_UINT64 },
{ "zil_itx_needcopy_count", KSTAT_DATA_UINT64 },
{ "zil_itx_needcopy_bytes", KSTAT_DATA_UINT64 },
{ "zil_itx_metaslab_normal_count", KSTAT_DATA_UINT64 },
{ "zil_itx_metaslab_normal_bytes", KSTAT_DATA_UINT64 },
{ "zil_itx_metaslab_slog_count", KSTAT_DATA_UINT64 },
{ "zil_itx_metaslab_slog_bytes", KSTAT_DATA_UINT64 },
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
};
static kstat_t *zil_ksp;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
Illumos #3742 3742 zfs comments need cleaner, more consistent style Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Eric Schrock <eric.schrock@delphix.com> Approved by: Christopher Siden <christopher.siden@delphix.com> References: https://www.illumos.org/issues/3742 illumos/illumos-gate@f7170741490edba9d1d9c697c177c887172bc741 Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. The change to zfs_vfsops.c was dropped because it involves zfs_mount_label_policy, which does not exist in the Linux port.
2013-06-11 17:12:34 +00:00
* Disable intent logging replay. This global ZIL switch affects all pools.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Illumos #3742 3742 zfs comments need cleaner, more consistent style Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Eric Schrock <eric.schrock@delphix.com> Approved by: Christopher Siden <christopher.siden@delphix.com> References: https://www.illumos.org/issues/3742 illumos/illumos-gate@f7170741490edba9d1d9c697c177c887172bc741 Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. The change to zfs_vfsops.c was dropped because it involves zfs_mount_label_policy, which does not exist in the Linux port.
2013-06-11 17:12:34 +00:00
int zil_replay_disable = 0;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 9963 - Separate tunable for disabling ZIL vdev flush Porting Notes: * Add options to zfs-module-parameters(5) man page. * zfs_nocacheflush move to vdev.c instead of vdev_disk.c, since the latter doesn't get built for user space. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Tom Caputi <tcaputi@datto.com> Reviewed by: George Melikov <mail@gmelikov.ru> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9963 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/f8fdf68125 Closes #8186
2018-10-23 16:44:37 +00:00
* Disable the DKIOCFLUSHWRITECACHE commands that are normally sent to
* the disk(s) by the ZIL after an LWB write has completed. Setting this
* will cause ZIL corruption on power loss if a volatile out-of-order
* write cache is enabled.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 9963 - Separate tunable for disabling ZIL vdev flush Porting Notes: * Add options to zfs-module-parameters(5) man page. * zfs_nocacheflush move to vdev.c instead of vdev_disk.c, since the latter doesn't get built for user space. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Tom Caputi <tcaputi@datto.com> Reviewed by: George Melikov <mail@gmelikov.ru> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9963 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/f8fdf68125 Closes #8186
2018-10-23 16:44:37 +00:00
int zil_nocacheflush = 0;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
/*
* Limit SLOG write size per commit executed with synchronous priority.
* Any writes above that will be executed with lower (asynchronous) priority
* to limit potential SLOG device abuse by single active ZIL writer.
*/
unsigned long zil_slog_bulk = 768 * 1024;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
static kmem_cache_t *zil_lwb_cache;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
static kmem_cache_t *zil_zcw_cache;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
#define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \
sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused))
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
static int
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_bp_compare(const void *x1, const void *x2)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva;
const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Reduce loaded range tree memory usage This patch implements a new tree structure for ZFS, and uses it to store range trees more efficiently. The new structure is approximately a B-tree, though there are some small differences from the usual characterizations. The tree has core nodes and leaf nodes; each contain data elements, which the elements in the core nodes acting as separators between its children. The difference between core and leaf nodes is that the core nodes have an array of children, while leaf nodes don't. Every node in the tree may be only partially full; in most cases, they are all at least 50% full (in terms of element count) except for the root node, which can be less full. Underfull nodes will steal from their neighbors or merge to remain full enough, while overfull nodes will split in two. The data elements are contained in tree-controlled buffers; they are copied into these on insertion, and overwritten on deletion. This means that the elements are not independently allocated, which reduces overhead, but also means they can't be shared between trees (and also that pointers to them are only valid until a side-effectful tree operation occurs). The overhead varies based on how dense the tree is, but is usually on the order of about 50% of the element size; the per-node overheads are very small, and so don't make a significant difference. The trees can accept arbitrary records; they accept a size and a comparator to allow them to be used for a variety of purposes. The new trees replace the AVL trees used in the range trees today. Currently, the range_seg_t structure contains three 8 byte integers of payload and two 24 byte avl_tree_node_ts to handle its storage in both an offset-sorted tree and a size-sorted tree (total size: 64 bytes). In the new model, the range seg structures are usually two 4 byte integers, but a separate one needs to exist for the size-sorted and offset-sorted tree. Between the raw size, the 50% overhead, and the double storage, the new btrees are expected to use 8*1.5*2 = 24 bytes per record, or 33.3% as much memory as the AVL trees (this is for the purposes of storing metaslab range trees; for other purposes, like scrubs, they use ~50% as much memory). We reduced the size of the payload in the range segments by teaching range trees about starting offsets and shifts; since metaslabs have a fixed starting offset, and they all operate in terms of disk sectors, we can store the ranges using 4-byte integers as long as the size of the metaslab divided by the sector size is less than 2^32. For 512-byte sectors, this is a 2^41 (or 2TB) metaslab, which with the default settings corresponds to a 256PB disk. 4k sector disks can handle metaslabs up to 2^46 bytes, or 2^63 byte disks. Since we do not anticipate disks of this size in the near future, there should be almost no cases where metaslabs need 64-byte integers to store their ranges. We do still have the capability to store 64-byte integer ranges to account for cases where we are storing per-vdev (or per-dnode) trees, which could reasonably go above the limits discussed. We also do not store fill information in the compact version of the node, since it is only used for sorted scrub. We also optimized the metaslab loading process in various other ways to offset some inefficiencies in the btree model. While individual operations (find, insert, remove_from) are faster for the btree than they are for the avl tree, remove usually requires a find operation, while in the AVL tree model the element itself suffices. Some clever changes actually caused an overall speedup in metaslab loading; we use approximately 40% less cpu to load metaslabs in our tests on Illumos. Another memory and performance optimization was achieved by changing what is stored in the size-sorted trees. When a disk is heavily fragmented, the df algorithm used by default in ZFS will almost always find a number of small regions in its initial cursor-based search; it will usually only fall back to the size-sorted tree to find larger regions. If we increase the size of the cursor-based search slightly, and don't store segments that are smaller than a tunable size floor in the size-sorted tree, we can further cut memory usage down to below 20% of what the AVL trees store. This also results in further reductions in CPU time spent loading metaslabs. The 16KiB size floor was chosen because it results in substantial memory usage reduction while not usually resulting in situations where we can't find an appropriate chunk with the cursor and are forced to use an oversized chunk from the size-sorted tree. In addition, even if we do have to use an oversized chunk from the size-sorted tree, the chunk would be too small to use for ZIL allocations, so it isn't as big of a loss as it might otherwise be. And often, more small allocations will follow the initial one, and the cursor search will now find the remainder of the chunk we didn't use all of and use it for subsequent allocations. Practical testing has shown little or no change in fragmentation as a result of this change. If the size-sorted tree becomes empty while the offset sorted one still has entries, it will load all the entries from the offset sorted tree and disregard the size floor until it is unloaded again. This operation occurs rarely with the default setting, only on incredibly thoroughly fragmented pools. There are some other small changes to zdb to teach it to handle btrees, but nothing major. Reviewed-by: George Wilson <gwilson@delphix.com> Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed by: Sebastien Roy seb@delphix.com Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Paul Dagnelie <pcd@delphix.com> Closes #9181
2019-10-09 17:36:03 +00:00
int cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
Performance optimization of AVL tree comparator functions perf: 2.75x faster ddt_entry_compare() First 256bits of ddt_key_t is a block checksum, which are expected to be close to random data. Hence, on average, comparison only needs to look at first few bytes of the keys. To reduce number of conditional jump instructions, the result is computed as: sign(memcmp(k1, k2)). Sign of an integer 'a' can be obtained as: `(0 < a) - (a < 0)` := {-1, 0, 1} , which is computed efficiently. Synthetic performance evaluation of original and new algorithm over 1G random keys on 2.6GHz Intel(R) Xeon(R) CPU E5-2660 v3: old 6.85789 s new 2.49089 s perf: 2.8x faster vdev_queue_offset_compare() and vdev_queue_timestamp_compare() Compute the result directly instead of using conditionals perf: zfs_range_compare() Speedup between 1.1x - 2.5x, depending on compiler version and optimization level. perf: spa_error_entry_compare() `bcmp()` is not suitable for comparator use. Use `memcmp()` instead. perf: 2.8x faster metaslab_compare() and metaslab_rangesize_compare() perf: 2.8x faster zil_bp_compare() perf: 2.8x faster mze_compare() perf: faster dbuf_compare() perf: faster compares in spa_misc perf: 2.8x faster layout_hash_compare() perf: 2.8x faster space_reftree_compare() perf: libzfs: faster avl tree comparators perf: guid_compare() perf: dsl_deadlist_compare() perf: perm_set_compare() perf: 2x faster range_tree_seg_compare() perf: faster unique_compare() perf: faster vdev_cache _compare() perf: faster vdev_uberblock_compare() perf: faster fuid _compare() perf: faster zfs_znode_hold_compare() Signed-off-by: Gvozden Neskovic <neskovic@gmail.com> Signed-off-by: Richard Elling <richard.elling@gmail.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #5033
2016-08-27 18:12:53 +00:00
if (likely(cmp))
return (cmp);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Reduce loaded range tree memory usage This patch implements a new tree structure for ZFS, and uses it to store range trees more efficiently. The new structure is approximately a B-tree, though there are some small differences from the usual characterizations. The tree has core nodes and leaf nodes; each contain data elements, which the elements in the core nodes acting as separators between its children. The difference between core and leaf nodes is that the core nodes have an array of children, while leaf nodes don't. Every node in the tree may be only partially full; in most cases, they are all at least 50% full (in terms of element count) except for the root node, which can be less full. Underfull nodes will steal from their neighbors or merge to remain full enough, while overfull nodes will split in two. The data elements are contained in tree-controlled buffers; they are copied into these on insertion, and overwritten on deletion. This means that the elements are not independently allocated, which reduces overhead, but also means they can't be shared between trees (and also that pointers to them are only valid until a side-effectful tree operation occurs). The overhead varies based on how dense the tree is, but is usually on the order of about 50% of the element size; the per-node overheads are very small, and so don't make a significant difference. The trees can accept arbitrary records; they accept a size and a comparator to allow them to be used for a variety of purposes. The new trees replace the AVL trees used in the range trees today. Currently, the range_seg_t structure contains three 8 byte integers of payload and two 24 byte avl_tree_node_ts to handle its storage in both an offset-sorted tree and a size-sorted tree (total size: 64 bytes). In the new model, the range seg structures are usually two 4 byte integers, but a separate one needs to exist for the size-sorted and offset-sorted tree. Between the raw size, the 50% overhead, and the double storage, the new btrees are expected to use 8*1.5*2 = 24 bytes per record, or 33.3% as much memory as the AVL trees (this is for the purposes of storing metaslab range trees; for other purposes, like scrubs, they use ~50% as much memory). We reduced the size of the payload in the range segments by teaching range trees about starting offsets and shifts; since metaslabs have a fixed starting offset, and they all operate in terms of disk sectors, we can store the ranges using 4-byte integers as long as the size of the metaslab divided by the sector size is less than 2^32. For 512-byte sectors, this is a 2^41 (or 2TB) metaslab, which with the default settings corresponds to a 256PB disk. 4k sector disks can handle metaslabs up to 2^46 bytes, or 2^63 byte disks. Since we do not anticipate disks of this size in the near future, there should be almost no cases where metaslabs need 64-byte integers to store their ranges. We do still have the capability to store 64-byte integer ranges to account for cases where we are storing per-vdev (or per-dnode) trees, which could reasonably go above the limits discussed. We also do not store fill information in the compact version of the node, since it is only used for sorted scrub. We also optimized the metaslab loading process in various other ways to offset some inefficiencies in the btree model. While individual operations (find, insert, remove_from) are faster for the btree than they are for the avl tree, remove usually requires a find operation, while in the AVL tree model the element itself suffices. Some clever changes actually caused an overall speedup in metaslab loading; we use approximately 40% less cpu to load metaslabs in our tests on Illumos. Another memory and performance optimization was achieved by changing what is stored in the size-sorted trees. When a disk is heavily fragmented, the df algorithm used by default in ZFS will almost always find a number of small regions in its initial cursor-based search; it will usually only fall back to the size-sorted tree to find larger regions. If we increase the size of the cursor-based search slightly, and don't store segments that are smaller than a tunable size floor in the size-sorted tree, we can further cut memory usage down to below 20% of what the AVL trees store. This also results in further reductions in CPU time spent loading metaslabs. The 16KiB size floor was chosen because it results in substantial memory usage reduction while not usually resulting in situations where we can't find an appropriate chunk with the cursor and are forced to use an oversized chunk from the size-sorted tree. In addition, even if we do have to use an oversized chunk from the size-sorted tree, the chunk would be too small to use for ZIL allocations, so it isn't as big of a loss as it might otherwise be. And often, more small allocations will follow the initial one, and the cursor search will now find the remainder of the chunk we didn't use all of and use it for subsequent allocations. Practical testing has shown little or no change in fragmentation as a result of this change. If the size-sorted tree becomes empty while the offset sorted one still has entries, it will load all the entries from the offset sorted tree and disregard the size floor until it is unloaded again. This operation occurs rarely with the default setting, only on incredibly thoroughly fragmented pools. There are some other small changes to zdb to teach it to handle btrees, but nothing major. Reviewed-by: George Wilson <gwilson@delphix.com> Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed by: Sebastien Roy seb@delphix.com Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Paul Dagnelie <pcd@delphix.com> Closes #9181
2019-10-09 17:36:03 +00:00
return (TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2)));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
static void
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_bp_tree_init(zilog_t *zilog)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
avl_create(&zilog->zl_bp_tree, zil_bp_compare,
sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
static void
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_bp_tree_fini(zilog_t *zilog)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
avl_tree_t *t = &zilog->zl_bp_tree;
zil_bp_node_t *zn;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
void *cookie = NULL;
while ((zn = avl_destroy_nodes(t, &cookie)) != NULL)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
kmem_free(zn, sizeof (zil_bp_node_t));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
avl_destroy(t);
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
int
zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
avl_tree_t *t = &zilog->zl_bp_tree;
Illumos 4757, 4913 4757 ZFS embedded-data block pointers ("zero block compression") 4913 zfs release should not be subject to space checks Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Max Grossman <max.grossman@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/4757 https://www.illumos.org/issues/4913 https://github.com/illumos/illumos-gate/commit/5d7b4d4 Porting notes: For compatibility with the fastpath code the zio_done() function needed to be updated. Because embedded-data block pointers do not require DVAs to be allocated the associated vdevs will not be marked and therefore should not be unmarked. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2544
2014-06-05 21:19:08 +00:00
const dva_t *dva;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_bp_node_t *zn;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
avl_index_t where;
Illumos 4757, 4913 4757 ZFS embedded-data block pointers ("zero block compression") 4913 zfs release should not be subject to space checks Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Max Grossman <max.grossman@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/4757 https://www.illumos.org/issues/4913 https://github.com/illumos/illumos-gate/commit/5d7b4d4 Porting notes: For compatibility with the fastpath code the zio_done() function needed to be updated. Because embedded-data block pointers do not require DVAs to be allocated the associated vdevs will not be marked and therefore should not be unmarked. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2544
2014-06-05 21:19:08 +00:00
if (BP_IS_EMBEDDED(bp))
return (0);
dva = BP_IDENTITY(bp);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (avl_find(t, dva, &where) != NULL)
Illumos #3598 3598 want to dtrace when errors are generated in zfs Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3598 illumos/illumos-gate@be6fd75a69ae679453d9cda5bff3326111e6d1ca Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. include/sys/zfs_context.h has been modified to render some new macros inert until dtrace is available on Linux. 2. Linux-specific changes have been adapted to use SET_ERROR(). 3. I'm NOT happy about this change. It does nothing but ugly up the code under Linux. Unfortunately we need to take it to avoid more merge conflicts in the future. -Brian
2013-03-08 18:41:28 +00:00
return (SET_ERROR(EEXIST));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zn->zn_dva = *dva;
avl_insert(t, zn, where);
return (0);
}
static zil_header_t *
zil_header_in_syncing_context(zilog_t *zilog)
{
return ((zil_header_t *)zilog->zl_header);
}
static void
zil_init_log_chain(zilog_t *zilog, blkptr_t *bp)
{
zio_cksum_t *zc = &bp->blk_cksum;
zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL);
zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL);
zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os);
zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
}
/*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* Read a log block and make sure it's valid.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
static int
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
zil_read_log_block(zilog_t *zilog, boolean_t decrypt, const blkptr_t *bp,
blkptr_t *nbp, void *dst, char **end)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
Illumos 5369 - arc flags should be an enum 5369 arc flags should be an enum 5370 consistent arc_buf_hdr_t naming scheme Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Alex Reece <alex.reece@delphix.com> Reviewed by: Sebastien Roy <sebastien.roy@delphix.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Approved by: Richard Lowe <richlowe@richlowe.net> Porting notes: ZoL has moved some ARC definitions into arc_impl.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported by: Tim Chase <tim@chase2k.com>
2014-12-06 17:24:32 +00:00
arc_flags_t aflags = ARC_FLAG_WAIT;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
arc_buf_t *abuf = NULL;
Illumos 4914 - zfs on-disk bookmark structure should be named *_phys_t 4914 zfs on-disk bookmark structure should be named *_phys_t Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Richard Lowe <richlowe@richlowe.net> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Approved by: Robert Mustacchi <rm@joyent.com> References: https://www.illumos.org/issues/4914 https://github.com/illumos/illumos-gate/commit/7802d7b Porting notes: There were a number of zfsonlinux-specific uses of zbookmark_t which needed to be updated. This should reduce the likelihood of further problems like issue #2094 from occurring. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2558
2014-06-25 18:37:59 +00:00
zbookmark_phys_t zb;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
int error;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (zilog->zl_header->zh_claim_txg == 0)
zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
zio_flags |= ZIO_FLAG_SPECULATIVE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
if (!decrypt)
zio_flags |= ZIO_FLAG_RAW;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET],
ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func,
&abuf, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (error == 0) {
zio_cksum_t cksum = bp->blk_cksum;
/*
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
* Validate the checksummed log block.
*
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
* Sequence numbers should be... sequential. The checksum
* verifier for the next block should be bp's checksum plus 1.
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
*
* Also check the log chain linkage and size used.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
cksum.zc_word[ZIL_ZC_SEQ]++;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
zil_chain_t *zilc = abuf->b_data;
char *lr = (char *)(zilc + 1);
uint64_t len = zilc->zc_nused - sizeof (zil_chain_t);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) {
Illumos #3598 3598 want to dtrace when errors are generated in zfs Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3598 illumos/illumos-gate@be6fd75a69ae679453d9cda5bff3326111e6d1ca Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. include/sys/zfs_context.h has been modified to render some new macros inert until dtrace is available on Linux. 2. Linux-specific changes have been adapted to use SET_ERROR(). 3. I'm NOT happy about this change. It does nothing but ugly up the code under Linux. Unfortunately we need to take it to avoid more merge conflicts in the future. -Brian
2013-03-08 18:41:28 +00:00
error = SET_ERROR(ECKSUM);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
} else {
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
ASSERT3U(len, <=, SPA_OLD_MAXBLOCKSIZE);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
bcopy(lr, dst, len);
*end = (char *)dst + len;
*nbp = zilc->zc_next_blk;
}
} else {
char *lr = abuf->b_data;
uint64_t size = BP_GET_LSIZE(bp);
zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1;
if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum,
sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) ||
(zilc->zc_nused > (size - sizeof (*zilc)))) {
Illumos #3598 3598 want to dtrace when errors are generated in zfs Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3598 illumos/illumos-gate@be6fd75a69ae679453d9cda5bff3326111e6d1ca Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. include/sys/zfs_context.h has been modified to render some new macros inert until dtrace is available on Linux. 2. Linux-specific changes have been adapted to use SET_ERROR(). 3. I'm NOT happy about this change. It does nothing but ugly up the code under Linux. Unfortunately we need to take it to avoid more merge conflicts in the future. -Brian
2013-03-08 18:41:28 +00:00
error = SET_ERROR(ECKSUM);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
} else {
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
ASSERT3U(zilc->zc_nused, <=,
SPA_OLD_MAXBLOCKSIZE);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
bcopy(lr, dst, zilc->zc_nused);
*end = (char *)dst + zilc->zc_nused;
*nbp = zilc->zc_next_blk;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
OpenZFS 6950 - ARC should cache compressed data Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: Paul Dagnelie <pcd@delphix.com> Reviewed by: Tom Caputi <tcaputi@datto.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Ported by: David Quigley <david.quigley@intel.com> This review covers the reading and writing of compressed arc headers, sharing data between the arc_hdr_t and the arc_buf_t, and the implementation of a new dbuf cache to keep frequently access data uncompressed. I've added a new member to l1 arc hdr called b_pdata. The b_pdata always hangs off the arc_buf_hdr_t (if an L1 hdr is in use) and points to the physical block for that DVA. The physical block may or may not be compressed. If compressed arc is enabled and the block on-disk is compressed, then the b_pdata will match the block on-disk and remain compressed in memory. If the block on disk is not compressed, then neither will the b_pdata. Lastly, if compressed arc is disabled, then b_pdata will always be an uncompressed version of the on-disk block. Typically the arc will cache only the arc_buf_hdr_t and will aggressively evict any arc_buf_t's that are no longer referenced. This means that the arc will primarily have compressed blocks as the arc_buf_t's are considered overhead and are always uncompressed. When a consumer reads a block we first look to see if the arc_buf_hdr_t is cached. If the hdr is cached then we allocate a new arc_buf_t and decompress the b_pdata contents into the arc_buf_t's b_data. If the hdr already has a arc_buf_t, then we will allocate an additional arc_buf_t and bcopy the uncompressed contents from the first arc_buf_t to the new one. Writing to the compressed arc requires that we first discard the b_pdata since the physical block is about to be rewritten. The new data contents will be passed in via an arc_buf_t (uncompressed) and during the I/O pipeline stages we will copy the physical block contents to a newly allocated b_pdata. When an l2arc is inuse it will also take advantage of the b_pdata. Now the l2arc will always write the contents of b_pdata to the l2arc. This means that when compressed arc is enabled that the l2arc blocks are identical to those stored in the main data pool. This provides a significant advantage since we can leverage the bp's checksum when reading from the l2arc to determine if the contents are valid. If the compressed arc is disabled, then we must first transform the read block to look like the physical block in the main data pool before comparing the checksum and determining it's valid. OpenZFS-issue: https://www.illumos.org/issues/6950 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7fc10f0 Issue #5078
2016-06-02 04:04:53 +00:00
arc_buf_destroy(abuf, &abuf);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
}
return (error);
}
/*
* Read a TX_WRITE log data block.
*/
static int
zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf)
{
enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
const blkptr_t *bp = &lr->lr_blkptr;
Illumos 5369 - arc flags should be an enum 5369 arc flags should be an enum 5370 consistent arc_buf_hdr_t naming scheme Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Alex Reece <alex.reece@delphix.com> Reviewed by: Sebastien Roy <sebastien.roy@delphix.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Approved by: Richard Lowe <richlowe@richlowe.net> Porting notes: ZoL has moved some ARC definitions into arc_impl.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported by: Tim Chase <tim@chase2k.com>
2014-12-06 17:24:32 +00:00
arc_flags_t aflags = ARC_FLAG_WAIT;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
arc_buf_t *abuf = NULL;
Illumos 4914 - zfs on-disk bookmark structure should be named *_phys_t 4914 zfs on-disk bookmark structure should be named *_phys_t Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Richard Lowe <richlowe@richlowe.net> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Approved by: Robert Mustacchi <rm@joyent.com> References: https://www.illumos.org/issues/4914 https://github.com/illumos/illumos-gate/commit/7802d7b Porting notes: There were a number of zfsonlinux-specific uses of zbookmark_t which needed to be updated. This should reduce the likelihood of further problems like issue #2094 from occurring. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2558
2014-06-25 18:37:59 +00:00
zbookmark_phys_t zb;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
int error;
if (BP_IS_HOLE(bp)) {
if (wbuf != NULL)
bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length));
return (0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (zilog->zl_header->zh_claim_txg == 0)
zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
/*
* If we are not using the resulting data, we are just checking that
* it hasn't been corrupted so we don't need to waste CPU time
* decompressing and decrypting it.
*/
if (wbuf == NULL)
zio_flags |= ZIO_FLAG_RAW;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid,
ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
Illumos #3498 panic in arc_read() 3498 panic in arc_read(): !refcount_is_zero(&pbuf->b_hdr->b_refcnt) Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Approved by: Richard Lowe <richlowe@richlowe.net> References: illumos/illumos-gate@1b912ec7100c10e7243bf0879af0fe580e08c73d https://www.illumos.org/issues/3498 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1249
2013-07-02 20:26:24 +00:00
error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
if (error == 0) {
if (wbuf != NULL)
bcopy(abuf->b_data, wbuf, arc_buf_size(abuf));
OpenZFS 6950 - ARC should cache compressed data Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: Paul Dagnelie <pcd@delphix.com> Reviewed by: Tom Caputi <tcaputi@datto.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Ported by: David Quigley <david.quigley@intel.com> This review covers the reading and writing of compressed arc headers, sharing data between the arc_hdr_t and the arc_buf_t, and the implementation of a new dbuf cache to keep frequently access data uncompressed. I've added a new member to l1 arc hdr called b_pdata. The b_pdata always hangs off the arc_buf_hdr_t (if an L1 hdr is in use) and points to the physical block for that DVA. The physical block may or may not be compressed. If compressed arc is enabled and the block on-disk is compressed, then the b_pdata will match the block on-disk and remain compressed in memory. If the block on disk is not compressed, then neither will the b_pdata. Lastly, if compressed arc is disabled, then b_pdata will always be an uncompressed version of the on-disk block. Typically the arc will cache only the arc_buf_hdr_t and will aggressively evict any arc_buf_t's that are no longer referenced. This means that the arc will primarily have compressed blocks as the arc_buf_t's are considered overhead and are always uncompressed. When a consumer reads a block we first look to see if the arc_buf_hdr_t is cached. If the hdr is cached then we allocate a new arc_buf_t and decompress the b_pdata contents into the arc_buf_t's b_data. If the hdr already has a arc_buf_t, then we will allocate an additional arc_buf_t and bcopy the uncompressed contents from the first arc_buf_t to the new one. Writing to the compressed arc requires that we first discard the b_pdata since the physical block is about to be rewritten. The new data contents will be passed in via an arc_buf_t (uncompressed) and during the I/O pipeline stages we will copy the physical block contents to a newly allocated b_pdata. When an l2arc is inuse it will also take advantage of the b_pdata. Now the l2arc will always write the contents of b_pdata to the l2arc. This means that when compressed arc is enabled that the l2arc blocks are identical to those stored in the main data pool. This provides a significant advantage since we can leverage the bp's checksum when reading from the l2arc to determine if the contents are valid. If the compressed arc is disabled, then we must first transform the read block to look like the physical block in the main data pool before comparing the checksum and determining it's valid. OpenZFS-issue: https://www.illumos.org/issues/6950 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7fc10f0 Issue #5078
2016-06-02 04:04:53 +00:00
arc_buf_destroy(abuf, &abuf);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (error);
}
/*
* Parse the intent log, and call parse_func for each valid record within.
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg,
boolean_t decrypt)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
const zil_header_t *zh = zilog->zl_header;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
boolean_t claimed = !!zh->zh_claim_txg;
uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX;
uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX;
uint64_t max_blk_seq = 0;
uint64_t max_lr_seq = 0;
uint64_t blk_count = 0;
uint64_t lr_count = 0;
blkptr_t blk, next_blk;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
char *lrbuf, *lrp;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
int error = 0;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
bzero(&next_blk, sizeof (blkptr_t));
Fix gcc uninitialized variable warnings Gcc -Wall warn: 'uninitialized variable' Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2010-08-26 16:58:04 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
/*
* Old logs didn't record the maximum zh_claim_lr_seq.
*/
if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
claim_lr_seq = UINT64_MAX;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Starting at the block pointed to by zh_log we read the log chain.
* For each block in the chain we strongly check that block to
* ensure its validity. We stop when an invalid block is found.
* For each block pointer in the chain we call parse_blk_func().
* For each record in each valid block we call parse_lr_func().
* If the log has been claimed, stop if we encounter a sequence
* number greater than the highest claimed sequence number.
*/
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
lrbuf = zio_buf_alloc(SPA_OLD_MAXBLOCKSIZE);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_bp_tree_init(zilog);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) {
uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
int reclen;
Fix gcc uninitialized variable warnings Gcc -Wall warn: 'uninitialized variable' Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2010-08-26 16:58:04 +00:00
char *end = NULL;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (blk_seq > claim_blk_seq)
break;
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
error = parse_blk_func(zilog, &blk, arg, txg);
if (error != 0)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
break;
ASSERT3U(max_blk_seq, <, blk_seq);
max_blk_seq = blk_seq;
blk_count++;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq)
break;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
error = zil_read_log_block(zilog, decrypt, &blk, &next_blk,
lrbuf, &end);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
break;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
for (lrp = lrbuf; lrp < end; lrp += reclen) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
lr_t *lr = (lr_t *)lrp;
reclen = lr->lrc_reclen;
ASSERT3U(reclen, >=, sizeof (lr_t));
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (lr->lrc_seq > claim_lr_seq)
goto done;
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
error = parse_lr_func(zilog, lr, arg, txg);
if (error != 0)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
goto done;
ASSERT3U(max_lr_seq, <, lr->lrc_seq);
max_lr_seq = lr->lrc_seq;
lr_count++;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
done:
zilog->zl_parse_error = error;
zilog->zl_parse_blk_seq = max_blk_seq;
zilog->zl_parse_lr_seq = max_lr_seq;
zilog->zl_parse_blk_count = blk_count;
zilog->zl_parse_lr_count = lr_count;
ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) ||
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
(max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq) ||
(decrypt && error == EIO));
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_bp_tree_fini(zilog);
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
zio_buf_free(lrbuf, SPA_OLD_MAXBLOCKSIZE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (error);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
/* ARGSUSED */
static int
zil_clear_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
{
ASSERT(!BP_IS_HOLE(bp));
/*
* As we call this function from the context of a rewind to a
* checkpoint, each ZIL block whose txg is later than the txg
* that we rewind to is invalid. Thus, we return -1 so
* zil_parse() doesn't attempt to read it.
*/
if (bp->blk_birth >= first_txg)
return (-1);
if (zil_bp_tree_add(zilog, bp) != 0)
return (0);
zio_free(zilog->zl_spa, first_txg, bp);
return (0);
}
/* ARGSUSED */
static int
zil_noop_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
{
return (0);
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
{
/*
* Claim log block if not already committed and not already claimed.
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* If tx == NULL, just verify that the block is claimable.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Illumos 4370, 4371 4370 avoid transmitting holes during zfs send 4371 DMU code clean up Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net> Approved by: Garrett D'Amore <garrett@damore.org>a References: https://www.illumos.org/issues/4370 https://www.illumos.org/issues/4371 https://github.com/illumos/illumos-gate/commit/43466aa Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2529
2013-12-09 18:37:51 +00:00
if (BP_IS_HOLE(bp) || bp->blk_birth < first_txg ||
zil_bp_tree_add(zilog, bp) != 0)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (0);
return (zio_wait(zio_claim(NULL, zilog->zl_spa,
tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB)));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lr_write_t *lr = (lr_write_t *)lrc;
int error;
if (lrc->lrc_txtype != TX_WRITE)
return (0);
/*
* If the block is not readable, don't claim it. This can happen
* in normal operation when a log block is written to disk before
* some of the dmu_sync() blocks it points to. In this case, the
* transaction cannot have been committed to anyone (we would have
* waited for all writes to be stable first), so it is semantically
* correct to declare this the end of the log.
*/
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
if (lr->lr_blkptr.blk_birth >= first_txg) {
error = zil_read_log_data(zilog, lr, NULL);
if (error != 0)
return (error);
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/* ARGSUSED */
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg)
{
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg)
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lr_write_t *lr = (lr_write_t *)lrc;
blkptr_t *bp = &lr->lr_blkptr;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* If we previously claimed it, we need to free it.
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE &&
Illumos 4370, 4371 4370 avoid transmitting holes during zfs send 4371 DMU code clean up Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net> Approved by: Garrett D'Amore <garrett@damore.org>a References: https://www.illumos.org/issues/4370 https://www.illumos.org/issues/4371 https://github.com/illumos/illumos-gate/commit/43466aa Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2529
2013-12-09 18:37:51 +00:00
bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0 &&
!BP_IS_HOLE(bp))
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
return (0);
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
static int
zil_lwb_vdev_compare(const void *x1, const void *x2)
{
const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev;
const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev;
Reduce loaded range tree memory usage This patch implements a new tree structure for ZFS, and uses it to store range trees more efficiently. The new structure is approximately a B-tree, though there are some small differences from the usual characterizations. The tree has core nodes and leaf nodes; each contain data elements, which the elements in the core nodes acting as separators between its children. The difference between core and leaf nodes is that the core nodes have an array of children, while leaf nodes don't. Every node in the tree may be only partially full; in most cases, they are all at least 50% full (in terms of element count) except for the root node, which can be less full. Underfull nodes will steal from their neighbors or merge to remain full enough, while overfull nodes will split in two. The data elements are contained in tree-controlled buffers; they are copied into these on insertion, and overwritten on deletion. This means that the elements are not independently allocated, which reduces overhead, but also means they can't be shared between trees (and also that pointers to them are only valid until a side-effectful tree operation occurs). The overhead varies based on how dense the tree is, but is usually on the order of about 50% of the element size; the per-node overheads are very small, and so don't make a significant difference. The trees can accept arbitrary records; they accept a size and a comparator to allow them to be used for a variety of purposes. The new trees replace the AVL trees used in the range trees today. Currently, the range_seg_t structure contains three 8 byte integers of payload and two 24 byte avl_tree_node_ts to handle its storage in both an offset-sorted tree and a size-sorted tree (total size: 64 bytes). In the new model, the range seg structures are usually two 4 byte integers, but a separate one needs to exist for the size-sorted and offset-sorted tree. Between the raw size, the 50% overhead, and the double storage, the new btrees are expected to use 8*1.5*2 = 24 bytes per record, or 33.3% as much memory as the AVL trees (this is for the purposes of storing metaslab range trees; for other purposes, like scrubs, they use ~50% as much memory). We reduced the size of the payload in the range segments by teaching range trees about starting offsets and shifts; since metaslabs have a fixed starting offset, and they all operate in terms of disk sectors, we can store the ranges using 4-byte integers as long as the size of the metaslab divided by the sector size is less than 2^32. For 512-byte sectors, this is a 2^41 (or 2TB) metaslab, which with the default settings corresponds to a 256PB disk. 4k sector disks can handle metaslabs up to 2^46 bytes, or 2^63 byte disks. Since we do not anticipate disks of this size in the near future, there should be almost no cases where metaslabs need 64-byte integers to store their ranges. We do still have the capability to store 64-byte integer ranges to account for cases where we are storing per-vdev (or per-dnode) trees, which could reasonably go above the limits discussed. We also do not store fill information in the compact version of the node, since it is only used for sorted scrub. We also optimized the metaslab loading process in various other ways to offset some inefficiencies in the btree model. While individual operations (find, insert, remove_from) are faster for the btree than they are for the avl tree, remove usually requires a find operation, while in the AVL tree model the element itself suffices. Some clever changes actually caused an overall speedup in metaslab loading; we use approximately 40% less cpu to load metaslabs in our tests on Illumos. Another memory and performance optimization was achieved by changing what is stored in the size-sorted trees. When a disk is heavily fragmented, the df algorithm used by default in ZFS will almost always find a number of small regions in its initial cursor-based search; it will usually only fall back to the size-sorted tree to find larger regions. If we increase the size of the cursor-based search slightly, and don't store segments that are smaller than a tunable size floor in the size-sorted tree, we can further cut memory usage down to below 20% of what the AVL trees store. This also results in further reductions in CPU time spent loading metaslabs. The 16KiB size floor was chosen because it results in substantial memory usage reduction while not usually resulting in situations where we can't find an appropriate chunk with the cursor and are forced to use an oversized chunk from the size-sorted tree. In addition, even if we do have to use an oversized chunk from the size-sorted tree, the chunk would be too small to use for ZIL allocations, so it isn't as big of a loss as it might otherwise be. And often, more small allocations will follow the initial one, and the cursor search will now find the remainder of the chunk we didn't use all of and use it for subsequent allocations. Practical testing has shown little or no change in fragmentation as a result of this change. If the size-sorted tree becomes empty while the offset sorted one still has entries, it will load all the entries from the offset sorted tree and disregard the size floor until it is unloaded again. This operation occurs rarely with the default setting, only on incredibly thoroughly fragmented pools. There are some other small changes to zdb to teach it to handle btrees, but nothing major. Reviewed-by: George Wilson <gwilson@delphix.com> Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed by: Sebastien Roy seb@delphix.com Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Paul Dagnelie <pcd@delphix.com> Closes #9181
2019-10-09 17:36:03 +00:00
return (TREE_CMP(v1, v2));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static lwb_t *
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, boolean_t slog, uint64_t txg,
boolean_t fastwrite)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
{
lwb_t *lwb;
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lwb->lwb_zilog = zilog;
lwb->lwb_blk = *bp;
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
lwb->lwb_fastwrite = fastwrite;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lwb->lwb_slog = slog;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_state = LWB_STATE_CLOSED;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp));
lwb->lwb_max_txg = txg;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_write_zio = NULL;
lwb->lwb_root_zio = NULL;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lwb->lwb_tx = NULL;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_issued_timestamp = 0;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) {
lwb->lwb_nused = sizeof (zil_chain_t);
lwb->lwb_sz = BP_GET_LSIZE(bp);
} else {
lwb->lwb_nused = 0;
lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
mutex_enter(&zilog->zl_lock);
list_insert_tail(&zilog->zl_lwb_list, lwb);
mutex_exit(&zilog->zl_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT(!MUTEX_HELD(&lwb->lwb_vdev_lock));
ASSERT(avl_is_empty(&lwb->lwb_vdev_tree));
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
VERIFY(list_is_empty(&lwb->lwb_waiters));
VERIFY(list_is_empty(&lwb->lwb_itxs));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (lwb);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
static void
zil_free_lwb(zilog_t *zilog, lwb_t *lwb)
{
ASSERT(MUTEX_HELD(&zilog->zl_lock));
ASSERT(!MUTEX_HELD(&lwb->lwb_vdev_lock));
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
VERIFY(list_is_empty(&lwb->lwb_waiters));
VERIFY(list_is_empty(&lwb->lwb_itxs));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT(avl_is_empty(&lwb->lwb_vdev_tree));
ASSERT3P(lwb->lwb_write_zio, ==, NULL);
ASSERT3P(lwb->lwb_root_zio, ==, NULL);
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
ASSERT3U(lwb->lwb_max_txg, <=, spa_syncing_txg(zilog->zl_spa));
ASSERT(lwb->lwb_state == LWB_STATE_CLOSED ||
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
lwb->lwb_state == LWB_STATE_FLUSH_DONE);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* Clear the zilog's field to indicate this lwb is no longer
* valid, and prevent use-after-free errors.
*/
if (zilog->zl_last_lwb_opened == lwb)
zilog->zl_last_lwb_opened = NULL;
kmem_cache_free(zil_lwb_cache, lwb);
}
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
/*
* Called when we create in-memory log transactions so that we know
* to cleanup the itxs at the end of spa_sync().
*/
void
zilog_dirty(zilog_t *zilog, uint64_t txg)
{
dsl_pool_t *dp = zilog->zl_dmu_pool;
dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT(spa_writeable(zilog->zl_spa));
Illumos 5056 - ZFS deadlock on db_mtx and dn_holds 5056 ZFS deadlock on db_mtx and dn_holds Author: Justin Gibbs <justing@spectralogic.com> Reviewed by: Will Andrews <willa@spectralogic.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5056 https://github.com/illumos/illumos-gate/commit/bc9014e Porting Notes: sa_handle_get_from_db(): - the original patch includes an otherwise unmentioned fix for a possible usage of an uninitialised variable dmu_objset_open_impl(): - Under Illumos list_link_init() is the same as filling a list_node_t with NULLs, so they don't notice if they miss doing list_link_init() on a zero'd containing structure (e.g. allocated with kmem_zalloc as here). Under Linux, not so much: an uninitialised list_node_t goes "Boom!" some time later when it's used or destroyed. dmu_objset_evict_dbufs(): - reduce stack usage using kmem_alloc() Ported-by: Chris Dunlop <chris@onthe.net.au> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2015-04-02 03:44:32 +00:00
if (ds->ds_is_snapshot)
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
panic("dirtying snapshot!");
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) {
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
/* up the hold count until we can be written out */
dmu_buf_add_ref(ds->ds_dbuf, zilog);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zilog->zl_dirty_max_txg = MAX(txg, zilog->zl_dirty_max_txg);
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
}
}
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
/*
* Determine if the zil is dirty in the specified txg. Callers wanting to
* ensure that the dirty state does not change must hold the itxg_lock for
* the specified txg. Holding the lock will ensure that the zil cannot be
* dirtied (zil_itx_assign) or cleaned (zil_clean) while we check its current
* state.
*/
boolean_t
zilog_is_dirty_in_txg(zilog_t *zilog, uint64_t txg)
{
dsl_pool_t *dp = zilog->zl_dmu_pool;
if (txg_list_member(&dp->dp_dirty_zilogs, zilog, txg & TXG_MASK))
return (B_TRUE);
return (B_FALSE);
}
/*
* Determine if the zil is dirty. The zil is considered dirty if it has
* any pending itx records that have not been cleaned by zil_clean().
*/
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
boolean_t
zilog_is_dirty(zilog_t *zilog)
{
dsl_pool_t *dp = zilog->zl_dmu_pool;
Undo c89 workarounds to match with upstream With PR 5756 the zfs module now supports c99 and the remaining past c89 workarounds can be undone. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: George Melikov <mail@gmelikov.ru> Signed-off-by: Don Brady <don.brady@delphix.com> Closes #6816
2017-11-04 20:25:13 +00:00
for (int t = 0; t < TXG_SIZE; t++) {
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t))
return (B_TRUE);
}
return (B_FALSE);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Create an on-disk intent log.
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static lwb_t *
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_create(zilog_t *zilog)
{
const zil_header_t *zh = zilog->zl_header;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lwb_t *lwb = NULL;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
uint64_t txg = 0;
dmu_tx_t *tx = NULL;
blkptr_t blk;
int error = 0;
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
boolean_t fastwrite = FALSE;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
boolean_t slog = FALSE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Wait for any previous destroy to complete.
*/
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
ASSERT(zh->zh_claim_txg == 0);
ASSERT(zh->zh_replay_seq == 0);
blk = zh->zh_log;
/*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* Allocate an initial log block if:
* - there isn't one already
Fix spelling Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Haakan T Johansson <f96hajo@chalmers.se> Closes #5547 Closes #5543
2017-01-03 17:31:18 +00:00
* - the existing block is the wrong endianness
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Rebase master to b105
2009-01-15 21:59:39 +00:00
if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
tx = dmu_tx_create(zilog->zl_os);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
Rebase master to b105
2009-01-15 21:59:39 +00:00
if (!BP_IS_HOLE(&blk)) {
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
zio_free(zilog->zl_spa, txg, &blk);
Rebase master to b105
2009-01-15 21:59:39 +00:00
BP_ZERO(&blk);
}
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
error = zio_alloc_zil(zilog->zl_spa, zilog->zl_os, txg, &blk,
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
ZIL_MIN_BLKSZ, &slog);
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
fastwrite = TRUE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (error == 0)
zil_init_log_chain(zilog, &blk);
}
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* Allocate a log write block (lwb) for the first log block.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (error == 0)
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lwb = zil_alloc_lwb(zilog, &blk, slog, txg, fastwrite);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* If we just allocated the first log block, commit our transaction
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* and wait for zil_sync() to stuff the block pointer into zh_log.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
* (zh is part of the MOS, so we cannot modify it in open context.)
*/
if (tx != NULL) {
dmu_tx_commit(tx);
txg_wait_synced(zilog->zl_dmu_pool, txg);
}
Fix ASSERT in zil_create() during ztest This patch corrects an ASSERT in zil_create() that will only be true if the call to zio_alloc_zil() does not fail. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed-by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #8010
2018-10-11 20:38:27 +00:00
ASSERT(error != 0 || bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0);
IMPLY(error == 0, lwb != NULL);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (lwb);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* In one tx, free all log blocks and clear the log header. If keep_first
* is set, then we're replaying a log with no content. We want to keep the
* first block, however, so that the first synchronous transaction doesn't
* require a txg_wait_synced() in zil_create(). We don't need to
* txg_wait_synced() here either when keep_first is set, because both
* zil_create() and zil_destroy() will wait for any in-progress destroys
* to complete.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
void
zil_destroy(zilog_t *zilog, boolean_t keep_first)
{
const zil_header_t *zh = zilog->zl_header;
lwb_t *lwb;
dmu_tx_t *tx;
uint64_t txg;
/*
* Wait for any previous destroy to complete.
*/
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilog->zl_old_header = *zh; /* debugging aid */
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (BP_IS_HOLE(&zh->zh_log))
return;
tx = dmu_tx_create(zilog->zl_os);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
mutex_enter(&zilog->zl_lock);
ASSERT3U(zilog->zl_destroy_txg, <, txg);
zilog->zl_destroy_txg = txg;
zilog->zl_keep_first = keep_first;
if (!list_is_empty(&zilog->zl_lwb_list)) {
ASSERT(zh->zh_claim_txg == 0);
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
VERIFY(!keep_first);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
if (lwb->lwb_fastwrite)
metaslab_fastwrite_unmark(zilog->zl_spa,
&lwb->lwb_blk);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
list_remove(&zilog->zl_lwb_list, lwb);
if (lwb->lwb_buf != NULL)
zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zio_free(zilog->zl_spa, txg, &lwb->lwb_blk);
zil_free_lwb(zilog, lwb);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
} else if (!keep_first) {
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
zil_destroy_sync(zilog, tx);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
mutex_exit(&zilog->zl_lock);
dmu_tx_commit(tx);
}
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
void
zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx)
{
ASSERT(list_is_empty(&zilog->zl_lwb_list));
(void) zil_parse(zilog, zil_free_log_block,
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
zil_free_log_record, tx, zilog->zl_header->zh_claim_txg, B_FALSE);
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
int
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
zil_claim(dsl_pool_t *dp, dsl_dataset_t *ds, void *txarg)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
dmu_tx_t *tx = txarg;
zilog_t *zilog;
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
uint64_t first_txg;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_header_t *zh;
objset_t *os;
int error;
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
error = dmu_objset_own_obj(dp, ds->ds_object,
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
DMU_OST_ANY, B_FALSE, B_FALSE, FTAG, &os);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0) {
Illumos 5140 - message about "%recv could not be opened" is printed when booting after crash Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Max Grossman <max.grossman@delphix.com> Reviewed by: Richard Elling <richard.elling@gmail.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/projects/illumos-gate//issues/5140 https://github.com/illumos/illumos-gate/commit/2243853 Ported by: Turbo Fredriksson <turbo@bayour.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2676
2014-09-07 15:37:25 +00:00
/*
* EBUSY indicates that the objset is inconsistent, in which
* case it can not have a ZIL.
*/
if (error != EBUSY) {
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
cmn_err(CE_WARN, "can't open objset for %llu, error %u",
(unsigned long long)ds->ds_object, error);
Illumos 5140 - message about "%recv could not be opened" is printed when booting after crash Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Max Grossman <max.grossman@delphix.com> Reviewed by: Richard Elling <richard.elling@gmail.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/projects/illumos-gate//issues/5140 https://github.com/illumos/illumos-gate/commit/2243853 Ported by: Turbo Fredriksson <turbo@bayour.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2676
2014-09-07 15:37:25 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (0);
}
zilog = dmu_objset_zil(os);
zh = zil_header_in_syncing_context(zilog);
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
ASSERT3U(tx->tx_txg, ==, spa_first_txg(zilog->zl_spa));
first_txg = spa_min_claim_txg(zilog->zl_spa);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
/*
* If the spa_log_state is not set to be cleared, check whether
* the current uberblock is a checkpoint one and if the current
* header has been claimed before moving on.
*
* If the current uberblock is a checkpointed uberblock then
* one of the following scenarios took place:
*
* 1] We are currently rewinding to the checkpoint of the pool.
* 2] We crashed in the middle of a checkpoint rewind but we
* did manage to write the checkpointed uberblock to the
* vdev labels, so when we tried to import the pool again
* the checkpointed uberblock was selected from the import
* procedure.
*
* In both cases we want to zero out all the ZIL blocks, except
* the ones that have been claimed at the time of the checkpoint
* (their zh_claim_txg != 0). The reason is that these blocks
* may be corrupted since we may have reused their locations on
* disk after we took the checkpoint.
*
* We could try to set spa_log_state to SPA_LOG_CLEAR earlier
* when we first figure out whether the current uberblock is
* checkpointed or not. Unfortunately, that would discard all
* the logs, including the ones that are claimed, and we would
* leak space.
*/
if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR ||
(zilog->zl_spa->spa_uberblock.ub_checkpoint_txg != 0 &&
zh->zh_claim_txg == 0)) {
if (!BP_IS_HOLE(&zh->zh_log)) {
(void) zil_parse(zilog, zil_clear_log_block,
zil_noop_log_record, tx, first_txg, B_FALSE);
}
Rebase master to b117
2009-07-02 22:44:48 +00:00
BP_ZERO(&zh->zh_log);
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
if (os->os_encrypted)
Change os->os_next_write_raw to work per txg Currently, os_next_write_raw is a single boolean used for determining whether or not the next call to dmu_objset_sync() should write out the objset_phys_t as a raw buffer. Since the boolean is not associated with a txg, the work simply happens during the next txg, which is not necessarily the correct one. In the current implementation this issue was misdiagnosed, resulting in a small hack in dmu_objset_sync() which seemed to resolve the problem. This patch changes os_next_write_raw to be an array of booleans, one for each txg in TXG_OFF and removes the hack. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #6864
2018-02-01 20:37:24 +00:00
os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE;
Rebase master to b117
2009-07-02 22:44:48 +00:00
dsl_dataset_dirty(dmu_objset_ds(os), tx);
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
dmu_objset_disown(os, B_FALSE, FTAG);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (0);
Rebase master to b117
2009-07-02 22:44:48 +00:00
}
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
/*
* If we are not rewinding and opening the pool normally, then
* the min_claim_txg should be equal to the first txg of the pool.
*/
ASSERT3U(first_txg, ==, spa_first_txg(zilog->zl_spa));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Claim all log blocks if we haven't already done so, and remember
* the highest claimed sequence number. This ensures that if we can
* read only part of the log now (e.g. due to a missing device),
* but we can read the entire log later, we will not try to replay
* or destroy beyond the last block we successfully claimed.
*/
ASSERT3U(zh->zh_claim_txg, <=, first_txg);
if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
(void) zil_parse(zilog, zil_claim_log_block,
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
zil_claim_log_record, tx, first_txg, B_FALSE);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zh->zh_claim_txg = first_txg;
zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq;
zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq;
if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1)
zh->zh_flags |= ZIL_REPLAY_NEEDED;
zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID;
Fix recovery import (-F) with encrypted pool When performing zil_claim() at pool import time, it is important that encrypted datasets set os_next_write_raw before writing to the zil_header_t. This prevents the code from attempting to re-authenticate the objset_phys_t when it writes it out, which is unnecessary because the zil_header_t is not protected by either objset MAC and impossible since the keys aren't loaded yet. Unfortunately, one of the code paths did not set this flag, which causes failed ASSERTs during 'zpool import -F'. This patch corrects this issue. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #6864 Closes #6916
2017-12-04 19:10:31 +00:00
if (os->os_encrypted)
Change os->os_next_write_raw to work per txg Currently, os_next_write_raw is a single boolean used for determining whether or not the next call to dmu_objset_sync() should write out the objset_phys_t as a raw buffer. Since the boolean is not associated with a txg, the work simply happens during the next txg, which is not necessarily the correct one. In the current implementation this issue was misdiagnosed, resulting in a small hack in dmu_objset_sync() which seemed to resolve the problem. This patch changes os_next_write_raw to be an array of booleans, one for each txg in TXG_OFF and removes the hack. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #6864
2018-02-01 20:37:24 +00:00
os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
dsl_dataset_dirty(dmu_objset_ds(os), tx);
}
ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
dmu_objset_disown(os, B_FALSE, FTAG);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (0);
}
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
/*
* Check the log by walking the log chain.
* Checksum errors are ok as they indicate the end of the chain.
* Any other error (no device or read failure) returns an error.
*/
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
/* ARGSUSED */
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
int
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds, void *tx)
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
{
zilog_t *zilog;
objset_t *os;
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
blkptr_t *bp;
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
int error;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
ASSERT(tx == NULL);
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
error = dmu_objset_from_ds(ds, &os);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0) {
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
cmn_err(CE_WARN, "can't open objset %llu, error %d",
(unsigned long long)ds->ds_object, error);
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
return (0);
}
zilog = dmu_objset_zil(os);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
bp = (blkptr_t *)&zilog->zl_header->zh_log;
if (!BP_IS_HOLE(bp)) {
vdev_t *vd;
boolean_t valid = B_TRUE;
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
/*
* Check the first block and determine if it's on a log device
* which may have been removed or faulted prior to loading this
* pool. If so, there's no point in checking the rest of the
* log as its content should have already been synced to the
* pool.
*/
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER);
vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0]));
if (vd->vdev_islog && vdev_is_dead(vd))
valid = vdev_log_state_valid(vd);
spa_config_exit(os->os_spa, SCL_STATE, FTAG);
Illumos 5269 - zpool import slow 5269 zpool import slow Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5269 https://github.com/illumos/illumos-gate/commit/12380e1e Ported-by: DHE <git@dehacked.net> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3396
2015-05-06 16:07:55 +00:00
if (!valid)
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
return (0);
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
/*
* Check whether the current uberblock is checkpointed (e.g.
* we are rewinding) and whether the current header has been
* claimed or not. If it hasn't then skip verifying it. We
* do this because its ZIL blocks may be part of the pool's
* state before the rewind, which is no longer valid.
*/
zil_header_t *zh = zil_header_in_syncing_context(zilog);
if (zilog->zl_spa->spa_uberblock.ub_checkpoint_txg != 0 &&
zh->zh_claim_txg == 0)
return (0);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
}
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
/*
* Because tx == NULL, zil_claim_log_block() will not actually claim
* any blocks, but just determine whether it is possible to do so.
* In addition to checking the log chain, zil_claim_log_block()
* will invoke zio_claim() with a done func of spa_claim_notify(),
* which will update spa_max_claim_txg. See spa_load() for details.
*/
error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx,
OpenZFS 9166 - zfs storage pool checkpoint Details about the motivation of this feature and its usage can be found in this blogpost: https://sdimitro.github.io/post/zpool-checkpoint/ A lightning talk of this feature can be found here: https://www.youtube.com/watch?v=fPQA8K40jAM Implementation details can be found in big block comment of spa_checkpoint.c Side-changes that are relevant to this commit but not explained elsewhere: * renames members of "struct metaslab trees to be shorter without losing meaning * space_map_{alloc,truncate}() accept a block size as a parameter. The reason is that in the current state all space maps that we allocate through the DMU use a global tunable (space_map_blksz) which defauls to 4KB. This is ok for metaslab space maps in terms of bandwirdth since they are scattered all over the disk. But for other space maps this default is probably not what we want. Examples are device removal's vdev_obsolete_sm or vdev_chedkpoint_sm from this review. Both of these have a 1:1 relationship with each vdev and could benefit from a bigger block size. Porting notes: * The part of dsl_scan_sync() which handles async destroys has been moved into the new dsl_process_async_destroys() function. * Remove "VERIFY(!(flags & FWRITE))" in "kernel.c" so zhack can write to block device backed pools. * ZTS: * Fix get_txg() in zpool_sync_001_pos due to "checkpoint_txg". * Don't use large dd block sizes on /dev/urandom under Linux in checkpoint_capacity. * Adopt Delphix-OS's setting of 4 (spa_asize_inflation = SPA_DVAS_PER_BP + 1) for the checkpoint_capacity test to speed its attempts to fill the pool * Create the base and nested pools with sync=disabled to speed up the "setup" phase. * Clear labels in test pool between checkpoint tests to avoid duplicate pool issues. * The import_rewind_device_replaced test has been marked as "known to fail" for the reasons listed in its DISCLAIMER. * New module parameters: zfs_spa_discard_memory_limit, zfs_remove_max_bytes_pause (not documented - debugging only) vdev_max_ms_count (formerly metaslabs_per_vdev) vdev_min_ms_count Authored by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: John Kennedy <john.kennedy@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://illumos.org/issues/9166 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7159fdb8 Closes #7570
2016-12-16 22:11:29 +00:00
zilog->zl_header->zh_claim_txg ? -1ULL :
spa_min_claim_txg(os->os_spa), B_FALSE);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return ((error == ECKSUM || error == ENOENT) ? 0 : error);
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* When an itx is "skipped", this function is used to properly mark the
* waiter as "done, and signal any thread(s) waiting on it. An itx can
* be skipped (and not committed to an lwb) for a variety of reasons,
* one of them being that the itx was committed via spa_sync(), prior to
* it being committed to an lwb; this can happen if a thread calling
* zil_commit() is racing with spa_sync().
*/
static void
zil_commit_waiter_skip(zil_commit_waiter_t *zcw)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_enter(&zcw->zcw_lock);
ASSERT3B(zcw->zcw_done, ==, B_FALSE);
zcw->zcw_done = B_TRUE;
cv_broadcast(&zcw->zcw_cv);
mutex_exit(&zcw->zcw_lock);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* This function is used when the given waiter is to be linked into an
* lwb's "lwb_waiter" list; i.e. when the itx is committed to the lwb.
* At this point, the waiter will no longer be referenced by the itx,
* and instead, will be referenced by the lwb.
*/
static void
zil_commit_waiter_link_lwb(zil_commit_waiter_t *zcw, lwb_t *lwb)
{
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
/*
* The lwb_waiters field of the lwb is protected by the zilog's
* zl_lock, thus it must be held when calling this function.
*/
ASSERT(MUTEX_HELD(&lwb->lwb_zilog->zl_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_enter(&zcw->zcw_lock);
ASSERT(!list_link_active(&zcw->zcw_node));
ASSERT3P(zcw->zcw_lwb, ==, NULL);
ASSERT3P(lwb, !=, NULL);
ASSERT(lwb->lwb_state == LWB_STATE_OPENED ||
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
lwb->lwb_state == LWB_STATE_ISSUED ||
lwb->lwb_state == LWB_STATE_WRITE_DONE);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
list_insert_tail(&lwb->lwb_waiters, zcw);
zcw->zcw_lwb = lwb;
mutex_exit(&zcw->zcw_lock);
}
/*
* This function is used when zio_alloc_zil() fails to allocate a ZIL
* block, and the given waiter must be linked to the "nolwb waiters"
* list inside of zil_process_commit_list().
*/
static void
zil_commit_waiter_link_nolwb(zil_commit_waiter_t *zcw, list_t *nolwb)
{
mutex_enter(&zcw->zcw_lock);
ASSERT(!list_link_active(&zcw->zcw_node));
ASSERT3P(zcw->zcw_lwb, ==, NULL);
list_insert_tail(nolwb, zcw);
mutex_exit(&zcw->zcw_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
void
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_lwb_add_block(lwb_t *lwb, const blkptr_t *bp)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
avl_tree_t *t = &lwb->lwb_vdev_tree;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
avl_index_t where;
zil_vdev_node_t *zv, zvsearch;
int ndvas = BP_GET_NDVAS(bp);
int i;
OpenZFS 9963 - Separate tunable for disabling ZIL vdev flush Porting Notes: * Add options to zfs-module-parameters(5) man page. * zfs_nocacheflush move to vdev.c instead of vdev_disk.c, since the latter doesn't get built for user space. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Tom Caputi <tcaputi@datto.com> Reviewed by: George Melikov <mail@gmelikov.ru> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9963 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/f8fdf68125 Closes #8186
2018-10-23 16:44:37 +00:00
if (zil_nocacheflush)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_enter(&lwb->lwb_vdev_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
for (i = 0; i < ndvas; i++) {
zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
if (avl_find(t, &zvsearch, &where) == NULL) {
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
zv = kmem_alloc(sizeof (*zv), KM_SLEEP);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zv->zv_vdev = zvsearch.zv_vdev;
avl_insert(t, zv, where);
}
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_exit(&lwb->lwb_vdev_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
static void
zil_lwb_flush_defer(lwb_t *lwb, lwb_t *nlwb)
{
avl_tree_t *src = &lwb->lwb_vdev_tree;
avl_tree_t *dst = &nlwb->lwb_vdev_tree;
void *cookie = NULL;
zil_vdev_node_t *zv;
ASSERT3S(lwb->lwb_state, ==, LWB_STATE_WRITE_DONE);
ASSERT3S(nlwb->lwb_state, !=, LWB_STATE_WRITE_DONE);
ASSERT3S(nlwb->lwb_state, !=, LWB_STATE_FLUSH_DONE);
/*
* While 'lwb' is at a point in its lifetime where lwb_vdev_tree does
* not need the protection of lwb_vdev_lock (it will only be modified
* while holding zilog->zl_lock) as its writes and those of its
* children have all completed. The younger 'nlwb' may be waiting on
* future writes to additional vdevs.
*/
mutex_enter(&nlwb->lwb_vdev_lock);
/*
* Tear down the 'lwb' vdev tree, ensuring that entries which do not
* exist in 'nlwb' are moved to it, freeing any would-be duplicates.
*/
while ((zv = avl_destroy_nodes(src, &cookie)) != NULL) {
avl_index_t where;
if (avl_find(dst, zv, &where) == NULL) {
avl_insert(dst, zv, where);
} else {
kmem_free(zv, sizeof (*zv));
}
}
mutex_exit(&nlwb->lwb_vdev_lock);
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
void
zil_lwb_add_txg(lwb_t *lwb, uint64_t txg)
{
lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg);
}
/*
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
* This function is a called after all vdevs associated with a given lwb
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* write have completed their DKIOCFLUSHWRITECACHE command; or as soon
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
* as the lwb write completes, if "zil_nocacheflush" is set. Further,
* all "previous" lwb's will have completed before this function is
* called; i.e. this function is called for all previous lwbs before
* it's called for "this" lwb (enforced via zio the dependencies
* configured in zil_lwb_set_zio_dependency()).
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
* The intention is for this function to be called as soon as the
* contents of an lwb are considered "stable" on disk, and will survive
* any sudden loss of power. At this point, any threads waiting for the
* lwb to reach this state are signalled, and the "waiter" structures
* are marked "done".
*/
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
static void
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_lwb_flush_vdevs_done(zio_t *zio)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb_t *lwb = zio->io_private;
zilog_t *zilog = lwb->lwb_zilog;
dmu_tx_t *tx = lwb->lwb_tx;
zil_commit_waiter_t *zcw;
itx_t *itx;
spa_config_exit(zilog->zl_spa, SCL_STATE, lwb);
zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_enter(&zilog->zl_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* Ensure the lwb buffer pointer is cleared before releasing the
* txg. If we have had an allocation failure and the txg is
* waiting to sync then we want zil_sync() to remove the lwb so
* that it's not picked up as the next new one in
* zil_process_commit_list(). zil_sync() will only remove the
* lwb if lwb_buf is null.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_buf = NULL;
lwb->lwb_tx = NULL;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3U(lwb->lwb_issued_timestamp, >, 0);
zilog->zl_last_lwb_latency = gethrtime() - lwb->lwb_issued_timestamp;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_root_zio = NULL;
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
ASSERT3S(lwb->lwb_state, ==, LWB_STATE_WRITE_DONE);
lwb->lwb_state = LWB_STATE_FLUSH_DONE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (zilog->zl_last_lwb_opened == lwb) {
/*
* Remember the highest committed log sequence number
* for ztest. We only update this value when all the log
* writes succeeded, because ztest wants to ASSERT that
* it got the whole log chain.
*/
zilog->zl_commit_lr_seq = zilog->zl_lr_seq;
}
while ((itx = list_head(&lwb->lwb_itxs)) != NULL) {
list_remove(&lwb->lwb_itxs, itx);
zil_itx_destroy(itx);
}
while ((zcw = list_head(&lwb->lwb_waiters)) != NULL) {
mutex_enter(&zcw->zcw_lock);
ASSERT(list_link_active(&zcw->zcw_node));
list_remove(&lwb->lwb_waiters, zcw);
ASSERT3P(zcw->zcw_lwb, ==, lwb);
zcw->zcw_lwb = NULL;
zcw->zcw_zio_error = zio->io_error;
ASSERT3B(zcw->zcw_done, ==, B_FALSE);
zcw->zcw_done = B_TRUE;
cv_broadcast(&zcw->zcw_cv);
mutex_exit(&zcw->zcw_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_exit(&zilog->zl_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* Now that we've written this log block, we have a stable pointer
* to the next block in the chain, so it's OK to let the txg in
* which we allocated the next block sync.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
dmu_tx_commit(tx);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/*
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
* This is called when an lwb's write zio completes. The callback's
* purpose is to issue the DKIOCFLUSHWRITECACHE commands for the vdevs
* in the lwb's lwb_vdev_tree. The tree will contain the vdevs involved
* in writing out this specific lwb's data, and in the case that cache
* flushes have been deferred, vdevs involved in writing the data for
* previous lwbs. The writes corresponding to all the vdevs in the
* lwb_vdev_tree will have completed by the time this is called, due to
* the zio dependencies configured in zil_lwb_set_zio_dependency(),
* which takes deferred flushes into account. The lwb will be "done"
* once zil_lwb_flush_vdevs_done() is called, which occurs in the zio
* completion callback for the lwb's root zio.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
static void
zil_lwb_write_done(zio_t *zio)
{
lwb_t *lwb = zio->io_private;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
spa_t *spa = zio->io_spa;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog_t *zilog = lwb->lwb_zilog;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
avl_tree_t *t = &lwb->lwb_vdev_tree;
void *cookie = NULL;
zil_vdev_node_t *zv;
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
lwb_t *nlwb;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), !=, 0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG);
ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER);
ASSERT(!BP_IS_GANG(zio->io_bp));
ASSERT(!BP_IS_HOLE(zio->io_bp));
Illumos 4757, 4913 4757 ZFS embedded-data block pointers ("zero block compression") 4913 zfs release should not be subject to space checks Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Max Grossman <max.grossman@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Dan McDonald <danmcd@omniti.com> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/4757 https://www.illumos.org/issues/4913 https://github.com/illumos/illumos-gate/commit/5d7b4d4 Porting notes: For compatibility with the fastpath code the zio_done() function needed to be updated. Because embedded-data block pointers do not require DVAs to be allocated the associated vdevs will not be marked and therefore should not be unmarked. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2544
2014-06-05 21:19:08 +00:00
ASSERT(BP_GET_FILL(zio->io_bp) == 0);
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
DLPX-44812 integrate EP-220 large memory scalability
2016-07-22 15:52:49 +00:00
abd_put(zio->io_abd);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_enter(&zilog->zl_lock);
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
ASSERT3S(lwb->lwb_state, ==, LWB_STATE_ISSUED);
lwb->lwb_state = LWB_STATE_WRITE_DONE;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_write_zio = NULL;
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
lwb->lwb_fastwrite = FALSE;
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
nlwb = list_next(&zilog->zl_lwb_list, lwb);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
mutex_exit(&zilog->zl_lock);
Rebase master to b117
2009-07-02 22:44:48 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (avl_numnodes(t) == 0)
return;
Rebase master to b117
2009-07-02 22:44:48 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* If there was an IO error, we're not going to call zio_flush()
* on these vdevs, so we simply empty the tree and free the
* nodes. We avoid calling zio_flush() since there isn't any
* good reason for doing so, after the lwb block failed to be
* written out.
Rebase master to b117
2009-07-02 22:44:48 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (zio->io_error != 0) {
while ((zv = avl_destroy_nodes(t, &cookie)) != NULL)
kmem_free(zv, sizeof (*zv));
return;
}
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
/*
* If this lwb does not have any threads waiting for it to
* complete, we want to defer issuing the DKIOCFLUSHWRITECACHE
* command to the vdevs written to by "this" lwb, and instead
* rely on the "next" lwb to handle the DKIOCFLUSHWRITECACHE
* command for those vdevs. Thus, we merge the vdev tree of
* "this" lwb with the vdev tree of the "next" lwb in the list,
* and assume the "next" lwb will handle flushing the vdevs (or
* deferring the flush(s) again).
*
* This is a useful performance optimization, especially for
* workloads with lots of async write activity and few sync
* write and/or fsync activity, as it has the potential to
* coalesce multiple flush commands to a vdev into one.
*/
if (list_head(&lwb->lwb_waiters) == NULL && nlwb != NULL) {
zil_lwb_flush_defer(lwb, nlwb);
ASSERT(avl_is_empty(&lwb->lwb_vdev_tree));
return;
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) {
vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev);
if (vd != NULL)
zio_flush(lwb->lwb_root_zio, vd);
kmem_free(zv, sizeof (*zv));
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
static void
zil_lwb_set_zio_dependency(zilog_t *zilog, lwb_t *lwb)
{
lwb_t *last_lwb_opened = zilog->zl_last_lwb_opened;
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
ASSERT(MUTEX_HELD(&zilog->zl_lock));
/*
* The zilog's "zl_last_lwb_opened" field is used to build the
* lwb/zio dependency chain, which is used to preserve the
* ordering of lwb completions that is required by the semantics
* of the ZIL. Each new lwb zio becomes a parent of the
* "previous" lwb zio, such that the new lwb's zio cannot
* complete until the "previous" lwb's zio completes.
*
* This is required by the semantics of zil_commit(); the commit
* waiters attached to the lwbs will be woken in the lwb zio's
* completion callback, so this zio dependency graph ensures the
* waiters are woken in the correct order (the same order the
* lwbs were created).
*/
if (last_lwb_opened != NULL &&
last_lwb_opened->lwb_state != LWB_STATE_FLUSH_DONE) {
ASSERT(last_lwb_opened->lwb_state == LWB_STATE_OPENED ||
last_lwb_opened->lwb_state == LWB_STATE_ISSUED ||
last_lwb_opened->lwb_state == LWB_STATE_WRITE_DONE);
ASSERT3P(last_lwb_opened->lwb_root_zio, !=, NULL);
zio_add_child(lwb->lwb_root_zio,
last_lwb_opened->lwb_root_zio);
/*
* If the previous lwb's write hasn't already completed,
* we also want to order the completion of the lwb write
* zios (above, we only order the completion of the lwb
* root zios). This is required because of how we can
* defer the DKIOCFLUSHWRITECACHE commands for each lwb.
*
Fix the spelling of deferred Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Richard Laager <rlaager@wiktel.com> Closes #8626
2019-04-15 02:07:36 +00:00
* When the DKIOCFLUSHWRITECACHE commands are deferred,
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
* the previous lwb will rely on this lwb to flush the
* vdevs written to by that previous lwb. Thus, we need
* to ensure this lwb doesn't issue the flush until
* after the previous lwb's write completes. We ensure
* this ordering by setting the zio parent/child
* relationship here.
*
* Without this relationship on the lwb's write zio,
* it's possible for this lwb's write to complete prior
* to the previous lwb's write completing; and thus, the
* vdevs for the previous lwb would be flushed prior to
* that lwb's data being written to those vdevs (the
* vdevs are flushed in the lwb write zio's completion
* handler, zil_lwb_write_done()).
*/
if (last_lwb_opened->lwb_state != LWB_STATE_WRITE_DONE) {
ASSERT(last_lwb_opened->lwb_state == LWB_STATE_OPENED ||
last_lwb_opened->lwb_state == LWB_STATE_ISSUED);
ASSERT3P(last_lwb_opened->lwb_write_zio, !=, NULL);
zio_add_child(lwb->lwb_write_zio,
last_lwb_opened->lwb_write_zio);
}
}
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* This function's purpose is to "open" an lwb such that it is ready to
* accept new itxs being committed to it. To do this, the lwb's zio
* structures are created, and linked to the lwb. This function is
* idempotent; if the passed in lwb has already been opened, this
* function is essentially a no-op.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
static void
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_lwb_write_open(zilog_t *zilog, lwb_t *lwb)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Illumos 4914 - zfs on-disk bookmark structure should be named *_phys_t 4914 zfs on-disk bookmark structure should be named *_phys_t Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Reviewed by: Richard Lowe <richlowe@richlowe.net> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Approved by: Robert Mustacchi <rm@joyent.com> References: https://www.illumos.org/issues/4914 https://github.com/illumos/illumos-gate/commit/7802d7b Porting notes: There were a number of zfsonlinux-specific uses of zbookmark_t which needed to be updated. This should reduce the likelihood of further problems like issue #2094 from occurring. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2558
2014-06-25 18:37:59 +00:00
zbookmark_phys_t zb;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
zio_priority_t prio;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb, !=, NULL);
EQUIV(lwb->lwb_root_zio == NULL, lwb->lwb_state == LWB_STATE_CLOSED);
EQUIV(lwb->lwb_root_zio != NULL, lwb->lwb_state == LWB_STATE_OPENED);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET],
ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
/* Lock so zil_sync() doesn't fastwrite_unmark after zio is created */
mutex_enter(&zilog->zl_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lwb->lwb_root_zio == NULL) {
DLPX-44812 integrate EP-220 large memory scalability
2016-07-22 15:52:49 +00:00
abd_t *lwb_abd = abd_get_from_buf(lwb->lwb_buf,
BP_GET_LSIZE(&lwb->lwb_blk));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
if (!lwb->lwb_fastwrite) {
metaslab_fastwrite_mark(zilog->zl_spa, &lwb->lwb_blk);
lwb->lwb_fastwrite = 1;
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
if (!lwb->lwb_slog || zilog->zl_cur_used <= zil_slog_bulk)
prio = ZIO_PRIORITY_SYNC_WRITE;
else
prio = ZIO_PRIORITY_ASYNC_WRITE;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb->lwb_root_zio = zio_root(zilog->zl_spa,
zil_lwb_flush_vdevs_done, lwb, ZIO_FLAG_CANFAIL);
ASSERT3P(lwb->lwb_root_zio, !=, NULL);
lwb->lwb_write_zio = zio_rewrite(lwb->lwb_root_zio,
zilog->zl_spa, 0, &lwb->lwb_blk, lwb_abd,
BP_GET_LSIZE(&lwb->lwb_blk), zil_lwb_write_done, lwb,
prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE |
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
ZIO_FLAG_FASTWRITE, &zb);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb->lwb_write_zio, !=, NULL);
lwb->lwb_state = LWB_STATE_OPENED;
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
zil_lwb_set_zio_dependency(zilog, lwb);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zilog->zl_last_lwb_opened = lwb;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
mutex_exit(&zilog->zl_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb->lwb_root_zio, !=, NULL);
ASSERT3P(lwb->lwb_write_zio, !=, NULL);
ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
/*
* Define a limited set of intent log block sizes.
Illumos #3742 3742 zfs comments need cleaner, more consistent style Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Eric Schrock <eric.schrock@delphix.com> Approved by: Christopher Siden <christopher.siden@delphix.com> References: https://www.illumos.org/issues/3742 illumos/illumos-gate@f7170741490edba9d1d9c697c177c887172bc741 Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. The change to zfs_vfsops.c was dropped because it involves zfs_mount_label_policy, which does not exist in the Linux port.
2013-06-11 17:12:34 +00:00
*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* These must be a multiple of 4KB. Note only the amount used (again
* aligned to 4KB) actually gets written. However, we can't always just
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 20:15:08 +00:00
* allocate SPA_OLD_MAXBLOCKSIZE as the slog space could be exhausted.
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
*/
Improve logging of 128KB writes Before my ZIL space optimization few years ago 128KB writes were logged as two 64KB+ records in two 128KB log blocks. After that change it became ~127KB+/1KB+ in two 128KB log blocks to free space in the second block for another record. Unfortunately in case of 128KB only writes, when space in the second block remained unused, that change increased write latency by unbalancing checksum computation and write times between parallel threads. It also didn't help with SLOG space efficiency in that case. This change introduces new 68KB log block size, used for both writes below 67KB and 128KB-sharp writes. Writes of 68-127KB are still using one 128KB block to not increase processing overhead. Writes above 131KB are still using full 128KB blocks, since possible saving there is small. Mixed loads will likely also fall back to previous 128KB, since code uses maximum of the last 16 requested block sizes. Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Closes #9409
2019-11-11 17:27:59 +00:00
struct {
uint64_t limit;
uint64_t blksz;
} zil_block_buckets[] = {
{ 4096, 4096 }, /* non TX_WRITE */
{ 8192 + 4096, 8192 + 4096 }, /* database */
{ 32768 + 4096, 32768 + 4096 }, /* NFS writes */
{ 65536 + 4096, 65536 + 4096 }, /* 64KB writes */
{ 131072, 131072 }, /* < 128KB writes */
{ 131072 +4096, 65536 + 4096 }, /* 128KB writes */
{ UINT64_MAX, SPA_OLD_MAXBLOCKSIZE}, /* > 128KB writes */
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
};
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
/*
* Maximum block size used by the ZIL. This is picked up when the ZIL is
* initialized. Otherwise this should not be used directly; see
* zl_max_block_size instead.
*/
int zil_maxblocksize = SPA_OLD_MAXBLOCKSIZE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Start a log block write and advance to the next log block.
* Calls are serialized.
*/
static lwb_t *
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_lwb_write_issue(zilog_t *zilog, lwb_t *lwb)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lwb_t *nlwb = NULL;
zil_chain_t *zilc;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
spa_t *spa = zilog->zl_spa;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
blkptr_t *bp;
dmu_tx_t *tx;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
uint64_t txg;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
uint64_t zil_blksz, wsz;
int i, error;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
boolean_t slog;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb->lwb_root_zio, !=, NULL);
ASSERT3P(lwb->lwb_write_zio, !=, NULL);
ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
zilc = (zil_chain_t *)lwb->lwb_buf;
bp = &zilc->zc_next_blk;
} else {
zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz);
bp = &zilc->zc_next_blk;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
ASSERT(lwb->lwb_nused <= lwb->lwb_sz);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Allocate the next block and save its address in this block
* before writing it in order to establish the log chain.
* Note that if the allocation of nlwb synced before we wrote
* the block that points at it (lwb), we'd leak it if we crashed.
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done().
* We dirty the dataset to ensure that zil_sync() will be called
* to clean up in the event of allocation failure or I/O failure.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
tx = dmu_tx_create(zilog->zl_os);
OpenZFS 8373 - TXG_WAIT in ZIL commit path Authored by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8373 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7f04961 Closes #6403
2017-07-17 12:31:30 +00:00
/*
OpenZFS 8997 - ztest assertion failure in zil_lwb_write_issue PROBLEM ======= When `dmu_tx_assign` is called from `zil_lwb_write_issue`, it's possible for either `ERESTART` or `EIO` to be returned. If `ERESTART` is returned, this will cause an assertion to fail directly in `zil_lwb_write_issue`, where the code assumes the return value is `EIO` if `dmu_tx_assign` returns a non-zero value. This can occur if the SPA is suspended when `dmu_tx_assign` is called, and most often occurs when running `zloop`. If `EIO` is returned, this can cause assertions to fail elsewhere in the ZIL code. For example, `zil_commit_waiter_timeout` contains the following logic: lwb_t *nlwb = zil_lwb_write_issue(zilog, lwb); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_OPENED); In this case, if `dmu_tx_assign` returned `EIO` from within `zil_lwb_write_issue`, the `lwb` variable passed in will not be issued to disk. Thus, it's `lwb_state` field will remain `LWB_STATE_OPENED` and this assertion will fail. `zil_commit_waiter_timeout` assumes that after it calls `zil_lwb_write_issue`, the `lwb` will be issued to disk, and doesn't handle the case where this is not true; i.e. it doesn't handle the case where `dmu_tx_assign` returns `EIO`. SOLUTION ======== This change modifies the `dmu_tx_assign` function such that `txg_how` is a bitmask, rather than of the `txg_how_t` enum type. Now, the previous `TXG_WAITED` semantics can be used via `TXG_NOTHROTTLE`, along with specifying either `TXG_NOWAIT` or `TXG_WAIT` semantics. Previously, when `TXG_WAITED` was specified, `TXG_NOWAIT` semantics was automatically invoked. This was not ideal when using `TXG_WAITED` within `zil_lwb_write_issued`, leading the problem described above. Rather, we want to achieve the semantics of `TXG_WAIT`, while also preventing the `tx` from being penalized via the dirty delay throttling. With this change, `zil_lwb_write_issued` can acheive the semtantics that it requires by passing in the value `TXG_WAIT | TXG_NOTHROTTLE` to `dmu_tx_assign`. Further, consumers of `dmu_tx_assign` wishing to achieve the old `TXG_WAITED` semantics can pass in the value `TXG_NOWAIT | TXG_NOTHROTTLE`. Authored by: Prakash Surya <prakash.surya@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Porting Notes: - Additionally updated `zfs_tmpfile` to use `TXG_NOTHROTTLE` OpenZFS-issue: https://www.illumos.org/issues/8997 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/19ea6cb0f9 Closes #7084
2018-01-08 21:45:53 +00:00
* Since we are not going to create any new dirty data, and we
* can even help with clearing the existing dirty data, we
* should not be subject to the dirty data based delays. We
* use TXG_NOTHROTTLE to bypass the delay mechanism.
OpenZFS 8373 - TXG_WAIT in ZIL commit path Authored by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8373 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7f04961 Closes #6403
2017-07-17 12:31:30 +00:00
*/
OpenZFS 8997 - ztest assertion failure in zil_lwb_write_issue PROBLEM ======= When `dmu_tx_assign` is called from `zil_lwb_write_issue`, it's possible for either `ERESTART` or `EIO` to be returned. If `ERESTART` is returned, this will cause an assertion to fail directly in `zil_lwb_write_issue`, where the code assumes the return value is `EIO` if `dmu_tx_assign` returns a non-zero value. This can occur if the SPA is suspended when `dmu_tx_assign` is called, and most often occurs when running `zloop`. If `EIO` is returned, this can cause assertions to fail elsewhere in the ZIL code. For example, `zil_commit_waiter_timeout` contains the following logic: lwb_t *nlwb = zil_lwb_write_issue(zilog, lwb); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_OPENED); In this case, if `dmu_tx_assign` returned `EIO` from within `zil_lwb_write_issue`, the `lwb` variable passed in will not be issued to disk. Thus, it's `lwb_state` field will remain `LWB_STATE_OPENED` and this assertion will fail. `zil_commit_waiter_timeout` assumes that after it calls `zil_lwb_write_issue`, the `lwb` will be issued to disk, and doesn't handle the case where this is not true; i.e. it doesn't handle the case where `dmu_tx_assign` returns `EIO`. SOLUTION ======== This change modifies the `dmu_tx_assign` function such that `txg_how` is a bitmask, rather than of the `txg_how_t` enum type. Now, the previous `TXG_WAITED` semantics can be used via `TXG_NOTHROTTLE`, along with specifying either `TXG_NOWAIT` or `TXG_WAIT` semantics. Previously, when `TXG_WAITED` was specified, `TXG_NOWAIT` semantics was automatically invoked. This was not ideal when using `TXG_WAITED` within `zil_lwb_write_issued`, leading the problem described above. Rather, we want to achieve the semantics of `TXG_WAIT`, while also preventing the `tx` from being penalized via the dirty delay throttling. With this change, `zil_lwb_write_issued` can acheive the semtantics that it requires by passing in the value `TXG_WAIT | TXG_NOTHROTTLE` to `dmu_tx_assign`. Further, consumers of `dmu_tx_assign` wishing to achieve the old `TXG_WAITED` semantics can pass in the value `TXG_NOWAIT | TXG_NOTHROTTLE`. Authored by: Prakash Surya <prakash.surya@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Porting Notes: - Additionally updated `zfs_tmpfile` to use `TXG_NOTHROTTLE` OpenZFS-issue: https://www.illumos.org/issues/8997 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/19ea6cb0f9 Closes #7084
2018-01-08 21:45:53 +00:00
VERIFY0(dmu_tx_assign(tx, TXG_WAIT | TXG_NOTHROTTLE));
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
lwb->lwb_tx = tx;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* Log blocks are pre-allocated. Here we select the size of the next
* block, based on size used in the last block.
* - first find the smallest bucket that will fit the block from a
* limited set of block sizes. This is because it's faster to write
* blocks allocated from the same metaslab as they are adjacent or
* close.
* - next find the maximum from the new suggested size and an array of
* previous sizes. This lessens a picket fence effect of wrongly
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* guessing the size if we have a stream of say 2k, 64k, 2k, 64k
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* requests.
*
* Note we only write what is used, but we can't just allocate
* the maximum block size because we can exhaust the available
* pool log space.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t);
Improve logging of 128KB writes Before my ZIL space optimization few years ago 128KB writes were logged as two 64KB+ records in two 128KB log blocks. After that change it became ~127KB+/1KB+ in two 128KB log blocks to free space in the second block for another record. Unfortunately in case of 128KB only writes, when space in the second block remained unused, that change increased write latency by unbalancing checksum computation and write times between parallel threads. It also didn't help with SLOG space efficiency in that case. This change introduces new 68KB log block size, used for both writes below 67KB and 128KB-sharp writes. Writes of 68-127KB are still using one 128KB block to not increase processing overhead. Writes above 131KB are still using full 128KB blocks, since possible saving there is small. Mixed loads will likely also fall back to previous 128KB, since code uses maximum of the last 16 requested block sizes. Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Closes #9409
2019-11-11 17:27:59 +00:00
for (i = 0; zil_blksz > zil_block_buckets[i].limit; i++)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
continue;
Improve logging of 128KB writes Before my ZIL space optimization few years ago 128KB writes were logged as two 64KB+ records in two 128KB log blocks. After that change it became ~127KB+/1KB+ in two 128KB log blocks to free space in the second block for another record. Unfortunately in case of 128KB only writes, when space in the second block remained unused, that change increased write latency by unbalancing checksum computation and write times between parallel threads. It also didn't help with SLOG space efficiency in that case. This change introduces new 68KB log block size, used for both writes below 67KB and 128KB-sharp writes. Writes of 68-127KB are still using one 128KB block to not increase processing overhead. Writes above 131KB are still using full 128KB blocks, since possible saving there is small. Mixed loads will likely also fall back to previous 128KB, since code uses maximum of the last 16 requested block sizes. Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Alexander Motin <mav@FreeBSD.org> Closes #9409
2019-11-11 17:27:59 +00:00
zil_blksz = MIN(zil_block_buckets[i].blksz, zilog->zl_max_block_size);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz;
for (i = 0; i < ZIL_PREV_BLKS; i++)
zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]);
zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
BP_ZERO(bp);
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
error = zio_alloc_zil(spa, zilog->zl_os, txg, bp, zil_blksz, &slog);
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
if (slog) {
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
ZIL_STAT_BUMP(zil_itx_metaslab_slog_count);
ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused);
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
} else {
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
ZIL_STAT_BUMP(zil_itx_metaslab_normal_count);
ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused);
}
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error == 0) {
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
ASSERT3U(bp->blk_birth, ==, txg);
bp->blk_cksum = lwb->lwb_blk.blk_cksum;
bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* Allocate a new log write block (lwb).
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
nlwb = zil_alloc_lwb(zilog, bp, slog, txg, TRUE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) {
/* For Slim ZIL only write what is used. */
wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t);
ASSERT3U(wsz, <=, lwb->lwb_sz);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zio_shrink(lwb->lwb_write_zio, wsz);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
} else {
wsz = lwb->lwb_sz;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilc->zc_pad = 0;
zilc->zc_nused = lwb->lwb_nused;
zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* clear unused data for security
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
spa_config_enter(zilog->zl_spa, SCL_STATE, lwb, RW_READER);
zil_lwb_add_block(lwb, &lwb->lwb_blk);
lwb->lwb_issued_timestamp = gethrtime();
lwb->lwb_state = LWB_STATE_ISSUED;
zio_nowait(lwb->lwb_root_zio);
zio_nowait(lwb->lwb_write_zio);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* If there was an allocation failure then nlwb will be null which
* forces a txg_wait_synced().
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
return (nlwb);
}
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
/*
* Maximum amount of write data that can be put into single log block.
*/
uint64_t
zil_max_log_data(zilog_t *zilog)
{
return (zilog->zl_max_block_size -
sizeof (zil_chain_t) - sizeof (lr_write_t));
}
/*
* Maximum amount of log space we agree to waste to reduce number of
* WR_NEED_COPY chunks to reduce zl_get_data() overhead (~12%).
*/
static inline uint64_t
zil_max_waste_space(zilog_t *zilog)
{
return (zil_max_log_data(zilog) / 8);
}
/*
* Maximum amount of write data for WR_COPIED. For correctness, consumers
* must fall back to WR_NEED_COPY if we can't fit the entire record into one
* maximum sized log block, because each WR_COPIED record must fit in a
* single log block. For space efficiency, we want to fit two records into a
* max-sized log block.
*/
uint64_t
zil_max_copied_data(zilog_t *zilog)
{
return ((zilog->zl_max_block_size - sizeof (zil_chain_t)) / 2 -
sizeof (lr_write_t));
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
static lwb_t *
zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
{
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lr_t *lrcb, *lrc;
lr_write_t *lrwb, *lrw;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
char *lr_buf;
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
uint64_t dlen, dnow, lwb_sp, reclen, txg, max_log_data;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb, !=, NULL);
ASSERT3P(lwb->lwb_buf, !=, NULL);
zil_lwb_write_open(zilog, lwb);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lrc = &itx->itx_lr;
lrw = (lr_write_t *)lrc;
/*
* A commit itx doesn't represent any on-disk state; instead
* it's simply used as a place holder on the commit list, and
* provides a mechanism for attaching a "commit waiter" onto the
* correct lwb (such that the waiter can be signalled upon
* completion of that lwb). Thus, we don't process this itx's
* log record if it's a commit itx (these itx's don't have log
* records), and instead link the itx's waiter onto the lwb's
* list of waiters.
*
* For more details, see the comment above zil_commit().
*/
if (lrc->lrc_txtype == TX_COMMIT) {
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
mutex_enter(&zilog->zl_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_commit_waiter_link_lwb(itx->itx_private, lwb);
itx->itx_private = NULL;
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
mutex_exit(&zilog->zl_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
return (lwb);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
dlen = P2ROUNDUP_TYPED(
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lrw->lr_length, sizeof (uint64_t), uint64_t);
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
} else {
dlen = 0;
}
reclen = lrc->lrc_reclen;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog->zl_cur_used += (reclen + dlen);
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
txg = lrc->lrc_txg;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3U(zilog->zl_cur_used, <, UINT64_MAX - (reclen + dlen));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
cont:
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* If this record won't fit in the current log block, start a new one.
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
* For WR_NEED_COPY optimize layout for minimal number of chunks.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lwb_sp = lwb->lwb_sz - lwb->lwb_nused;
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
max_log_data = zil_max_log_data(zilog);
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
if (reclen > lwb_sp || (reclen + dlen > lwb_sp &&
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
lwb_sp < zil_max_waste_space(zilog) &&
(dlen % max_log_data == 0 ||
lwb_sp < reclen + dlen % max_log_data))) {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lwb = zil_lwb_write_issue(zilog, lwb);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (lwb == NULL)
return (NULL);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_lwb_write_open(zilog, lwb);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
ASSERT(LWB_EMPTY(lwb));
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lwb_sp = lwb->lwb_sz - lwb->lwb_nused;
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
/*
* There must be enough space in the new, empty log block to
* hold reclen. For WR_COPIED, we need to fit the whole
* record in one block, and reclen is the header size + the
* data size. For WR_NEED_COPY, we can create multiple
* records, splitting the data into multiple blocks, so we
* only need to fit one word of data per block; in this case
* reclen is just the header size (no data).
*/
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
ASSERT3U(reclen + MIN(dlen, sizeof (uint64_t)), <=, lwb_sp);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
dnow = MIN(dlen, lwb_sp - reclen);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
lr_buf = lwb->lwb_buf + lwb->lwb_nused;
bcopy(lrc, lr_buf, reclen);
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lrcb = (lr_t *)lr_buf; /* Like lrc, but inside lwb. */
lrwb = (lr_write_t *)lrcb; /* Like lrw, but inside lwb. */
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
ZIL_STAT_BUMP(zil_itx_count);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* If it's a write, fetch the data or get its blkptr as appropriate.
*/
if (lrc->lrc_txtype == TX_WRITE) {
if (txg > spa_freeze_txg(zilog->zl_spa))
txg_wait_synced(zilog->zl_dmu_pool, txg);
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
if (itx->itx_wr_state == WR_COPIED) {
ZIL_STAT_BUMP(zil_itx_copied_count);
ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length);
} else {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
char *dbuf;
int error;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
if (itx->itx_wr_state == WR_NEED_COPY) {
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
dbuf = lr_buf + reclen;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lrcb->lrc_reclen += dnow;
if (lrwb->lr_length > dnow)
lrwb->lr_length = dnow;
lrw->lr_offset += dnow;
lrw->lr_length -= dnow;
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
ZIL_STAT_BUMP(zil_itx_needcopy_count);
Increment zil_itx_needcopy_bytes properly In zil_lwb_commit() with TX_WRITE, we copy the log write record (lrw) into the log write block (lwb) and send it off using zil_lwb_add_txg(). If we also have WR_NEED_COPY, we additionally copy the lwr's data into the lwb to be sent off. If the lwr + data doesn't fit into the lwb, we send the lrw and as much data as will fit (dnow bytes), then go back and do the same with the remaining data. Each time through this loop we're sending dnow data bytes. I.e. zil_itx_needcopy_bytes should be incremented by dnow. Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Chris Dunlop <chris@onthe.net.au> Closes #6988 Closes #7176
2018-03-02 18:01:53 +00:00
ZIL_STAT_INCR(zil_itx_needcopy_bytes, dnow);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
} else {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3S(itx->itx_wr_state, ==, WR_INDIRECT);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
dbuf = NULL;
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
ZIL_STAT_BUMP(zil_itx_indirect_count);
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
ZIL_STAT_INCR(zil_itx_indirect_bytes,
lrw->lr_length);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* We pass in the "lwb_write_zio" rather than
* "lwb_root_zio" so that the "lwb_write_zio"
* becomes the parent of any zio's created by
* the "zl_get_data" callback. The vdevs are
* flushed after the "lwb_write_zio" completes,
* so we want to make sure that completion
* callback waits for these additional zio's,
* such that the vdevs used by those zio's will
* be included in the lwb's vdev tree, and those
* vdevs will be properly flushed. If we passed
* in "lwb_root_zio" here, then these additional
* vdevs may not be flushed; e.g. if these zio's
* completed after "lwb_write_zio" completed.
*/
error = zilog->zl_get_data(itx->itx_private,
lrwb, dbuf, lwb, lwb->lwb_write_zio);
Rebase master to b121
2009-08-18 18:43:27 +00:00
if (error == EIO) {
txg_wait_synced(zilog->zl_dmu_pool, txg);
return (lwb);
}
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
ASSERT(error == ENOENT || error == EEXIST ||
error == EALREADY);
return (lwb);
}
}
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
/*
* We're actually making an entry, so update lrc_seq to be the
* log record sequence number. Note that this is generally not
* equal to the itx sequence number because not all transactions
* are synchronous, and sometimes spa_sync() gets there first.
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
lrcb->lrc_seq = ++zilog->zl_lr_seq;
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
lwb->lwb_nused += reclen + dnow;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_lwb_add_txg(lwb, txg);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz);
Illumos #3006 3006 VERIFY[S,U,P] and ASSERT[S,U,P] frequently check if first argument is zero Reviewed by Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by George Wilson <george.wilson@delphix.com> Approved by Eric Schrock <eric.schrock@delphix.com> References: illumos/illumos-gate@fb09f5aad449c97fe309678f3f604982b563a96f https://illumos.org/issues/3006 Requires: zfsonlinux/spl@1c6d149feb4033e4a56fb987004edc5d45288bcb Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1509
2013-05-10 21:17:03 +00:00
ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
dlen -= dnow;
if (dlen > 0) {
zilog->zl_cur_used += reclen;
goto cont;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (lwb);
}
itx_t *
zil_itx_create(uint64_t txtype, size_t lrsize)
{
Preserve itx alloc size for zio_data_buf_free() Using zio_data_buf_alloc() to allocate the itx's may be unsafe because the itx->itx_lr.lrc_reclen field is not constant from allocation to free. Using a different itx->itx_lr.lrc_reclen size in zio_data_buf_free() can result in the allocation being returned to the wrong kmem cache. This issue can be avoided entirely by storing the allocation size in itx->itx_size and using that for zio_data_buf_free(). Reviewed by: Prakash Surya <prakash.surya@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #6912
2017-12-04 19:44:39 +00:00
size_t itxsize;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
itx_t *itx;
lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t);
Preserve itx alloc size for zio_data_buf_free() Using zio_data_buf_alloc() to allocate the itx's may be unsafe because the itx->itx_lr.lrc_reclen field is not constant from allocation to free. Using a different itx->itx_lr.lrc_reclen size in zio_data_buf_free() can result in the allocation being returned to the wrong kmem cache. This issue can be avoided entirely by storing the allocation size in itx->itx_size and using that for zio_data_buf_free(). Reviewed by: Prakash Surya <prakash.surya@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #6912
2017-12-04 19:44:39 +00:00
itxsize = offsetof(itx_t, itx_lr) + lrsize;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Preserve itx alloc size for zio_data_buf_free() Using zio_data_buf_alloc() to allocate the itx's may be unsafe because the itx->itx_lr.lrc_reclen field is not constant from allocation to free. Using a different itx->itx_lr.lrc_reclen size in zio_data_buf_free() can result in the allocation being returned to the wrong kmem cache. This issue can be avoided entirely by storing the allocation size in itx->itx_size and using that for zio_data_buf_free(). Reviewed by: Prakash Surya <prakash.surya@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #6912
2017-12-04 19:44:39 +00:00
itx = zio_data_buf_alloc(itxsize);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
itx->itx_lr.lrc_txtype = txtype;
itx->itx_lr.lrc_reclen = lrsize;
itx->itx_lr.lrc_seq = 0; /* defensive */
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
itx->itx_sync = B_TRUE; /* default is synchronous */
Only commit the ZIL once in zpl_writepages() (msync() case). Currently, using msync() results in the following code path: sys_msync -> zpl_fsync -> filemap_write_and_wait_range -> zpl_writepages -> write_cache_pages -> zpl_putpage In such a code path, zil_commit() is called as part of zpl_putpage(). This means that for each page, the write is handed to the DMU, the ZIL is committed, and only then do we move on to the next page. As one might imagine, this results in atrocious performance where there is a large number of pages to write: instead of committing a batch of N writes, we do N commits containing one page each. In some extreme cases this can result in msync() being ~700 times slower than it should be, as well as very inefficient use of ZIL resources. This patch fixes this issue by making sure that the requested writes are batched and then committed only once. Unfortunately, the implementation is somewhat non-trivial because there is no way to run write_cache_pages in SYNC mode (so that we get all pages) without making it wait on the writeback tag for each page. The solution implemented here is composed of two parts: - I added a new callback system to the ZIL, which allows the caller to be notified when its ITX gets written to stable storage. One nice thing is that the callback is called not only in zil_commit() but in zil_sync() as well, which means that the caller doesn't have to care whether the write ended up in the ZIL or the DMU: it will get notified as soon as it's safe, period. This is an improvement over dmu_tx_callback_register() that was used previously, which only supports DMU writes. The rationale for this change is to allow zpl_putpage() to be notified when a ZIL commit is completed without having to block on zil_commit() itself. - zpl_writepages() now calls write_cache_pages in non-SYNC mode, which will prevent (1) write_cache_pages from blocking, and (2) zpl_putpage from issuing ZIL commits. zpl_writepages() will issue the commit itself instead of relying on zpl_putpage() to do it, thus nicely batching the writes. Note, however, that we still have to call write_cache_pages() again in SYNC mode because there is an edge case documented in the implementation of write_cache_pages() whereas it will not give us all dirty pages when running in non-SYNC mode. Thus we need to run it at least once in SYNC mode to make sure we honor persistency guarantees. This only happens when the pages are modified at the same time msync() is running, which should be rare. In most cases there won't be any additional pages and this second call will do nothing. Note that this change also fixes a bug related to #907 whereas calling msync() on pages that were already handed over to the DMU in a previous writepages() call would make msync() block until the next TXG sync instead of returning as soon as the ZIL commit is complete. The new callback system fixes that problem. Signed-off-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1849 Closes #907
2013-11-10 15:00:11 +00:00
itx->itx_callback = NULL;
itx->itx_callback_data = NULL;
Preserve itx alloc size for zio_data_buf_free() Using zio_data_buf_alloc() to allocate the itx's may be unsafe because the itx->itx_lr.lrc_reclen field is not constant from allocation to free. Using a different itx->itx_lr.lrc_reclen size in zio_data_buf_free() can result in the allocation being returned to the wrong kmem cache. This issue can be avoided entirely by storing the allocation size in itx->itx_size and using that for zio_data_buf_free(). Reviewed by: Prakash Surya <prakash.surya@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #6912
2017-12-04 19:44:39 +00:00
itx->itx_size = itxsize;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (itx);
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
void
zil_itx_destroy(itx_t *itx)
{
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
IMPLY(itx->itx_lr.lrc_txtype == TX_COMMIT, itx->itx_callback == NULL);
IMPLY(itx->itx_callback != NULL, itx->itx_lr.lrc_txtype != TX_COMMIT);
if (itx->itx_callback != NULL)
itx->itx_callback(itx->itx_callback_data);
Preserve itx alloc size for zio_data_buf_free() Using zio_data_buf_alloc() to allocate the itx's may be unsafe because the itx->itx_lr.lrc_reclen field is not constant from allocation to free. Using a different itx->itx_lr.lrc_reclen size in zio_data_buf_free() can result in the allocation being returned to the wrong kmem cache. This issue can be avoided entirely by storing the allocation size in itx->itx_size and using that for zio_data_buf_free(). Reviewed by: Prakash Surya <prakash.surya@delphix.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #6912
2017-12-04 19:44:39 +00:00
zio_data_buf_free(itx, itx->itx_size);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
}
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
/*
* Free up the sync and async itxs. The itxs_t has already been detached
* so no locks are needed.
*/
static void
zil_itxg_clean(itxs_t *itxs)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
itx_t *itx;
list_t *list;
avl_tree_t *t;
void *cookie;
itx_async_node_t *ian;
list = &itxs->i_sync_list;
while ((itx = list_head(list)) != NULL) {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* In the general case, commit itxs will not be found
* here, as they'll be committed to an lwb via
* zil_lwb_commit(), and free'd in that function. Having
* said that, it is still possible for commit itxs to be
* found here, due to the following race:
*
* - a thread calls zil_commit() which assigns the
* commit itx to a per-txg i_sync_list
* - zil_itxg_clean() is called (e.g. via spa_sync())
* while the waiter is still on the i_sync_list
*
* There's nothing to prevent syncing the txg while the
* waiter is on the i_sync_list. This normally doesn't
* happen because spa_sync() is slower than zil_commit(),
* but if zil_commit() calls txg_wait_synced() (e.g.
* because zil_create() or zil_commit_writer_stall() is
* called) we will hit this case.
*/
if (itx->itx_lr.lrc_txtype == TX_COMMIT)
zil_commit_waiter_skip(itx->itx_private);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
list_remove(list, itx);
Use zio buffers in zil_itx_create() The zil_itx_create() function uses the vmem_alloc() allocator for its buffers because when logging a write that buffer may be as large as 64K. This is non-optimal because we may need to allocate many of of these buffers and this interface has the potential to be slow. Instead, use zio_data_buf_alloc() which is specifically designed to be able to efficiently allocate a wide range of buffer sizes. In addition, do some cleanup and use the zil_itx_destroy() function to always free an itx structure. This way we're always sure the right allocation functions are used. Notice that in the current code kmem_free() and vmem_free() were both used. This happened to work because these wrappers map to the same internal SPL function. This was identified as a potential problem when a low-end memory constrained system began logging the following warnings. There was no deadlock here just repeated allocation failures resulting in increased latency. Possible memory allocation deadlock: size=65792 lflags=0x42d0 Pid: 20118, comm: kvm Tainted: P O 3.2.0-0.bpo.4-amd64 Call Trace: [<ffffffffa040b834>] ? spl_kmem_alloc_impl+0x115/0x127 [spl] [<ffffffffa040b84f>] ? spl_kmem_alloc_debug+0x9/0x36 [spl] [<ffffffffa05d8a0b>] ? zil_itx_create+0x2d/0x59 [zfs] [<ffffffffa05c71e6>] ? zfs_log_write+0x13a/0x2f0 [zfs] [<ffffffffa05d41bc>] ? zfs_write+0x85b/0x9bb [zfs] [<ffffffffa05e37ec>] ? zpl_aio_write+0xca/0x110 [zfs] [<ffffffff811088e5>] ? do_sync_readv_writev+0xa3/0xde [<ffffffff81108f41>] ? do_readv_writev+0xaf/0x125 [<ffffffff81109055>] ? sys_pwritev+0x55/0x9a [<ffffffff813721d2>] ? system_call_fastpath+0x16/0x1b Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Richard Yao <ryao@gentoo.org> Closes #3059
2015-01-29 23:09:51 +00:00
zil_itx_destroy(itx);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
cookie = NULL;
t = &itxs->i_async_tree;
while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
list = &ian->ia_list;
while ((itx = list_head(list)) != NULL) {
list_remove(list, itx);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/* commit itxs should never be on the async lists. */
ASSERT3U(itx->itx_lr.lrc_txtype, !=, TX_COMMIT);
Use zio buffers in zil_itx_create() The zil_itx_create() function uses the vmem_alloc() allocator for its buffers because when logging a write that buffer may be as large as 64K. This is non-optimal because we may need to allocate many of of these buffers and this interface has the potential to be slow. Instead, use zio_data_buf_alloc() which is specifically designed to be able to efficiently allocate a wide range of buffer sizes. In addition, do some cleanup and use the zil_itx_destroy() function to always free an itx structure. This way we're always sure the right allocation functions are used. Notice that in the current code kmem_free() and vmem_free() were both used. This happened to work because these wrappers map to the same internal SPL function. This was identified as a potential problem when a low-end memory constrained system began logging the following warnings. There was no deadlock here just repeated allocation failures resulting in increased latency. Possible memory allocation deadlock: size=65792 lflags=0x42d0 Pid: 20118, comm: kvm Tainted: P O 3.2.0-0.bpo.4-amd64 Call Trace: [<ffffffffa040b834>] ? spl_kmem_alloc_impl+0x115/0x127 [spl] [<ffffffffa040b84f>] ? spl_kmem_alloc_debug+0x9/0x36 [spl] [<ffffffffa05d8a0b>] ? zil_itx_create+0x2d/0x59 [zfs] [<ffffffffa05c71e6>] ? zfs_log_write+0x13a/0x2f0 [zfs] [<ffffffffa05d41bc>] ? zfs_write+0x85b/0x9bb [zfs] [<ffffffffa05e37ec>] ? zpl_aio_write+0xca/0x110 [zfs] [<ffffffff811088e5>] ? do_sync_readv_writev+0xa3/0xde [<ffffffff81108f41>] ? do_readv_writev+0xaf/0x125 [<ffffffff81109055>] ? sys_pwritev+0x55/0x9a [<ffffffff813721d2>] ? system_call_fastpath+0x16/0x1b Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Richard Yao <ryao@gentoo.org> Closes #3059
2015-01-29 23:09:51 +00:00
zil_itx_destroy(itx);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
}
list_destroy(list);
kmem_free(ian, sizeof (itx_async_node_t));
}
avl_destroy(t);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
kmem_free(itxs, sizeof (itxs_t));
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
static int
zil_aitx_compare(const void *x1, const void *x2)
{
const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid;
const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid;
Reduce loaded range tree memory usage This patch implements a new tree structure for ZFS, and uses it to store range trees more efficiently. The new structure is approximately a B-tree, though there are some small differences from the usual characterizations. The tree has core nodes and leaf nodes; each contain data elements, which the elements in the core nodes acting as separators between its children. The difference between core and leaf nodes is that the core nodes have an array of children, while leaf nodes don't. Every node in the tree may be only partially full; in most cases, they are all at least 50% full (in terms of element count) except for the root node, which can be less full. Underfull nodes will steal from their neighbors or merge to remain full enough, while overfull nodes will split in two. The data elements are contained in tree-controlled buffers; they are copied into these on insertion, and overwritten on deletion. This means that the elements are not independently allocated, which reduces overhead, but also means they can't be shared between trees (and also that pointers to them are only valid until a side-effectful tree operation occurs). The overhead varies based on how dense the tree is, but is usually on the order of about 50% of the element size; the per-node overheads are very small, and so don't make a significant difference. The trees can accept arbitrary records; they accept a size and a comparator to allow them to be used for a variety of purposes. The new trees replace the AVL trees used in the range trees today. Currently, the range_seg_t structure contains three 8 byte integers of payload and two 24 byte avl_tree_node_ts to handle its storage in both an offset-sorted tree and a size-sorted tree (total size: 64 bytes). In the new model, the range seg structures are usually two 4 byte integers, but a separate one needs to exist for the size-sorted and offset-sorted tree. Between the raw size, the 50% overhead, and the double storage, the new btrees are expected to use 8*1.5*2 = 24 bytes per record, or 33.3% as much memory as the AVL trees (this is for the purposes of storing metaslab range trees; for other purposes, like scrubs, they use ~50% as much memory). We reduced the size of the payload in the range segments by teaching range trees about starting offsets and shifts; since metaslabs have a fixed starting offset, and they all operate in terms of disk sectors, we can store the ranges using 4-byte integers as long as the size of the metaslab divided by the sector size is less than 2^32. For 512-byte sectors, this is a 2^41 (or 2TB) metaslab, which with the default settings corresponds to a 256PB disk. 4k sector disks can handle metaslabs up to 2^46 bytes, or 2^63 byte disks. Since we do not anticipate disks of this size in the near future, there should be almost no cases where metaslabs need 64-byte integers to store their ranges. We do still have the capability to store 64-byte integer ranges to account for cases where we are storing per-vdev (or per-dnode) trees, which could reasonably go above the limits discussed. We also do not store fill information in the compact version of the node, since it is only used for sorted scrub. We also optimized the metaslab loading process in various other ways to offset some inefficiencies in the btree model. While individual operations (find, insert, remove_from) are faster for the btree than they are for the avl tree, remove usually requires a find operation, while in the AVL tree model the element itself suffices. Some clever changes actually caused an overall speedup in metaslab loading; we use approximately 40% less cpu to load metaslabs in our tests on Illumos. Another memory and performance optimization was achieved by changing what is stored in the size-sorted trees. When a disk is heavily fragmented, the df algorithm used by default in ZFS will almost always find a number of small regions in its initial cursor-based search; it will usually only fall back to the size-sorted tree to find larger regions. If we increase the size of the cursor-based search slightly, and don't store segments that are smaller than a tunable size floor in the size-sorted tree, we can further cut memory usage down to below 20% of what the AVL trees store. This also results in further reductions in CPU time spent loading metaslabs. The 16KiB size floor was chosen because it results in substantial memory usage reduction while not usually resulting in situations where we can't find an appropriate chunk with the cursor and are forced to use an oversized chunk from the size-sorted tree. In addition, even if we do have to use an oversized chunk from the size-sorted tree, the chunk would be too small to use for ZIL allocations, so it isn't as big of a loss as it might otherwise be. And often, more small allocations will follow the initial one, and the cursor search will now find the remainder of the chunk we didn't use all of and use it for subsequent allocations. Practical testing has shown little or no change in fragmentation as a result of this change. If the size-sorted tree becomes empty while the offset sorted one still has entries, it will load all the entries from the offset sorted tree and disregard the size floor until it is unloaded again. This operation occurs rarely with the default setting, only on incredibly thoroughly fragmented pools. There are some other small changes to zdb to teach it to handle btrees, but nothing major. Reviewed-by: George Wilson <gwilson@delphix.com> Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed by: Sebastien Roy seb@delphix.com Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Paul Dagnelie <pcd@delphix.com> Closes #9181
2019-10-09 17:36:03 +00:00
return (TREE_CMP(o1, o2));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/*
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
* Remove all async itx with the given oid.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Fix out-of-order ZIL txtype lost on hardlinked files We should only call zil_remove_async when an object is removed. However, in current implementation, it is called whenever TX_REMOVE is called. In the case of hardlinked file, every unlink will generate TX_REMOVE and causing operations to be dropped even when the object is not removed. We fix this by only calling zil_remove_async when the file is fully unlinked. Reviewed-by: George Wilson <gwilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Prakash Surya <prakash.surya@delphix.com> Signed-off-by: Chunwei Chen <david.chen@nutanix.com> Closes #8769 Closes #9061
2019-08-14 03:21:27 +00:00
void
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
zil_remove_async(zilog_t *zilog, uint64_t oid)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
uint64_t otxg, txg;
itx_async_node_t *ian;
avl_tree_t *t;
avl_index_t where;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
list_t clean_list;
itx_t *itx;
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
ASSERT(oid != 0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node));
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
otxg = ZILTEST_TXG;
else
otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
mutex_enter(&itxg->itxg_lock);
if (itxg->itxg_txg != txg) {
mutex_exit(&itxg->itxg_lock);
continue;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
/*
* Locate the object node and append its list.
*/
t = &itxg->itxg_itxs->i_async_tree;
ian = avl_find(t, &oid, &where);
if (ian != NULL)
list_move_tail(&clean_list, &ian->ia_list);
mutex_exit(&itxg->itxg_lock);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
while ((itx = list_head(&clean_list)) != NULL) {
list_remove(&clean_list, itx);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/* commit itxs should never be on the async lists. */
ASSERT3U(itx->itx_lr.lrc_txtype, !=, TX_COMMIT);
Use zio buffers in zil_itx_create() The zil_itx_create() function uses the vmem_alloc() allocator for its buffers because when logging a write that buffer may be as large as 64K. This is non-optimal because we may need to allocate many of of these buffers and this interface has the potential to be slow. Instead, use zio_data_buf_alloc() which is specifically designed to be able to efficiently allocate a wide range of buffer sizes. In addition, do some cleanup and use the zil_itx_destroy() function to always free an itx structure. This way we're always sure the right allocation functions are used. Notice that in the current code kmem_free() and vmem_free() were both used. This happened to work because these wrappers map to the same internal SPL function. This was identified as a potential problem when a low-end memory constrained system began logging the following warnings. There was no deadlock here just repeated allocation failures resulting in increased latency. Possible memory allocation deadlock: size=65792 lflags=0x42d0 Pid: 20118, comm: kvm Tainted: P O 3.2.0-0.bpo.4-amd64 Call Trace: [<ffffffffa040b834>] ? spl_kmem_alloc_impl+0x115/0x127 [spl] [<ffffffffa040b84f>] ? spl_kmem_alloc_debug+0x9/0x36 [spl] [<ffffffffa05d8a0b>] ? zil_itx_create+0x2d/0x59 [zfs] [<ffffffffa05c71e6>] ? zfs_log_write+0x13a/0x2f0 [zfs] [<ffffffffa05d41bc>] ? zfs_write+0x85b/0x9bb [zfs] [<ffffffffa05e37ec>] ? zpl_aio_write+0xca/0x110 [zfs] [<ffffffff811088e5>] ? do_sync_readv_writev+0xa3/0xde [<ffffffff81108f41>] ? do_readv_writev+0xaf/0x125 [<ffffffff81109055>] ? sys_pwritev+0x55/0x9a [<ffffffff813721d2>] ? system_call_fastpath+0x16/0x1b Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Richard Yao <ryao@gentoo.org> Closes #3059
2015-01-29 23:09:51 +00:00
zil_itx_destroy(itx);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
list_destroy(&clean_list);
}
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
void
zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
{
uint64_t txg;
itxg_t *itxg;
itxs_t *itxs, *clean = NULL;
/*
* Ensure the data of a renamed file is committed before the rename.
*/
if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME)
zil_async_to_sync(zilog, itx->itx_oid);
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX)
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
txg = ZILTEST_TXG;
else
txg = dmu_tx_get_txg(tx);
itxg = &zilog->zl_itxg[txg & TXG_MASK];
mutex_enter(&itxg->itxg_lock);
itxs = itxg->itxg_itxs;
if (itxg->itxg_txg != txg) {
if (itxs != NULL) {
/*
* The zil_clean callback hasn't got around to cleaning
* this itxg. Save the itxs for release below.
* This should be rare.
*/
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
zfs_dbgmsg("zil_itx_assign: missed itx cleanup for "
"txg %llu", itxg->itxg_txg);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
clean = itxg->itxg_itxs;
}
itxg->itxg_txg = txg;
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t),
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
KM_SLEEP);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
list_create(&itxs->i_sync_list, sizeof (itx_t),
offsetof(itx_t, itx_node));
avl_create(&itxs->i_async_tree, zil_aitx_compare,
sizeof (itx_async_node_t),
offsetof(itx_async_node_t, ia_node));
}
if (itx->itx_sync) {
list_insert_tail(&itxs->i_sync_list, itx);
} else {
avl_tree_t *t = &itxs->i_async_tree;
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
uint64_t foid =
LR_FOID_GET_OBJ(((lr_ooo_t *)&itx->itx_lr)->lr_foid);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
itx_async_node_t *ian;
avl_index_t where;
ian = avl_find(t, &foid, &where);
if (ian == NULL) {
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
ian = kmem_alloc(sizeof (itx_async_node_t),
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
KM_SLEEP);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
list_create(&ian->ia_list, sizeof (itx_t),
offsetof(itx_t, itx_node));
ian->ia_foid = foid;
avl_insert(t, ian, where);
}
list_insert_tail(&ian->ia_list, itx);
}
itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* We don't want to dirty the ZIL using ZILTEST_TXG, because
* zil_clean() will never be called using ZILTEST_TXG. Thus, we
* need to be careful to always dirty the ZIL using the "real"
* TXG (not itxg_txg) even when the SPA is frozen.
*/
zilog_dirty(zilog, dmu_tx_get_txg(tx));
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
mutex_exit(&itxg->itxg_lock);
/* Release the old itxs now we've dropped the lock */
if (clean != NULL)
zil_itxg_clean(clean);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* If there are any in-memory intent log transactions which have now been
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
* synced then start up a taskq to free them. We should only do this after we
Fix typos in module/zfs/ Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Ryan Moeller <ryan@ixsystems.com> Reviewed-by: Richard Laager <rlaager@wiktel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Andrea Gelmini <andrea.gelmini@gelma.net> Closes #9240
2019-09-03 00:56:41 +00:00
* have written out the uberblocks (i.e. txg has been committed) so that
Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async 3086 unnecessarily setting DS_FLAG_INCONSISTENT on async destroyed datasets Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: illumos/illumos-gate@ce636f8b38e8c9ff484e880d9abb27251a882860 illumos changeset: 13776:cd512c80fd75 https://www.illumos.org/issues/3086 Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-12-15 00:13:40 +00:00
* don't inadvertently clean out in-memory log records that would be required
* by zil_commit().
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
void
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
zil_clean(zilog_t *zilog, uint64_t synced_txg)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK];
itxs_t *clean_me;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3U(synced_txg, <, ZILTEST_TXG);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
mutex_enter(&itxg->itxg_lock);
if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) {
mutex_exit(&itxg->itxg_lock);
return;
}
ASSERT3U(itxg->itxg_txg, <=, synced_txg);
OpenZFS 8558, 8602 - lwp_create() returns EAGAIN 8558 lwp_create() returns EAGAIN on system with more than 80K ZFS filesystems On a system with more than 80K ZFS filesystems, we've seen cases where lwp_create() will start to fail by returning EAGAIN. The problem being, for each of those 80K ZFS filesystems, a taskq will be created for each dataset as part of the ZIL for each dataset. Porting Notes: - The new nomem taskq kstat was dropped. - Added module options and documentation for new tunings zfs_zil_clean_taskq_nthr_pct, zfs_zil_clean_taskq_minalloc, zfs_zil_clean_taskq_maxalloc, and zfs_sync_taskq_batch_pct. Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Sebastien Roy <sebastien.roy@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8558 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/216d772 8602 remove unused "dp_early_sync_tasks" field from "dsl_pool" structure Reviewed by: Serapheim Dimitropoulos <serapheim@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8602 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2bcb545 Closes #6779
2017-10-26 19:57:53 +00:00
ASSERT3U(itxg->itxg_txg, !=, 0);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
clean_me = itxg->itxg_itxs;
itxg->itxg_itxs = NULL;
itxg->itxg_txg = 0;
mutex_exit(&itxg->itxg_lock);
/*
* Preferably start a task queue to free up the old itxs but
* if taskq_dispatch can't allocate resources to do that then
* free it in-line. This should be rare. Note, using TQ_SLEEP
* created a bad performance problem.
*/
OpenZFS 8558, 8602 - lwp_create() returns EAGAIN 8558 lwp_create() returns EAGAIN on system with more than 80K ZFS filesystems On a system with more than 80K ZFS filesystems, we've seen cases where lwp_create() will start to fail by returning EAGAIN. The problem being, for each of those 80K ZFS filesystems, a taskq will be created for each dataset as part of the ZIL for each dataset. Porting Notes: - The new nomem taskq kstat was dropped. - Added module options and documentation for new tunings zfs_zil_clean_taskq_nthr_pct, zfs_zil_clean_taskq_minalloc, zfs_zil_clean_taskq_maxalloc, and zfs_sync_taskq_batch_pct. Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Sebastien Roy <sebastien.roy@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8558 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/216d772 8602 remove unused "dp_early_sync_tasks" field from "dsl_pool" structure Reviewed by: Serapheim Dimitropoulos <serapheim@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Approved by: Robert Mustacchi <rm@joyent.com> Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed-by: George Melikov <mail@gmelikov.ru> Reviewed-by: Chris Dunlop <chris@onthe.net.au> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8602 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2bcb545 Closes #6779
2017-10-26 19:57:53 +00:00
ASSERT3P(zilog->zl_dmu_pool, !=, NULL);
ASSERT3P(zilog->zl_dmu_pool->dp_zil_clean_taskq, !=, NULL);
taskqid_t id = taskq_dispatch(zilog->zl_dmu_pool->dp_zil_clean_taskq,
(void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP);
if (id == TASKQID_INVALID)
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
zil_itxg_clean(clean_me);
}
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* This function will traverse the queue of itxs that need to be
* committed, and move them onto the ZIL's zl_itx_commit_list.
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
*/
static void
zil_get_commit_list(zilog_t *zilog)
{
uint64_t otxg, txg;
list_t *commit_list = &zilog->zl_itx_commit_list;
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
otxg = ZILTEST_TXG;
else
otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
/*
* This is inherently racy, since there is nothing to prevent
* the last synced txg from changing. That's okay since we'll
* only commit things in the future.
*/
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
mutex_enter(&itxg->itxg_lock);
if (itxg->itxg_txg != txg) {
mutex_exit(&itxg->itxg_lock);
continue;
}
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
/*
* If we're adding itx records to the zl_itx_commit_list,
* then the zil better be dirty in this "txg". We can assert
* that here since we're holding the itxg_lock which will
* prevent spa_sync from cleaning it. Once we add the itxs
* to the zl_itx_commit_list we must commit it to disk even
* if it's unnecessary (i.e. the txg was synced).
*/
ASSERT(zilog_is_dirty_in_txg(zilog, txg) ||
spa_freeze_txg(zilog->zl_spa) != UINT64_MAX);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list);
mutex_exit(&itxg->itxg_lock);
}
}
/*
* Move the async itxs for a specified object to commit into sync lists.
*/
Make `zil_async_to_sync` visible to platform code FreeBSD's zvol platform code requires access to the zil_async_to_sync() function. Reviewed-by: Ryan Moeller <ryan@ixsystems.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Closes #9440
2019-10-10 22:39:44 +00:00
void
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
zil_async_to_sync(zilog_t *zilog, uint64_t foid)
{
uint64_t otxg, txg;
itx_async_node_t *ian;
avl_tree_t *t;
avl_index_t where;
if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */
otxg = ZILTEST_TXG;
else
otxg = spa_last_synced_txg(zilog->zl_spa) + 1;
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
/*
* This is inherently racy, since there is nothing to prevent
* the last synced txg from changing.
*/
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) {
itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK];
mutex_enter(&itxg->itxg_lock);
if (itxg->itxg_txg != txg) {
mutex_exit(&itxg->itxg_lock);
continue;
}
/*
* If a foid is specified then find that node and append its
* list. Otherwise walk the tree appending all the lists
* to the sync list. We add to the end rather than the
* beginning to ensure the create has happened.
*/
t = &itxg->itxg_itxs->i_async_tree;
if (foid != 0) {
ian = avl_find(t, &foid, &where);
if (ian != NULL) {
list_move_tail(&itxg->itxg_itxs->i_sync_list,
&ian->ia_list);
}
} else {
void *cookie = NULL;
while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) {
list_move_tail(&itxg->itxg_itxs->i_sync_list,
&ian->ia_list);
list_destroy(&ian->ia_list);
kmem_free(ian, sizeof (itx_async_node_t));
}
}
mutex_exit(&itxg->itxg_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* This function will prune commit itxs that are at the head of the
* commit list (it won't prune past the first non-commit itx), and
* either: a) attach them to the last lwb that's still pending
* completion, or b) skip them altogether.
*
* This is used as a performance optimization to prevent commit itxs
* from generating new lwbs when it's unnecessary to do so.
*/
Rebase to OpenSolaris b103, in the process we are removing any code which did not originate from the OpenSolaris source. These changes will be reintroduced in topic branches for easier tracking
2008-12-03 20:09:06 +00:00
static void
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_prune_commit_list(zilog_t *zilog)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
itx_t *itx;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
while ((itx = list_head(&zilog->zl_itx_commit_list)) != NULL) {
lr_t *lrc = &itx->itx_lr;
if (lrc->lrc_txtype != TX_COMMIT)
break;
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_enter(&zilog->zl_lock);
lwb_t *last_lwb = zilog->zl_last_lwb_opened;
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
if (last_lwb == NULL ||
last_lwb->lwb_state == LWB_STATE_FLUSH_DONE) {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* All of the itxs this waiter was waiting on
* must have already completed (or there were
* never any itx's for it to wait on), so it's
* safe to skip this waiter and mark it done.
*/
zil_commit_waiter_skip(itx->itx_private);
} else {
zil_commit_waiter_link_lwb(itx->itx_private, last_lwb);
itx->itx_private = NULL;
}
mutex_exit(&zilog->zl_lock);
list_remove(&zilog->zl_itx_commit_list, itx);
zil_itx_destroy(itx);
}
IMPLY(itx != NULL, itx->itx_lr.lrc_txtype != TX_COMMIT);
}
static void
zil_commit_writer_stall(zilog_t *zilog)
{
/*
* When zio_alloc_zil() fails to allocate the next lwb block on
* disk, we must call txg_wait_synced() to ensure all of the
* lwbs in the zilog's zl_lwb_list are synced and then freed (in
* zil_sync()), such that any subsequent ZIL writer (i.e. a call
* to zil_process_commit_list()) will have to call zil_create(),
* and start a new ZIL chain.
*
* Since zil_alloc_zil() failed, the lwb that was previously
* issued does not have a pointer to the "next" lwb on disk.
* Thus, if another ZIL writer thread was to allocate the "next"
* on-disk lwb, that block could be leaked in the event of a
* crash (because the previous lwb on-disk would not point to
* it).
*
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
* We must hold the zilog's zl_issuer_lock while we do this, to
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* ensure no new threads enter zil_process_commit_list() until
* all lwb's in the zl_lwb_list have been synced and freed
* (which is achieved via the txg_wait_synced() call).
*/
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
txg_wait_synced(zilog->zl_dmu_pool, 0);
ASSERT3P(list_tail(&zilog->zl_lwb_list), ==, NULL);
}
/*
* This function will traverse the commit list, creating new lwbs as
* needed, and committing the itxs from the commit list to these newly
* created lwbs. Additionally, as a new lwb is created, the previous
* lwb will be issued to the zio layer to be written to disk.
*/
static void
zil_process_commit_list(zilog_t *zilog)
{
spa_t *spa = zilog->zl_spa;
list_t nolwb_itxs;
list_t nolwb_waiters;
lwb_t *lwb;
itx_t *itx;
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock));
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
/*
* Return if there's nothing to commit before we dirty the fs by
* calling zil_create().
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (list_head(&zilog->zl_itx_commit_list) == NULL)
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
return;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
list_create(&nolwb_itxs, sizeof (itx_t), offsetof(itx_t, itx_node));
list_create(&nolwb_waiters, sizeof (zil_commit_waiter_t),
offsetof(zil_commit_waiter_t, zcw_node));
lwb = list_tail(&zilog->zl_lwb_list);
if (lwb == NULL) {
lwb = zil_create(zilog);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
} else {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_ISSUED);
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_WRITE_DONE);
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_FLUSH_DONE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
while ((itx = list_head(&zilog->zl_itx_commit_list)) != NULL) {
lr_t *lrc = &itx->itx_lr;
uint64_t txg = lrc->lrc_txg;
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
ASSERT3U(txg, !=, 0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lrc->lrc_txtype == TX_COMMIT) {
DTRACE_PROBE2(zil__process__commit__itx,
zilog_t *, zilog, itx_t *, itx);
} else {
DTRACE_PROBE2(zil__process__normal__itx,
zilog_t *, zilog, itx_t *, itx);
}
list_remove(&zilog->zl_itx_commit_list, itx);
boolean_t synced = txg <= spa_last_synced_txg(spa);
boolean_t frozen = txg > spa_freeze_txg(spa);
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
/*
* If the txg of this itx has already been synced out, then
* we don't need to commit this itx to an lwb. This is
* because the data of this itx will have already been
* written to the main pool. This is inherently racy, and
* it's still ok to commit an itx whose txg has already
* been synced; this will result in a write that's
* unnecessary, but will do no harm.
*
* With that said, we always want to commit TX_COMMIT itxs
* to an lwb, regardless of whether or not that itx's txg
* has been synced out. We do this to ensure any OPENED lwb
* will always have at least one zil_commit_waiter_t linked
* to the lwb.
*
* As a counter-example, if we skipped TX_COMMIT itx's
* whose txg had already been synced, the following
* situation could occur if we happened to be racing with
* spa_sync:
*
* 1. We commit a non-TX_COMMIT itx to an lwb, where the
* itx's txg is 10 and the last synced txg is 9.
* 2. spa_sync finishes syncing out txg 10.
* 3. We move to the next itx in the list, it's a TX_COMMIT
* whose txg is 10, so we skip it rather than committing
* it to the lwb used in (1).
*
* If the itx that is skipped in (3) is the last TX_COMMIT
* itx in the commit list, than it's possible for the lwb
* used in (1) to remain in the OPENED state indefinitely.
*
* To prevent the above scenario from occurring, ensuring
* that once an lwb is OPENED it will transition to ISSUED
* and eventually DONE, we always commit TX_COMMIT itx's to
* an lwb here, even if that itx's txg has already been
* synced.
*
* Finally, if the pool is frozen, we _always_ commit the
* itx. The point of freezing the pool is to prevent data
* from being written to the main pool via spa_sync, and
* instead rely solely on the ZIL to persistently store the
* data; i.e. when the pool is frozen, the last synced txg
* value can't be trusted.
*/
if (frozen || !synced || lrc->lrc_txtype == TX_COMMIT) {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lwb != NULL) {
lwb = zil_lwb_commit(zilog, itx, lwb);
if (lwb == NULL)
list_insert_tail(&nolwb_itxs, itx);
else
list_insert_tail(&lwb->lwb_itxs, itx);
} else {
if (lrc->lrc_txtype == TX_COMMIT) {
zil_commit_waiter_link_nolwb(
itx->itx_private, &nolwb_waiters);
}
list_insert_tail(&nolwb_itxs, itx);
}
} else {
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
ASSERT3S(lrc->lrc_txtype, !=, TX_COMMIT);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_itx_destroy(itx);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lwb == NULL) {
/*
* This indicates zio_alloc_zil() failed to allocate the
* "next" lwb on-disk. When this happens, we must stall
* the ZIL write pipeline; see the comment within
* zil_commit_writer_stall() for more details.
*/
zil_commit_writer_stall(zilog);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* Additionally, we have to signal and mark the "nolwb"
* waiters as "done" here, since without an lwb, we
* can't do this via zil_lwb_flush_vdevs_done() like
* normal.
*/
zil_commit_waiter_t *zcw;
while ((zcw = list_head(&nolwb_waiters)) != NULL) {
zil_commit_waiter_skip(zcw);
list_remove(&nolwb_waiters, zcw);
}
/*
* And finally, we have to destroy the itx's that
* couldn't be committed to an lwb; this will also call
* the itx's callback if one exists for the itx.
*/
while ((itx = list_head(&nolwb_itxs)) != NULL) {
list_remove(&nolwb_itxs, itx);
zil_itx_destroy(itx);
}
} else {
ASSERT(list_is_empty(&nolwb_waiters));
ASSERT3P(lwb, !=, NULL);
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_ISSUED);
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_WRITE_DONE);
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_FLUSH_DONE);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* At this point, the ZIL block pointed at by the "lwb"
* variable is in one of the following states: "closed"
* or "open".
*
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* If it's "closed", then no itxs have been committed to
* it, so there's no point in issuing its zio (i.e. it's
* "empty").
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* If it's "open", then it contains one or more itxs that
* eventually need to be committed to stable storage. In
* this case we intentionally do not issue the lwb's zio
* to disk yet, and instead rely on one of the following
* two mechanisms for issuing the zio:
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* 1. Ideally, there will be more ZIL activity occurring
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* on the system, such that this function will be
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* immediately called again (not necessarily by the same
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* thread) and this lwb's zio will be issued via
* zil_lwb_commit(). This way, the lwb is guaranteed to
* be "full" when it is issued to disk, and we'll make
* use of the lwb's size the best we can.
*
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* 2. If there isn't sufficient ZIL activity occurring on
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* the system, such that this lwb's zio isn't issued via
* zil_lwb_commit(), zil_commit_waiter() will issue the
* lwb's zio. If this occurs, the lwb is not guaranteed
* to be "full" by the time its zio is issued, and means
* the size of the lwb was "too large" given the amount
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* of ZIL activity occurring on the system at that time.
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
* We do this for a couple of reasons:
*
* 1. To try and reduce the number of IOPs needed to
* write the same number of itxs. If an lwb has space
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* available in its buffer for more itxs, and more itxs
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* will be committed relatively soon (relative to the
* latency of performing a write), then it's beneficial
* to wait for these "next" itxs. This way, more itxs
* can be committed to stable storage with fewer writes.
*
* 2. To try and use the largest lwb block size that the
* incoming rate of itxs can support. Again, this is to
* try and pack as many itxs into as few lwbs as
* possible, without significantly impacting the latency
* of each individual itx.
*/
}
}
/*
* This function is responsible for ensuring the passed in commit waiter
* (and associated commit itx) is committed to an lwb. If the waiter is
* not already committed to an lwb, all itxs in the zilog's queue of
* itxs will be processed. The assumption is the passed in waiter's
* commit itx will found in the queue just like the other non-commit
* itxs, such that when the entire queue is processed, the waiter will
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* have been committed to an lwb.
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
* The lwb associated with the passed in waiter is not guaranteed to
* have been issued by the time this function completes. If the lwb is
* not issued, we rely on future calls to zil_commit_writer() to issue
* the lwb, or the timeout mechanism found in zil_commit_waiter().
*/
static void
zil_commit_writer(zilog_t *zilog, zil_commit_waiter_t *zcw)
{
ASSERT(!MUTEX_HELD(&zilog->zl_lock));
ASSERT(spa_writeable(zilog->zl_spa));
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
mutex_enter(&zilog->zl_issuer_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (zcw->zcw_lwb != NULL || zcw->zcw_done) {
/*
* It's possible that, while we were waiting to acquire
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
* the "zl_issuer_lock", another thread committed this
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* waiter to an lwb. If that occurs, we bail out early,
* without processing any of the zilog's queue of itxs.
*
* On certain workloads and system configurations, the
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
* "zl_issuer_lock" can become highly contended. In an
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* attempt to reduce this contention, we immediately drop
* the lock if the waiter has already been processed.
*
* We've measured this optimization to reduce CPU spent
* contending on this lock by up to 5%, using a system
* with 32 CPUs, low latency storage (~50 usec writes),
* and 1024 threads performing sync writes.
*/
goto out;
}
ZIL_STAT_BUMP(zil_commit_writer_count);
zil_get_commit_list(zilog);
zil_prune_commit_list(zilog);
zil_process_commit_list(zilog);
out:
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
mutex_exit(&zilog->zl_issuer_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
}
static void
zil_commit_waiter_timeout(zilog_t *zilog, zil_commit_waiter_t *zcw)
{
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(!MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT(MUTEX_HELD(&zcw->zcw_lock));
ASSERT3B(zcw->zcw_done, ==, B_FALSE);
lwb_t *lwb = zcw->zcw_lwb;
ASSERT3P(lwb, !=, NULL);
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_CLOSED);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* If the lwb has already been issued by another thread, we can
* immediately return since there's no work to be done (the
* point of this function is to issue the lwb). Additionally, we
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
* do this prior to acquiring the zl_issuer_lock, to avoid
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* acquiring it when it's not necessary to do so.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lwb->lwb_state == LWB_STATE_ISSUED ||
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
lwb->lwb_state == LWB_STATE_WRITE_DONE ||
lwb->lwb_state == LWB_STATE_FLUSH_DONE)
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
return;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* In order to call zil_lwb_write_issue() we must hold the
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
* zilog's "zl_issuer_lock". We can't simply acquire that lock,
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* since we're already holding the commit waiter's "zcw_lock",
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* and those two locks are acquired in the opposite order
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* elsewhere.
*/
mutex_exit(&zcw->zcw_lock);
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
mutex_enter(&zilog->zl_issuer_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
mutex_enter(&zcw->zcw_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* Since we just dropped and re-acquired the commit waiter's
* lock, we have to re-check to see if the waiter was marked
* "done" during that process. If the waiter was marked "done",
* the "lwb" pointer is no longer valid (it can be free'd after
* the waiter is marked "done"), so without this check we could
* wind up with a use-after-free error below.
*/
if (zcw->zcw_done)
goto out;
Only commit the ZIL once in zpl_writepages() (msync() case). Currently, using msync() results in the following code path: sys_msync -> zpl_fsync -> filemap_write_and_wait_range -> zpl_writepages -> write_cache_pages -> zpl_putpage In such a code path, zil_commit() is called as part of zpl_putpage(). This means that for each page, the write is handed to the DMU, the ZIL is committed, and only then do we move on to the next page. As one might imagine, this results in atrocious performance where there is a large number of pages to write: instead of committing a batch of N writes, we do N commits containing one page each. In some extreme cases this can result in msync() being ~700 times slower than it should be, as well as very inefficient use of ZIL resources. This patch fixes this issue by making sure that the requested writes are batched and then committed only once. Unfortunately, the implementation is somewhat non-trivial because there is no way to run write_cache_pages in SYNC mode (so that we get all pages) without making it wait on the writeback tag for each page. The solution implemented here is composed of two parts: - I added a new callback system to the ZIL, which allows the caller to be notified when its ITX gets written to stable storage. One nice thing is that the callback is called not only in zil_commit() but in zil_sync() as well, which means that the caller doesn't have to care whether the write ended up in the ZIL or the DMU: it will get notified as soon as it's safe, period. This is an improvement over dmu_tx_callback_register() that was used previously, which only supports DMU writes. The rationale for this change is to allow zpl_putpage() to be notified when a ZIL commit is completed without having to block on zil_commit() itself. - zpl_writepages() now calls write_cache_pages in non-SYNC mode, which will prevent (1) write_cache_pages from blocking, and (2) zpl_putpage from issuing ZIL commits. zpl_writepages() will issue the commit itself instead of relying on zpl_putpage() to do it, thus nicely batching the writes. Note, however, that we still have to call write_cache_pages() again in SYNC mode because there is an edge case documented in the implementation of write_cache_pages() whereas it will not give us all dirty pages when running in non-SYNC mode. Thus we need to run it at least once in SYNC mode to make sure we honor persistency guarantees. This only happens when the pages are modified at the same time msync() is running, which should be rare. In most cases there won't be any additional pages and this second call will do nothing. Note that this change also fixes a bug related to #907 whereas calling msync() on pages that were already handed over to the DMU in a previous writepages() call would make msync() block until the next TXG sync instead of returning as soon as the ZIL commit is complete. The new callback system fixes that problem. Signed-off-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1849 Closes #907
2013-11-10 15:00:11 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb, ==, zcw->zcw_lwb);
/*
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* We've already checked this above, but since we hadn't acquired
* the zilog's zl_issuer_lock, we have to perform this check a
* second time while holding the lock.
*
* We don't need to hold the zl_lock since the lwb cannot transition
* from OPENED to ISSUED while we hold the zl_issuer_lock. The lwb
* _can_ transition from ISSUED to DONE, but it's OK to race with
* that transition since we treat the lwb the same, whether it's in
* the ISSUED or DONE states.
*
* The important thing, is we treat the lwb differently depending on
* if it's ISSUED or OPENED, and block any other threads that might
* attempt to issue this lwb. For that reason we hold the
* zl_issuer_lock when checking the lwb_state; we must not call
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* zil_lwb_write_issue() if the lwb had already been issued.
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
*
* See the comment above the lwb_state_t structure definition for
* more details on the lwb states, and locking requirements.
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*/
if (lwb->lwb_state == LWB_STATE_ISSUED ||
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
lwb->lwb_state == LWB_STATE_WRITE_DONE ||
lwb->lwb_state == LWB_STATE_FLUSH_DONE)
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
goto out;
ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED);
/*
* As described in the comments above zil_commit_waiter() and
* zil_process_commit_list(), we need to issue this lwb's zio
* since we've reached the commit waiter's timeout and it still
* hasn't been issued.
*/
lwb_t *nlwb = zil_lwb_write_issue(zilog, lwb);
OpenZFS 9456 - ztest failure in zil_commit_waiter_timeout Problem ======= Illumos bug 8373 was integrated, which now presents a code path where "dmu_tx_assign" can fail. When "dmu_tx_assign" fails, it will not issue the lwb that was passed in to "zil_lwb_write_issue". As a result, when "zil_lwb_write_issue" returns, the lwb will still be in the "opened" state, just as it was when "zil_lwb_write_issue" was originally called. Solution ======== As a result of this new call path, the failed assertion needs to be modified to be aware of this new possibility. Thus, we can only assert that the lwb is no longer in the "opened" state if the returned lwb is non-null, since we cannot differentiate between the case of "dmu_tx_assign" failing or "zio_alloc_zil" failing within the call to "zil_lwb_write_issue". Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: George Melikov <mail@gmelikov.ru> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Matt Ahrens <mahrens@delphix.com> OpenZFS-issue: https://www.illumos.org/issues/9456 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/a8b09f4e Closes #7695
2017-09-07 22:12:35 +00:00
IMPLY(nlwb != NULL, lwb->lwb_state != LWB_STATE_OPENED);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* Since the lwb's zio hadn't been issued by the time this thread
* reached its timeout, we reset the zilog's "zl_cur_used" field
* to influence the zil block size selection algorithm.
*
* By having to issue the lwb's zio here, it means the size of the
* lwb was too large, given the incoming throughput of itxs. By
* setting "zl_cur_used" to zero, we communicate this fact to the
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* block size selection algorithm, so it can take this information
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* into account, and potentially select a smaller size for the
* next lwb block that is allocated.
*/
zilog->zl_cur_used = 0;
if (nlwb == NULL) {
/*
* When zil_lwb_write_issue() returns NULL, this
* indicates zio_alloc_zil() failed to allocate the
* "next" lwb on-disk. When this occurs, the ZIL write
* pipeline must be stalled; see the comment within the
* zil_commit_writer_stall() function for more details.
*
* We must drop the commit waiter's lock prior to
* calling zil_commit_writer_stall() or else we can wind
* up with the following deadlock:
*
* - This thread is waiting for the txg to sync while
* holding the waiter's lock; txg_wait_synced() is
* used within txg_commit_writer_stall().
*
* - The txg can't sync because it is waiting for this
* lwb's zio callback to call dmu_tx_commit().
*
* - The lwb's zio callback can't call dmu_tx_commit()
* because it's blocked trying to acquire the waiter's
* lock, which occurs prior to calling dmu_tx_commit()
*/
mutex_exit(&zcw->zcw_lock);
zil_commit_writer_stall(zilog);
mutex_enter(&zcw->zcw_lock);
Only commit the ZIL once in zpl_writepages() (msync() case). Currently, using msync() results in the following code path: sys_msync -> zpl_fsync -> filemap_write_and_wait_range -> zpl_writepages -> write_cache_pages -> zpl_putpage In such a code path, zil_commit() is called as part of zpl_putpage(). This means that for each page, the write is handed to the DMU, the ZIL is committed, and only then do we move on to the next page. As one might imagine, this results in atrocious performance where there is a large number of pages to write: instead of committing a batch of N writes, we do N commits containing one page each. In some extreme cases this can result in msync() being ~700 times slower than it should be, as well as very inefficient use of ZIL resources. This patch fixes this issue by making sure that the requested writes are batched and then committed only once. Unfortunately, the implementation is somewhat non-trivial because there is no way to run write_cache_pages in SYNC mode (so that we get all pages) without making it wait on the writeback tag for each page. The solution implemented here is composed of two parts: - I added a new callback system to the ZIL, which allows the caller to be notified when its ITX gets written to stable storage. One nice thing is that the callback is called not only in zil_commit() but in zil_sync() as well, which means that the caller doesn't have to care whether the write ended up in the ZIL or the DMU: it will get notified as soon as it's safe, period. This is an improvement over dmu_tx_callback_register() that was used previously, which only supports DMU writes. The rationale for this change is to allow zpl_putpage() to be notified when a ZIL commit is completed without having to block on zil_commit() itself. - zpl_writepages() now calls write_cache_pages in non-SYNC mode, which will prevent (1) write_cache_pages from blocking, and (2) zpl_putpage from issuing ZIL commits. zpl_writepages() will issue the commit itself instead of relying on zpl_putpage() to do it, thus nicely batching the writes. Note, however, that we still have to call write_cache_pages() again in SYNC mode because there is an edge case documented in the implementation of write_cache_pages() whereas it will not give us all dirty pages when running in non-SYNC mode. Thus we need to run it at least once in SYNC mode to make sure we honor persistency guarantees. This only happens when the pages are modified at the same time msync() is running, which should be rare. In most cases there won't be any additional pages and this second call will do nothing. Note that this change also fixes a bug related to #907 whereas calling msync() on pages that were already handed over to the DMU in a previous writepages() call would make msync() block until the next TXG sync instead of returning as soon as the ZIL commit is complete. The new callback system fixes that problem. Signed-off-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1849 Closes #907
2013-11-10 15:00:11 +00:00
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
out:
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
mutex_exit(&zilog->zl_issuer_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT(MUTEX_HELD(&zcw->zcw_lock));
}
/*
* This function is responsible for performing the following two tasks:
*
* 1. its primary responsibility is to block until the given "commit
* waiter" is considered "done".
*
* 2. its secondary responsibility is to issue the zio for the lwb that
* the given "commit waiter" is waiting on, if this function has
* waited "long enough" and the lwb is still in the "open" state.
*
* Given a sufficient amount of itxs being generated and written using
* the ZIL, the lwb's zio will be issued via the zil_lwb_commit()
* function. If this does not occur, this secondary responsibility will
* ensure the lwb is issued even if there is not other synchronous
* activity on the system.
*
* For more details, see zil_process_commit_list(); more specifically,
* the comment at the bottom of that function.
*/
static void
zil_commit_waiter(zilog_t *zilog, zil_commit_waiter_t *zcw)
{
ASSERT(!MUTEX_HELD(&zilog->zl_lock));
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
ASSERT(!MUTEX_HELD(&zilog->zl_issuer_lock));
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT(spa_writeable(zilog->zl_spa));
mutex_enter(&zcw->zcw_lock);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* The timeout is scaled based on the lwb latency to avoid
* significantly impacting the latency of each individual itx.
* For more details, see the comment at the bottom of the
* zil_process_commit_list() function.
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
*/
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
int pct = MAX(zfs_commit_timeout_pct, 1);
hrtime_t sleep = (zilog->zl_last_lwb_latency * pct) / 100;
hrtime_t wakeup = gethrtime() + sleep;
boolean_t timedout = B_FALSE;
while (!zcw->zcw_done) {
ASSERT(MUTEX_HELD(&zcw->zcw_lock));
lwb_t *lwb = zcw->zcw_lwb;
/*
* Usually, the waiter will have a non-NULL lwb field here,
* but it's possible for it to be NULL as a result of
* zil_commit() racing with spa_sync().
*
* When zil_clean() is called, it's possible for the itxg
* list (which may be cleaned via a taskq) to contain
* commit itxs. When this occurs, the commit waiters linked
* off of these commit itxs will not be committed to an
* lwb. Additionally, these commit waiters will not be
* marked done until zil_commit_waiter_skip() is called via
* zil_itxg_clean().
*
* Thus, it's possible for this commit waiter (i.e. the
* "zcw" variable) to be found in this "in between" state;
* where it's "zcw_lwb" field is NULL, and it hasn't yet
* been skipped, so it's "zcw_done" field is still B_FALSE.
*/
IMPLY(lwb != NULL, lwb->lwb_state != LWB_STATE_CLOSED);
if (lwb != NULL && lwb->lwb_state == LWB_STATE_OPENED) {
ASSERT3B(timedout, ==, B_FALSE);
/*
* If the lwb hasn't been issued yet, then we
* need to wait with a timeout, in case this
* function needs to issue the lwb after the
* timeout is reached; responsibility (2) from
* the comment above this function.
*/
clock_t timeleft = cv_timedwait_hires(&zcw->zcw_cv,
&zcw->zcw_lock, wakeup, USEC2NSEC(1),
CALLOUT_FLAG_ABSOLUTE);
if (timeleft >= 0 || zcw->zcw_done)
continue;
timedout = B_TRUE;
zil_commit_waiter_timeout(zilog, zcw);
if (!zcw->zcw_done) {
/*
* If the commit waiter has already been
* marked "done", it's possible for the
* waiter's lwb structure to have already
* been freed. Thus, we can only reliably
* make these assertions if the waiter
* isn't done.
*/
ASSERT3P(lwb, ==, zcw->zcw_lwb);
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_OPENED);
}
} else {
/*
* If the lwb isn't open, then it must have already
* been issued. In that case, there's no need to
* use a timeout when waiting for the lwb to
* complete.
*
* Additionally, if the lwb is NULL, the waiter
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* will soon be signaled and marked done via
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* zil_clean() and zil_itxg_clean(), so no timeout
* is required.
*/
IMPLY(lwb != NULL,
lwb->lwb_state == LWB_STATE_ISSUED ||
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
lwb->lwb_state == LWB_STATE_WRITE_DONE ||
lwb->lwb_state == LWB_STATE_FLUSH_DONE);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
cv_wait(&zcw->zcw_cv, &zcw->zcw_lock);
}
}
mutex_exit(&zcw->zcw_lock);
}
static zil_commit_waiter_t *
zil_alloc_commit_waiter(void)
{
zil_commit_waiter_t *zcw = kmem_cache_alloc(zil_zcw_cache, KM_SLEEP);
cv_init(&zcw->zcw_cv, NULL, CV_DEFAULT, NULL);
mutex_init(&zcw->zcw_lock, NULL, MUTEX_DEFAULT, NULL);
list_link_init(&zcw->zcw_node);
zcw->zcw_lwb = NULL;
zcw->zcw_done = B_FALSE;
zcw->zcw_zio_error = 0;
return (zcw);
}
static void
zil_free_commit_waiter(zil_commit_waiter_t *zcw)
{
ASSERT(!list_link_active(&zcw->zcw_node));
ASSERT3P(zcw->zcw_lwb, ==, NULL);
ASSERT3B(zcw->zcw_done, ==, B_TRUE);
mutex_destroy(&zcw->zcw_lock);
cv_destroy(&zcw->zcw_cv);
kmem_cache_free(zil_zcw_cache, zcw);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* This function is used to create a TX_COMMIT itx and assign it. This
* way, it will be linked into the ZIL's list of synchronous itxs, and
* then later committed to an lwb (or skipped) when
* zil_process_commit_list() is called.
*/
static void
zil_commit_itx_assign(zilog_t *zilog, zil_commit_waiter_t *zcw)
{
dmu_tx_t *tx = dmu_tx_create(zilog->zl_os);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
itx_t *itx = zil_itx_create(TX_COMMIT, sizeof (lr_t));
itx->itx_sync = B_TRUE;
itx->itx_private = zcw;
zil_itx_assign(zilog, itx, tx);
dmu_tx_commit(tx);
}
/*
* Commit ZFS Intent Log transactions (itxs) to stable storage.
*
* When writing ZIL transactions to the on-disk representation of the
* ZIL, the itxs are committed to a Log Write Block (lwb). Multiple
* itxs can be committed to a single lwb. Once a lwb is written and
* committed to stable storage (i.e. the lwb is written, and vdevs have
* been flushed), each itx that was committed to that lwb is also
* considered to be committed to stable storage.
*
* When an itx is committed to an lwb, the log record (lr_t) contained
* by the itx is copied into the lwb's zio buffer, and once this buffer
* is written to disk, it becomes an on-disk ZIL block.
*
* As itxs are generated, they're inserted into the ZIL's queue of
* uncommitted itxs. The semantics of zil_commit() are such that it will
* block until all itxs that were in the queue when it was called, are
* committed to stable storage.
*
* If "foid" is zero, this means all "synchronous" and "asynchronous"
* itxs, for all objects in the dataset, will be committed to stable
* storage prior to zil_commit() returning. If "foid" is non-zero, all
* "synchronous" itxs for all objects, but only "asynchronous" itxs
* that correspond to the foid passed in, will be committed to stable
* storage prior to zil_commit() returning.
*
* Generally speaking, when zil_commit() is called, the consumer doesn't
* actually care about _all_ of the uncommitted itxs. Instead, they're
* simply trying to waiting for a specific itx to be committed to disk,
* but the interface(s) for interacting with the ZIL don't allow such
* fine-grained communication. A better interface would allow a consumer
* to create and assign an itx, and then pass a reference to this itx to
* zil_commit(); such that zil_commit() would return as soon as that
* specific itx was committed to disk (instead of waiting for _all_
* itxs to be committed).
*
* When a thread calls zil_commit() a special "commit itx" will be
* generated, along with a corresponding "waiter" for this commit itx.
* zil_commit() will wait on this waiter's CV, such that when the waiter
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* is marked done, and signaled, zil_commit() will return.
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
* This commit itx is inserted into the queue of uncommitted itxs. This
* provides an easy mechanism for determining which itxs were in the
* queue prior to zil_commit() having been called, and which itxs were
* added after zil_commit() was called.
*
* The commit it is special; it doesn't have any on-disk representation.
* When a commit itx is "committed" to an lwb, the waiter associated
* with it is linked onto the lwb's list of waiters. Then, when that lwb
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* completes, each waiter on the lwb's list is marked done and signaled
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* -- allowing the thread waiting on the waiter to return from zil_commit().
*
* It's important to point out a few critical factors that allow us
* to make use of the commit itxs, commit waiters, per-lwb lists of
* commit waiters, and zio completion callbacks like we're doing:
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* 1. The list of waiters for each lwb is traversed, and each commit
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* waiter is marked "done" and signaled, in the zio completion
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* callback of the lwb's zio[*].
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
*
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* * Actually, the waiters are signaled in the zio completion
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* callback of the root zio for the DKIOCFLUSHWRITECACHE commands
* that are sent to the vdevs upon completion of the lwb zio.
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* 2. When the itxs are inserted into the ZIL's queue of uncommitted
* itxs, the order in which they are inserted is preserved[*]; as
* itxs are added to the queue, they are added to the tail of
* in-memory linked lists.
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* When committing the itxs to lwbs (to be written to disk), they
* are committed in the same order in which the itxs were added to
* the uncommitted queue's linked list(s); i.e. the linked list of
* itxs to commit is traversed from head to tail, and each itx is
* committed to an lwb in that order.
*
* * To clarify:
*
* - the order of "sync" itxs is preserved w.r.t. other
* "sync" itxs, regardless of the corresponding objects.
* - the order of "async" itxs is preserved w.r.t. other
* "async" itxs corresponding to the same object.
* - the order of "async" itxs is *not* preserved w.r.t. other
* "async" itxs corresponding to different objects.
* - the order of "sync" itxs w.r.t. "async" itxs (or vice
* versa) is *not* preserved, even for itxs that correspond
* to the same object.
*
* For more details, see: zil_itx_assign(), zil_async_to_sync(),
* zil_get_commit_list(), and zil_process_commit_list().
*
* 3. The lwbs represent a linked list of blocks on disk. Thus, any
* lwb cannot be considered committed to stable storage, until its
* "previous" lwb is also committed to stable storage. This fact,
* coupled with the fact described above, means that itxs are
* committed in (roughly) the order in which they were generated.
* This is essential because itxs are dependent on prior itxs.
* Thus, we *must not* deem an itx as being committed to stable
* storage, until *all* prior itxs have also been committed to
* stable storage.
*
* To enforce this ordering of lwb zio's, while still leveraging as
* much of the underlying storage performance as possible, we rely
* on two fundamental concepts:
*
* 1. The creation and issuance of lwb zio's is protected by
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
* the zilog's "zl_issuer_lock", which ensures only a single
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* thread is creating and/or issuing lwb's at a time
* 2. The "previous" lwb is a child of the "current" lwb
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* (leveraging the zio parent-child dependency graph)
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
*
* By relying on this parent-child zio relationship, we can have
* many lwb zio's concurrently issued to the underlying storage,
* but the order in which they complete will be the same order in
* which they were created.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
void
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
zil_commit(zilog_t *zilog, uint64_t foid)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* We should never attempt to call zil_commit on a snapshot for
* a couple of reasons:
*
* 1. A snapshot may never be modified, thus it cannot have any
* in-flight itxs that would have modified the dataset.
*
* 2. By design, when zil_commit() is called, a commit itx will
* be assigned to this zilog; as a result, the zilog will be
* dirtied. We must not dirty the zilog of a snapshot; there's
* checks in the code that enforce this invariant, and will
* cause a panic if it's not upheld.
*/
ASSERT3B(dmu_objset_is_snapshot(zilog->zl_os), ==, B_FALSE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
if (zilog->zl_sync == ZFS_SYNC_DISABLED)
return;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (!spa_writeable(zilog->zl_spa)) {
/*
* If the SPA is not writable, there should never be any
* pending itxs waiting to be committed to disk. If that
* weren't true, we'd skip writing those itxs out, and
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
* would break the semantics of zil_commit(); thus, we're
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* verifying that truth before we return to the caller.
*/
ASSERT(list_is_empty(&zilog->zl_lwb_list));
ASSERT3P(zilog->zl_last_lwb_opened, ==, NULL);
for (int i = 0; i < TXG_SIZE; i++)
ASSERT3P(zilog->zl_itxg[i].itxg_itxs, ==, NULL);
return;
}
/*
* If the ZIL is suspended, we don't want to dirty it by calling
* zil_commit_itx_assign() below, nor can we write out
* lwbs like would be done in zil_commit_write(). Thus, we
* simply rely on txg_wait_synced() to maintain the necessary
* semantics, and avoid calling those functions altogether.
*/
if (zilog->zl_suspend > 0) {
txg_wait_synced(zilog->zl_dmu_pool, 0);
return;
}
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
zil_commit_impl(zilog, foid);
}
void
zil_commit_impl(zilog_t *zilog, uint64_t foid)
{
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
ZIL_STAT_BUMP(zil_commit_count);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* Move the "async" itxs for the specified foid to the "sync"
* queues, such that they will be later committed (or skipped)
* to an lwb when zil_process_commit_list() is called.
*
* Since these "async" itxs must be committed prior to this
* call to zil_commit returning, we must perform this operation
* before we call zil_commit_itx_assign().
*/
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
zil_async_to_sync(zilog, foid);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* We allocate a new "waiter" structure which will initially be
* linked to the commit itx using the itx's "itx_private" field.
* Since the commit itx doesn't represent any on-disk state,
* when it's committed to an lwb, rather than copying the its
* lr_t into the lwb's buffer, the commit itx's "waiter" will be
* added to the lwb's list of waiters. Then, when the lwb is
* committed to stable storage, each waiter in the lwb's list of
* waiters will be marked "done", and signalled.
*
* We must create the waiter and assign the commit itx prior to
* calling zil_commit_writer(), or else our specific commit itx
* is not guaranteed to be committed to an lwb prior to calling
* zil_commit_waiter().
*/
zil_commit_waiter_t *zcw = zil_alloc_commit_waiter();
zil_commit_itx_assign(zilog, zcw);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_commit_writer(zilog, zcw);
zil_commit_waiter(zilog, zcw);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (zcw->zcw_zio_error != 0) {
/*
* If there was an error writing out the ZIL blocks that
* this thread is waiting on, then we fallback to
* relying on spa_sync() to write out the data this
* thread is waiting on. Obviously this has performance
* implications, but the expectation is for this to be
* an exceptional case, and shouldn't occur often.
*/
DTRACE_PROBE2(zil__commit__io__error,
zilog_t *, zilog, zil_commit_waiter_t *, zcw);
txg_wait_synced(zilog->zl_dmu_pool, 0);
}
Condition variable usage, zilog->zl_cv_batch The following incorrect usage of cv_signal and cv_broadcast() was caught by code inspection. The cv_signal and cv_broadcast() functions must be called under the associated mutex to preventing racing with cv_wait(). Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2012-10-15 20:40:07 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_free_commit_waiter(zcw);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Called in syncing context to free committed log blocks and update log header.
*/
void
zil_sync(zilog_t *zilog, dmu_tx_t *tx)
{
zil_header_t *zh = zil_header_in_syncing_context(zilog);
uint64_t txg = dmu_tx_get_txg(tx);
spa_t *spa = zilog->zl_spa;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK];
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
lwb_t *lwb;
Rebase master to b117
2009-07-02 22:44:48 +00:00
/*
* We don't zero out zl_destroy_txg, so make sure we don't try
* to destroy it twice.
*/
if (spa_sync_pass(spa) != 1)
return;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_enter(&zilog->zl_lock);
ASSERT(zilog->zl_stop_sync == 0);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (*replayed_seq != 0) {
ASSERT(zh->zh_replay_seq < *replayed_seq);
zh->zh_replay_seq = *replayed_seq;
*replayed_seq = 0;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (zilog->zl_destroy_txg == txg) {
blkptr_t blk = zh->zh_log;
ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
bzero(zh, sizeof (zil_header_t));
Rebase master to b105
2009-01-15 21:59:39 +00:00
bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (zilog->zl_keep_first) {
/*
* If this block was part of log chain that couldn't
* be claimed because a device was missing during
* zil_claim(), but that device later returns,
* then this block could erroneously appear valid.
* To guard against this, assign a new GUID to the new
* log chain so it doesn't matter what blk points to.
*/
zil_init_log_chain(zilog, &blk);
zh->zh_log = blk;
}
}
Rebase master to b117
2009-07-02 22:44:48 +00:00
while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zh->zh_log = lwb->lwb_blk;
if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
break;
list_remove(&zilog->zl_lwb_list, lwb);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zio_free(spa, txg, &lwb->lwb_blk);
zil_free_lwb(zilog, lwb);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* If we don't have anything left in the lwb list then
* we've had an allocation failure and we need to zero
* out the zil_header blkptr so that we don't end
* up freeing the same block twice.
*/
if (list_head(&zilog->zl_lwb_list) == NULL)
BP_ZERO(&zh->zh_log);
}
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
/*
* Remove fastwrite on any blocks that have been pre-allocated for
* the next commit. This prevents fastwrite counter pollution by
* unused, long-lived LWBs.
*/
for (; lwb != NULL; lwb = list_next(&zilog->zl_lwb_list, lwb)) {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lwb->lwb_fastwrite && !lwb->lwb_write_zio) {
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk);
lwb->lwb_fastwrite = 0;
}
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_exit(&zilog->zl_lock);
}
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/* ARGSUSED */
static int
zil_lwb_cons(void *vbuf, void *unused, int kmflag)
{
lwb_t *lwb = vbuf;
list_create(&lwb->lwb_itxs, sizeof (itx_t), offsetof(itx_t, itx_node));
list_create(&lwb->lwb_waiters, sizeof (zil_commit_waiter_t),
offsetof(zil_commit_waiter_t, zcw_node));
avl_create(&lwb->lwb_vdev_tree, zil_lwb_vdev_compare,
sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node));
mutex_init(&lwb->lwb_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
return (0);
}
/* ARGSUSED */
static void
zil_lwb_dest(void *vbuf, void *unused)
{
lwb_t *lwb = vbuf;
mutex_destroy(&lwb->lwb_vdev_lock);
avl_destroy(&lwb->lwb_vdev_tree);
list_destroy(&lwb->lwb_waiters);
list_destroy(&lwb->lwb_itxs);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
void
zil_init(void)
{
zil_lwb_cache = kmem_cache_create("zil_lwb_cache",
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
sizeof (lwb_t), 0, zil_lwb_cons, zil_lwb_dest, NULL, NULL, NULL, 0);
zil_zcw_cache = kmem_cache_create("zil_zcw_cache",
sizeof (zil_commit_waiter_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
zil_ksp = kstat_create("zfs", 0, "zil", "misc",
cstyle: Resolve C style issues The vast majority of these changes are in Linux specific code. They are the result of not having an automated style checker to validate the code when it was originally written. Others were caused when the common code was slightly adjusted for Linux. This patch contains no functional changes. It only refreshes the code to conform to style guide. Everyone submitting patches for inclusion upstream should now run 'make checkstyle' and resolve any warning prior to opening a pull request. The automated builders have been updated to fail a build if when 'make checkstyle' detects an issue. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1821
2013-11-01 19:26:11 +00:00
KSTAT_TYPE_NAMED, sizeof (zil_stats) / sizeof (kstat_named_t),
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
KSTAT_FLAG_VIRTUAL);
if (zil_ksp != NULL) {
zil_ksp->ks_data = &zil_stats;
kstat_install(zil_ksp);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
void
zil_fini(void)
{
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
kmem_cache_destroy(zil_zcw_cache);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
kmem_cache_destroy(zil_lwb_cache);
Add ZIL statistics. The performance of the ZIL is usually the main bottleneck when dealing with synchronous, write-heavy workloads (e.g. databases). Understanding the behavior of the ZIL is required to diagnose performance issues for these workloads, and to tune ZIL parameters (like zil_slog_limit) accordingly. This commit adds a new kstat page dedicated to the ZIL with some counters which, hopefully, scheds some light into what the ZIL is doing, and how it is doing it. Currently, these statistics are available in /proc/spl/kstat/zfs/zil. A description of the fields can be found in zil.h. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #786
2012-06-15 14:22:14 +00:00
if (zil_ksp != NULL) {
kstat_delete(zil_ksp);
zil_ksp = NULL;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
void
zil_set_sync(zilog_t *zilog, uint64_t sync)
{
zilog->zl_sync = sync;
}
void
zil_set_logbias(zilog_t *zilog, uint64_t logbias)
{
zilog->zl_logbias = logbias;
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog_t *
zil_alloc(objset_t *os, zil_header_t *zh_phys)
{
zilog_t *zilog;
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog->zl_header = zh_phys;
zilog->zl_os = os;
zilog->zl_spa = dmu_objset_spa(os);
zilog->zl_dmu_pool = dmu_objset_pool(os);
zilog->zl_destroy_txg = TXG_INITIAL - 1;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilog->zl_logbias = dmu_objset_logbias(os);
zilog->zl_sync = dmu_objset_syncprop(os);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zilog->zl_dirty_max_txg = 0;
zilog->zl_last_lwb_opened = NULL;
zilog->zl_last_lwb_latency = 0;
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
zilog->zl_max_block_size = zil_maxblocksize;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL);
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
mutex_init(&zilog->zl_issuer_lock, NULL, MUTEX_DEFAULT, NULL);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Undo c89 workarounds to match with upstream With PR 5756 the zfs module now supports c99 and the remaining past c89 workarounds can be undone. Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: George Melikov <mail@gmelikov.ru> Signed-off-by: Don Brady <don.brady@delphix.com> Closes #6816
2017-11-04 20:25:13 +00:00
for (int i = 0; i < TXG_SIZE; i++) {
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL,
MUTEX_DEFAULT, NULL);
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
list_create(&zilog->zl_lwb_list, sizeof (lwb_t),
offsetof(lwb_t, lwb_node));
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
list_create(&zilog->zl_itx_commit_list, sizeof (itx_t),
offsetof(itx_t, itx_node));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL);
return (zilog);
}
void
zil_free(zilog_t *zilog)
{
Fix gcc c90 compliance warnings Fix non-c90 compliant code, for the most part these changes simply deal with where a particular variable is declared. Under c90 it must alway be done at the very start of a block. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2010-08-26 16:52:39 +00:00
int i;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog->zl_stop_sync = 1;
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
ASSERT0(zilog->zl_suspend);
ASSERT0(zilog->zl_suspending);
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
ASSERT(list_is_empty(&zilog->zl_lwb_list));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
list_destroy(&zilog->zl_lwb_list);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
ASSERT(list_is_empty(&zilog->zl_itx_commit_list));
list_destroy(&zilog->zl_itx_commit_list);
Fix gcc c90 compliance warnings Fix non-c90 compliant code, for the most part these changes simply deal with where a particular variable is declared. Under c90 it must alway be done at the very start of a block. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2010-08-26 16:52:39 +00:00
for (i = 0; i < TXG_SIZE; i++) {
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
/*
* It's possible for an itx to be generated that doesn't dirty
* a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean()
* callback to remove the entry. We remove those here.
*
* Also free up the ziltest itxs.
*/
if (zilog->zl_itxg[i].itxg_itxs)
zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs);
mutex_destroy(&zilog->zl_itxg[i].itxg_lock);
}
OpenZFS 8603 - rename zilog's "zl_writer_lock" to "zl_issuer_lock" This is a purely cosmetic change. The zilog's "zl_writer_lock" field is being renamed to "zl_issuer_lock" to try and make the code easier to understand; no other changes are made. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: C Fraire <cfraire@me.com> Approved by: Dan McDonald <danmcd@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8603 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/2daf06546b Closes #6927
2017-09-01 18:04:26 +00:00
mutex_destroy(&zilog->zl_issuer_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_destroy(&zilog->zl_lock);
cv_destroy(&zilog->zl_cv_suspend);
kmem_free(zilog, sizeof (zilog_t));
}
/*
* Open an intent log.
*/
zilog_t *
zil_open(objset_t *os, zil_get_data_t *get_data)
{
zilog_t *zilog = dmu_objset_zil(os);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(zilog->zl_get_data, ==, NULL);
ASSERT3P(zilog->zl_last_lwb_opened, ==, NULL);
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
ASSERT(list_is_empty(&zilog->zl_lwb_list));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog->zl_get_data = get_data;
return (zilog);
}
/*
* Close an intent log.
*/
void
zil_close(zilog_t *zilog)
{
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
lwb_t *lwb;
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
uint64_t txg;
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (!dmu_objset_is_snapshot(zilog->zl_os)) {
zil_commit(zilog, 0);
} else {
ASSERT3P(list_tail(&zilog->zl_lwb_list), ==, NULL);
ASSERT0(zilog->zl_dirty_max_txg);
ASSERT3B(zilog_is_dirty(zilog), ==, B_FALSE);
}
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
mutex_enter(&zilog->zl_lock);
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
lwb = list_tail(&zilog->zl_lwb_list);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
if (lwb == NULL)
txg = zilog->zl_dirty_max_txg;
else
txg = MAX(zilog->zl_dirty_max_txg, lwb->lwb_max_txg);
Update to onnv_147 This is the last official OpenSolaris tag before the public development tree was closed.
2010-08-26 21:24:34 +00:00
mutex_exit(&zilog->zl_lock);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
/*
* We need to use txg_wait_synced() to wait long enough for the
* ZIL to be clean, and to wait for all pending lwbs to be
* written out.
*/
if (txg != 0)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
txg_wait_synced(zilog->zl_dmu_pool, txg);
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
if (zilog_is_dirty(zilog))
Restrict kstats and print real pointers There are several places where we use zfs_dbgmsg and %p to print pointers. In the Linux kernel, these values obfuscated to prevent information leaks which means the pointers aren't very useful for debugging crash dumps. We decided to restrict the permissions of dbgmsg (and some other kstats while we were at it) and print pointers with %px in zfs_dbgmsg as well as spl_dumpstack Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: John Gallagher <john.gallagher@delphix.com> Signed-off-by: sara hartse <sara.hartse@delphix.com> Closes #8467 Closes #8476
2019-04-05 01:57:06 +00:00
zfs_dbgmsg("zil (%px) is dirty, txg %llu", zilog, txg);
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
if (txg < spa_freeze_txg(zilog->zl_spa))
OpenZFS 3821 - Race in rollback, zil close, and zil flush Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Approved by: Richard Lowe <richlowe@richlowe.net> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Ported-by: George Melikov <mail@gmelikov.ru> OpenZFS-issue: https://www.illumos.org/issues/3821 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/43297f9 Closes #5905
2016-11-06 03:43:56 +00:00
VERIFY(!zilog_is_dirty(zilog));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog->zl_get_data = NULL;
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
/*
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
* We should have only one lwb left on the list; remove it now.
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
*/
mutex_enter(&zilog->zl_lock);
lwb = list_head(&zilog->zl_lwb_list);
if (lwb != NULL) {
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
ASSERT3P(lwb, ==, list_tail(&zilog->zl_lwb_list));
ASSERT3S(lwb->lwb_state, !=, LWB_STATE_ISSUED);
Add FASTWRITE algorithm for synchronous writes. Currently, ZIL blocks are spread over vdevs using hint block pointers managed by the ZIL commit code and passed to metaslab_alloc(). Spreading log blocks accross vdevs is important for performance: indeed, using mutliple disks in parallel decreases the ZIL commit latency, which is the main performance metric for synchronous writes. However, the current implementation suffers from the following issues: 1) It would be best if the ZIL module was not aware of such low-level details. They should be handled by the ZIO and metaslab modules; 2) Because the hint block pointer is managed per log, simultaneous commits from multiple logs might use the same vdevs at the same time, which is inefficient; 3) Because dmu_write() does not honor the block pointer hint, indirect writes are not spread. The naive solution of rotating the metaslab rotor each time a block is allocated for the ZIL or dmu_sync() doesn't work in practice because the first ZIL block to be written is actually allocated during the previous commit. Consequently, when metaslab_alloc() decides the vdev for this block, it will do so while a bunch of other allocations are happening at the same time (from dmu_sync() and other ZILs). This means the vdev for this block is chosen more or less at random. When the next commit happens, there is a high chance (especially when the number of blocks per commit is slightly less than the number of the disks) that one disk will have to write two blocks (with a potential seek) while other disks are sitting idle, which defeats spreading and increases the commit latency. This commit introduces a new concept in the metaslab allocator: fastwrites. Basically, each top-level vdev maintains a counter indicating the number of synchronous writes (from dmu_sync() and the ZIL) which have been allocated but not yet completed. When the metaslab is called with the FASTWRITE flag, it will choose the vdev with the least amount of pending synchronous writes. If there are multiple vdevs with the same value, the first matching vdev (starting from the rotor) is used. Once metaslab_alloc() has decided which vdev the block is allocated to, it updates the fastwrite counter for this vdev. The rationale goes like this: when an allocation is done with FASTWRITE, it "reserves" the vdev until the data is written. Until then, all future allocations will naturally avoid this vdev, even after a full rotation of the rotor. As a result, pending synchronous writes at a given point in time will be nicely spread over all vdevs. This contrasts with the previous algorithm, which is based on the implicit assumption that blocks are written instantaneously after they're allocated. metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to manually increase or decrease fastwrite counters, respectively. They should be used with caution, as there is no per-BP tracking of fastwrite information, so leaks and "double-unmarks" are possible. There is, however, an assert in the vdev teardown code which will fire if the fastwrite counters are not zero when the pool is exported or the vdev removed. Note that as stated above, marking is also done implictly by metaslab_alloc(). ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to the metaslab when allocating (assuming ZIO does the allocation, which is only true in the case of dmu_sync). This flag will also trigger an unmark when zio_done() fires. A side-effect of the new algorithm is that when a ZIL stops being used, its last block can stay in the pending state (allocated but not yet written) for a long time, polluting the fastwrite counters. To avoid that, I've implemented a somewhat crude but working solution which unmarks these pending blocks in zil_sync(), thus guaranteeing that linguering fastwrites will get pruned at each sync event. The best performance improvements are observed with pools using a large number of top-level vdevs and heavy synchronous write workflows (especially indirect writes and concurrent writes from multiple ZILs). Real-life testing shows a 200% to 300% performance increase with indirect writes and various commit sizes. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1013
2012-06-27 13:20:20 +00:00
if (lwb->lwb_fastwrite)
metaslab_fastwrite_unmark(zilog->zl_spa, &lwb->lwb_blk);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
list_remove(&zilog->zl_lwb_list, lwb);
zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
zil_free_lwb(zilog, lwb);
Illumos #883: ZIL reuse during remount corruption Moving the zil_free() cleanup to zil_close() prevents this problem from occurring in the first place. There is a very good description of the issue and fix in Illumus #883. Reviewed by: Matt Ahrens <Matt.Ahrens@delphix.com> Reviewed by: Adam Leventhal <Adam.Leventhal@delphix.com> Reviewed by: Albert Lee <trisk@nexenta.com> Reviewed by: Gordon Ross <gwr@nexenta.com> Reviewed by: Garrett D'Amore <garrett@nexenta.com> Reivewed by: Dan McDonald <danmcd@nexenta.com> Approved by: Gordon Ross <gwr@nexenta.com> References to Illumos issue and patch: - https://www.illumos.org/issues/883 - https://github.com/illumos/illumos-gate/commit/c9ba2a43cb Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #340
2011-07-26 19:41:53 +00:00
}
mutex_exit(&zilog->zl_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
static char *suspend_tag = "zil suspending";
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Suspend an intent log. While in suspended mode, we still honor
* synchronous semantics, but we rely on txg_wait_synced() to do it.
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
* On old version pools, we suspend the log briefly when taking a
* snapshot so that it will have an empty intent log.
*
* Long holds are not really intended to be used the way we do here --
* held for such a short time. A concurrent caller of dsl_dataset_long_held()
* could fail. Therefore we take pains to only put a long hold if it is
* actually necessary. Fortunately, it will only be necessary if the
* objset is currently mounted (or the ZVOL equivalent). In that case it
* will already have a long hold, so we are not really making things any worse.
*
* Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or
* zvol_state_t), and use their mechanism to prevent their hold from being
* dropped (e.g. VFS_HOLD()). However, that would be even more pain for
* very little gain.
*
* if cookiep == NULL, this does both the suspend & resume.
* Otherwise, it returns with the dataset "long held", and the cookie
* should be passed into zil_resume().
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
int
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
zil_suspend(const char *osname, void **cookiep)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
objset_t *os;
zilog_t *zilog;
const zil_header_t *zh;
int error;
error = dmu_objset_hold(osname, suspend_tag, &os);
if (error != 0)
return (error);
zilog = dmu_objset_zil(os);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_enter(&zilog->zl_lock);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
zh = zilog->zl_header;
Rebase master to b117
2009-07-02 22:44:48 +00:00
if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_exit(&zilog->zl_lock);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
dmu_objset_rele(os, suspend_tag);
Illumos #3598 3598 want to dtrace when errors are generated in zfs Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3598 illumos/illumos-gate@be6fd75a69ae679453d9cda5bff3326111e6d1ca Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. include/sys/zfs_context.h has been modified to render some new macros inert until dtrace is available on Linux. 2. Linux-specific changes have been adapted to use SET_ERROR(). 3. I'm NOT happy about this change. It does nothing but ugly up the code under Linux. Unfortunately we need to take it to avoid more merge conflicts in the future. -Brian
2013-03-08 18:41:28 +00:00
return (SET_ERROR(EBUSY));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
/*
* Don't put a long hold in the cases where we can avoid it. This
* is when there is no cookie so we are doing a suspend & resume
* (i.e. called from zil_vdev_offline()), and there's nothing to do
* for the suspend because it's already suspended, or there's no ZIL.
*/
if (cookiep == NULL && !zilog->zl_suspending &&
(zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) {
mutex_exit(&zilog->zl_lock);
dmu_objset_rele(os, suspend_tag);
return (0);
}
dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag);
dsl_pool_rele(dmu_objset_pool(os), suspend_tag);
zilog->zl_suspend++;
if (zilog->zl_suspend > 1) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
* Someone else is already suspending it.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
* Just wait for them to finish.
*/
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
while (zilog->zl_suspending)
cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock);
mutex_exit(&zilog->zl_lock);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (cookiep == NULL)
zil_resume(os);
else
*cookiep = os;
return (0);
}
/*
* If there is no pointer to an on-disk block, this ZIL must not
* be active (e.g. filesystem not mounted), so there's nothing
* to clean up.
*/
if (BP_IS_HOLE(&zh->zh_log)) {
ASSERT(cookiep != NULL); /* fast path already handled */
*cookiep = os;
mutex_exit(&zilog->zl_lock);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (0);
}
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
Encryption patch follow-up * PBKDF2 implementation changed to OpenSSL implementation. * HKDF implementation moved to its own file and tests added to ensure correctness. * Removed libzfs's now unnecessary dependency on libzpool and libicp. * Ztest can now create and test encrypted datasets. This is currently disabled until issue #6526 is resolved, but otherwise functions as advertised. * Several small bug fixes discovered after enabling ztest to run on encrypted datasets. * Fixed coverity defects added by the encryption patch. * Updated man pages for encrypted send / receive behavior. * Fixed a bug where encrypted datasets could receive DRR_WRITE_EMBEDDED records. * Minor code cleanups / consolidation. Signed-off-by: Tom Caputi <tcaputi@datto.com>
2017-09-12 20:15:11 +00:00
/*
* The ZIL has work to do. Ensure that the associated encryption
* key will remain mapped while we are committing the log by
* grabbing a reference to it. If the key isn't loaded we have no
* choice but to return an error until the wrapping key is loaded.
*/
Refcounted DSL Crypto Key Mappings Since native ZFS encryption was merged, we have been fighting against a series of bugs that come down to the same problem: Key mappings (which must be present during all I/O operations) are created and destroyed based on dataset ownership, but I/Os can have traditionally been allowed to "leak" into the next txg after the dataset is disowned. In the past we have attempted to solve this problem by trying to ensure that datasets are disowned ater all I/O is finished by calling txg_wait_synced(), but we have repeatedly found edge cases that need to be squashed and code paths that might incur a high number of txg syncs. This patch attempts to resolve this issue differently, by adding a reference to the key mapping for each txg it is dirtied in. By doing so, we can remove many of the unnecessary calls to txg_wait_synced() we have added in the past and ensure we don't need to deal with this problem in the future. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #7949
2018-10-03 16:47:11 +00:00
if (os->os_encrypted &&
dsl_dataset_create_key_mapping(dmu_objset_ds(os)) != 0) {
Encryption patch follow-up * PBKDF2 implementation changed to OpenSSL implementation. * HKDF implementation moved to its own file and tests added to ensure correctness. * Removed libzfs's now unnecessary dependency on libzpool and libicp. * Ztest can now create and test encrypted datasets. This is currently disabled until issue #6526 is resolved, but otherwise functions as advertised. * Several small bug fixes discovered after enabling ztest to run on encrypted datasets. * Fixed coverity defects added by the encryption patch. * Updated man pages for encrypted send / receive behavior. * Fixed a bug where encrypted datasets could receive DRR_WRITE_EMBEDDED records. * Minor code cleanups / consolidation. Signed-off-by: Tom Caputi <tcaputi@datto.com>
2017-09-12 20:15:11 +00:00
zilog->zl_suspend--;
mutex_exit(&zilog->zl_lock);
dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag);
dsl_dataset_rele(dmu_objset_ds(os), suspend_tag);
Wrong error message when removing log device In the case where the pool is loaded without the crypto keys necessary to playback the intent log, and log device removal is attempted, a generic busy message is received. Change the message to inform the user that the datasets must be mounted. Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Tom Caputi <tcaputi@datto.com> Signed-off-by: Paul Zuchowski <pzuchowski@datto.com> Closes #7518
2018-06-08 01:07:29 +00:00
return (SET_ERROR(EACCES));
Encryption patch follow-up * PBKDF2 implementation changed to OpenSSL implementation. * HKDF implementation moved to its own file and tests added to ensure correctness. * Removed libzfs's now unnecessary dependency on libzpool and libicp. * Ztest can now create and test encrypted datasets. This is currently disabled until issue #6526 is resolved, but otherwise functions as advertised. * Several small bug fixes discovered after enabling ztest to run on encrypted datasets. * Fixed coverity defects added by the encryption patch. * Updated man pages for encrypted send / receive behavior. * Fixed a bug where encrypted datasets could receive DRR_WRITE_EMBEDDED records. * Minor code cleanups / consolidation. Signed-off-by: Tom Caputi <tcaputi@datto.com>
2017-09-12 20:15:11 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zilog->zl_suspending = B_TRUE;
mutex_exit(&zilog->zl_lock);
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
/*
* We need to use zil_commit_impl to ensure we wait for all
* LWB_STATE_OPENED and LWB_STATE_ISSUED lwbs to be committed
* to disk before proceeding. If we used zil_commit instead, it
* would just call txg_wait_synced(), because zl_suspend is set.
* txg_wait_synced() doesn't wait for these lwb's to be
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
* LWB_STATE_FLUSH_DONE before returning.
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
*/
zil_commit_impl(zilog, 0);
/*
OpenZFS 9962 - zil_commit should omit cache thrash As a result of the changes made in 8585, it's possible for an excessive amount of vdev flush commands to be issued under some workloads. Specifically, when the workload consists of mostly async write activity, interspersed with some sync write and/or fsync activity, we can end up issuing more flush commands to the underlying storage than is actually necessary. As a result of these flush commands, the write latency and overall throughput of the pool can be poorly impacted (latency increases, throughput decreases). Currently, any time an lwb completes, the vdev(s) written to as a result of that lwb will be issued a flush command. The intenion is so the data written to that vdev is on stable storage, prior to communicating to any waiting threads that their data is safe on disk. The problem with this scheme, is that sometimes an lwb will not have any threads waiting for it to complete. This can occur when there's async activity that gets "converted" to sync requests, as a result of calling the zil_async_to_sync() function via zil_commit_impl(). When this occurs, the current code may issue many lwbs that don't have waiters associated with them, resulting in many flush commands, potentially to the same vdev(s). For example, given a pool with a single vdev, and a single fsync() call that results in 10 lwbs being written out (e.g. due to other async writes), that will result in 10 flush commands to that single vdev (a flush issued after each lwb write completes). Ideally, we'd only issue a single flush command to that vdev, after all 10 lwb writes completed. Further, and most important as it pertains to this change, since the flush commands are often very impactful to the performance of the pool's underlying storage, unnecessarily issuing these flush commands can poorly impact the performance of the lwb writes themselves. Thus, we need to avoid issuing flush commands when possible, in order to acheive the best possible performance out of the pool's underlying storage. This change attempts to address this problem by changing the ZIL's logic to only issue a vdev flush command when it detects an lwb that has a thread waiting for it to complete. When an lwb does not have threads waiting for it, the responsibility of issuing the flush command to the vdevs involved with that lwb's write is passed on to the "next" lwb. It's only once a write for an lwb with waiters completes, do we issue the vdev flush command(s). As a result, now when we issue the flush(s), we will issue them to the vdevs involved with that specific lwb's write, but potentially also to vdevs involved with "previous" lwb writes (i.e. if the previous lwbs did not have waiters associated with them). Thus, in our prior example with 10 lwbs, it's only once the last lwb completes (which will be the lwb containing the waiter for the thread that called fsync) will we issue the vdev flush command; all of the other lwbs will find they have no waiters, so they'll pass the responsibility of the flush to the "next" lwb (until reaching the last lwb that has the waiter). Porting Notes: * Reconciled conflicts with the fastwrite feature. Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Matt Ahrens <matt@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Patrick Mooney <patrick.mooney@joyent.com> Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com> Approved by: Joshua M. Clulow <josh@sysmgr.org> Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/9962 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6 Closes #8188
2018-10-23 21:14:27 +00:00
* Now that we've ensured all lwb's are LWB_STATE_FLUSH_DONE, we
* use txg_wait_synced() to ensure the data from the zilog has
* migrated to the main pool before calling zil_destroy().
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
*/
txg_wait_synced(zilog->zl_dmu_pool, 0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_destroy(zilog, B_FALSE);
mutex_enter(&zilog->zl_lock);
zilog->zl_suspending = B_FALSE;
cv_broadcast(&zilog->zl_cv_suspend);
mutex_exit(&zilog->zl_lock);
Refcounted DSL Crypto Key Mappings Since native ZFS encryption was merged, we have been fighting against a series of bugs that come down to the same problem: Key mappings (which must be present during all I/O operations) are created and destroyed based on dataset ownership, but I/Os can have traditionally been allowed to "leak" into the next txg after the dataset is disowned. In the past we have attempted to solve this problem by trying to ensure that datasets are disowned ater all I/O is finished by calling txg_wait_synced(), but we have repeatedly found edge cases that need to be squashed and code paths that might incur a high number of txg syncs. This patch attempts to resolve this issue differently, by adding a reference to the key mapping for each txg it is dirtied in. By doing so, we can remove many of the unnecessary calls to txg_wait_synced() we have added in the past and ensure we don't need to deal with this problem in the future. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #7949
2018-10-03 16:47:11 +00:00
if (os->os_encrypted)
dsl_dataset_remove_key_mapping(dmu_objset_ds(os));
Encryption patch follow-up * PBKDF2 implementation changed to OpenSSL implementation. * HKDF implementation moved to its own file and tests added to ensure correctness. * Removed libzfs's now unnecessary dependency on libzpool and libicp. * Ztest can now create and test encrypted datasets. This is currently disabled until issue #6526 is resolved, but otherwise functions as advertised. * Several small bug fixes discovered after enabling ztest to run on encrypted datasets. * Fixed coverity defects added by the encryption patch. * Updated man pages for encrypted send / receive behavior. * Fixed a bug where encrypted datasets could receive DRR_WRITE_EMBEDDED records. * Minor code cleanups / consolidation. Signed-off-by: Tom Caputi <tcaputi@datto.com>
2017-09-12 20:15:11 +00:00
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (cookiep == NULL)
zil_resume(os);
else
*cookiep = os;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
return (0);
}
void
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
zil_resume(void *cookie)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
objset_t *os = cookie;
zilog_t *zilog = dmu_objset_zil(os);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
mutex_enter(&zilog->zl_lock);
ASSERT(zilog->zl_suspend != 0);
zilog->zl_suspend--;
mutex_exit(&zilog->zl_lock);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag);
dsl_dataset_rele(dmu_objset_ds(os), suspend_tag);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
typedef struct zil_replay_arg {
OpenZFS 8081 - Compiler warnings in zdb Fix compiler warnings in zdb. With these changes, FreeBSD can compile zdb with all compiler warnings enabled save -Wunused-parameter. usr/src/cmd/zdb/zdb.c usr/src/cmd/zdb/zdb_il.c usr/src/uts/common/fs/zfs/sys/sa.h usr/src/uts/common/fs/zfs/sys/spa.h Fix numerous warnings, including: * const-correctness * shadowing global definitions * signed vs unsigned comparisons * missing prototypes, or missing static declarations * unused variables and functions * Unreadable array initializations * Missing struct initializers usr/src/cmd/zdb/zdb.h Add a header file to declare common symbols usr/src/lib/libzpool/common/sys/zfs_context.h usr/src/uts/common/fs/zfs/arc.c usr/src/uts/common/fs/zfs/dbuf.c usr/src/uts/common/fs/zfs/spa.c usr/src/uts/common/fs/zfs/txg.c Add a function prototype for zk_thread_create, and ensure that every callback supplied to this function actually matches the prototype. usr/src/cmd/ztest/ztest.c usr/src/uts/common/fs/zfs/sys/zil.h usr/src/uts/common/fs/zfs/zfs_replay.c usr/src/uts/common/fs/zfs/zvol.c Add a function prototype for zil_replay_func_t, and ensure that every function of this type actually matches the prototype. usr/src/uts/common/fs/zfs/sys/refcount.h Change FTAG so it discards any constness of __func__, necessary since existing APIs expect it passed as void *. Porting Notes: - Many of these fixes have already been applied to Linux. For consistency the OpenZFS version of a change was applied if the warning was addressed in an equivalent but different fashion. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Authored by: Alan Somers <asomers@gmail.com> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8081 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/843abe1b8a Closes #6787
2017-10-27 19:46:35 +00:00
zil_replay_func_t **zr_replay;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
void *zr_arg;
boolean_t zr_byteswap;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
char *zr_lr;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
} zil_replay_arg_t;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static int
zil_replay_error(zilog_t *zilog, lr_t *lr, int error)
{
OpenZFS 6314 - buffer overflow in dsl_dataset_name Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Igor Kozhukhov <ikozhukhov@gmail.com> Approved by: Dan McDonald <danmcd@omniti.com> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/6314 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/d6160ee
2016-06-15 21:28:36 +00:00
char name[ZFS_MAX_DATASET_NAME_LEN];
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilog->zl_replaying_seq--; /* didn't actually replay this one */
dmu_objset_name(zilog->zl_os, name);
cmn_err(CE_WARN, "ZFS replay transaction error %d, "
"dataset %s, seq 0x%llx, txtype %llu %s\n", error, name,
(u_longlong_t)lr->lrc_seq,
(u_longlong_t)(lr->lrc_txtype & ~TX_CI),
(lr->lrc_txtype & TX_CI) ? "CI" : "");
return (error);
}
static int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
{
zil_replay_arg_t *zr = zra;
const zil_header_t *zh = zilog->zl_header;
uint64_t reclen = lr->lrc_reclen;
uint64_t txtype = lr->lrc_txtype;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
int error = 0;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilog->zl_replaying_seq = lr->lrc_seq;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (0);
if (lr->lrc_txg < claim_txg) /* already committed */
return (0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/* Strip case-insensitive bit, still present in log record */
txtype &= ~TX_CI;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (txtype == 0 || txtype >= TX_MAX_TYPE)
return (zil_replay_error(zilog, lr, EINVAL));
/*
* If this record type can be logged out of order, the object
* (lr_foid) may no longer exist. That's legitimate, not an error.
*/
if (TX_OOO(txtype)) {
error = dmu_object_info(zilog->zl_os,
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 01:25:34 +00:00
LR_FOID_GET_OBJ(((lr_ooo_t *)lr)->lr_foid), NULL);
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (error == ENOENT || error == EEXIST)
return (0);
Rebase master to b105
2009-01-15 21:59:39 +00:00
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Make a copy of the data so we can revise and extend it.
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
bcopy(lr, zr->zr_lr, reclen);
/*
* If this is a TX_WRITE with a blkptr, suck in the data.
*/
if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
error = zil_read_log_data(zilog, (lr_write_t *)lr,
zr->zr_lr + reclen);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (zil_replay_error(zilog, lr, error));
}
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* The log block containing this lr may have been byteswapped
* so that we can easily examine common fields like lrc_txtype.
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* However, the log is a mix of different record types, and only the
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
* replay vectors know how to byteswap their records. Therefore, if
* the lr was byteswapped, undo it before invoking the replay vector.
*/
if (zr->zr_byteswap)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
byteswap_uint64_array(zr->zr_lr, reclen);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* We must now do two things atomically: replay this log record,
Rebase master to b105
2009-01-15 21:59:39 +00:00
* and update the log header sequence number to reflect the fact that
* we did so. At the end of each replay function the sequence number
* is updated if we are in replay mode.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* The DMU's dnode layer doesn't see removes until the txg
* commits, so a subsequent claim can spuriously fail with
Rebase master to b105
2009-01-15 21:59:39 +00:00
* EEXIST. So if we receive any error we try syncing out
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
* any removes then retry the transaction. Note that we
* specify B_FALSE for byteswap now, so we don't do it twice.
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
*/
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE);
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0)
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (zil_replay_error(zilog, lr, error));
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/* ARGSUSED */
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
static int
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
{
zilog->zl_replay_blks++;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (0);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
/*
* If this dataset has a non-empty intent log, replay it and destroy it.
*/
void
OpenZFS 8081 - Compiler warnings in zdb Fix compiler warnings in zdb. With these changes, FreeBSD can compile zdb with all compiler warnings enabled save -Wunused-parameter. usr/src/cmd/zdb/zdb.c usr/src/cmd/zdb/zdb_il.c usr/src/uts/common/fs/zfs/sys/sa.h usr/src/uts/common/fs/zfs/sys/spa.h Fix numerous warnings, including: * const-correctness * shadowing global definitions * signed vs unsigned comparisons * missing prototypes, or missing static declarations * unused variables and functions * Unreadable array initializations * Missing struct initializers usr/src/cmd/zdb/zdb.h Add a header file to declare common symbols usr/src/lib/libzpool/common/sys/zfs_context.h usr/src/uts/common/fs/zfs/arc.c usr/src/uts/common/fs/zfs/dbuf.c usr/src/uts/common/fs/zfs/spa.c usr/src/uts/common/fs/zfs/txg.c Add a function prototype for zk_thread_create, and ensure that every callback supplied to this function actually matches the prototype. usr/src/cmd/ztest/ztest.c usr/src/uts/common/fs/zfs/sys/zil.h usr/src/uts/common/fs/zfs/zfs_replay.c usr/src/uts/common/fs/zfs/zvol.c Add a function prototype for zil_replay_func_t, and ensure that every function of this type actually matches the prototype. usr/src/uts/common/fs/zfs/sys/refcount.h Change FTAG so it discards any constness of __func__, necessary since existing APIs expect it passed as void *. Porting Notes: - Many of these fixes have already been applied to Linux. For consistency the OpenZFS version of a change was applied if the warning was addressed in an equivalent but different fashion. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Authored by: Alan Somers <asomers@gmail.com> Approved by: Richard Lowe <richlowe@richlowe.net> Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/8081 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/843abe1b8a Closes #6787
2017-10-27 19:46:35 +00:00
zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE])
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
zilog_t *zilog = dmu_objset_zil(os);
const zil_header_t *zh = zilog->zl_header;
zil_replay_arg_t zr;
Rebase master to b117
2009-07-02 22:44:48 +00:00
if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) {
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_destroy(zilog, B_TRUE);
return;
}
zr.zr_replay = replay_func;
zr.zr_arg = arg;
zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log);
Change KM_PUSHPAGE -> KM_SLEEP By marking DMU transaction processing contexts with PF_FSTRANS we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings us back in line with upstream. In some cases this means simply swapping the flags back. For others fnvlist_alloc() was replaced by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to fnvlist_alloc() which assumes KM_SLEEP. The one place KM_PUSHPAGE is kept is when allocating ARC buffers which allows us to dip in to reserved memory. This is again the same as upstream. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2014-11-21 00:09:39 +00:00
zr.zr_lr = vmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
/*
* Wait for in-progress removes to sync before starting replay.
*/
txg_wait_synced(zilog->zl_dmu_pool, 0);
Rebase master to b105
2009-01-15 21:59:39 +00:00
zilog->zl_replay = B_TRUE;
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
zilog->zl_replay_time = ddi_get_lbolt();
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
ASSERT(zilog->zl_replay_blks == 0);
(void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr,
Native Encryption for ZFS on Linux This change incorporates three major pieces: The first change is a keystore that manages wrapping and encryption keys for encrypted datasets. These commands mostly involve manipulating the new DSL Crypto Key ZAP Objects that live in the MOS. Each encrypted dataset has its own DSL Crypto Key that is protected with a user's key. This level of indirection allows users to change their keys without re-encrypting their entire datasets. The change implements the new subcommands "zfs load-key", "zfs unload-key" and "zfs change-key" which allow the user to manage their encryption keys and settings. In addition, several new flags and properties have been added to allow dataset creation and to make mounting and unmounting more convenient. The second piece of this patch provides the ability to encrypt, decyrpt, and authenticate protected datasets. Each object set maintains a Merkel tree of Message Authentication Codes that protect the lower layers, similarly to how checksums are maintained. This part impacts the zio layer, which handles the actual encryption and generation of MACs, as well as the ARC and DMU, which need to be able to handle encrypted buffers and protected data. The last addition is the ability to do raw, encrypted sends and receives. The idea here is to send raw encrypted and compressed data and receive it exactly as is on a backup system. This means that the dataset on the receiving system is protected using the same user key that is in use on the sending side. By doing so, datasets can be efficiently backed up to an untrusted system without fear of data being compromised. Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Jorgen Lundman <lundman@lundman.net> Signed-off-by: Tom Caputi <tcaputi@datto.com> Closes #494 Closes #5769
2017-08-14 17:36:48 +00:00
zh->zh_claim_txg, B_TRUE);
Add linux kernel memory support Required kmem/vmem changes Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2010-08-26 18:46:09 +00:00
vmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
zil_destroy(zilog, B_FALSE);
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
Rebase master to b105
2009-01-15 21:59:39 +00:00
zilog->zl_replay = B_FALSE;
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
boolean_t
zil_replaying(zilog_t *zilog, dmu_tx_t *tx)
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
{
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (zilog->zl_sync == ZFS_SYNC_DISABLED)
return (B_TRUE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
if (zilog->zl_replay) {
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] =
zilog->zl_replaying_seq;
return (B_TRUE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Update core ZFS code from build 121 to build 141.
2010-05-28 20:45:14 +00:00
return (B_FALSE);
Initial Linux ZFS GIT Repo
2008-11-20 20:01:55 +00:00
}
Rebase master to b117
2009-07-02 22:44:48 +00:00
/* ARGSUSED */
int
OpenZFS 7614, 9064 - zfs device evacuation/removal OpenZFS 7614 - zfs device evacuation/removal OpenZFS 9064 - remove_mirror should wait for device removal to complete This project allows top-level vdevs to be removed from the storage pool with "zpool remove", reducing the total amount of storage in the pool. This operation copies all allocated regions of the device to be removed onto other devices, recording the mapping from old to new location. After the removal is complete, read and free operations to the removed (now "indirect") vdev must be remapped and performed at the new location on disk. The indirect mapping table is kept in memory whenever the pool is loaded, so there is minimal performance overhead when doing operations on the indirect vdev. The size of the in-memory mapping table will be reduced when its entries become "obsolete" because they are no longer used by any block pointers in the pool. An entry becomes obsolete when all the blocks that use it are freed. An entry can also become obsolete when all the snapshots that reference it are deleted, and the block pointers that reference it have been "remapped" in all filesystems/zvols (and clones). Whenever an indirect block is written, all the block pointers in it will be "remapped" to their new (concrete) locations if possible. This process can be accelerated by using the "zfs remap" command to proactively rewrite all indirect blocks that reference indirect (removed) vdevs. Note that when a device is removed, we do not verify the checksum of the data that is copied. This makes the process much faster, but if it were used on redundant vdevs (i.e. mirror or raidz vdevs), it would be possible to copy the wrong data, when we have the correct data on e.g. the other side of the mirror. At the moment, only mirrors and simple top-level vdevs can be removed and no removal is allowed if any of the top-level vdevs are raidz. Porting Notes: * Avoid zero-sized kmem_alloc() in vdev_compact_children(). The device evacuation code adds a dependency that vdev_compact_children() be able to properly empty the vdev_child array by setting it to NULL and zeroing vdev_children. Under Linux, kmem_alloc() and related functions return a sentinel pointer rather than NULL for zero-sized allocations. * Remove comment regarding "mpt" driver where zfs_remove_max_segment is initialized to SPA_MAXBLOCKSIZE. Change zfs_condense_indirect_commit_entry_delay_ticks to zfs_condense_indirect_commit_entry_delay_ms for consistency with most other tunables in which delays are specified in ms. * ZTS changes: Use set_tunable rather than mdb Use zpool sync as appropriate Use sync_pool instead of sync Kill jobs during test_removal_with_operation to allow unmount/export Don't add non-disk names such as "mirror" or "raidz" to $DISKS Use $TEST_BASE_DIR instead of /tmp Increase HZ from 100 to 1000 which is more common on Linux removal_multiple_indirection.ksh Reduce iterations in order to not time out on the code coverage builders. removal_resume_export: Functionally, the test case is correct but there exists a race where the kernel thread hasn't been fully started yet and is not visible. Wait for up to 1 second for the removal thread to be started before giving up on it. Also, increase the amount of data copied in order that the removal not finish before the export has a chance to fail. * MMP compatibility, the concept of concrete versus non-concrete devices has slightly changed the semantics of vdev_writeable(). Update mmp_random_leaf_impl() accordingly. * Updated dbuf_remap() to handle the org.zfsonlinux:large_dnode pool feature which is not supported by OpenZFS. * Added support for new vdev removal tracepoints. * Test cases removal_with_zdb and removal_condense_export have been intentionally disabled. When run manually they pass as intended, but when running in the automated test environment they produce unreliable results on the latest Fedora release. They may work better once the upstream pool import refectoring is merged into ZoL at which point they will be re-enabled. Authored by: Matthew Ahrens <mahrens@delphix.com> Reviewed-by: Alex Reece <alex@delphix.com> Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: John Kennedy <john.kennedy@delphix.com> Reviewed-by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Richard Laager <rlaager@wiktel.com> Reviewed by: Tim Chase <tim@chase2k.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Garrett D'Amore <garrett@damore.org> Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Tim Chase <tim@chase2k.com> OpenZFS-issue: https://www.illumos.org/issues/7614 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/f539f1eb Closes #6900
2016-09-22 16:30:13 +00:00
zil_reset(const char *osname, void *arg)
Rebase master to b117
2009-07-02 22:44:48 +00:00
{
int error;
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
error = zil_suspend(osname, NULL);
Wrong error message when removing log device In the case where the pool is loaded without the crypto keys necessary to playback the intent log, and log device removal is attempted, a generic busy message is received. Change the message to inform the user that the datasets must be mounted. Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Tom Caputi <tcaputi@datto.com> Signed-off-by: Paul Zuchowski <pzuchowski@datto.com> Closes #7518
2018-06-08 01:07:29 +00:00
/* EACCES means crypto key not loaded */
if ((error == EACCES) || (error == EBUSY))
return (SET_ERROR(error));
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
if (error != 0)
Illumos #3598 3598 want to dtrace when errors are generated in zfs Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3598 illumos/illumos-gate@be6fd75a69ae679453d9cda5bff3326111e6d1ca Ported-by: Richard Yao <ryao@gentoo.org> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Issue #1775 Porting notes: 1. include/sys/zfs_context.h has been modified to render some new macros inert until dtrace is available on Linux. 2. Linux-specific changes have been adapted to use SET_ERROR(). 3. I'm NOT happy about this change. It does nothing but ugly up the code under Linux. Unfortunately we need to take it to avoid more merge conflicts in the future. -Brian
2013-03-08 18:41:28 +00:00
return (SET_ERROR(EEXIST));
Illumos #3464 3464 zfs synctask code needs restructuring Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/3464 illumos/illumos-gate@3b2aab18808792cbd248a12f1edf139b89833c13 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1495
2013-09-04 12:00:57 +00:00
return (0);
Rebase master to b117
2009-07-02 22:44:48 +00:00
}
Add missing ZFS tunables This commit adds module options for all existing zfs tunables. Ideally the average user should never need to modify any of these values. However, in practice sometimes you do need to tweak these values for one reason or another. In those cases it's nice not to have to resort to rebuilding from source. All tunables are visable to modinfo and the list is as follows: $ modinfo module/zfs/zfs.ko filename: module/zfs/zfs.ko license: CDDL author: Sun Microsystems/Oracle, Lawrence Livermore National Laboratory description: ZFS srcversion: 8EAB1D71DACE05B5AA61567 depends: spl,znvpair,zcommon,zunicode,zavl vermagic: 2.6.32-131.0.5.el6.x86_64 SMP mod_unload modversions parm: zvol_major:Major number for zvol device (uint) parm: zvol_threads:Number of threads for zvol device (uint) parm: zio_injection_enabled:Enable fault injection (int) parm: zio_bulk_flags:Additional flags to pass to bulk buffers (int) parm: zio_delay_max:Max zio millisec delay before posting event (int) parm: zio_requeue_io_start_cut_in_line:Prioritize requeued I/O (bool) parm: zil_replay_disable:Disable intent logging replay (int) parm: zfs_nocacheflush:Disable cache flushes (bool) parm: zfs_read_chunk_size:Bytes to read per chunk (long) parm: zfs_vdev_max_pending:Max pending per-vdev I/Os (int) parm: zfs_vdev_min_pending:Min pending per-vdev I/Os (int) parm: zfs_vdev_aggregation_limit:Max vdev I/O aggregation size (int) parm: zfs_vdev_time_shift:Deadline time shift for vdev I/O (int) parm: zfs_vdev_ramp_rate:Exponential I/O issue ramp-up rate (int) parm: zfs_vdev_read_gap_limit:Aggregate read I/O over gap (int) parm: zfs_vdev_write_gap_limit:Aggregate write I/O over gap (int) parm: zfs_vdev_scheduler:I/O scheduler (charp) parm: zfs_vdev_cache_max:Inflate reads small than max (int) parm: zfs_vdev_cache_size:Total size of the per-disk cache (int) parm: zfs_vdev_cache_bshift:Shift size to inflate reads too (int) parm: zfs_scrub_limit:Max scrub/resilver I/O per leaf vdev (int) parm: zfs_recover:Set to attempt to recover from fatal errors (int) parm: spa_config_path:SPA config file (/etc/zfs/zpool.cache) (charp) parm: zfs_zevent_len_max:Max event queue length (int) parm: zfs_zevent_cols:Max event column width (int) parm: zfs_zevent_console:Log events to the console (int) parm: zfs_top_maxinflight:Max I/Os per top-level (int) parm: zfs_resilver_delay:Number of ticks to delay resilver (int) parm: zfs_scrub_delay:Number of ticks to delay scrub (int) parm: zfs_scan_idle:Idle window in clock ticks (int) parm: zfs_scan_min_time_ms:Min millisecs to scrub per txg (int) parm: zfs_free_min_time_ms:Min millisecs to free per txg (int) parm: zfs_resilver_min_time_ms:Min millisecs to resilver per txg (int) parm: zfs_no_scrub_io:Set to disable scrub I/O (bool) parm: zfs_no_scrub_prefetch:Set to disable scrub prefetching (bool) parm: zfs_txg_timeout:Max seconds worth of delta per txg (int) parm: zfs_no_write_throttle:Disable write throttling (int) parm: zfs_write_limit_shift:log2(fraction of memory) per txg (int) parm: zfs_txg_synctime_ms:Target milliseconds between tgx sync (int) parm: zfs_write_limit_min:Min tgx write limit (ulong) parm: zfs_write_limit_max:Max tgx write limit (ulong) parm: zfs_write_limit_inflated:Inflated tgx write limit (ulong) parm: zfs_write_limit_override:Override tgx write limit (ulong) parm: zfs_prefetch_disable:Disable all ZFS prefetching (int) parm: zfetch_max_streams:Max number of streams per zfetch (uint) parm: zfetch_min_sec_reap:Min time before stream reclaim (uint) parm: zfetch_block_cap:Max number of blocks to fetch at a time (uint) parm: zfetch_array_rd_sz:Number of bytes in a array_read (ulong) parm: zfs_pd_blks_max:Max number of blocks to prefetch (int) parm: zfs_dedup_prefetch:Enable prefetching dedup-ed blks (int) parm: zfs_arc_min:Min arc size (ulong) parm: zfs_arc_max:Max arc size (ulong) parm: zfs_arc_meta_limit:Meta limit for arc size (ulong) parm: zfs_arc_reduce_dnlc_percent:Meta reclaim percentage (int) parm: zfs_arc_grow_retry:Seconds before growing arc size (int) parm: zfs_arc_shrink_shift:log2(fraction of arc to reclaim) (int) parm: zfs_arc_p_min_shift:arc_c shift to calc min/max arc_p (int)
2011-05-03 22:09:28 +00:00
Export symbols for ZIL interface These symbols are needed by consumers (i.e. Lustre) who wish to integrate with the ZIL. In addition the zil_rollback_destroy() prototype was removed because the implementation of this function was removed long ago. Signed-off-by: Alex Zhuravlev <alexey.zhuravlev@intel.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2892
2014-11-13 18:09:05 +00:00
EXPORT_SYMBOL(zil_alloc);
EXPORT_SYMBOL(zil_free);
EXPORT_SYMBOL(zil_open);
EXPORT_SYMBOL(zil_close);
EXPORT_SYMBOL(zil_replay);
EXPORT_SYMBOL(zil_replaying);
EXPORT_SYMBOL(zil_destroy);
EXPORT_SYMBOL(zil_destroy_sync);
EXPORT_SYMBOL(zil_itx_create);
EXPORT_SYMBOL(zil_itx_destroy);
EXPORT_SYMBOL(zil_itx_assign);
EXPORT_SYMBOL(zil_commit);
EXPORT_SYMBOL(zil_claim);
EXPORT_SYMBOL(zil_check_log_chain);
EXPORT_SYMBOL(zil_sync);
EXPORT_SYMBOL(zil_clean);
EXPORT_SYMBOL(zil_suspend);
EXPORT_SYMBOL(zil_resume);
OpenZFS 8585 - improve batching done in zil_commit() Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> Problem ======= The current implementation of zil_commit() can introduce significant latency, beyond what is inherent due to the latency of the underlying storage. The additional latency comes from two main problems: 1. When there's outstanding ZIL blocks being written (i.e. there's already a "writer thread" in progress), then any new calls to zil_commit() will block waiting for the currently oustanding ZIL blocks to complete. The blocks written for each "writer thread" is coined a "batch", and there can only ever be a single "batch" being written at a time. When a batch is being written, any new ZIL transactions will have to wait for the next batch to be written, which won't occur until the current batch finishes. As a result, the underlying storage may not be used as efficiently as possible. While "new" threads enter zil_commit() and are blocked waiting for the next batch, it's possible that the underlying storage isn't fully utilized by the current batch of ZIL blocks. In that case, it'd be better to allow these new threads to generate (and issue) a new ZIL block, such that it could be serviced by the underlying storage concurrently with the other ZIL blocks that are being serviced. 2. Any call to zil_commit() must wait for all ZIL blocks in its "batch" to complete, prior to zil_commit() returning. The size of any given batch is proportional to the number of ZIL transaction in the queue at the time that the batch starts processing the queue; which doesn't occur until the previous batch completes. Thus, if there's a lot of transactions in the queue, the batch could be composed of many ZIL blocks, and each call to zil_commit() will have to wait for all of these writes to complete (even if the thread calling zil_commit() only cared about one of the transactions in the batch). To further complicate the situation, these two issues result in the following side effect: 3. If a given batch takes longer to complete than normal, this results in larger batch sizes, which then take longer to complete and further drive up the latency of zil_commit(). This can occur for a number of reasons, including (but not limited to): transient changes in the workload, and storage latency irregularites. Solution ======== The solution attempted by this change has the following goals: 1. no on-disk changes; maintain current on-disk format. 2. modify the "batch size" to be equal to the "ZIL block size". 3. allow new batches to be generated and issued to disk, while there's already batches being serviced by the disk. 4. allow zil_commit() to wait for as few ZIL blocks as possible. 5. use as few ZIL blocks as possible, for the same amount of ZIL transactions, without introducing significant latency to any individual ZIL transaction. i.e. use fewer, but larger, ZIL blocks. In theory, with these goals met, the new allgorithm will allow the following improvements: 1. new ZIL blocks can be generated and issued, while there's already oustanding ZIL blocks being serviced by the storage. 2. the latency of zil_commit() should be proportional to the underlying storage latency, rather than the incoming synchronous workload. Porting Notes ============= Due to the changes made in commit 119a394ab0, the lifetime of an itx structure differs than in OpenZFS. Specifically, the itx structure is kept around until the data associated with the itx is considered to be safe on disk; this is so that the itx's callback can be called after the data is committed to stable storage. Since OpenZFS doesn't have this itx callback mechanism, it's able to destroy the itx structure immediately after the itx is committed to an lwb (before the lwb is written to disk). To support this difference, and to ensure the itx's callbacks can still be called after the itx's data is on disk, a few changes had to be made: * A list of itxs was added to the lwb structure. This list contains all of the itxs that have been committed to the lwb, such that the callbacks for these itxs can be called from zil_lwb_flush_vdevs_done(), after the data for the itxs is committed to disk. * A list of itxs was added on the stack of the zil_process_commit_list() function; the "nolwb_itxs" list. In some circumstances, an itx may not be committed to an lwb (e.g. if allocating the "next" ZIL block on disk fails), so this list is used to keep track of which itxs fall into this state, such that their callbacks can be called after the ZIL's writer pipeline is "stalled". * The logic to actually call the itx's callback was moved into the zil_itx_destroy() function. Since all consumers of zil_itx_destroy() were effectively performing the same logic (i.e. if callback is non-null, call the callback), it seemed like useful code cleanup to consolidate this logic into a single function. Additionally, the existing Linux tracepoint infrastructure dealing with the ZIL's probes and structures had to be updated to reflect these code changes. Specifically: * The "zil__cw1" and "zil__cw2" probes were removed, so they had to be removed from "trace_zil.h" as well. * Some of the zilog structure's fields were removed, which affected the tracepoint definitions of the structure. * New tracepoints had to be added for the following 3 new probes: * zil__process__commit__itx * zil__process__normal__itx * zil__commit__io__error OpenZFS-issue: https://www.illumos.org/issues/8585 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5d95a3a Closes #6566
2017-12-05 17:39:16 +00:00
EXPORT_SYMBOL(zil_lwb_add_block);
Export symbols for ZIL interface These symbols are needed by consumers (i.e. Lustre) who wish to integrate with the ZIL. In addition the zil_rollback_destroy() prototype was removed because the implementation of this function was removed long ago. Signed-off-by: Alex Zhuravlev <alexey.zhuravlev@intel.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2892
2014-11-13 18:09:05 +00:00
EXPORT_SYMBOL(zil_bp_tree_add);
EXPORT_SYMBOL(zil_set_sync);
EXPORT_SYMBOL(zil_set_logbias);
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
/* BEGIN CSTYLED */
Make module tunables cross platform Adds ZFS_MODULE_PARAM to abstract module parameter setting to operating systems other than Linux. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Signed-off-by: Ryan Moeller <ryan@ixsystems.com> Closes #9230
2019-09-05 21:49:49 +00:00
ZFS_MODULE_PARAM(zfs, zfs_, commit_timeout_pct, INT, ZMOD_RW,
"ZIL block open timeout percentage");
OpenZFS 8909 - 8585 can cause a use-after-free kernel panic Authored by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: John Kennedy <jwk404@gmail.com> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Igor Kozhukhov <igor@dilos.org> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Robert Mustacchi <rm@joyent.com> Ported-by: Prakash Surya <prakash.surya@delphix.com> PROBLEM ======= There's a race condition that exists if `zil_free_lwb` races with either `zil_commit_waiter_timeout` and/or `zil_lwb_flush_vdevs_done`. Here's an example panic due to this bug: > ::status debugging crash dump vmcore.0 (64-bit) from ip-10-110-205-40 operating system: 5.11 dlpx-5.2.2.0_2017-12-04-17-28-32b6ba51fb (i86pc) image uuid: 4af0edfb-e58e-6ed8-cafc-d3e9167c7513 panic message: BAD TRAP: type=e (#pf Page fault) rp=ffffff0010555970 addr=60 occurred in module "zfs" due to a NULL pointer dereference dump content: kernel pages only > $c zio_shrink+0x12() zil_lwb_write_issue+0x30d(ffffff03dcd15cc0, ffffff03e0730e20) zil_commit_waiter_timeout+0xa2(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit_waiter+0xf3(ffffff03dcd15cc0, ffffff03d97ffcf8) zil_commit+0x80(ffffff03dcd15cc0, 9a9) zfs_write+0xc34(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) fop_write+0x5b(ffffff03dc38b140, ffffff0010555e60, 40, ffffff03e00fb758, 0) write+0x250(42, fffffd7ff4832000, 2000) sys_syscall+0x177() If there's an outstanding lwb that's in `zil_commit_waiter_timeout` waiting to timeout, waiting on it's waiter's CV, we must be sure not to call `zil_free_lwb`. If we end up calling `zil_free_lwb`, then that LWB may be freed and can result in a use-after-free situation where the stale lwb pointer stored in the `zil_commit_waiter_t` structure of the thread waiting on the waiter's CV is used. A similar situation can occur if an lwb is issued to disk, and thus in the `LWB_STATE_ISSUED` state, and `zil_free_lwb` is called while the disk is servicing that lwb. In this situation, the lwb will be freed by `zil_free_lwb`, which will result in a use-after-free situation when the lwb's zio completes, and `zil_lwb_flush_vdevs_done` is called. This race condition is prevented in `zil_close` by calling `zil_commit` before `zil_free_lwb` is called, which will ensure all outstanding (i.e. all lwb's in the `LWB_STATE_OPEN` and/or `LWB_STATE_ISSUED` states) reach the `LWB_STATE_DONE` state before the lwb's are freed (`zil_commit` will not return untill all the lwb's are `LWB_STATE_DONE`). Further, this race condition is prevented in `zil_sync` by only calling `zil_free_lwb` for lwb's that do not have their `lwb_buf` pointer set. All lwb's not in the `LWB_STATE_DONE` state will have a non-null value for this pointer; the pointer is only cleared in `zil_lwb_flush_vdevs_done`, at which point the lwb's state will be changed to `LWB_STATE_DONE`. This race *is* present in `zil_suspend`, leading to this bug. At first glance, it would appear as though this would not be true because `zil_suspend` will call `zil_commit`, just like `zil_close`, but the problem is that `zil_suspend` will set the zilog's `zl_suspend` field prior to calling `zil_commit`. Further, in `zil_commit`, if `zl_suspend` is set, `zil_commit` will take a special branch of logic and use `txg_wait_synced` instead of performing the normal `zil_commit` logic. This call to `txg_wait_synced` might be good enough for the data to reach disk safely before it returns, but it does not ensure that all outstanding lwb's reach the `LWB_STATE_DONE` state before it returns. This is because, if there's an lwb "stuck" in `zil_commit_waiter_timeout`, waiting for it's lwb to timeout, it will maintain a non-null value for it's `lwb_buf` field and thus `zil_sync` will not free that lwb. Thus, even though the lwb's data is already on disk, the lwb will be left lingering, waiting on the CV, and will eventually timeout and be issued to disk even though the write is unnecessary. So, after `zil_commit` is called from `zil_suspend`, we incorrectly assume that there are not outstanding lwb's, and proceed to free all lwb's found on the zilog's lwb list. As a result, we free the lwb that will later be used `zil_commit_waiter_timeout`. SOLUTION ======== The solution to this, is to ensure all outstanding lwb's complete before calling `zil_free_lwb` via `zil_destroy` in `zil_suspend`. This patch accomplishes this goal by forcing the normal `zil_commit` logic when called from `zil_sync`. Now, `zil_suspend` will call `zil_commit_impl` which will always use the normal logic of waiting/issuing lwb's to disk before it returns. As a result, any lwb's outstanding when `zil_commit_impl` is called will be guaranteed to reach the `LWB_STATE_DONE` state by the time it returns. Further, no new lwb's will be created via `zil_commit` since the zilog's `zl_suspend` flag will be set. This will force all new callers of `zil_commit` to use `txg_wait_synced` instead of creating and issuing new lwb's. Thus, all lwb's left on the zilog's lwb list when `zil_destroy` is called will be in the `LWB_STATE_DONE` state, and we'll avoid this race condition. OpenZFS-issue: https://www.illumos.org/issues/8909 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/ece62b6f8d Closes #6940
2017-12-07 19:26:32 +00:00
Make module tunables cross platform Adds ZFS_MODULE_PARAM to abstract module parameter setting to operating systems other than Linux. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Signed-off-by: Ryan Moeller <ryan@ixsystems.com> Closes #9230
2019-09-05 21:49:49 +00:00
ZFS_MODULE_PARAM(zfs_zil, zil_, replay_disable, INT, ZMOD_RW,
"Disable intent logging replay");
Add missing ZFS tunables This commit adds module options for all existing zfs tunables. Ideally the average user should never need to modify any of these values. However, in practice sometimes you do need to tweak these values for one reason or another. In those cases it's nice not to have to resort to rebuilding from source. All tunables are visable to modinfo and the list is as follows: $ modinfo module/zfs/zfs.ko filename: module/zfs/zfs.ko license: CDDL author: Sun Microsystems/Oracle, Lawrence Livermore National Laboratory description: ZFS srcversion: 8EAB1D71DACE05B5AA61567 depends: spl,znvpair,zcommon,zunicode,zavl vermagic: 2.6.32-131.0.5.el6.x86_64 SMP mod_unload modversions parm: zvol_major:Major number for zvol device (uint) parm: zvol_threads:Number of threads for zvol device (uint) parm: zio_injection_enabled:Enable fault injection (int) parm: zio_bulk_flags:Additional flags to pass to bulk buffers (int) parm: zio_delay_max:Max zio millisec delay before posting event (int) parm: zio_requeue_io_start_cut_in_line:Prioritize requeued I/O (bool) parm: zil_replay_disable:Disable intent logging replay (int) parm: zfs_nocacheflush:Disable cache flushes (bool) parm: zfs_read_chunk_size:Bytes to read per chunk (long) parm: zfs_vdev_max_pending:Max pending per-vdev I/Os (int) parm: zfs_vdev_min_pending:Min pending per-vdev I/Os (int) parm: zfs_vdev_aggregation_limit:Max vdev I/O aggregation size (int) parm: zfs_vdev_time_shift:Deadline time shift for vdev I/O (int) parm: zfs_vdev_ramp_rate:Exponential I/O issue ramp-up rate (int) parm: zfs_vdev_read_gap_limit:Aggregate read I/O over gap (int) parm: zfs_vdev_write_gap_limit:Aggregate write I/O over gap (int) parm: zfs_vdev_scheduler:I/O scheduler (charp) parm: zfs_vdev_cache_max:Inflate reads small than max (int) parm: zfs_vdev_cache_size:Total size of the per-disk cache (int) parm: zfs_vdev_cache_bshift:Shift size to inflate reads too (int) parm: zfs_scrub_limit:Max scrub/resilver I/O per leaf vdev (int) parm: zfs_recover:Set to attempt to recover from fatal errors (int) parm: spa_config_path:SPA config file (/etc/zfs/zpool.cache) (charp) parm: zfs_zevent_len_max:Max event queue length (int) parm: zfs_zevent_cols:Max event column width (int) parm: zfs_zevent_console:Log events to the console (int) parm: zfs_top_maxinflight:Max I/Os per top-level (int) parm: zfs_resilver_delay:Number of ticks to delay resilver (int) parm: zfs_scrub_delay:Number of ticks to delay scrub (int) parm: zfs_scan_idle:Idle window in clock ticks (int) parm: zfs_scan_min_time_ms:Min millisecs to scrub per txg (int) parm: zfs_free_min_time_ms:Min millisecs to free per txg (int) parm: zfs_resilver_min_time_ms:Min millisecs to resilver per txg (int) parm: zfs_no_scrub_io:Set to disable scrub I/O (bool) parm: zfs_no_scrub_prefetch:Set to disable scrub prefetching (bool) parm: zfs_txg_timeout:Max seconds worth of delta per txg (int) parm: zfs_no_write_throttle:Disable write throttling (int) parm: zfs_write_limit_shift:log2(fraction of memory) per txg (int) parm: zfs_txg_synctime_ms:Target milliseconds between tgx sync (int) parm: zfs_write_limit_min:Min tgx write limit (ulong) parm: zfs_write_limit_max:Max tgx write limit (ulong) parm: zfs_write_limit_inflated:Inflated tgx write limit (ulong) parm: zfs_write_limit_override:Override tgx write limit (ulong) parm: zfs_prefetch_disable:Disable all ZFS prefetching (int) parm: zfetch_max_streams:Max number of streams per zfetch (uint) parm: zfetch_min_sec_reap:Min time before stream reclaim (uint) parm: zfetch_block_cap:Max number of blocks to fetch at a time (uint) parm: zfetch_array_rd_sz:Number of bytes in a array_read (ulong) parm: zfs_pd_blks_max:Max number of blocks to prefetch (int) parm: zfs_dedup_prefetch:Enable prefetching dedup-ed blks (int) parm: zfs_arc_min:Min arc size (ulong) parm: zfs_arc_max:Max arc size (ulong) parm: zfs_arc_meta_limit:Meta limit for arc size (ulong) parm: zfs_arc_reduce_dnlc_percent:Meta reclaim percentage (int) parm: zfs_arc_grow_retry:Seconds before growing arc size (int) parm: zfs_arc_shrink_shift:log2(fraction of arc to reclaim) (int) parm: zfs_arc_p_min_shift:arc_c shift to calc min/max arc_p (int)
2011-05-03 22:09:28 +00:00
Make module tunables cross platform Adds ZFS_MODULE_PARAM to abstract module parameter setting to operating systems other than Linux. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Signed-off-by: Ryan Moeller <ryan@ixsystems.com> Closes #9230
2019-09-05 21:49:49 +00:00
ZFS_MODULE_PARAM(zfs_zil, zil_, nocacheflush, INT, ZMOD_RW,
"Disable ZIL cache flushes");
Make zil_slog_limit a tunable module parameter. zil_slog_limit specifies the maximum commit size to be written to the separate log device. Larger commits bypass the separate log device and go directly to the data devices. The optimal value for zil_slog_limit directly depends on the latency and throughput characteristics of both the separate log device and the data disks. Small synchronous writes are faster on low-latency separate log devices (e.g. SSDs) whereas large synchronous writes are faster on high-latency data disks (e.g. spindles) because of higher throughput, especially with a large array. The point is, the line between "small" and "large" synchronous writes in this scenario is heavily dependent on the hardware used. That's why it should be made configurable. Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #783
2012-06-12 09:40:36 +00:00
Make module tunables cross platform Adds ZFS_MODULE_PARAM to abstract module parameter setting to operating systems other than Linux. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Signed-off-by: Ryan Moeller <ryan@ixsystems.com> Closes #9230
2019-09-05 21:49:49 +00:00
ZFS_MODULE_PARAM(zfs_zil, zil_, slog_bulk, ULONG, ZMOD_RW,
"Limit in bytes slog sync writes per commit");
make zil max block size tunable We've observed that on some highly fragmented pools, most metaslab allocations are small (~2-8KB), but there are some large, 128K allocations. The large allocations are for ZIL blocks. If there is a lot of fragmentation, the large allocations can be hard to satisfy. The most common impact of this is that we need to check (and thus load) lots of metaslabs from the ZIL allocation code path, causing sync writes to wait for metaslabs to load, which can take a second or more. In the worst case, we may not be able to satisfy the allocation, in which case the ZIL will resort to txg_wait_synced() to ensure the change is on disk. To provide a workaround for this, this change adds a tunable that can reduce the size of ZIL blocks. External-issue: DLPX-61719 Reviewed-by: George Wilson <george.wilson@delphix.com> Reviewed-by: Paul Dagnelie <pcd@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matthew Ahrens <mahrens@delphix.com> Closes #8865
2019-06-10 18:48:42 +00:00
Make module tunables cross platform Adds ZFS_MODULE_PARAM to abstract module parameter setting to operating systems other than Linux. Reviewed-by: Jorgen Lundman <lundman@lundman.net> Reviewed-by: Igor Kozhukhov <igor@dilos.org> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Matt Macy <mmacy@FreeBSD.org> Signed-off-by: Ryan Moeller <ryan@ixsystems.com> Closes #9230
2019-09-05 21:49:49 +00:00
ZFS_MODULE_PARAM(zfs_zil, zil_, maxblocksize, INT, ZMOD_RW,
"Limit in bytes of ZIL log block size");
OpenZFS 7578 - Fix/improve some aspects of ZIL writing - After some ZIL changes 6 years ago zil_slog_limit got partially broken due to zl_itx_list_sz not updated when async itx'es upgraded to sync. Actually because of other changes about that time zl_itx_list_sz is not really required to implement the functionality, so this patch removes some unneeded broken code and variables. - Original idea of zil_slog_limit was to reduce chance of SLOG abuse by single heavy logger, that increased latency for other (more latency critical) loggers, by pushing heavy log out into the main pool instead of SLOG. Beside huge latency increase for heavy writers, this implementation caused double write of all data, since the log records were explicitly prepared for SLOG. Since we now have I/O scheduler, I've found it can be much more efficient to reduce priority of heavy logger SLOG writes from ZIO_PRIORITY_SYNC_WRITE to ZIO_PRIORITY_ASYNC_WRITE, while still leave them on SLOG. - Existing ZIL implementation had problem with space efficiency when it has to write large chunks of data into log blocks of limited size. In some cases efficiency stopped to almost as low as 50%. In case of ZIL stored on spinning rust, that also reduced log write speed in half, since head had to uselessly fly over allocated but not written areas. This change improves the situation by offloading problematic operations from z*_log_write() to zil_lwb_commit(), which knows real situation of log blocks allocation and can split large requests into pieces much more efficiently. Also as side effect it removes one of two data copy operations done by ZIL code WR_COPIED case. - While there, untangle and unify code of z*_log_write() functions. Also zfs_log_write() alike to zvol_log_write() can now handle writes crossing block boundary, that may also improve efficiency if ZPL is made to do that. Sponsored by: iXsystems, Inc. Authored by: Alexander Motin <mav@FreeBSD.org> Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Andriy Gapon <avg@FreeBSD.org> Reviewed by: Steven Hartland <steven.hartland@multiplay.co.uk> Reviewed by: Brad Lewis <brad.lewis@delphix.com> Reviewed by: Richard Elling <Richard.Elling@RichardElling.com> Approved by: Robert Mustacchi <rm@joyent.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Reviewed-by: Richard Yao <ryao@gentoo.org> Ported-by: Giuseppe Di Natale <dinatale2@llnl.gov> OpenZFS-issue: https://www.illumos.org/issues/7578 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/aeb13ac Closes #6191
2017-06-09 16:15:37 +00:00
/* END CSTYLED */
Reference in New Issue Copy Permalink