d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
40152db5d3
freebsd-dev/sys/kern/vfs_bio.c

5501 lines
144 KiB
C
Raw Normal View History

Give dev_strategy() an explict cdev argument in preparation for removing buf->b-dev. Put a bio between the buf passed to dev_strategy() and the device driver strategy routine in order to not clobber fields in the buf. Assert copyright on vfs_bio.c and update copyright message to canonical text. There is no legal difference between John Dysons two-clause abbreviated BSD license and the canonical text.
2004-10-29 07:16:37 +00:00
/*-
sys/kern: adoption of SPDX licensing ID tags. Mainly focus on files that use BSD 2-Clause license, however the tool I was using misidentified many licenses so this was mostly a manual - error prone - task. The Software Package Data Exchange (SPDX) group provides a specification to make it easier for automated tools to detect and summarize well known opensource licenses. We are gradually adopting the specification, noting that the tags are considered only advisory and do not, in any way, superceed or replace the license texts.
2017-11-27 15:20:12 +00:00
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
Give dev_strategy() an explict cdev argument in preparation for removing buf->b-dev. Put a bio between the buf passed to dev_strategy() and the device driver strategy routine in order to not clobber fields in the buf. Assert copyright on vfs_bio.c and update copyright message to canonical text. There is no legal difference between John Dysons two-clause abbreviated BSD license and the canonical text.
2004-10-29 07:16:37 +00:00
* Copyright (c) 2004 Poul-Henning Kamp
Improve my copyright.
1997-12-22 11:54:00 +00:00
* Copyright (c) 1994,1997 John S. Dyson
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
* Copyright (c) 2013 The FreeBSD Foundation
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
* All rights reserved.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
* Portions of this software were developed by Konstantin Belousov
* under sponsorship from the FreeBSD Foundation.
*
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
Give dev_strategy() an explict cdev argument in preparation for removing buf->b-dev. Put a bio between the buf passed to dev_strategy() and the device driver strategy routine in order to not clobber fields in the buf. Assert copyright on vfs_bio.c and update copyright message to canonical text. There is no legal difference between John Dysons two-clause abbreviated BSD license and the canonical text.
2004-10-29 07:16:37 +00:00
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
/*
* this file contains a new buffer I/O scheme implementing a coherent
* VM object and buffer cache scheme. Pains have been taken to make
* sure that the performance degradation associated with schemes such
* as this is not realized.
*
* Author: John S. Dyson
* Significant help during the development and debugging phases
* had been provided by David Greenman, also of the FreeBSD core team.
Adjust some comments to prevent future confusion on the implementation. Also add a reference to the buf(9) manual page.
1998-12-22 18:57:30 +00:00
*
* see man buf(9) for more info.
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
Separate the struct bio related stuff out of <sys/buf.h> into <sys/bio.h>. <sys/bio.h> is now a prerequisite for <sys/buf.h> but it shall not be made a nested include according to bdes teachings on the subject of nested includes. Diskdrivers and similar stuff below specfs::strategy() should no longer need to include <sys/buf.> unless they need caching of data. Still a few bogus uses of struct buf to track down. Repocopy by: peter
2000-05-05 09:59:14 +00:00
#include <sys/bio.h>
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
#include <sys/bitset.h>
The size and contents of the DEV_STRATEGY() macro has progressed to the point where it being a macro is no longer sensible, and it will only be more so in days to come. BIO_STRATEGY() is now only used from DEV_STRATEGY() and should not be used directly anymore. Put the contents of both in the new function dev_strategy() and make DEV_STRATEGY() call that function. In addition, this allows us to make the rather magic bufdonebio() helper function static. This alse saves hunderedandsome bytes of code in a typical kernel.
2003-10-18 09:03:15 +00:00
#include <sys/conf.h>
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
#include <sys/counter.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/buf.h>
Un-inline the non-trivial "trivial" bio* functions. Untangle devstat_end_transaction_bio()
2002-09-14 19:34:11 +00:00
#include <sys/devicestat.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/eventhandler.h>
fail(9) support: Add support for kernel fault injection using KFAIL_POINT_* macros and fail_point_* infrastructure. Add example fail point in vfs_bio.c to simulate VM buf pressure. Approved by: dfr (mentor)
2009-05-27 16:36:54 +00:00
#include <sys/fail.h>
Include ktr.h in more compilation units Similar to r348026, exhaustive search for uses of CTRn() and cross reference ktr.h includes. Where it was obvious that an OS compat header of some kind included ktr.h indirectly, .c files were left alone. Some of these files clearly got ktr.h via header pollution in some scenarios, or tinderbox would not be passing prior to this revision, but go ahead and explicitly include it in files using it anyway. Like r348026, these CUs did not show up in tinderbox as missing the include. Reported by: peterj (arm64/mp_machdep.c) X-MFC-With: r347984 Sponsored by: Dell EMC Isilon
2019-05-21 20:38:48 +00:00
#include <sys/ktr.h>
Work around an integer overflow in expression `3 * maxbufspace / 4', when maxbufspace is larger than INT_MAX / 3. The overflow causes a hard hang on ia64 when physical memory is sufficiently large (8GB).
2007-06-09 23:41:14 +00:00
#include <sys/limits.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mount.h>
Catch up to moving headers: - machine/ipl.h -> sys/ipl.h - machine/mutex.h -> sys/mutex.h
2000-10-20 07:58:15 +00:00
#include <sys/mutex.h>
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
#include <sys/kernel.h>
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
#include <sys/kthread.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/proc.h>
Add four new RCTL resources - readbps, readiops, writebps and writeiops, for limiting disk (actually filesystem) IO. Note that in some cases these limits are not quite precise. It's ok, as long as it's within some reasonable bounds. Testing - and review of the code, in particular the VFS and VM parts - is very welcome. MFC after: 1 month Relnotes: yes Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D5080
2016-04-07 04:23:25 +00:00
#include <sys/racct.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/resourcevar.h>
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
#include <sys/rwlock.h>
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
#include <sys/smp.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/sysctl.h>
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
#include <sys/sysproto.h>
- Add a general purpose resource allocator, vmem, from NetBSD. It was originally inspired by the Solaris vmem detailed in the proceedings of usenix 2001. The NetBSD version was heavily refactored for bugs and simplicity. - Use this resource allocator to allocate the buffer and transient maps. Buffer cache defrags are reduced by 25% when used by filesystems with mixed block sizes. Ultimately this may permit dynamic buffer cache sizing on low KVA machines. Discussed with: alc, kib, attilio Tested by: pho Sponsored by: EMC / Isilon Storage Division
2013-06-28 03:51:20 +00:00
#include <sys/vmem.h>
Untangled the vm.h include file spaghetti.
1995-12-07 12:48:31 +00:00
#include <sys/vmmeter.h>
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
#include <sys/vnode.h>
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
#include <sys/watchdog.h>
Give dev_strategy() an explict cdev argument in preparation for removing buf->b-dev. Put a bio between the buf passed to dev_strategy() and the device driver strategy routine in order to not clobber fields in the buf. Assert copyright on vfs_bio.c and update copyright message to canonical text. There is no legal difference between John Dysons two-clause abbreviated BSD license and the canonical text.
2004-10-29 07:16:37 +00:00
#include <geom/geom.h>
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
#include <vm/vm.h>
Untangled the vm.h include file spaghetti.
1995-12-07 12:48:31 +00:00
#include <vm/vm_param.h>
Add and move declarations to fix all of the warnings from `gcc -Wimplicit' (except in netccitt, netiso and netns) and most of the warnings from `gcc -Wnested-externs'. Fix all the bugs found. There were no serious ones.
1995-03-16 18:17:34 +00:00
#include <vm/vm_kern.h>
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
#include <vm/vm_object.h>
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_pager.h>
Untangled the vm.h include file spaghetti.
1995-12-07 12:48:31 +00:00
#include <vm/vm_extern.h>
Implement a new totally dynamic (up to MAXPHYS) buffer kva allocation scheme. Additionally, add the capability for checking for unexpected kernel page faults. The maximum amount of kva space for buffers hasn't been decreased from where it is, but it will now be possible to do so. This scheme manages the kva space similar to the buffers themselves. If there isn't enough kva space because of usage or fragementation, buffers will be reclaimed until a buffer allocation is successful. This scheme should be very resistant to fragmentation problems until/if the LFS code is fixed and uses the bogus buffer locking scheme -- but a 'fixed' LFS is not likely to use such a scheme. Now there should be NO problem allocating buffers up to MAXPHYS.
1996-11-30 22:41:49 +00:00
#include <vm/vm_map.h>
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
#include <vm/swap_pager.h>
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
Normalize a significant number of kernel malloc type names: - Prefer '_' to ' ', as it results in more easily parsed results in memory monitoring tools such as vmstat. - Remove punctuation that is incompatible with using memory type names as file names, such as '/' characters. - Disambiguate some collisions by adding subsystem prefixes to some memory types. - Generally prefer lower case to upper case. - If the same type is defined in multiple architecture directories, attempt to use the same name in additional cases. Not all instances were caught in this change, so more work is required to finish this conversion. Similar changes are required for UMA zone names.
2005-10-31 15:41:29 +00:00
static MALLOC_DEFINE(M_BIOBUF, "biobuf", "BIO buffer");
Distribute and statizice a lot of the malloc M_* types. Substantial input from: bde
1997-10-11 18:31:40 +00:00
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
struct bio_ops bioops; /* I/O operation notification */
This patch removes the VOP_BWRITE() vector. VOP_BWRITE() was a hack which made it possible for NFS client side to use struct buf with non-bio backing. This patch takes a more general approach and adds a bp->b_op vector where more methods can be added. The success of this patch depends on bp->b_op being initialized all relevant places for some value of "relevant" which is not easy to determine. For now the buffers have grown a b_magic element which will make such issues a tiny bit easier to debug.
2001-04-17 08:56:39 +00:00
struct buf_ops buf_ops_bio = {
Move the buffer method vector (buf->b_op) to the bufobj. Extend it with a strategy method. Add bufstrategy() which do the usual VOP_SPECSTRATEGY/VOP_STRATEGY song and dance. Rename ibwrite to bufwrite(). Move the two NFS buf_ops to more sensible places, add bufstrategy to them. Add inlines for bwrite() and bstrategy() which calls through buf->b_bufobj->b_ops->b_{write,strategy}(). Replace almost all VOP_STRATEGY()/VOP_SPECSTRATEGY() calls with bstrategy().
2004-10-24 20:03:41 +00:00
.bop_name = "buf_ops_bio",
.bop_write = bufwrite,
.bop_strategy = bufstrategy,
Add BO_SYNC() and add a default which uses the secret vnode pointer and VOP_FSYNC() for now.
2005-01-11 10:43:08 +00:00
.bop_sync = bufsync,
Cylinder group bitmaps and blocks containing inode for a snapshot file are after snaplock, while other ffs device buffers are before snaplock in global lock order. By itself, this could cause deadlock when bdwrite() tries to flush dirty buffers on snapshotted ffs. If, during the flush, COW activity for snapshot needs to allocate block and ffs_alloccg() selects the cylinder group that is being written by bdwrite(), then kernel would panic due to recursive buffer lock acquision. Avoid dealing with buffers in bdwrite() that are from other side of snaplock divisor in the lock order then the buffer being written. Add new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in the bdwrite(). Default implementation, bufbdflush(), refactors the code from bdwrite(). For ffs device buffers, specialized implementation is used. Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes) Tested by: Peter Holm X-MFC after: 3 weeks (if ever: it changes ABI)
2007-01-23 10:01:19 +00:00
.bop_bdflush = bufbdflush,
This patch removes the VOP_BWRITE() vector. VOP_BWRITE() was a hack which made it possible for NFS client side to use struct buf with non-bio backing. This patch takes a more general approach and adds a bp->b_op vector where more methods can be added. The success of this patch depends on bp->b_op being initialized all relevant places for some value of "relevant" which is not easy to determine. For now the buffers have grown a b_magic element which will make such issues a tiny bit easier to debug.
2001-04-17 08:56:39 +00:00
};
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
struct bufqueue {
struct mtx_padalign bq_lock;
TAILQ_HEAD(, buf) bq_queue;
uint8_t bq_index;
uint16_t bq_subqueue;
int bq_len;
} __aligned(CACHE_LINE_SIZE);
#define BQ_LOCKPTR(bq) (&(bq)->bq_lock)
#define BQ_LOCK(bq) mtx_lock(BQ_LOCKPTR((bq)))
#define BQ_UNLOCK(bq) mtx_unlock(BQ_LOCKPTR((bq)))
#define BQ_ASSERT_LOCKED(bq) mtx_assert(BQ_LOCKPTR((bq)), MA_OWNED)
struct bufdomain {
struct bufqueue bd_subq[MAXCPU + 1]; /* Per-cpu sub queues + global */
struct bufqueue bd_dirtyq;
struct bufqueue *bd_cleanq;
struct mtx_padalign bd_run_lock;
/* Constants */
long bd_maxbufspace;
long bd_hibufspace;
long bd_lobufspace;
long bd_bufspacethresh;
int bd_hifreebuffers;
int bd_lofreebuffers;
int bd_hidirtybuffers;
int bd_lodirtybuffers;
int bd_dirtybufthresh;
int bd_lim;
/* atomics */
int bd_wanted;
int __aligned(CACHE_LINE_SIZE) bd_numdirtybuffers;
int __aligned(CACHE_LINE_SIZE) bd_running;
long __aligned(CACHE_LINE_SIZE) bd_bufspace;
int __aligned(CACHE_LINE_SIZE) bd_freebuffers;
} __aligned(CACHE_LINE_SIZE);
#define BD_LOCKPTR(bd) (&(bd)->bd_cleanq->bq_lock)
#define BD_LOCK(bd) mtx_lock(BD_LOCKPTR((bd)))
#define BD_UNLOCK(bd) mtx_unlock(BD_LOCKPTR((bd)))
#define BD_ASSERT_LOCKED(bd) mtx_assert(BD_LOCKPTR((bd)), MA_OWNED)
#define BD_RUN_LOCKPTR(bd) (&(bd)->bd_run_lock)
#define BD_RUN_LOCK(bd) mtx_lock(BD_RUN_LOCKPTR((bd)))
#define BD_RUN_UNLOCK(bd) mtx_unlock(BD_RUN_LOCKPTR((bd)))
#define BD_DOMAIN(bd) (bd - bdomain)
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
static struct buf *buf; /* buffer header pool */
extern struct buf *swbuf; /* Swap buffer header pool. */
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
caddr_t unmapped_buf;
Moved header definitions to buf.h, and added missing splx() - found by Johannes Helander.
1994-05-26 08:45:29 +00:00
Add support for multi-threading of soft updates. Replace a single soft updates thread with a thread per FFS-filesystem mount point. The threads are associated with the bufdaemon process. Reviewed by: kib Tested by: Peter Holm and Scott Long MFC after: 2 weeks Sponsored by: Netflix
2014-08-04 22:03:58 +00:00
/* Used below and for softdep flushing threads in ufs/ffs/ffs_softdep.c */
struct proc *bufdaemonproc;
Allow the bufdaemon and update daemon processes to skip the waitrunningbufspace() calls so that they are always able to proceed and clean up buffer space. Submitted by: Brian Fundakowski Feldman <green@freebsd.org>
2003-11-04 06:30:00 +00:00
whitespace
2004-11-04 08:25:52 +00:00
static int inmem(struct vnode *vp, daddr_t blkno);
Change the implementation of vm_hold_free_pages() so that it performs at most one call to pmap_qremove(), and thus one TLB shootdown, instead of one call and TLB shootdown per page. Simplify the interface to vm_hold_free_pages(). MFC after: 3 weeks
2010-07-11 20:11:44 +00:00
static void vm_hold_free_pages(struct buf *bp, int newbsize);
stylistic polishing.
2004-09-15 20:54:23 +00:00
static void vm_hold_load_pages(struct buf *bp, vm_offset_t from,
A Major staticize sweep. Generates a couple of warnings that I'll deal with later. A number of unused vars removed. A number of unused procs removed or #ifdefed.
1995-12-14 08:32:45 +00:00
vm_offset_t to);
Eliminate page queues locking from bufdone_finish() through the following changes: Rename vfs_page_set_valid() to vfs_page_set_validclean() to reflect what this function actually does. Suggested by: tegge Introduce a new version of vfs_page_set_valid() that does no more than what the function's name implies. Specifically, it does not update the page's dirty mask, and thus it does not require the page queues lock to be held. Update two of the three callers to the old vfs_page_set_valid() to call vfs_page_set_validclean() instead because they actually require the page's dirty mask to be cleared. Introduce vm_page_set_valid(). Reviewed by: tegge
2009-05-13 05:39:39 +00:00
static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m);
static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off,
Eliminate vfs_page_set_valid()'s unused argument.
2007-12-02 01:28:35 +00:00
vm_page_t m);
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
static void vfs_clean_pages_dirty_buf(struct buf *bp);
Refactor vfs_setdirty(), creating vfs_setdirty_locked_object(). Call vfs_setdirty_locked_object() from vfs_busy_pages() instead of vfs_setdirty(), thereby eliminating a second acquisition and release of the same vm object lock.
2006-10-29 00:04:39 +00:00
static void vfs_setdirty_locked_object(struct buf *bp);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
static void vfs_vmio_invalidate(struct buf *bp);
static void vfs_vmio_truncate(struct buf *bp, int npages);
static void vfs_vmio_extend(struct buf *bp, int npages, int size);
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
static int vfs_bio_clcheck(struct vnode *vp, int size,
daddr_t lblkno, daddr_t blkno);
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
static void breada(struct vnode *, daddr_t *, int *, int, struct ucred *, int,
void (*)(struct buf *));
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static int buf_flush(struct vnode *vp, struct bufdomain *, int);
static int flushbufqueues(struct vnode *, struct bufdomain *, int, int);
Remove __P.
2002-03-19 21:25:46 +00:00
static void buf_daemon(void);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
static __inline void bd_wakeup(void);
Don't allow vfs.lorunningspace or vfs.hirunningspace to be set such that lorunningspace is greater than hirunningspace as the system performs terribly if it is mistuned in this fashion. MFC after: 1 week
2013-11-15 15:29:53 +00:00
static int sysctl_runningspace(SYSCTL_HANDLER_ARGS);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
static void bufkva_reclaim(vmem_t *, int);
static void bufkva_free(struct buf *);
Implement NUMA support in uma(9) and malloc(9). Allocations from specific domains can be done by the _domain() API variants. UMA also supports a first-touch policy via the NUMA zone flag. The slab layer is now segregated by VM domains and is precise. It handles iteration for round-robin directly. The per-cpu cache layer remains a mix of domains according to where memory is allocated and freed. Well behaved clients can achieve perfect locality with no performance penalty. The direct domain allocation functions have to visit the slab layer and so require per-zone locks which come at some expense. Reviewed by: Attilio (a slightly older version) Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon
2018-01-12 23:25:05 +00:00
static int buf_import(void *, void **, int, int, int);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
static void buf_release(void *, void **, int);
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
static void maxbcachebuf_adjust(void);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static inline struct bufdomain *bufdomain(struct buf *);
static void bq_remove(struct bufqueue *bq, struct buf *bp);
static void bq_insert(struct bufqueue *bq, struct buf *bp, bool unlock);
static int buf_recycle(struct bufdomain *, bool kva);
static void bq_init(struct bufqueue *bq, int qindex, int cpu,
const char *lockname);
static void bd_init(struct bufdomain *bd);
static int bd_flushall(struct bufdomain *bd);
static int sysctl_bufdomain_long(SYSCTL_HANDLER_ARGS);
static int sysctl_bufdomain_int(SYSCTL_HANDLER_ARGS);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
Add an ABI compat shim for the vfs.bufspace sysctl for sysctl requests that try to fetch it as an int rather than a long. If the current value is greater than INT_MAX it reports a value of INT_MAX.
2009-03-10 15:26:50 +00:00
static int sysctl_bufspace(SYSCTL_HANDLER_ARGS);
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
int vmiodirenable = TRUE;
SYSCTL_INT(_vfs, OID_AUTO, vmiodirenable, CTLFLAG_RW, &vmiodirenable, 0,
"Use the VM system for directory writes");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
long runningbufspace;
SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0,
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Amount of presently outstanding async buffer io");
Add an ABI compat shim for the vfs.bufspace sysctl for sysctl requests that try to fetch it as an int rather than a long. If the current value is greater than INT_MAX it reports a value of INT_MAX.
2009-03-10 15:26:50 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, bufspace, CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RD,
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
NULL, 0, sysctl_bufspace, "L", "Physical memory used for buffers");
static counter_u64_t bufkvaspace;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, bufkvaspace, CTLFLAG_RD, &bufkvaspace,
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
"Kernel virtual memory used for buffers");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
static long maxbufspace;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, maxbufspace,
CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &maxbufspace,
__offsetof(struct bufdomain, bd_maxbufspace), sysctl_bufdomain_long, "L",
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
"Maximum allowed value of bufspace (including metadata)");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
static long bufmallocspace;
SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0,
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Amount of malloced memory for buffers");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
static long maxbufmallocspace;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RW, &maxbufmallocspace,
0, "Maximum amount of malloced memory for buffers");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
static long lobufspace;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, lobufspace,
CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &lobufspace,
__offsetof(struct bufdomain, bd_lobufspace), sysctl_bufdomain_long, "L",
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Minimum amount of buffers we want to have");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
long hibufspace;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, hibufspace,
CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &hibufspace,
__offsetof(struct bufdomain, bd_hibufspace), sysctl_bufdomain_long, "L",
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
"Maximum allowed value of bufspace (excluding metadata)");
long bufspacethresh;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, bufspacethresh,
CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &bufspacethresh,
__offsetof(struct bufdomain, bd_bufspacethresh), sysctl_bufdomain_long, "L",
"Bufspace consumed before waking the daemon to free some");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static counter_u64_t buffreekvacnt;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, buffreekvacnt, CTLFLAG_RW, &buffreekvacnt,
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Number of times we have freed the KVA space from some buffer");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static counter_u64_t bufdefragcnt;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, bufdefragcnt, CTLFLAG_RW, &bufdefragcnt,
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Number of times we have had to repeat buffer allocation to defragment");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
static long lorunningspace;
Don't allow vfs.lorunningspace or vfs.hirunningspace to be set such that lorunningspace is greater than hirunningspace as the system performs terribly if it is mistuned in this fashion. MFC after: 1 week
2013-11-15 15:29:53 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, lorunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE |
CTLFLAG_RW, &lorunningspace, 0, sysctl_runningspace, "L",
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Minimum preferred space used for in-progress I/O");
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
static long hirunningspace;
Don't allow vfs.lorunningspace or vfs.hirunningspace to be set such that lorunningspace is greater than hirunningspace as the system performs terribly if it is mistuned in this fashion. MFC after: 1 week
2013-11-15 15:29:53 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, hirunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE |
CTLFLAG_RW, &hirunningspace, 0, sysctl_runningspace, "L",
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Maximum amount of space to use for in-progress I/O");
Cylinder group bitmaps and blocks containing inode for a snapshot file are after snaplock, while other ffs device buffers are before snaplock in global lock order. By itself, this could cause deadlock when bdwrite() tries to flush dirty buffers on snapshotted ffs. If, during the flush, COW activity for snapshot needs to allocate block and ffs_alloccg() selects the cylinder group that is being written by bdwrite(), then kernel would panic due to recursive buffer lock acquision. Avoid dealing with buffers in bdwrite() that are from other side of snaplock divisor in the lock order then the buffer being written. Add new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in the bdwrite(). Default implementation, bufbdflush(), refactors the code from bdwrite(). For ffs device buffers, specialized implementation is used. Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes) Tested by: Peter Holm X-MFC after: 3 weeks (if ever: it changes ABI)
2007-01-23 10:01:19 +00:00
int dirtybufferflushes;
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
SYSCTL_INT(_vfs, OID_AUTO, dirtybufferflushes, CTLFLAG_RW, &dirtybufferflushes,
0, "Number of bdwrite to bawrite conversions to limit dirty buffers");
Cylinder group bitmaps and blocks containing inode for a snapshot file are after snaplock, while other ffs device buffers are before snaplock in global lock order. By itself, this could cause deadlock when bdwrite() tries to flush dirty buffers on snapshotted ffs. If, during the flush, COW activity for snapshot needs to allocate block and ffs_alloccg() selects the cylinder group that is being written by bdwrite(), then kernel would panic due to recursive buffer lock acquision. Avoid dealing with buffers in bdwrite() that are from other side of snaplock divisor in the lock order then the buffer being written. Add new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in the bdwrite(). Default implementation, bufbdflush(), refactors the code from bdwrite(). For ffs device buffers, specialized implementation is used. Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes) Tested by: Peter Holm X-MFC after: 3 weeks (if ever: it changes ABI)
2007-01-23 10:01:19 +00:00
int bdwriteskip;
SYSCTL_INT(_vfs, OID_AUTO, bdwriteskip, CTLFLAG_RW, &bdwriteskip,
0, "Number of buffers supplied to bdwrite with snapshot deadlock risk");
int altbufferflushes;
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
SYSCTL_INT(_vfs, OID_AUTO, altbufferflushes, CTLFLAG_RW, &altbufferflushes,
0, "Number of fsync flushes to limit dirty buffers");
When doing cleanup of excessive buffers in bdwrite (see kern/vfs_bio.c delta 1.371) we must ensure that we do not get ourselves into a recursive trap endlessly trying to clean up after ourselves. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 23:59:09 +00:00
static int recursiveflushes;
SYSCTL_INT(_vfs, OID_AUTO, recursiveflushes, CTLFLAG_RW, &recursiveflushes,
0, "Number of flushes skipped due to being recursive");
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static int sysctl_numdirtybuffers(SYSCTL_HANDLER_ARGS);
SYSCTL_PROC(_vfs, OID_AUTO, numdirtybuffers,
CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RD, NULL, 0, sysctl_numdirtybuffers, "I",
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"Number of buffers that are dirty (has unwritten changes) at the moment");
static int lodirtybuffers;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, lodirtybuffers,
Fix sysctl types broken in r329612.
2018-03-21 23:21:32 +00:00
CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &lodirtybuffers,
Redo r331328. We need to fix not only type but also format. While here again notice that we are fixing regression from r331106.
2018-03-22 05:26:27 +00:00
__offsetof(struct bufdomain, bd_lodirtybuffers), sysctl_bufdomain_int, "I",
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"How many buffers we want to have free before bufdaemon can sleep");
static int hidirtybuffers;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, hidirtybuffers,
Fix sysctl types broken in r329612.
2018-03-21 23:21:32 +00:00
CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &hidirtybuffers,
Redo r331328. We need to fix not only type but also format. While here again notice that we are fixing regression from r331106.
2018-03-22 05:26:27 +00:00
__offsetof(struct bufdomain, bd_hidirtybuffers), sysctl_bufdomain_int, "I",
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
"When the number of dirty buffers is considered severe");
Cylinder group bitmaps and blocks containing inode for a snapshot file are after snaplock, while other ffs device buffers are before snaplock in global lock order. By itself, this could cause deadlock when bdwrite() tries to flush dirty buffers on snapshotted ffs. If, during the flush, COW activity for snapshot needs to allocate block and ffs_alloccg() selects the cylinder group that is being written by bdwrite(), then kernel would panic due to recursive buffer lock acquision. Avoid dealing with buffers in bdwrite() that are from other side of snaplock divisor in the lock order then the buffer being written. Add new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in the bdwrite(). Default implementation, bufbdflush(), refactors the code from bdwrite(). For ffs device buffers, specialized implementation is used. Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes) Tested by: Peter Holm X-MFC after: 3 weeks (if ever: it changes ABI)
2007-01-23 10:01:19 +00:00
int dirtybufthresh;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, dirtybufthresh,
Fix sysctl types broken in r329612.
2018-03-21 23:21:32 +00:00
CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &dirtybufthresh,
Redo r331328. We need to fix not only type but also format. While here again notice that we are fixing regression from r331106.
2018-03-22 05:26:27 +00:00
__offsetof(struct bufdomain, bd_dirtybufthresh), sysctl_bufdomain_int, "I",
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
"Number of bdwrite to bawrite conversions to clear dirty buffers");
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
static int numfreebuffers;
SYSCTL_INT(_vfs, OID_AUTO, numfreebuffers, CTLFLAG_RD, &numfreebuffers, 0,
"Number of free buffers");
static int lofreebuffers;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, lofreebuffers,
Fix sysctl types broken in r329612.
2018-03-21 23:21:32 +00:00
CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &lofreebuffers,
Redo r331328. We need to fix not only type but also format. While here again notice that we are fixing regression from r331106.
2018-03-22 05:26:27 +00:00
__offsetof(struct bufdomain, bd_lofreebuffers), sysctl_bufdomain_int, "I",
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
"Target number of free buffers");
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
static int hifreebuffers;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
SYSCTL_PROC(_vfs, OID_AUTO, hifreebuffers,
Fix sysctl types broken in r329612.
2018-03-21 23:21:32 +00:00
CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &hifreebuffers,
Redo r331328. We need to fix not only type but also format. While here again notice that we are fixing regression from r331106.
2018-03-22 05:26:27 +00:00
__offsetof(struct bufdomain, bd_hifreebuffers), sysctl_bufdomain_int, "I",
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
"Threshold for clean buffer recycling");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static counter_u64_t getnewbufcalls;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RD,
&getnewbufcalls, "Number of calls to getnewbuf");
static counter_u64_t getnewbufrestarts;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, getnewbufrestarts, CTLFLAG_RD,
&getnewbufrestarts,
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
"Number of times getnewbuf has had to restart a buffer acquisition");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static counter_u64_t mappingrestarts;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, mappingrestarts, CTLFLAG_RD,
&mappingrestarts,
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
"Number of times getblk has had to restart a buffer mapping for "
"unmapped buffer");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static counter_u64_t numbufallocfails;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, numbufallocfails, CTLFLAG_RW,
&numbufallocfails, "Number of times buffer allocations failed");
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
static int flushbufqtarget = 100;
SYSCTL_INT(_vfs, OID_AUTO, flushbufqtarget, CTLFLAG_RW, &flushbufqtarget, 0,
"Amount of work to do in flushbufqueues when helping bufdaemon");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static counter_u64_t notbufdflushes;
SYSCTL_COUNTER_U64(_vfs, OID_AUTO, notbufdflushes, CTLFLAG_RD, &notbufdflushes,
Export the number of times bufdaemon got help from the normal threads.
2009-04-16 09:33:52 +00:00
"Number of dirty buffer flushes done by the bufdaemon helpers");
Add barrier write capability to the VFS buffer interface. A barrier write is a disk write request that tells the disk that the buffer being written must be committed to the media along with any writes that preceeded it before any future blocks may be written to the drive. Barrier writes are provided by adding the functions bbarrierwrite (bwrite with barrier) and babarrierwrite (bawrite with barrier). Following a bbarrierwrite the client knows that the requested buffer is on the media. It does not ensure that buffers written before that buffer are on the media. It only ensure that buffers written before that buffer will get to the media before any buffers written after that buffer. A flush command must be sent to the disk to ensure that all earlier written buffers are on the media. Reviewed by: kib Tested by: Peter Holm
2013-02-16 14:51:30 +00:00
static long barrierwrites;
SYSCTL_LONG(_vfs, OID_AUTO, barrierwrites, CTLFLAG_RW, &barrierwrites, 0,
"Number of barrier writes");
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
SYSCTL_INT(_vfs, OID_AUTO, unmapped_buf_allowed, CTLFLAG_RD,
&unmapped_buf_allowed, 0,
"Permit the use of the unmapped i/o");
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
int maxbcachebuf = MAXBCACHEBUF;
SYSCTL_INT(_vfs, OID_AUTO, maxbcachebuf, CTLFLAG_RDTUN, &maxbcachebuf, 0,
"Maximum size of a buffer cache block");
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* This lock synchronizes access to bd_request.
*/
Start annotating global _padalign locks with __exclusive_cache_line While these locks are guarnteed to not share their respective cache lines, their current placement leaves unnecessary holes in lines which preceeded them. For instance the annotation of vm_page_queue_free_mtx allows 2 neighbour cachelines (previously separate by the lock) to be collapsed into 1. The annotation is only effective on architectures which have it implemented in their linker script (currently only amd64). Thus locks are not converted to their not-padaligned variants as to not affect the rest. MFC after: 1 week
2017-09-06 20:28:18 +00:00
static struct mtx_padalign __exclusive_cache_line bdlock;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* This lock protects the runningbufreq and synchronizes runningbufwakeup and
* waitrunningbufspace().
*/
Start annotating global _padalign locks with __exclusive_cache_line While these locks are guarnteed to not share their respective cache lines, their current placement leaves unnecessary holes in lines which preceeded them. For instance the annotation of vm_page_queue_free_mtx allows 2 neighbour cachelines (previously separate by the lock) to be collapsed into 1. The annotation is only effective on architectures which have it implemented in their linker script (currently only amd64). Thus locks are not converted to their not-padaligned variants as to not affect the rest. MFC after: 1 week
2017-09-06 20:28:18 +00:00
static struct mtx_padalign __exclusive_cache_line rbreqlock;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* Lock that protects bdirtywait.
*/
Start annotating global _padalign locks with __exclusive_cache_line While these locks are guarnteed to not share their respective cache lines, their current placement leaves unnecessary holes in lines which preceeded them. For instance the annotation of vm_page_queue_free_mtx allows 2 neighbour cachelines (previously separate by the lock) to be collapsed into 1. The annotation is only effective on architectures which have it implemented in their linker script (currently only amd64). Thus locks are not converted to their not-padaligned variants as to not affect the rest. MFC after: 1 week
2017-09-06 20:28:18 +00:00
static struct mtx_padalign __exclusive_cache_line bdirtylock;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/*
* Wakeup point for bufdaemon, as well as indicator of whether it is already
* active. Set to 1 when the bufdaemon is already "on" the queue, 0 when it
* is idling.
*/
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
static int bd_request;
- Merge soft-updates journaling from projects/suj/head into head. This brings in support for an optional intent log which eliminates the need for background fsck on unclean shutdown. Sponsored by: iXsystems, Yahoo!, and Juniper. With help from: McKusick and Peter Holm
2010-04-24 07:05:35 +00:00
/*
* Request for the buf daemon to write more buffers than is indicated by
* lodirtybuf. This may be necessary to push out excess dependencies or
* defragment the address space where a simple count of the number of dirty
* buffers is insufficient to characterize the demand for flushing them.
*/
static int bd_speedupreq;
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/*
* Synchronization (sleep/wakeup) variable for active buffer space requests.
* Set when wait starts, cleared prior to wakeup().
* Used in runningbufwakeup() and waitrunningbufspace().
*/
static int runningbufreq;
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
/*
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
* Synchronization for bwillwrite() waiters.
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
*/
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
static int bdirtywait;
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
* Move bswlist declaration and initialization from kern/vfs_bio.c to vm/vm_pager.c, which is the only place it is used. * Make the QUEUE_* definitions and bufqueues local to vfs_bio.c. * constify buf_wmesg.
2002-03-05 18:20:58 +00:00
/*
* Definitions for the buffer free lists.
*/
#define QUEUE_NONE 0 /* on no queue */
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
#define QUEUE_EMPTY 1 /* empty buffer headers */
- Move BX_BKGRDWAIT and BX_BKGRDINPROG to BV_ and the b_vflags field. - Surround all accesses of the BKGRD{WAIT,INPROG} flags with the vnode interlock. - Don't use the B_LOCKED flag and QUEUE_LOCKED for background write buffers. Check for the BKGRDINPROG flag before recycling or throwing away a buffer. We do this instead because it is not safe for us to move the original buffer to a new queue from the callback on the background write buffer. - Remove the B_LOCKED flag and the locked buffer queue. They are no longer used. - The vnode interlock is used around checks for BKGRDINPROG where it may not be strictly necessary. If we hold the buf lock the a back-ground write will not be started without our knowledge, one may only be completed while we're not looking. Rather than remove the code, Document two of the places where this extra locking is done. A pass should be done to verify and minimize the locking later.
2003-08-28 06:55:18 +00:00
#define QUEUE_DIRTY 2 /* B_DELWRI buffers */
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
#define QUEUE_CLEAN 3 /* non-B_DELWRI buffers */
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
#define QUEUE_SENTINEL 4 /* not an queue index, but mark for sentinel */
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
/* Maximum number of buffer domains. */
#define BUF_DOMAINS 8
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
struct bufdomainset bdlodirty; /* Domains > lodirty */
struct bufdomainset bdhidirty; /* Domains > hidirty */
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/* Configured number of clean queues. */
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static int __read_mostly buf_domains;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
BITSET_DEFINE(bufdomainset, BUF_DOMAINS);
struct bufdomain __exclusive_cache_line bdomain[BUF_DOMAINS];
struct bufqueue __exclusive_cache_line bqempty;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* per-cpu empty buffer cache.
*/
uma_zone_t buf_zone;
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/*
* Single global constant for BUF_WMESG, to avoid getting multiple references.
* buf_wmesg is referred from macros.
*/
* Move bswlist declaration and initialization from kern/vfs_bio.c to vm/vm_pager.c, which is the only place it is used. * Make the QUEUE_* definitions and bufqueues local to vfs_bio.c. * constify buf_wmesg.
2002-03-05 18:20:58 +00:00
const char *buf_wmesg = BUF_WMESG;
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
Don't allow vfs.lorunningspace or vfs.hirunningspace to be set such that lorunningspace is greater than hirunningspace as the system performs terribly if it is mistuned in this fashion. MFC after: 1 week
2013-11-15 15:29:53 +00:00
static int
sysctl_runningspace(SYSCTL_HANDLER_ARGS)
{
long value;
int error;
value = *(long *)arg1;
error = sysctl_handle_long(oidp, &value, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
mtx_lock(&rbreqlock);
if (arg1 == &hirunningspace) {
if (value < lorunningspace)
error = EINVAL;
else
hirunningspace = value;
} else {
KASSERT(arg1 == &lorunningspace,
("%s: unknown arg1", __func__));
if (value > hirunningspace)
error = EINVAL;
else
lorunningspace = value;
}
mtx_unlock(&rbreqlock);
return (error);
}
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static int
sysctl_bufdomain_int(SYSCTL_HANDLER_ARGS)
{
int error;
int value;
int i;
value = *(int *)arg1;
error = sysctl_handle_int(oidp, &value, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
*(int *)arg1 = value;
for (i = 0; i < buf_domains; i++)
Cast through uintptr_t to narrow the buf domain pointer on 32-bit archs arg2 is an intmax_t, which on 32-bit architectures is 64 bits, wider than a pointer. When &bdomain[i] is added to arg2 it widens from uintptr_t to intmax_t, then gcc whines when it gets cast to a pointer. Casting through uintptr_t silences this warning.
2018-03-20 02:01:30 +00:00
*(int *)(uintptr_t)(((uintptr_t)&bdomain[i]) + arg2) =
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
value / buf_domains;
return (error);
}
static int
sysctl_bufdomain_long(SYSCTL_HANDLER_ARGS)
{
long value;
int error;
int i;
value = *(long *)arg1;
error = sysctl_handle_long(oidp, &value, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
*(long *)arg1 = value;
for (i = 0; i < buf_domains; i++)
Cast through uintptr_t to narrow the buf domain pointer on 32-bit archs arg2 is an intmax_t, which on 32-bit architectures is 64 bits, wider than a pointer. When &bdomain[i] is added to arg2 it widens from uintptr_t to intmax_t, then gcc whines when it gets cast to a pointer. Casting through uintptr_t silences this warning.
2018-03-20 02:01:30 +00:00
*(long *)(uintptr_t)(((uintptr_t)&bdomain[i]) + arg2) =
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
value / buf_domains;
return (error);
}
Add an ABI compat shim for the vfs.bufspace sysctl for sysctl requests that try to fetch it as an int rather than a long. If the current value is greater than INT_MAX it reports a value of INT_MAX.
2009-03-10 15:26:50 +00:00
#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \
defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7)
static int
sysctl_bufspace(SYSCTL_HANDLER_ARGS)
{
long lvalue;
int ivalue;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
int i;
Add an ABI compat shim for the vfs.bufspace sysctl for sysctl requests that try to fetch it as an int rather than a long. If the current value is greater than INT_MAX it reports a value of INT_MAX.
2009-03-10 15:26:50 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
lvalue = 0;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
for (i = 0; i < buf_domains; i++)
lvalue += bdomain[i].bd_bufspace;
Only use the ABI compat shim for vfs.bufspace if the old buffer is smaller than a long. PR: amd64/134786 Submitted by: Emil Mikulic emikulic| gmail MFC after: 3 days
2009-05-21 16:18:45 +00:00
if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long))
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
return (sysctl_handle_long(oidp, &lvalue, 0, req));
In the ABI shim for vfs.bufspace, rather than truncating values larger than INT_MAX to INT_MAX, just go ahead and write out the full long to give an error of ENOMEM to the user process. Requested by: bde
2009-03-10 21:27:15 +00:00
if (lvalue > INT_MAX)
/* On overflow, still write out a long to trigger ENOMEM. */
return (sysctl_handle_long(oidp, &lvalue, 0, req));
ivalue = lvalue;
Add an ABI compat shim for the vfs.bufspace sysctl for sysctl requests that try to fetch it as an int rather than a long. If the current value is greater than INT_MAX it reports a value of INT_MAX.
2009-03-10 15:26:50 +00:00
return (sysctl_handle_int(oidp, &ivalue, 0, req));
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
#else
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
static int
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
sysctl_bufspace(SYSCTL_HANDLER_ARGS)
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
long lvalue;
int i;
stylistic polishing.
2004-09-15 20:54:23 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
lvalue = 0;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
for (i = 0; i < buf_domains; i++)
lvalue += bdomain[i].bd_bufspace;
Fix up sysctl vfs.buffercache broken in r329612 Sample problem: top: sysctl(vfs.bufspace...) expected 8, got 4 Reported by: O. Hartmann <ohartmann walstatt.org>
2018-02-22 20:39:25 +00:00
return (sysctl_handle_long(oidp, &lvalue, 0, req));
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
#endif
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static int
sysctl_numdirtybuffers(SYSCTL_HANDLER_ARGS)
{
int value;
int i;
value = 0;
for (i = 0; i < buf_domains; i++)
value += bdomain[i].bd_numdirtybuffers;
return (sysctl_handle_int(oidp, &value, 0, req));
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* bdirtywakeup:
*
* Wakeup any bwillwrite() waiters.
*/
static void
bdirtywakeup(void)
{
mtx_lock(&bdirtylock);
if (bdirtywait) {
bdirtywait = 0;
wakeup(&bdirtywait);
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
mtx_unlock(&bdirtylock);
}
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
/*
* bd_clear:
*
* Clear a domain from the appropriate bitsets when dirtybuffers
* is decremented.
*/
static void
bd_clear(struct bufdomain *bd)
{
mtx_lock(&bdirtylock);
if (bd->bd_numdirtybuffers <= bd->bd_lodirtybuffers)
BIT_CLR(BUF_DOMAINS, BD_DOMAIN(bd), &bdlodirty);
if (bd->bd_numdirtybuffers <= bd->bd_hidirtybuffers)
BIT_CLR(BUF_DOMAINS, BD_DOMAIN(bd), &bdhidirty);
mtx_unlock(&bdirtylock);
}
/*
* bd_set:
*
* Set a domain in the appropriate bitsets when dirtybuffers
* is incremented.
*/
static void
bd_set(struct bufdomain *bd)
{
mtx_lock(&bdirtylock);
if (bd->bd_numdirtybuffers > bd->bd_lodirtybuffers)
BIT_SET(BUF_DOMAINS, BD_DOMAIN(bd), &bdlodirty);
if (bd->bd_numdirtybuffers > bd->bd_hidirtybuffers)
BIT_SET(BUF_DOMAINS, BD_DOMAIN(bd), &bdhidirty);
mtx_unlock(&bdirtylock);
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* bdirtysub:
*
* Decrement the numdirtybuffers count by one and wakeup any
* threads blocked in bwillwrite().
*/
static void
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bdirtysub(struct buf *bp)
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
{
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
struct bufdomain *bd;
int num;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = bufdomain(bp);
num = atomic_fetchadd_int(&bd->bd_numdirtybuffers, -1);
if (num == (bd->bd_lodirtybuffers + bd->bd_hidirtybuffers) / 2)
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bdirtywakeup();
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
if (num == bd->bd_lodirtybuffers || num == bd->bd_hidirtybuffers)
bd_clear(bd);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
}
/*
* bdirtyadd:
*
* Increment the numdirtybuffers count by one and wakeup the buf
* daemon if needed.
*/
static void
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bdirtyadd(struct buf *bp)
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
{
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
struct bufdomain *bd;
int num;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* Only do the wakeup once as we cross the boundary. The
* buf daemon will keep running until the condition clears.
*/
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = bufdomain(bp);
num = atomic_fetchadd_int(&bd->bd_numdirtybuffers, 1);
if (num == (bd->bd_lodirtybuffers + bd->bd_hidirtybuffers) / 2)
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bd_wakeup();
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
if (num == bd->bd_lodirtybuffers || num == bd->bd_hidirtybuffers)
bd_set(bd);
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
}
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
/*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* bufspace_daemon_wakeup:
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* Wakeup the daemons responsible for freeing clean bufs.
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*/
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
static void
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_daemon_wakeup(struct bufdomain *bd)
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
/*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* avoid the lock if the daemon is running.
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*/
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (atomic_fetchadd_int(&bd->bd_running, 1) == 0) {
BD_RUN_LOCK(bd);
atomic_store_int(&bd->bd_running, 1);
wakeup(&bd->bd_running);
BD_RUN_UNLOCK(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
}
/*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* bufspace_daemon_wait:
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* Sleep until the domain falls below a limit or one second passes.
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
*/
static void
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_daemon_wait(struct bufdomain *bd)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
/*
* Re-check our limits and sleep. bd_running must be
* cleared prior to checking the limits to avoid missed
* wakeups. The waker will adjust one of bufspace or
* freebuffers prior to checking bd_running.
*/
BD_RUN_LOCK(bd);
atomic_store_int(&bd->bd_running, 0);
if (bd->bd_bufspace < bd->bd_bufspacethresh &&
bd->bd_freebuffers > bd->bd_lofreebuffers) {
msleep(&bd->bd_running, BD_RUN_LOCKPTR(bd), PRIBIO|PDROP,
"-", hz);
} else {
/* Avoid spurious wakeups while running. */
atomic_store_int(&bd->bd_running, 1);
BD_RUN_UNLOCK(bd);
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
}
}
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
/*
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* bufspace_adjust:
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
*
* Adjust the reported bufspace for a KVA managed buffer, possibly
* waking any waiters.
*/
static void
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufspace_adjust(struct buf *bp, int bufsize)
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufdomain *bd;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
long space;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
int diff;
KASSERT((bp->b_flags & B_MALLOC) == 0,
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
("bufspace_adjust: malloc buf %p", bp));
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = bufdomain(bp);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
diff = bufsize - bp->b_bufsize;
if (diff < 0) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
atomic_subtract_long(&bd->bd_bufspace, -diff);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
} else if (diff > 0) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
space = atomic_fetchadd_long(&bd->bd_bufspace, diff);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/* Wake up the daemon on the transition. */
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (space < bd->bd_bufspacethresh &&
space + diff >= bd->bd_bufspacethresh)
bufspace_daemon_wakeup(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
bp->b_bufsize = bufsize;
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* bufspace_reserve:
*
* Reserve bufspace before calling allocbuf(). metadata has a
* different space limit than data.
*/
static int
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_reserve(struct bufdomain *bd, int size, bool metadata)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
long limit, new;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
long space;
if (metadata)
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
limit = bd->bd_maxbufspace;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
else
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
limit = bd->bd_hibufspace;
space = atomic_fetchadd_long(&bd->bd_bufspace, size);
new = space + size;
if (new > limit) {
atomic_subtract_long(&bd->bd_bufspace, size);
return (ENOSPC);
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/* Wake up the daemon on the transition. */
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (space < bd->bd_bufspacethresh && new >= bd->bd_bufspacethresh)
bufspace_daemon_wakeup(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return (0);
}
/*
* bufspace_release:
*
* Release reserved bufspace after bufspace_adjust() has consumed it.
*/
static void
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_release(struct bufdomain *bd, int size)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
atomic_subtract_long(&bd->bd_bufspace, size);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
/*
* bufspace_wait:
*
* Wait for bufspace, acting as the buf daemon if a locked vnode is
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* supplied. bd_wanted must be set prior to polling for space. The
* operation must be re-tried on return.
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
*/
static void
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_wait(struct bufdomain *bd, struct vnode *vp, int gbflags,
int slpflag, int slptimeo)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
struct thread *td;
int error, fl, norunbuf;
if ((gbflags & GB_NOWAIT_BD) != 0)
return;
td = curthread;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BD_LOCK(bd);
while (bd->bd_wanted) {
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (vp != NULL && vp->v_type != VCHR &&
(td->td_pflags & TDP_BUFNEED) == 0) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BD_UNLOCK(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* getblk() is called with a vnode locked, and
* some majority of the dirty buffers may as
* well belong to the vnode. Flushing the
* buffers there would make a progress that
* cannot be achieved by the buf_daemon, that
* cannot lock the vnode.
*/
norunbuf = ~(TDP_BUFNEED | TDP_NORUNNINGBUF) |
(td->td_pflags & TDP_NORUNNINGBUF);
/*
* Play bufdaemon. The getnewbuf() function
* may be called while the thread owns lock
* for another dirty buffer for the same
* vnode, which makes it impossible to use
* VOP_FSYNC() there, due to the buffer lock
* recursion.
*/
td->td_pflags |= TDP_BUFNEED | TDP_NORUNNINGBUF;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
fl = buf_flush(vp, bd, flushbufqtarget);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
td->td_pflags &= norunbuf;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BD_LOCK(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (fl != 0)
continue;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (bd->bd_wanted == 0)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
break;
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
error = msleep(&bd->bd_wanted, BD_LOCKPTR(bd),
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
(PRIBIO + 4) | slpflag, "newbuf", slptimeo);
if (error != 0)
break;
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BD_UNLOCK(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
/*
* bufspace_daemon:
*
* buffer space management daemon. Tries to maintain some marginal
* amount of free buffer space so that requesting processes neither
* block nor work to reclaim buffers.
*/
static void
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_daemon(void *arg)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufdomain *bd;
Make bufdaemon and bufspacedaemon use kthread_suspend_check instead of kproc_suspend_check. In r329612 bufspacedaemon was turned into a thread of the bufdaemon process causing both to call kproc_suspend_check with the same proc argument and that function contains the following while loop: while (SIGISMEMBER(p->p_siglist, SIGSTOP)) { wakeup(&p->p_siglist); msleep(&p->p_siglist, &p->p_mtx, PPAUSE, "kpsusp", 0); } So one thread wakes up the other and the other wakes up the first again, locking up UP machines on shutdown. Also register the shutdown handlers with SHUTDOWN_PRI_LAST + 100 so they run after the syncer has shutdown, because the syncer can cause a situation where bufdaemon help is needed to proceed. PR: 227404 Reviewed by: kib Tested by: cy, rmacklem
2018-04-22 16:05:29 +00:00
EVENTHANDLER_REGISTER(shutdown_pre_sync, kthread_shutdown, curthread,
SHUTDOWN_PRI_LAST + 100);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bd = arg;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
for (;;) {
Make bufdaemon and bufspacedaemon use kthread_suspend_check instead of kproc_suspend_check. In r329612 bufspacedaemon was turned into a thread of the bufdaemon process causing both to call kproc_suspend_check with the same proc argument and that function contains the following while loop: while (SIGISMEMBER(p->p_siglist, SIGSTOP)) { wakeup(&p->p_siglist); msleep(&p->p_siglist, &p->p_mtx, PPAUSE, "kpsusp", 0); } So one thread wakes up the other and the other wakes up the first again, locking up UP machines on shutdown. Also register the shutdown handlers with SHUTDOWN_PRI_LAST + 100 so they run after the syncer has shutdown, because the syncer can cause a situation where bufdaemon help is needed to proceed. PR: 227404 Reviewed by: kib Tested by: cy, rmacklem
2018-04-22 16:05:29 +00:00
kthread_suspend_check();
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* Free buffers from the clean queue until we meet our
* targets.
*
* Theory of operation: The buffer cache is most efficient
* when some free buffer headers and space are always
* available to getnewbuf(). This daemon attempts to prevent
* the excessive blocking and synchronization associated
* with shortfall. It goes through three phases according
* demand:
*
* 1) The daemon wakes up voluntarily once per-second
* during idle periods when the counters are below
* the wakeup thresholds (bufspacethresh, lofreebuffers).
*
* 2) The daemon wakes up as we cross the thresholds
* ahead of any potential blocking. This may bounce
* slightly according to the rate of consumption and
* release.
*
* 3) The daemon and consumers are starved for working
* clean buffers. This is the 'bufspace' sleep below
* which will inefficiently trade bufs with bqrelse
* until we return to condition 2.
*/
Fix a bug introduced in r329612 that slowly invalidates all clean bufs. Reported by: bde Reviewed by: bde Sponsored by: Netflix, Dell/EMC Isilon
2018-03-26 18:36:17 +00:00
while (bd->bd_bufspace > bd->bd_lobufspace ||
bd->bd_freebuffers < bd->bd_hifreebuffers) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (buf_recycle(bd, false) != 0) {
if (bd_flushall(bd))
continue;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
/*
* Speedup dirty if we've run out of clean
* buffers. This is possible in particular
* because softdep may held many bufs locked
* pending writes to other bufs which are
* marked for delayed write, exhausting
* clean space until they are written.
*/
bd_speedup();
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BD_LOCK(bd);
if (bd->bd_wanted) {
msleep(&bd->bd_wanted, BD_LOCKPTR(bd),
PRIBIO|PDROP, "bufspace", hz/10);
} else
BD_UNLOCK(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
maybe_yield();
Fix a bug introduced in r329612 that slowly invalidates all clean bufs. Reported by: bde Reviewed by: bde Sponsored by: Netflix, Dell/EMC Isilon
2018-03-26 18:36:17 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_daemon_wait(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
}
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
/*
* bufmallocadjust:
*
* Adjust the reported bufspace for a malloc managed buffer, possibly
* waking any waiters.
*/
static void
bufmallocadjust(struct buf *bp, int bufsize)
{
int diff;
KASSERT((bp->b_flags & B_MALLOC) != 0,
("bufmallocadjust: non-malloc buf %p", bp));
diff = bufsize - bp->b_bufsize;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (diff < 0)
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
atomic_subtract_long(&bufmallocspace, -diff);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
else
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
atomic_add_long(&bufmallocspace, diff);
bp->b_bufsize = bufsize;
}
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
/*
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
* runningwakeup:
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
*
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
* Wake up processes that are waiting on asynchronous writes to fall
* below lorunningspace.
*/
static void
runningwakeup(void)
{
mtx_lock(&rbreqlock);
if (runningbufreq) {
runningbufreq = 0;
wakeup(&runningbufreq);
}
mtx_unlock(&rbreqlock);
}
/*
* runningbufwakeup:
*
* Decrement the outstanding write count according.
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
*/
Un-staticize runningbufwakeup() and staticize updateproc. Add a new private thread flag to indicate that the thread should not sleep if runningbufspace is too large. Set this flag on the bufdaemon and syncer threads so that they skip the waitrunningbufspace() call in bufwrite() rather than than checking the proc pointer vs. the known proc pointers for these two threads. A way of preventing these threads from being starved for I/O but still placing limits on their outstanding I/O would be desirable. Set this flag in ffs_copyonwrite() to prevent bufwrite() calls from blocking on the runningbufspace check while holding snaplk. This prevents snaplk from being held for an arbitrarily long period of time if runningbufspace is high and greatly reduces the contention for snaplk. The disadvantage is that ffs_copyonwrite() can start a large amount of I/O if there are a large number of snapshots, which could cause a deadlock in other parts of the code. Call runningbufwakeup() in ffs_copyonwrite() to decrement runningbufspace before attempting to grab snaplk so that I/O requests waiting on snaplk are not counted in runningbufspace as being in-progress. Increment runningbufspace again before actually launching the original I/O request. Prior to the above two changes, the system could deadlock if enough I/O requests were blocked by snaplk to prevent runningbufspace from falling below lorunningspace and one of the bawrite() calls in ffs_copyonwrite() blocked in waitrunningbufspace() while holding snaplk. See <http://www.holm.cc/stress/log/cons143.html>
2005-09-30 01:30:01 +00:00
void
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
runningbufwakeup(struct buf *bp)
{
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
long space, bspace;
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bspace = bp->b_runningbufspace;
Assert that runningbufspace does not underflow. Sponsored by: The FreeBSD Foundation
2013-07-13 19:36:18 +00:00
if (bspace == 0)
return;
space = atomic_fetchadd_long(&runningbufspace, -bspace);
KASSERT(space >= bspace, ("runningbufspace underflow %ld %ld",
space, bspace));
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bp->b_runningbufspace = 0;
/*
* Only acquire the lock and wakeup on the transition from exceeding
* the threshold to falling below it.
*/
if (space < lorunningspace)
return;
if (space - bspace > lorunningspace)
return;
runningwakeup();
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
}
/*
* waitrunningbufspace()
*
* runningbufspace is a measure of the amount of I/O currently
* running. This routine is used in async-write situations to
* prevent creating huge backups of pending writes to a device.
* Only asynchronous writes are governed by this function.
*
* This does NOT turn an async write into a sync write. It waits
* for earlier writes to complete and generally returns before the
* caller's write has reached the device.
*/
Un-staticize waitrunningbufspace() and call it before returning from ffs_copyonwrite() if any async writes were launched. Restore the threads previous TDP_NORUNNINGBUF state before returning from ffs_copyonwrite().
2005-09-30 18:07:41 +00:00
void
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
waitrunningbufspace(void)
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_lock(&rbreqlock);
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
while (runningbufspace > hirunningspace) {
There is no need to count waiters for the runningbufspace. Sponsored by: The FreeBSD Foundation
2013-07-13 19:34:34 +00:00
runningbufreq = 1;
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
msleep(&runningbufreq, &rbreqlock, PVM, "wdrain", 0);
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
}
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_unlock(&rbreqlock);
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
}
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* vfs_buf_test_cache:
*
* Called when a buffer is extended. This function clears the B_CACHE
* bit if the newly extended portion of the buffer does not contain
* valid data.
*/
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
static __inline void
vfs_buf_test_cache(struct buf *bp, vm_ooffset_t foff, vm_offset_t off,
vm_offset_t size, vm_page_t m)
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
VM_OBJECT_ASSERT_LOCKED(m->object);
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
if (bp->b_flags & B_CACHE) {
int base = (foff + off) & PAGE_MASK;
if (vm_page_is_valid(m, base, size) == 0)
bp->b_flags &= ~B_CACHE;
}
}
Correct a comment typo
2007-03-25 10:07:23 +00:00
/* Wake up the buffer daemon if necessary */
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static void
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bd_wakeup(void)
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_lock(&bdlock);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
if (bd_request == 0) {
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
bd_request = 1;
wakeup(&bd_request);
}
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_unlock(&bdlock);
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
}
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
/*
* Adjust the maxbcachbuf tunable.
*/
static void
maxbcachebuf_adjust(void)
{
int i;
/*
* maxbcachebuf must be a power of 2 >= MAXBSIZE.
*/
i = 2;
while (i * 2 <= maxbcachebuf)
i *= 2;
maxbcachebuf = i;
if (maxbcachebuf < MAXBSIZE)
maxbcachebuf = MAXBSIZE;
if (maxbcachebuf > MAXPHYS)
maxbcachebuf = MAXPHYS;
if (bootverbose != 0 && maxbcachebuf != MAXBCACHEBUF)
printf("maxbcachebuf=%d\n", maxbcachebuf);
}
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
/*
* bd_speedup - speedup the buffer cache flushing code
*/
void
bd_speedup(void)
{
- Merge soft-updates journaling from projects/suj/head into head. This brings in support for an optional intent log which eliminates the need for background fsck on unclean shutdown. Sponsored by: iXsystems, Yahoo!, and Juniper. With help from: McKusick and Peter Holm
2010-04-24 07:05:35 +00:00
int needwake;
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Merge soft-updates journaling from projects/suj/head into head. This brings in support for an optional intent log which eliminates the need for background fsck on unclean shutdown. Sponsored by: iXsystems, Yahoo!, and Juniper. With help from: McKusick and Peter Holm
2010-04-24 07:05:35 +00:00
mtx_lock(&bdlock);
needwake = 0;
if (bd_speedupreq == 0 || bd_request == 0)
needwake = 1;
bd_speedupreq = 1;
bd_request = 1;
if (needwake)
wakeup(&bd_request);
mtx_unlock(&bdlock);
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
}
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
On i386, double the default size of the bio transient map. With the maxbcache size fixed, the auto-tuned transient map is too small for real-world load on i386. Tested by: David Wolfskill Sponsored by: The FreeBSD Foundation
2013-03-27 10:56:15 +00:00
#ifdef __i386__
#define TRANSIENT_DENOM 5
#else
#define TRANSIENT_DENOM 10
#endif
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
/*
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
* Calculating buffer cache scaling values and reserve space for buffer
* headers. This is called during low level kernel initialization and
* may be called more then once. We CANNOT write to the memory area
* being reserved at this time.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
bufhashinit() is called with a caddr_t and is expected to return the same in both the alpha and i386 ports.
1999-07-09 16:41:19 +00:00
caddr_t
Change hw.physmem and hw.usermem to unsigned long like they used to be in the original hardwired sysctl implementation. The buf size calculator still overflows an integer on machines with large KVA (eg: ia64) where the number of pages does not fit into an int. Use 'long' there. Change Maxmem and physmem and related variables to 'long', mostly for completeness. Machines are not likely to overflow 'int' pages in the near term, but then again, 640K ought to be enough for anybody. This comes for free on 32 bit machines, so why not?
2002-08-30 04:04:37 +00:00
kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est)
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
{
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
int tuned_nbuf;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
long maxbuf, maxbuf_sz, buf_sz, biotmap_sz;
stylistic polishing.
2004-09-15 20:54:23 +00:00
The nbuf calculation was assuming that PAGE_SIZE = 4096 bytes, which is bogus. The calculation has been adjusted to use units of kilobytes. Noticed by: Chad David <davidc@acns.ab.ca> MFC after: 1 week
2001-12-08 20:37:08 +00:00
/*
* physmem_est is in pages. Convert it to kilobytes (assumes
* PAGE_SIZE is >= 1K)
*/
physmem_est = physmem_est * (PAGE_SIZE / 1024);
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
maxbcachebuf_adjust();
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
/*
* The nominal buffer size (and minimum KVA allocation) is BKVASIZE.
* For the first 64MB of ram nominally allocate sufficient buffers to
* cover 1/4 of our ram. Beyond the first 64MB allocate additional
Fix the description of the formula used to autosize the number of buffers in the buffer cache. Approved by: re (kensmith)
2007-09-26 11:22:23 +00:00
* buffers to cover 1/10 of our ram over 64MB. When auto-sizing
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
* the buffer cache we limit the eventual kva reservation to
* maxbcache bytes.
*
* factor represents the 1/4 x ram conversion.
*/
if (nbuf == 0) {
The nbuf calculation was assuming that PAGE_SIZE = 4096 bytes, which is bogus. The calculation has been adjusted to use units of kilobytes. Noticed by: Chad David <davidc@acns.ab.ca> MFC after: 1 week
2001-12-08 20:37:08 +00:00
int factor = 4 * BKVASIZE / 1024;
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
nbuf = 50;
The nbuf calculation was assuming that PAGE_SIZE = 4096 bytes, which is bogus. The calculation has been adjusted to use units of kilobytes. Noticed by: Chad David <davidc@acns.ab.ca> MFC after: 1 week
2001-12-08 20:37:08 +00:00
if (physmem_est > 4096)
nbuf += min((physmem_est - 4096) / factor,
65536 / factor);
if (physmem_est > 65536)
When auto-sizing the buffer cache, limit the amount of physical memory used as the estimation of size, to 32GB. This provides around 100K of buffer headers and corresponding KVA for buffer map at the peak. Sizing the cache larger is not useful, also resulting in the wasting and exhausting of KVA for large machines. Reported and tested by: bdrewery Sponsored by: The FreeBSD Foundation
2013-06-03 04:16:48 +00:00
nbuf += min((physmem_est - 65536) * 2 / (factor * 5),
32 * 1024 * 1024 / (factor * 5));
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
if (maxbcache && nbuf > maxbcache / BKVASIZE)
nbuf = maxbcache / BKVASIZE;
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
tuned_nbuf = 1;
} else
tuned_nbuf = 0;
/* XXX Avoid unsigned long overflows later on with maxbufspace. */
maxbuf = (LONG_MAX / 3) / BKVASIZE;
if (nbuf > maxbuf) {
if (!tuned_nbuf)
printf("Warning: nbufs lowered from %d to %ld\n", nbuf,
maxbuf);
nbuf = maxbuf;
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
Fix a typo.
2014-01-21 03:24:52 +00:00
* Ideal allocation size for the transient bio submap is 10%
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
* of the maximal space buffer map. This roughly corresponds
* to the amount of the buffer mapped for typical UFS load.
*
* Clip the buffer map to reserve space for the transient
On i386, double the default size of the bio transient map. With the maxbcache size fixed, the auto-tuned transient map is too small for real-world load on i386. Tested by: David Wolfskill Sponsored by: The FreeBSD Foundation
2013-03-27 10:56:15 +00:00
* BIOs, if its extent is bigger than 90% (80% on i386) of the
* maximum buffer map extent on the platform.
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
*
* The fall-back to the maxbuf in case of maxbcache unset,
* allows to not trim the buffer KVA for the architectures
* with ample KVA space.
*/
Only size and create the bio_transient_map when unmapped buffers are enabled. Now, disabling the unmapped buffers should result in the kernel memory map identical to pre-r248550. Sponsored by: The FreeBSD Foundation
2013-03-21 07:28:15 +00:00
if (bio_transient_maxcnt == 0 && unmapped_buf_allowed) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
maxbuf_sz = maxbcache != 0 ? maxbcache : maxbuf * BKVASIZE;
buf_sz = (long)nbuf * BKVASIZE;
On i386, double the default size of the bio transient map. With the maxbcache size fixed, the auto-tuned transient map is too small for real-world load on i386. Tested by: David Wolfskill Sponsored by: The FreeBSD Foundation
2013-03-27 10:56:15 +00:00
if (buf_sz < maxbuf_sz / TRANSIENT_DENOM *
(TRANSIENT_DENOM - 1)) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
* There is more KVA than memory. Do not
* adjust buffer map size, and assign the rest
* of maxbuf to transient map.
*/
biotmap_sz = maxbuf_sz - buf_sz;
} else {
/*
* Buffer map spans all KVA we could afford on
On i386, double the default size of the bio transient map. With the maxbcache size fixed, the auto-tuned transient map is too small for real-world load on i386. Tested by: David Wolfskill Sponsored by: The FreeBSD Foundation
2013-03-27 10:56:15 +00:00
* this platform. Give 10% (20% on i386) of
* the buffer map to the transient bio map.
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
*/
On i386, double the default size of the bio transient map. With the maxbcache size fixed, the auto-tuned transient map is too small for real-world load on i386. Tested by: David Wolfskill Sponsored by: The FreeBSD Foundation
2013-03-27 10:56:15 +00:00
biotmap_sz = buf_sz / TRANSIENT_DENOM;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
buf_sz -= biotmap_sz;
}
if (biotmap_sz / INT_MAX > MAXPHYS)
bio_transient_maxcnt = INT_MAX;
else
bio_transient_maxcnt = biotmap_sz / MAXPHYS;
/*
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* Artificially limit to 1024 simultaneous in-flight I/Os
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
* using the transient mapping.
*/
if (bio_transient_maxcnt > 1024)
bio_transient_maxcnt = 1024;
if (tuned_nbuf)
nbuf = buf_sz / BKVASIZE;
}
Fix mistake in r343030: move nswbuf calculation back to kern_vfs_bio_buffer_alloc(), because in init_param2() nbuf isn't really initialized yet. Pointed out by: bde
2019-01-16 20:20:38 +00:00
if (nswbuf == 0) {
nswbuf = min(nbuf / 4, 256);
if (nswbuf < NSWBUF_MIN)
nswbuf = NSWBUF_MIN;
}
Move most of the kernel submap initialization code, including the timeout callwheel and buffer cache, out of the platform specific areas and into the machine independant area. i386 and alpha adjusted here. Other cpus can be fixed piecemeal. Reviewed by: freebsd-smp, jake
2001-08-22 04:07:27 +00:00
/*
* Reserve space for the buffer cache buffers
*/
buf = (void *)v;
v = (caddr_t)(buf + nbuf);
return(v);
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
}
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/* Initialize the buffer subsystem. Called before use of any buffers. */
While in the real world, I had a bad case of being swapped out for a lot of cycles. While waiting there I added a lot of the extra ()'s I have, (I have never used LISP to any extent). So I compiled the kernel with -Wall and shut up a lot of "suggest you add ()'s", removed a bunch of unused var's and added a couple of declarations here and there. Having a lap-top is highly recommended. My kernel still runs, yell at me if you kernel breaks.
1994-09-25 19:34:02 +00:00
void
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
bufinit(void)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
struct buf *bp;
int i;
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
KASSERT(maxbcachebuf >= MAXBSIZE,
("maxbcachebuf (%d) must be >= MAXBSIZE (%d)\n", maxbcachebuf,
MAXBSIZE));
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq_init(&bqempty, QUEUE_EMPTY, -1, "bufq empty lock");
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_init(&rbreqlock, "runningbufspace lock", NULL, MTX_DEF);
mtx_init(&bdlock, "buffer daemon lock", NULL, MTX_DEF);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
mtx_init(&bdirtylock, "dirty buf lock", NULL, MTX_DEF);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
unmapped_buf = (caddr_t)kva_alloc(MAXPHYS);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/* finally, initialize each buffer header and stick on empty q */
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
for (i = 0; i < nbuf; i++) {
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
bp = &buf[i];
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
bzero(bp, sizeof *bp);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bp->b_flags = B_INVAL;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
bp->b_rcred = NOCRED;
bp->b_wcred = NOCRED;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bp->b_qindex = QUEUE_NONE;
bp->b_domain = -1;
Fix issues with sparse cpu allocation. Consistently use mp_maxid + 1. Reported by: pho Reviewed by: markj Sponsored by: Netflix, Dell/EMC Isilon
2018-02-25 00:35:21 +00:00
bp->b_subqueue = mp_maxid + 1;
Use TAILQ macros for clean/dirty block list processing. Set b_xflags rather than abusing the list next pointer with a magic number.
1998-10-31 15:31:29 +00:00
bp->b_xflags = 0;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
bp->b_data = bp->b_kvabase = unmapped_buf;
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
LIST_INIT(&bp->b_dep);
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
BUF_LOCKINIT(bp);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq_insert(&bqempty, bp, false);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
/*
Commit the buffer cache cleanup patch to 4.x and 5.x. This patch fixes a fragmentation problem due to geteblk() reserving too much space for the buffer and imposes a larger granularity (16K) on KVA reservations for the buffer cache to avoid fragmentation issues. The buffer cache size calculations have been redone to simplify them (fewer defines, better comments, less chance of running out of KVA). The geteblk() fix solves a performance problem that DG was able reproduce. This patch does not completely fix the KVA fragmentation problems, but it goes a long way Mostly Reviewed by: bde and others Approved by: jkh
2000-03-27 21:29:33 +00:00
* maxbufspace is the absolute maximum amount of buffer space we are
* allowed to reserve in KVM and in real terms. The absolute maximum
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* is nominally used by metadata. hibufspace is the nominal maximum
* used by most other requests. The differential is required to
* ensure that metadata deadlocks don't occur.
Commit the buffer cache cleanup patch to 4.x and 5.x. This patch fixes a fragmentation problem due to geteblk() reserving too much space for the buffer and imposes a larger granularity (16K) on KVA reservations for the buffer cache to avoid fragmentation issues. The buffer cache size calculations have been redone to simplify them (fewer defines, better comments, less chance of running out of KVA). The geteblk() fix solves a performance problem that DG was able reproduce. This patch does not completely fix the KVA fragmentation problems, but it goes a long way Mostly Reviewed by: bde and others Approved by: jkh
2000-03-27 21:29:33 +00:00
*
* maxbufspace is based on BKVASIZE. Allocating buffers larger then
* this may result in KVM fragmentation which is not handled optimally
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* by the system. XXX This is less true with vmem. We could use
* PAGE_SIZE.
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*/
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
maxbufspace = (long)nbuf * BKVASIZE;
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - maxbcachebuf * 10);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
lobufspace = (hibufspace / 20) * 19; /* 95% */
bufspacethresh = lobufspace + (hibufspace - lobufspace) / 2;
Commit the buffer cache cleanup patch to 4.x and 5.x. This patch fixes a fragmentation problem due to geteblk() reserving too much space for the buffer and imposes a larger granularity (16K) on KVA reservations for the buffer cache to avoid fragmentation issues. The buffer cache size calculations have been redone to simplify them (fewer defines, better comments, less chance of running out of KVA). The geteblk() fix solves a performance problem that DG was able reproduce. This patch does not completely fix the KVA fragmentation problems, but it goes a long way Mostly Reviewed by: bde and others Approved by: jkh
2000-03-27 21:29:33 +00:00
Make lorunningspace catch up with hirunningspace. While there, add comment about the magic numbers. Prodded by: alc
2010-07-23 12:30:29 +00:00
/*
Reduce the difference between hirunningspace and lorunningspace, it should help interactivity in edge cases.
2010-10-25 14:05:25 +00:00
* Note: The 16 MiB upper limit for hirunningspace was chosen
Elaborate on how hirunningspace was chosen.
2010-08-09 22:22:46 +00:00
* arbitrarily and may need further tuning. It corresponds to
* 128 outstanding write IO requests (if IO size is 128 KiB),
Fix (hopefully) the spelling of "queuing." Submitted by: bf1783 at gmail com
2010-08-09 23:32:37 +00:00
* which fits with many RAID controllers' tagged queuing limits.
Reduce the difference between hirunningspace and lorunningspace, it should help interactivity in edge cases.
2010-10-25 14:05:25 +00:00
* The lower 1 MiB limit is the historical upper limit for
Elaborate on how hirunningspace was chosen.
2010-08-09 22:22:46 +00:00
* hirunningspace.
Make lorunningspace catch up with hirunningspace. While there, add comment about the magic numbers. Prodded by: alc
2010-07-23 12:30:29 +00:00
*/
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
hirunningspace = lmax(lmin(roundup(hibufspace / 64, maxbcachebuf),
Fix expression style. Prodded by: jhb
2010-07-20 13:59:51 +00:00
16 * 1024 * 1024), 1024 * 1024);
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
lorunningspace = roundup((hirunningspace * 2) / 3, maxbcachebuf);
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* Limit the amount of malloc memory since it is wired permanently into
* the kernel space. Even though this is accounted for in the buffer
* allocation, we don't want the malloced region to grow uncontrolled.
* The malloc scheme improves memory utilization significantly on
* average (small) directories.
*/
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
maxbufmallocspace = hibufspace / 20;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* Reduce the chance of a deadlock occurring by limiting the number
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* of delayed-write dirty buffers we allow to stack up.
*/
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
hidirtybuffers = nbuf / 4 + 20;
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
dirtybufthresh = hidirtybuffers * 9 / 10;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* To support extreme low-memory systems, make sure hidirtybuffers
* cannot eat up all available buffer space. This occurs when our
* minimum cannot be met. We try to size hidirtybuffers to 3/4 our
* buffer space assuming BKVASIZE'd buffers.
*/
Adjust some variables (mostly related to the buffer cache) that hold address space sizes to be longs instead of ints. Specifically, the follow values are now longs: runningbufspace, bufspace, maxbufspace, bufmallocspace, maxbufmallocspace, lobufspace, hibufspace, lorunningspace, hirunningspace, maxswzone, maxbcache, and maxpipekva. Previously, a relatively small number (~ 44000) of buffers set in kern.nbuf would result in integer overflows resulting either in hangs or bogus values of hidirtybuffers and lodirtybuffers. Now one has to overflow a long to see such problems. There was a check for a nbuf setting that would cause overflows in the auto-tuning of nbuf. I've changed it to always check and cap nbuf but warn if a user-supplied tunable would cause overflow. Note that this changes the ABI of several sysctls that are used by things like top(1), etc., so any MFC would probably require a some gross shims to allow for that. MFC after: 1 month
2009-03-09 19:35:20 +00:00
while ((long)hidirtybuffers * BKVASIZE > 3 * hibufspace / 4) {
Adjust the buffer cache to better handle small-memory machines. A slightly older version of this code was tested by BDE and I. Also fixes a lockup situation when kva gets too fragmented. Remove the maxvmiobufspace variable and sysctl, they are no longer used. Also cleanup (remove) #if 0 sections from prior commits. This code is more of a hack, but presumably the whole buffer cache implementation is going to be rewritten in the next year so it's no big deal.
1999-10-24 03:27:28 +00:00
hidirtybuffers >>= 1;
}
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
lodirtybuffers = hidirtybuffers / 2;
Adjust the buffer cache to better handle small-memory machines. A slightly older version of this code was tested by BDE and I. Also fixes a lockup situation when kva gets too fragmented. Remove the maxvmiobufspace variable and sysctl, they are no longer used. Also cleanup (remove) #if 0 sections from prior commits. This code is more of a hack, but presumably the whole buffer cache implementation is going to be rewritten in the next year so it's no big deal.
1999-10-24 03:27:28 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* lofreebuffers should be sufficient to avoid stalling waiting on
* buf headers under heavy utilization. The bufs in per-cpu caches
* are counted as free but will be unavailable to threads executing
* on other cpus.
*
* hifreebuffers is the free target for the bufspace daemon. This
* should be set appropriately to limit work per-iteration.
*/
lofreebuffers = MIN((nbuf / 25) + (20 * mp_ncpus), 128 * mp_ncpus);
hifreebuffers = (3 * lofreebuffers) / 2;
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
numfreebuffers = nbuf;
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/* Setup the kva and free list allocators. */
vmem_set_reclaim(buffer_arena, bufkva_reclaim);
buf_zone = uma_zcache_create("buf free cache", sizeof(struct buf),
NULL, NULL, NULL, NULL, buf_import, buf_release, NULL, 0);
/*
* Size the clean queue according to the amount of buffer space.
* One queue per-256mb up to the max. More queues gives better
* concurrency but less accurate LRU.
*/
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
buf_domains = MIN(howmany(maxbufspace, 256*1024*1024), BUF_DOMAINS);
for (i = 0 ; i < buf_domains; i++) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufdomain *bd;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = &bdomain[i];
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bd_init(bd);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd->bd_freebuffers = nbuf / buf_domains;
bd->bd_hifreebuffers = hifreebuffers / buf_domains;
bd->bd_lofreebuffers = lofreebuffers / buf_domains;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bd->bd_bufspace = 0;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd->bd_maxbufspace = maxbufspace / buf_domains;
bd->bd_hibufspace = hibufspace / buf_domains;
bd->bd_lobufspace = lobufspace / buf_domains;
bd->bd_bufspacethresh = bufspacethresh / buf_domains;
bd->bd_numdirtybuffers = 0;
bd->bd_hidirtybuffers = hidirtybuffers / buf_domains;
bd->bd_lodirtybuffers = lodirtybuffers / buf_domains;
bd->bd_dirtybufthresh = dirtybufthresh / buf_domains;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
/* Don't allow more than 2% of bufs in the per-cpu caches. */
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd->bd_lim = nbuf / buf_domains / 50 / mp_ncpus;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
}
getnewbufcalls = counter_u64_alloc(M_WAITOK);
getnewbufrestarts = counter_u64_alloc(M_WAITOK);
mappingrestarts = counter_u64_alloc(M_WAITOK);
numbufallocfails = counter_u64_alloc(M_WAITOK);
notbufdflushes = counter_u64_alloc(M_WAITOK);
buffreekvacnt = counter_u64_alloc(M_WAITOK);
bufdefragcnt = counter_u64_alloc(M_WAITOK);
bufkvaspace = counter_u64_alloc(M_WAITOK);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
#ifdef INVARIANTS
static inline void
vfs_buf_check_mapped(struct buf *bp)
{
KASSERT(bp->b_kvabase != unmapped_buf,
("mapped buf: b_kvabase was not updated %p", bp));
KASSERT(bp->b_data != unmapped_buf,
("mapped buf: b_data was not updated %p", bp));
vfs: fill fallout from r286076 This right operator is >= not =>. Reported by: cem
2015-07-30 15:43:26 +00:00
KASSERT(bp->b_data < unmapped_buf || bp->b_data >= unmapped_buf +
With the removal of b_saveaddr in the r285819, b_data must be reset to b_kvabase when the buffer is reclaimed. Otherwise, if b_data for the mapped buffer was adjusted with the page-offset portion of b_offset, nothing would re-adjust the b_data, which breaks buffer management code which expects page-aligned b_data (see e.g. bpman_qenter(), which skips partial pages). Fix a minor issue with the GB_KVAALLOC requests, which could result in returning the mapped buffer if the reused buffer is mapped and have the right amount of KVA reserved. Improve assertion in the vfs_buf_check_mapped() to catch unmapped buffers which have their b_data incorrectly adjusted with offset. Reported and tested by: pho (previous version) Reviewed by: jeff (previous version) Sponsored by: The FreeBSD Foundation
2015-07-25 15:00:14 +00:00
MAXPHYS, ("b_data + b_offset unmapped %p", bp));
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
Move bufshutdown() out of the #ifdef INVARIANTS block.
2015-07-29 09:57:34 +00:00
static inline void
vfs_buf_check_unmapped(struct buf *bp)
{
KASSERT(bp->b_data == unmapped_buf,
("unmapped buf: corrupted b_data %p", bp));
}
#define BUF_CHECK_MAPPED(bp) vfs_buf_check_mapped(bp)
#define BUF_CHECK_UNMAPPED(bp) vfs_buf_check_unmapped(bp)
#else
#define BUF_CHECK_MAPPED(bp) do {} while (0)
#define BUF_CHECK_UNMAPPED(bp) do {} while (0)
#endif
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
static int
isbufbusy(struct buf *bp)
{
bufs: make B_DIRTY and B_PERSISTENT flags available It appears these flags were related to ext2fs but are completely unused nowadays. Retire them. Suggested by: mckusick
2016-04-29 16:32:28 +00:00
if (((bp->b_flags & B_INVAL) == 0 && BUF_ISLOCKED(bp)) ||
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI))
return (1);
return (0);
}
/*
* Shutdown the system cleanly to prepare for reboot, halt, or power off.
*/
void
bufshutdown(int show_busybufs)
{
static int first_buf_printf = 1;
struct buf *bp;
int iter, nbusy, pbusy;
#ifndef PREEMPTION
int subiter;
#endif
/*
* Sync filesystems for shutdown
*/
wdog_kern_pat(WD_LASTVAL);
sys_sync(curthread, NULL);
/*
* With soft updates, some buffers that are
* written will be remarked as dirty until other
* buffers are written.
*/
for (iter = pbusy = 0; iter < 20; iter++) {
nbusy = 0;
for (bp = &buf[nbuf]; --bp >= buf; )
if (isbufbusy(bp))
nbusy++;
if (nbusy == 0) {
if (first_buf_printf)
printf("All buffers synced.");
break;
}
if (first_buf_printf) {
printf("Syncing disks, buffers remaining... ");
first_buf_printf = 0;
}
printf("%d ", nbusy);
if (nbusy < pbusy)
iter = 0;
pbusy = nbusy;
wdog_kern_pat(WD_LASTVAL);
sys_sync(curthread, NULL);
#ifdef PREEMPTION
/*
bufshutdown is no longer called with Giant held, so there's no need to drop or pickup Giant anymore. Remove that code and adjust comments.
2018-03-21 14:46:59 +00:00
* Spin for a while to allow interrupt threads to run.
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
*/
DELAY(50000 * iter);
#else
/*
bufshutdown is no longer called with Giant held, so there's no need to drop or pickup Giant anymore. Remove that code and adjust comments.
2018-03-21 14:46:59 +00:00
* Context switch several times to allow interrupt
* threads to run.
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
*/
for (subiter = 0; subiter < 50 * iter; subiter++) {
thread_lock(curthread);
mi_switch(SW_VOL, NULL);
thread_unlock(curthread);
DELAY(1000);
}
#endif
}
printf("\n");
/*
* Count only busy local buffers to prevent forcing
* a fsck if we're just a client of a wedged NFS server
*/
nbusy = 0;
for (bp = &buf[nbuf]; --bp >= buf; ) {
if (isbufbusy(bp)) {
#if 0
/* XXX: This is bogus. We should probably have a BO_REMOTE flag instead */
if (bp->b_dev == NULL) {
TAILQ_REMOVE(&mountlist,
bp->b_vp->v_mount, mnt_list);
continue;
}
#endif
nbusy++;
if (show_busybufs > 0) {
printf(
"%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
nbusy, bp, bp->b_vp, bp->b_flags,
(intmax_t)bp->b_blkno,
(intmax_t)bp->b_lblkno);
BUF_LOCKPRINTINFO(bp);
if (show_busybufs > 1)
vn_printf(bp->b_vp,
"vnode content: ");
}
}
}
if (nbusy) {
/*
* Failed to sync all blocks. Indicate this and don't
* unmount filesystems (thus forcing an fsck on reboot).
*/
printf("Giving up on %d buffers\n", nbusy);
DELAY(5000000); /* 5 seconds */
} else {
if (!first_buf_printf)
printf("Final sync complete\n");
/*
* Unmount filesystems
*/
Don't use 0 for pointer comparison Use NULL instead of 0 for comparison with panicstr. MFC after: 1 week Sponsored by: Multiplay
2015-12-08 18:38:33 +00:00
if (panicstr == NULL)
- Make 'struct buf *buf' private to vfs_bio.c. Having a global variable 'buf' is inconvenient and has lead me to some irritating to discover bugs over the years. It also makes it more challenging to refactor the buf allocation system. - Move swbuf and declare it as an extern in vfs_bio.c. This is still not perfect but better than it was before. - Eliminate the unused ffs function that relied on knowledge of the buf array. - Move the shutdown code that iterates over the buf array into vfs_bio.c. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2015-07-29 02:26:57 +00:00
vfs_unmountall();
}
swapoff_all();
DELAY(100000); /* wait for console output to finish */
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
static void
bpmap_qenter(struct buf *bp)
{
BUF_CHECK_MAPPED(bp);
/*
* bp->b_data is relative to bp->b_offset, but
* bp->b_offset may be offset into the first page.
*/
bp->b_data = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages);
bp->b_data = (caddr_t)((vm_offset_t)bp->b_data |
(vm_offset_t)(bp->b_offset & PAGE_MASK));
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
static inline struct bufdomain *
bufdomain(struct buf *bp)
{
return (&bdomain[bp->b_domain]);
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
static struct bufqueue *
bufqueue(struct buf *bp)
{
switch (bp->b_qindex) {
case QUEUE_NONE:
/* FALLTHROUGH */
case QUEUE_SENTINEL:
return (NULL);
case QUEUE_EMPTY:
return (&bqempty);
case QUEUE_DIRTY:
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
return (&bufdomain(bp)->bd_dirtyq);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
case QUEUE_CLEAN:
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
return (&bufdomain(bp)->bd_subq[bp->b_subqueue]);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
default:
break;
}
panic("bufqueue(%p): Unhandled type %d\n", bp, bp->b_qindex);
}
/*
* Return the locked bufqueue that bp is a member of.
*/
static struct bufqueue *
bufqueue_acquire(struct buf *bp)
{
struct bufqueue *bq, *nbq;
/*
* bp can be pushed from a per-cpu queue to the
* cleanq while we're waiting on the lock. Retry
* if the queues don't match.
*/
bq = bufqueue(bp);
BQ_LOCK(bq);
for (;;) {
nbq = bufqueue(bp);
if (bq == nbq)
break;
BQ_UNLOCK(bq);
BQ_LOCK(nbq);
bq = nbq;
}
return (bq);
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* binsfree:
*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* Insert the buffer into the appropriate free list. Requires a
* locked buffer on entry and buffer is unlocked before return.
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
*/
static void
binsfree(struct buf *bp, int qindex)
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufdomain *bd;
struct bufqueue *bq;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
KASSERT(qindex == QUEUE_CLEAN || qindex == QUEUE_DIRTY,
("binsfree: Invalid qindex %d", qindex));
BUF_ASSERT_XLOCKED(bp);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* Handle delayed bremfree() processing.
*/
Only take previous buffer queue lock (olock) when needed for REMFREE in binsfree(). Submitted by: Conrad Meyer Sponsored by: EMC / Isilon Storage Division Review: https://reviews.freebsd.org/D2882 MFC after: 1 week
2015-06-23 06:12:14 +00:00
if (bp->b_flags & B_REMFREE) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (bp->b_qindex == qindex) {
bp->b_flags |= B_REUSE;
bp->b_flags &= ~B_REMFREE;
BUF_UNLOCK(bp);
return;
Only take previous buffer queue lock (olock) when needed for REMFREE in binsfree(). Submitted by: Conrad Meyer Sponsored by: EMC / Isilon Storage Division Review: https://reviews.freebsd.org/D2882 MFC after: 1 week
2015-06-23 06:12:14 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq = bufqueue_acquire(bp);
bq_remove(bq, bp);
BQ_UNLOCK(bq);
}
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = bufdomain(bp);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (qindex == QUEUE_CLEAN) {
if (bd->bd_lim != 0)
bq = &bd->bd_subq[PCPU_GET(cpuid)];
else
bq = bd->bd_cleanq;
Only take previous buffer queue lock (olock) when needed for REMFREE in binsfree(). Submitted by: Conrad Meyer Sponsored by: EMC / Isilon Storage Division Review: https://reviews.freebsd.org/D2882 MFC after: 1 week
2015-06-23 06:12:14 +00:00
} else
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bq = &bd->bd_dirtyq;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq_insert(bq, bp, true);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
/*
* buf_free:
*
* Free a buffer to the buf zone once it no longer has valid contents.
*/
static void
buf_free(struct buf *bp)
{
if (bp->b_flags & B_REMFREE)
bremfreef(bp);
if (bp->b_vflags & BV_BKGRDINPROG)
panic("losing buffer 1");
if (bp->b_rcred != NOCRED) {
crfree(bp->b_rcred);
bp->b_rcred = NOCRED;
}
if (bp->b_wcred != NOCRED) {
crfree(bp->b_wcred);
bp->b_wcred = NOCRED;
}
if (!LIST_EMPTY(&bp->b_dep))
buf_deallocate(bp);
bufkva_free(bp);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
atomic_add_int(&bufdomain(bp)->bd_freebuffers, 1);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
BUF_UNLOCK(bp);
uma_zfree(buf_zone, bp);
}
/*
* buf_import:
*
* Import bufs into the uma cache from the buf list. The system still
* expects a static array of bufs and much of the synchronization
* around bufs assumes type stable storage. As a result, UMA is used
* only as a per-cpu cache of bufs still maintained on a global list.
*/
static int
Implement NUMA support in uma(9) and malloc(9). Allocations from specific domains can be done by the _domain() API variants. UMA also supports a first-touch policy via the NUMA zone flag. The slab layer is now segregated by VM domains and is precise. It handles iteration for round-robin directly. The per-cpu cache layer remains a mix of domains according to where memory is allocated and freed. Well behaved clients can achieve perfect locality with no performance penalty. The direct domain allocation functions have to visit the slab layer and so require per-zone locks which come at some expense. Reviewed by: Attilio (a slightly older version) Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon
2018-01-12 23:25:05 +00:00
buf_import(void *arg, void **store, int cnt, int domain, int flags)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
struct buf *bp;
int i;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_LOCK(&bqempty);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
for (i = 0; i < cnt; i++) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bp = TAILQ_FIRST(&bqempty.bq_queue);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (bp == NULL)
break;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq_remove(&bqempty, bp);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
store[i] = bp;
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_UNLOCK(&bqempty);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return (i);
}
/*
* buf_release:
*
* Release bufs from the uma cache back to the buffer queues.
*/
static void
buf_release(void *arg, void **store, int cnt)
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufqueue *bq;
struct buf *bp;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
int i;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq = &bqempty;
BQ_LOCK(bq);
for (i = 0; i < cnt; i++) {
bp = store[i];
/* Inline bq_insert() to batch locking. */
TAILQ_INSERT_TAIL(&bq->bq_queue, bp, b_freelist);
bp->b_flags &= ~(B_AGE | B_REUSE);
bq->bq_len++;
bp->b_qindex = bq->bq_index;
}
BQ_UNLOCK(bq);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
}
/*
* buf_alloc:
*
* Allocate an empty buffer header.
*/
static struct buf *
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
buf_alloc(struct bufdomain *bd)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
struct buf *bp;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
int freebufs;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
/*
* We can only run out of bufs in the buf zone if the average buf
* is less than BKVASIZE. In this case the actual wait/block will
* come from buf_reycle() failing to flush one of these small bufs.
*/
bp = NULL;
freebufs = atomic_fetchadd_int(&bd->bd_freebuffers, -1);
if (freebufs > 0)
bp = uma_zalloc(buf_zone, M_NOWAIT);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (bp == NULL) {
Use plain atomic_{add,subtract} when that's sufficient. CID: 1386920 MFC after: 2 weeks
2018-11-06 17:32:25 +00:00
atomic_add_int(&bd->bd_freebuffers, 1);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_daemon_wakeup(bd);
counter_u64_add(numbufallocfails, 1);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return (NULL);
}
/*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* Wake-up the bufspace daemon on transition below threshold.
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
*/
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (freebufs == bd->bd_lofreebuffers)
bufspace_daemon_wakeup(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
panic("getnewbuf_empty: Locked buf %p on free queue.", bp);
KASSERT(bp->b_vp == NULL,
("bp: %p still has vnode %p.", bp, bp->b_vp));
KASSERT((bp->b_flags & (B_DELWRI | B_NOREUSE)) == 0,
("invalid buffer %p flags %#x", bp, bp->b_flags));
KASSERT((bp->b_xflags & (BX_VNCLEAN|BX_VNDIRTY)) == 0,
("bp: %p still on a buffer list. xflags %X", bp, bp->b_xflags));
KASSERT(bp->b_npages == 0,
("bp: %p still has %d vm pages\n", bp, bp->b_npages));
KASSERT(bp->b_kvasize == 0, ("bp: %p still has kva\n", bp));
KASSERT(bp->b_bufsize == 0, ("bp: %p still has bufspace\n", bp));
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bp->b_domain = BD_DOMAIN(bd);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bp->b_flags = 0;
bp->b_ioflags = 0;
bp->b_xflags = 0;
bp->b_vflags = 0;
bp->b_vp = NULL;
bp->b_blkno = bp->b_lblkno = 0;
bp->b_offset = NOOFFSET;
bp->b_iodone = 0;
bp->b_error = 0;
bp->b_resid = 0;
bp->b_bcount = 0;
bp->b_npages = 0;
bp->b_dirtyoff = bp->b_dirtyend = 0;
bp->b_bufobj = NULL;
bp->b_data = bp->b_kvabase = unmapped_buf;
bp->b_fsprivate1 = NULL;
bp->b_fsprivate2 = NULL;
bp->b_fsprivate3 = NULL;
LIST_INIT(&bp->b_dep);
return (bp);
}
/*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* buf_recycle:
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
*
* Free a buffer from the given bufqueue. kva controls whether the
* freed buf must own some kva resources. This is used for
* defragmenting.
*/
static int
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
buf_recycle(struct bufdomain *bd, bool kva)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufqueue *bq;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
struct buf *bp, *nbp;
if (kva)
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
counter_u64_add(bufdefragcnt, 1);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
nbp = NULL;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq = bd->bd_cleanq;
BQ_LOCK(bq);
KASSERT(BQ_LOCKPTR(bq) == BD_LOCKPTR(bd),
("buf_recycle: Locks don't match"));
nbp = TAILQ_FIRST(&bq->bq_queue);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* Run scan, possibly freeing data and/or kva mappings on the fly
* depending.
*/
while ((bp = nbp) != NULL) {
/*
* Calculate next bp (we can only use it if we do not
* release the bqlock).
*/
nbp = TAILQ_NEXT(bp, b_freelist);
/*
* If we are defragging then we need a buffer with
* some kva to reclaim.
*/
if (kva && bp->b_kvasize == 0)
continue;
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
continue;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
/*
* Implement a second chance algorithm for frequently
* accessed buffers.
*/
if ((bp->b_flags & B_REUSE) != 0) {
TAILQ_REMOVE(&bq->bq_queue, bp, b_freelist);
TAILQ_INSERT_TAIL(&bq->bq_queue, bp, b_freelist);
bp->b_flags &= ~B_REUSE;
BUF_UNLOCK(bp);
continue;
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* Skip buffers with background writes in progress.
*/
if ((bp->b_vflags & BV_BKGRDINPROG) != 0) {
BUF_UNLOCK(bp);
continue;
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
KASSERT(bp->b_qindex == QUEUE_CLEAN,
("buf_recycle: inconsistent queue %d bp %p",
bp->b_qindex, bp));
KASSERT(bp->b_domain == BD_DOMAIN(bd),
("getnewbuf: queue domain %d doesn't match request %d",
bp->b_domain, (int)BD_DOMAIN(bd)));
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* NOTE: nbp is now entirely invalid. We can only restart
* the scan from this point on.
*/
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq_remove(bq, bp);
BQ_UNLOCK(bq);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* Requeue the background write buffer with error and
* restart the scan.
*/
if ((bp->b_vflags & BV_BKGRDERR) != 0) {
bqrelse(bp);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_LOCK(bq);
nbp = TAILQ_FIRST(&bq->bq_queue);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
continue;
}
bp->b_flags |= B_INVAL;
brelse(bp);
return (0);
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bd->bd_wanted = 1;
BQ_UNLOCK(bq);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return (ENOBUFS);
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
* bremfree:
*
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
* Mark the buffer for removal from the appropriate free list.
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
*
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
bremfree(struct buf *bp)
- Introduce a new function bremfreel() that does a bremfree with the buf queue lock already held. - In getblk() and flushbufqueues() use bremfreel() while we still have the buf queue lock held to keep the lists consistent. - Add LK_NOWAIT to two cases where we're essentially asserting that the bufs are not locked while acquiring the locks. This will make sure that we get the appropriate panic() and not another one for sleeping with a lock held.
2003-02-16 10:43:06 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bremfree(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
- Split one KASSERT in bremfree() into two to aid in debugging. Sponsored by: Isilon Systems, Inc.
2005-06-13 00:45:05 +00:00
KASSERT((bp->b_flags & B_REMFREE) == 0,
("bremfree: buffer %p already marked for delayed removal.", bp));
KASSERT(bp->b_qindex != QUEUE_NONE,
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
("bremfree: buffer %p not on a queue.", bp));
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BUF_ASSERT_XLOCKED(bp);
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
bp->b_flags |= B_REMFREE;
}
/*
* bremfreef:
*
* Force an immediate removal from a free list. Used only in nfs when
* it abuses the b_freelist pointer.
*/
void
bremfreef(struct buf *bp)
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufqueue *bq;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq = bufqueue_acquire(bp);
bq_remove(bq, bp);
BQ_UNLOCK(bq);
}
static void
bq_init(struct bufqueue *bq, int qindex, int subqueue, const char *lockname)
{
mtx_init(&bq->bq_lock, lockname, NULL, MTX_DEF);
TAILQ_INIT(&bq->bq_queue);
bq->bq_len = 0;
bq->bq_index = qindex;
bq->bq_subqueue = subqueue;
}
static void
bd_init(struct bufdomain *bd)
{
int i;
Fix issues with sparse cpu allocation. Consistently use mp_maxid + 1. Reported by: pho Reviewed by: markj Sponsored by: Netflix, Dell/EMC Isilon
2018-02-25 00:35:21 +00:00
bd->bd_cleanq = &bd->bd_subq[mp_maxid + 1];
bq_init(bd->bd_cleanq, QUEUE_CLEAN, mp_maxid + 1, "bufq clean lock");
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bq_init(&bd->bd_dirtyq, QUEUE_DIRTY, -1, "bufq dirty lock");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
for (i = 0; i <= mp_maxid; i++)
bq_init(&bd->bd_subq[i], QUEUE_CLEAN, i,
"bufq clean subqueue lock");
mtx_init(&bd->bd_run_lock, "bufspace daemon run lock", NULL, MTX_DEF);
- Introduce a new function bremfreel() that does a bremfree with the buf queue lock already held. - In getblk() and flushbufqueues() use bremfreel() while we still have the buf queue lock held to keep the lists consistent. - Add LK_NOWAIT to two cases where we're essentially asserting that the bufs are not locked while acquiring the locks. This will make sure that we get the appropriate panic() and not another one for sleeping with a lock held.
2003-02-16 10:43:06 +00:00
}
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
/*
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
* bq_remove:
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
*
* Removes a buffer from the free list, must be called with the
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
* correct qlock held.
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
*/
Make various vnode related functions static
2005-02-10 12:28:58 +00:00
static void
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq_remove(struct bufqueue *bq, struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Add INVARIANTS checking that numfreebufs values are sane. Also add a per-buf flag to catch if a buf is double-counted in the free count. This code was useful to debug an instance where a local patch at Isilon was incorrectly managing numfreebufs for a new buf state. Reviewed by: jeff Approved by: zml (mentor)
2010-06-11 17:03:26 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
CTR3(KTR_BUF, "bq_remove(%p) vp %p flags %X",
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
bp, bp->b_vp, bp->b_flags);
KASSERT(bp->b_qindex != QUEUE_NONE,
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
("bq_remove: buffer %p not on a queue.", bp));
KASSERT(bufqueue(bp) == bq,
("bq_remove: Remove buffer %p from wrong queue.", bp));
BQ_ASSERT_LOCKED(bq);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (bp->b_qindex != QUEUE_EMPTY) {
BUF_ASSERT_XLOCKED(bp);
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
KASSERT(bq->bq_len >= 1,
("queue %d underflow", bp->b_qindex));
TAILQ_REMOVE(&bq->bq_queue, bp, b_freelist);
bq->bq_len--;
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
bp->b_qindex = QUEUE_NONE;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bp->b_flags &= ~(B_REMFREE | B_REUSE);
}
static void
bd_flush(struct bufdomain *bd, struct bufqueue *bq)
{
struct buf *bp;
BQ_ASSERT_LOCKED(bq);
if (bq != bd->bd_cleanq) {
BD_LOCK(bd);
while ((bp = TAILQ_FIRST(&bq->bq_queue)) != NULL) {
TAILQ_REMOVE(&bq->bq_queue, bp, b_freelist);
TAILQ_INSERT_TAIL(&bd->bd_cleanq->bq_queue, bp,
b_freelist);
Fix issues with sparse cpu allocation. Consistently use mp_maxid + 1. Reported by: pho Reviewed by: markj Sponsored by: Netflix, Dell/EMC Isilon
2018-02-25 00:35:21 +00:00
bp->b_subqueue = bd->bd_cleanq->bq_subqueue;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
}
bd->bd_cleanq->bq_len += bq->bq_len;
bq->bq_len = 0;
}
if (bd->bd_wanted) {
bd->bd_wanted = 0;
wakeup(&bd->bd_wanted);
}
if (bq != bd->bd_cleanq)
BD_UNLOCK(bd);
}
static int
bd_flushall(struct bufdomain *bd)
{
struct bufqueue *bq;
int flushed;
int i;
if (bd->bd_lim == 0)
return (0);
flushed = 0;
Fix issues with sparse cpu allocation. Consistently use mp_maxid + 1. Reported by: pho Reviewed by: markj Sponsored by: Netflix, Dell/EMC Isilon
2018-02-25 00:35:21 +00:00
for (i = 0; i <= mp_maxid; i++) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bq = &bd->bd_subq[i];
if (bq->bq_len == 0)
continue;
BQ_LOCK(bq);
bd_flush(bd, bq);
BQ_UNLOCK(bq);
flushed++;
}
return (flushed);
}
static void
bq_insert(struct bufqueue *bq, struct buf *bp, bool unlock)
{
struct bufdomain *bd;
if (bp->b_qindex != QUEUE_NONE)
panic("bq_insert: free buffer %p onto another queue?", bp);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = bufdomain(bp);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (bp->b_flags & B_AGE) {
/* Place this buf directly on the real queue. */
if (bq->bq_index == QUEUE_CLEAN)
bq = bd->bd_cleanq;
BQ_LOCK(bq);
TAILQ_INSERT_HEAD(&bq->bq_queue, bp, b_freelist);
} else {
BQ_LOCK(bq);
TAILQ_INSERT_TAIL(&bq->bq_queue, bp, b_freelist);
}
bp->b_flags &= ~(B_AGE | B_REUSE);
bq->bq_len++;
bp->b_qindex = bq->bq_index;
bp->b_subqueue = bq->bq_subqueue;
/*
* Unlock before we notify so that we don't wakeup a waiter that
* fails a trylock on the buf and sleeps again.
*/
if (unlock)
BUF_UNLOCK(bp);
if (bp->b_qindex == QUEUE_CLEAN) {
/*
* Flush the per-cpu queue and notify any waiters.
*/
if (bd->bd_wanted || (bq != bd->bd_cleanq &&
bq->bq_len >= bd->bd_lim))
bd_flush(bd, bq);
}
BQ_UNLOCK(bq);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
/*
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* bufkva_free:
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
*
* Free the kva allocation for a buffer.
*
*/
static void
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufkva_free(struct buf *bp)
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
{
#ifdef INVARIANTS
if (bp->b_kvasize == 0) {
KASSERT(bp->b_kvabase == unmapped_buf &&
bp->b_data == unmapped_buf,
("Leaked KVA space on %p", bp));
} else if (buf_mapped(bp))
BUF_CHECK_MAPPED(bp);
else
BUF_CHECK_UNMAPPED(bp);
#endif
if (bp->b_kvasize == 0)
return;
vmem_free(buffer_arena, (vm_offset_t)bp->b_kvabase, bp->b_kvasize);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
counter_u64_add(bufkvaspace, -bp->b_kvasize);
counter_u64_add(buffreekvacnt, 1);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
bp->b_data = bp->b_kvabase = unmapped_buf;
bp->b_kvasize = 0;
}
/*
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* bufkva_alloc:
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
*
* Allocate the buffer KVA and set b_kvasize and b_kvabase.
*/
static int
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufkva_alloc(struct buf *bp, int maxsize, int gbflags)
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
{
vm_offset_t addr;
int error;
KASSERT((gbflags & GB_UNMAPPED) == 0 || (gbflags & GB_KVAALLOC) != 0,
("Invalid gbflags 0x%x in %s", gbflags, __func__));
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufkva_free(bp);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
addr = 0;
error = vmem_alloc(buffer_arena, maxsize, M_BESTFIT | M_NOWAIT, &addr);
if (error != 0) {
/*
* Buffer map is too fragmented. Request the caller
* to defragment the map.
*/
return (error);
}
bp->b_kvabase = (caddr_t)addr;
bp->b_kvasize = maxsize;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
counter_u64_add(bufkvaspace, bp->b_kvasize);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if ((gbflags & GB_UNMAPPED) != 0) {
bp->b_data = unmapped_buf;
BUF_CHECK_UNMAPPED(bp);
} else {
bp->b_data = bp->b_kvabase;
BUF_CHECK_MAPPED(bp);
}
return (0);
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
/*
* bufkva_reclaim:
*
* Reclaim buffer kva by freeing buffers holding kva. This is a vmem
* callback that fires to avoid returning failure.
*/
static void
bufkva_reclaim(vmem_t *vmem, int flags)
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bool done;
int q;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
int i;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
done = false;
for (i = 0; i < 5; i++) {
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
for (q = 0; q < buf_domains; q++)
if (buf_recycle(&bdomain[q], true) != 0)
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
done = true;
if (done)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
break;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return;
}
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
/*
* Attempt to initiate asynchronous I/O on read-ahead blocks. We must
* clear BIO_ERROR and B_INVAL prior to initiating I/O . If B_CACHE is set,
* the buffer is valid and we do not have to do anything.
*/
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
static void
breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt,
struct ucred * cred, int flags, void (*ckhashfunc)(struct buf *))
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
{
struct buf *rabp;
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
struct thread *td;
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
int i;
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
td = curthread;
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
for (i = 0; i < cnt; i++, rablkno++, rabsize++) {
if (inmem(vp, *rablkno))
continue;
rabp = getblk(vp, *rablkno, *rabsize, 0, 0, 0);
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
if ((rabp->b_flags & B_CACHE) != 0) {
brelse(rabp);
continue;
}
Add four new RCTL resources - readbps, readiops, writebps and writeiops, for limiting disk (actually filesystem) IO. Note that in some cases these limits are not quite precise. It's ok, as long as it's within some reasonable bounds. Testing - and review of the code, in particular the VFS and VM parts - is very welcome. MFC after: 1 month Relnotes: yes Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D5080
2016-04-07 04:23:25 +00:00
#ifdef RACCT
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
if (racct_enable) {
PROC_LOCK(curproc);
racct_add_buf(curproc, rabp, 0);
PROC_UNLOCK(curproc);
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
}
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
#endif /* RACCT */
td->td_ru.ru_inblock++;
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
rabp->b_flags |= B_ASYNC;
rabp->b_flags &= ~B_INVAL;
if ((flags & GB_CKHASH) != 0) {
rabp->b_flags |= B_CKHASH;
rabp->b_ckhashcalc = ckhashfunc;
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
}
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
rabp->b_ioflags &= ~BIO_ERROR;
rabp->b_iocmd = BIO_READ;
if (rabp->b_rcred == NOCRED && cred != NOCRED)
rabp->b_rcred = crhold(cred);
vfs_busy_pages(rabp, 0);
BUF_KERNPROC(rabp);
rabp->b_iooffset = dbtob(rabp->b_blkno);
bstrategy(rabp);
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
}
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
This change avoids a kernel deadlock on "snaplk" when using snapshots on UFS filesystems running with journaled soft updates. This is the first of several bugs that need to be fixed before removing the restriction added in -r230250 to prevent the use of snapshots on filesystems running with journaled soft updates. The deadlock occurs when holding the snapshot lock (snaplk) and then trying to flush an inode via ffs_update(). We become blocked by another process trying to flush a different inode contained in the same inode block that we need. It holds the inode block for which we are waiting locked. When it tries to write the inode block, it gets blocked waiting for the our snaplk when it calls ffs_copyonwrite() to see if the inode block needs to be copied in our snapshot. The most obvious place that this deadlock arises is in the ffs_copyonwrite() routine when it updates critical metadata in a snapshot and tries to write it out before proceeding. The fix here is to write the data and indirect block pointer for the snapshot, but to skip the call to ffs_update() to write the snapshot inode. To ensure that we will never have to update a pointer in the inode itself, the ffs_snapshot() routine that creates the snapshot has to ensure that all the direct blocks are allocated as part of the creation of the snapshot. A less obvious place that this deadlock occurs is when we hold the snaplk because we are deleting a snapshot. In the course of doing the deletion, we need to allocate various soft update dependency structures and allocate some journal space. If we hit a resource limit while doing this we decrease the resources in use by flushing out an existing dirty file to get it to give up the soft dependency resources that it holds. The flush can cause an ffs_update() to be done on the inode for the file that we have selected to flush resulting in the same deadlock as described above when the inode that we have chosen to flush resides in the same inode block as the snapshot inode that we hold. The fix is to defer cleaning up any time that the inode on which we are operating is a snapshot. Help and review by: Jeff Roberson Tested by: Peter Holm MFC (to 9 only) after: 2 weeks
2012-03-01 18:45:25 +00:00
* Entry point for bread() and breadn() via #defines in sys/buf.h.
*
* Get a buffer with the specified data. Look in the cache first. We
* must clear BIO_ERROR and B_INVAL prior to initiating I/O. If B_CACHE
* is set, the buffer is valid and we do not have to do anything, see
* getblk(). Also starts asynchronous I/O on read-ahead blocks.
The bread() function was inconsistent about whether it would return a buffer pointer in the event of an error (for some errors it would return a buffer pointer and for other errors it would not return a buffer pointer). The cluster_read() function was similarly inconsistent. Clients of these functions were inconsistent in handling errors. Some would assume that no buffer was returned after an error and would thus lose buffers under certain error conditions. Others would assume that brelse() should always be called after an error and would thus panic the system under certain error conditions. To correct both of these problems with minimal code churn, bread() and cluster_write() now always free the buffer when returning an error thus ensuring that buffers will never be lost. The brelse() routine checks for being passed a NULL buffer pointer and silently returns to avoid panics. Thus both approaches to handling error returns from bread() and cluster_read() will work correctly. Future code should be written assuming that bread() and cluster_read() will never return a buffer with an error, so should not attempt to brelse() the buffer when an error is returned. Reviewed by: kib
2016-01-27 21:23:01 +00:00
*
* Always return a NULL buffer pointer (in bpp) when returning an error.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
int
Some style fixes. Sponsored by: The FreeBSD Foundation
2013-03-14 20:31:39 +00:00
breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno,
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
int *rabsize, int cnt, struct ucred *cred, int flags,
void (*ckhashfunc)(struct buf *), struct buf **bpp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
struct buf *bp;
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
struct thread *td;
int error, readwait, rv;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "breadn(%p, %jd, %d)", vp, blkno, size);
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
td = curthread;
This change avoids a kernel deadlock on "snaplk" when using snapshots on UFS filesystems running with journaled soft updates. This is the first of several bugs that need to be fixed before removing the restriction added in -r230250 to prevent the use of snapshots on filesystems running with journaled soft updates. The deadlock occurs when holding the snapshot lock (snaplk) and then trying to flush an inode via ffs_update(). We become blocked by another process trying to flush a different inode contained in the same inode block that we need. It holds the inode block for which we are waiting locked. When it tries to write the inode block, it gets blocked waiting for the our snaplk when it calls ffs_copyonwrite() to see if the inode block needs to be copied in our snapshot. The most obvious place that this deadlock arises is in the ffs_copyonwrite() routine when it updates critical metadata in a snapshot and tries to write it out before proceeding. The fix here is to write the data and indirect block pointer for the snapshot, but to skip the call to ffs_update() to write the snapshot inode. To ensure that we will never have to update a pointer in the inode itself, the ffs_snapshot() routine that creates the snapshot has to ensure that all the direct blocks are allocated as part of the creation of the snapshot. A less obvious place that this deadlock occurs is when we hold the snaplk because we are deleting a snapshot. In the course of doing the deletion, we need to allocate various soft update dependency structures and allocate some journal space. If we hit a resource limit while doing this we decrease the resources in use by flushing out an existing dirty file to get it to give up the soft dependency resources that it holds. The flush can cause an ffs_update() to be done on the inode for the file that we have selected to flush resulting in the same deadlock as described above when the inode that we have chosen to flush resides in the same inode block as the snapshot inode that we hold. The fix is to defer cleaning up any time that the inode on which we are operating is a snapshot. Help and review by: Jeff Roberson Tested by: Peter Holm MFC (to 9 only) after: 2 weeks
2012-03-01 18:45:25 +00:00
/*
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
* Can only return NULL if GB_LOCK_NOWAIT or GB_SPARSE flags
* are specified.
This change avoids a kernel deadlock on "snaplk" when using snapshots on UFS filesystems running with journaled soft updates. This is the first of several bugs that need to be fixed before removing the restriction added in -r230250 to prevent the use of snapshots on filesystems running with journaled soft updates. The deadlock occurs when holding the snapshot lock (snaplk) and then trying to flush an inode via ffs_update(). We become blocked by another process trying to flush a different inode contained in the same inode block that we need. It holds the inode block for which we are waiting locked. When it tries to write the inode block, it gets blocked waiting for the our snaplk when it calls ffs_copyonwrite() to see if the inode block needs to be copied in our snapshot. The most obvious place that this deadlock arises is in the ffs_copyonwrite() routine when it updates critical metadata in a snapshot and tries to write it out before proceeding. The fix here is to write the data and indirect block pointer for the snapshot, but to skip the call to ffs_update() to write the snapshot inode. To ensure that we will never have to update a pointer in the inode itself, the ffs_snapshot() routine that creates the snapshot has to ensure that all the direct blocks are allocated as part of the creation of the snapshot. A less obvious place that this deadlock occurs is when we hold the snaplk because we are deleting a snapshot. In the course of doing the deletion, we need to allocate various soft update dependency structures and allocate some journal space. If we hit a resource limit while doing this we decrease the resources in use by flushing out an existing dirty file to get it to give up the soft dependency resources that it holds. The flush can cause an ffs_update() to be done on the inode for the file that we have selected to flush resulting in the same deadlock as described above when the inode that we have chosen to flush resides in the same inode block as the snapshot inode that we hold. The fix is to defer cleaning up any time that the inode on which we are operating is a snapshot. Help and review by: Jeff Roberson Tested by: Peter Holm MFC (to 9 only) after: 2 weeks
2012-03-01 18:45:25 +00:00
*/
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
error = getblkx(vp, blkno, size, 0, 0, flags, &bp);
if (error != 0) {
*bpp = NULL;
return (error);
}
flags &= ~GB_NOSPARSE;
*bpp = bp;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
/*
* If not found in cache, do some I/O
*/
readwait = 0;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
if ((bp->b_flags & B_CACHE) == 0) {
Add four new RCTL resources - readbps, readiops, writebps and writeiops, for limiting disk (actually filesystem) IO. Note that in some cases these limits are not quite precise. It's ok, as long as it's within some reasonable bounds. Testing - and review of the code, in particular the VFS and VM parts - is very welcome. MFC after: 1 month Relnotes: yes Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D5080
2016-04-07 04:23:25 +00:00
#ifdef RACCT
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
if (racct_enable) {
PROC_LOCK(td->td_proc);
racct_add_buf(td->td_proc, bp, 0);
PROC_UNLOCK(td->td_proc);
Add four new RCTL resources - readbps, readiops, writebps and writeiops, for limiting disk (actually filesystem) IO. Note that in some cases these limits are not quite precise. It's ok, as long as it's within some reasonable bounds. Testing - and review of the code, in particular the VFS and VM parts - is very welcome. MFC after: 1 month Relnotes: yes Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D5080
2016-04-07 04:23:25 +00:00
}
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
#endif /* RACCT */
td->td_ru.ru_inblock++;
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
bp->b_iocmd = BIO_READ;
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
bp->b_flags &= ~B_INVAL;
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
if ((flags & GB_CKHASH) != 0) {
bp->b_flags |= B_CKHASH;
bp->b_ckhashcalc = ckhashfunc;
}
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
bp->b_ioflags &= ~BIO_ERROR;
Change the kernel's ucred API as follows: - crhold() returns a reference to the ucred whose refcount it bumps. - crcopy() now simply copies the credentials from one credential to another and has no return value. - a new crshared() primitive is added which returns true if a ucred's refcount is > 1 and false (0) otherwise.
2001-10-11 23:38:17 +00:00
if (bp->b_rcred == NOCRED && cred != NOCRED)
bp->b_rcred = crhold(cred);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
vfs_busy_pages(bp, 0);
Initialize b_iooffset before calling VOP_[SPEC]STRATEGY
2003-10-18 19:49:46 +00:00
bp->b_iooffset = dbtob(bp->b_blkno);
Move the buffer method vector (buf->b_op) to the bufobj. Extend it with a strategy method. Add bufstrategy() which do the usual VOP_SPECSTRATEGY/VOP_STRATEGY song and dance. Rename ibwrite to bufwrite(). Move the two NFS buf_ops to more sensible places, add bufstrategy to them. Add inlines for bwrite() and bstrategy() which calls through buf->b_bufobj->b_ops->b_{write,strategy}(). Replace almost all VOP_STRATEGY()/VOP_SPECSTRATEGY() calls with bstrategy().
2004-10-24 20:03:41 +00:00
bstrategy(bp);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
++readwait;
}
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
/*
* Attempt to initiate asynchronous I/O on read-ahead blocks.
*/
breada(vp, rablkno, rabsize, cnt, cred, flags, ckhashfunc);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Continuing efforts to provide hardening of FFS, this change adds a check hash to cylinder groups. If a check hash fails when a cylinder group is read, no further allocations are attempted in that cylinder group until it has been fixed by fsck. This avoids a class of filesystem panics related to corrupted cylinder group maps. The hash is done using crc32c. Check hases are added only to UFS2 and not to UFS1 as UFS1 is primarily used in embedded systems with small memories and low-powered processors which need as light-weight a filesystem as possible. Specifics of the changes: sys/sys/buf.h: Add BX_FSPRIV to reserve a set of eight b_xflags that may be used by individual filesystems for their own purpose. Their specific definitions are found in the header files for each filesystem that uses them. Also add fields to struct buf as noted below. sys/kern/vfs_bio.c: It is only necessary to compute a check hash for a cylinder group when it is actually read from disk. When calling bread, you do not know whether the buffer was found in the cache or read. So a new flag (GB_CKHASH) and a pointer to a function to perform the hash has been added to breadn_flags to say that the function should be called to calculate a hash if the data has been read. The check hash is placed in b_ckhash and the B_CKHASH flag is set to indicate that a read was done and a check hash calculated. Though a rather elaborate mechanism, it should also work for check hashing other metadata in the future. A kernel internal API change was to change breada into a static fucntion and add flags and a function pointer to a check-hash function. sys/ufs/ffs/fs.h: Add flags for types of check hashes; stored in a new word in the superblock. Define corresponding BX_ flags for the different types of check hashes. Add a check hash word in the cylinder group. sys/ufs/ffs/ffs_alloc.c: In ffs_getcg do the dance with breadn_flags to get a check hash and if one is provided, check it. sys/ufs/ffs/ffs_vfsops.c: Copy across the BX_FFSTYPES flags in background writes. Update the check hash when writing out buffers that need them. sys/ufs/ffs/ffs_snapshot.c: Recompute check hash when updating snapshot cylinder groups. sys/libkern/crc32.c: lib/libufs/Makefile: lib/libufs/libufs.h: lib/libufs/cgroup.c: Include libkern/crc32.c in libufs and use it to compute check hashes when updating cylinder groups. Four utilities are affected: sbin/newfs/mkfs.c: Add the check hashes when building the cylinder groups. sbin/fsck_ffs/fsck.h: sbin/fsck_ffs/fsutil.c: Verify and update check hashes when checking and writing cylinder groups. sbin/fsck_ffs/pass5.c: Offer to add check hashes to existing filesystems. Precompute check hashes when rebuilding cylinder group (although this will be done when it is written in fsutil.c it is necessary to do it early before comparing with the old cylinder group) sbin/dumpfs/dumpfs.c Print out the new check hash flag(s) sbin/fsdb/Makefile: Needs to add libufs now used by pass5.c imported from fsck_ffs. Reviewed by: kib Tested by: Peter Holm (pho)
2017-09-22 12:45:15 +00:00
rv = 0;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (readwait) {
s/biowait/bufwait/g Prodded by: several.
2000-04-29 16:25:22 +00:00
rv = bufwait(bp);
The bread() function was inconsistent about whether it would return a buffer pointer in the event of an error (for some errors it would return a buffer pointer and for other errors it would not return a buffer pointer). The cluster_read() function was similarly inconsistent. Clients of these functions were inconsistent in handling errors. Some would assume that no buffer was returned after an error and would thus lose buffers under certain error conditions. Others would assume that brelse() should always be called after an error and would thus panic the system under certain error conditions. To correct both of these problems with minimal code churn, bread() and cluster_write() now always free the buffer when returning an error thus ensuring that buffers will never be lost. The brelse() routine checks for being passed a NULL buffer pointer and silently returns to avoid panics. Thus both approaches to handling error returns from bread() and cluster_read() will work correctly. Future code should be written assuming that bread() and cluster_read() will never return a buffer with an error, so should not attempt to brelse() the buffer when an error is returned. Reviewed by: kib
2016-01-27 21:23:01 +00:00
if (rv != 0) {
brelse(bp);
*bpp = NULL;
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
}
return (rv);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
* Write, release buffer on completion. (Done by iodone
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* if async). Do not bother writing anything if the buffer
* is invalid.
*
* Note that we set B_CACHE here, indicating that buffer is
* fully valid and thus cacheable. This is true even of NFS
* now so we set it generally. This could be set either here
* or in biodone() since the I/O is synchronous. We put it
* here.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
int
Move the buffer method vector (buf->b_op) to the bufobj. Extend it with a strategy method. Add bufstrategy() which do the usual VOP_SPECSTRATEGY/VOP_STRATEGY song and dance. Rename ibwrite to bufwrite(). Move the two NFS buf_ops to more sensible places, add bufstrategy to them. Add inlines for bwrite() and bstrategy() which calls through buf->b_bufobj->b_ops->b_{write,strategy}(). Replace almost all VOP_STRATEGY()/VOP_SPECSTRATEGY() calls with bstrategy().
2004-10-24 20:03:41 +00:00
bufwrite(struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
- Remove long dead splbio() calls and comments relating to the old synchronization mechanism.
2005-04-30 12:18:50 +00:00
int oldflags;
In rev. 1.514, iodone on async buffer may happen before code checks the vnode v_flag. For cluster buffers this would result in dereferencing NULL b_vp. To prevent the panic, cache relevant vnode flag before calling bstrategy. Reported by: Peter Holm, kris Tested by: Peter Holm Reviewed by: tegge Pointy hat to: kib
2006-12-20 09:22:31 +00:00
struct vnode *vp;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
long space;
In rev. 1.514, iodone on async buffer may happen before code checks the vnode v_flag. For cluster buffers this would result in dereferencing NULL b_vp. To prevent the panic, cache relevant vnode flag before calling bstrategy. Reported by: Peter Holm, kris Tested by: Peter Holm Reviewed by: tegge Pointy hat to: kib
2006-12-20 09:22:31 +00:00
int vp_md;
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
Do not allow creation of the dirty buffers for the dead buffer objects, i.e. for buffer objects which vnode was reclaimed. Buffer cache cannot write such buffers. Return the error and discard the buffer immediately on write attempt. BO_DIRTY now always set during vnode reclamation, since it is used not only for the INVARIANTS checks. Do allow placement of the clean buffers on dead bufobj list, otherwise filesystems cannot use bufcache at all after the devvp reclaim. Reported and tested by: trasz Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2015-07-11 11:21:56 +00:00
if ((bp->b_bufobj->bo_flag & BO_DEAD) != 0) {
bp->b_flags |= B_INVAL | B_RELBUF;
bp->b_flags &= ~B_CACHE;
brelse(bp);
return (ENXIO);
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (bp->b_flags & B_INVAL) {
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
brelse(bp);
return (0);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
In kern_physio.c fix tsleep priority messup. In vfs_bio.c, remove b_generation count usage, remove redundant reassignbuf, remove redundant spl(s), manage page PG_ZERO flags more correctly, utilize in invalid value for b_offset until it is properly initialized. Add asserts for #ifdef DIAGNOSTIC, when b_offset is improperly used. when a process is not performing I/O, and just waiting on a buffer generally, make the sleep priority low. only check page validity in getblk for B_VMIO buffers. In vfs_cluster, add b_offset asserts, correct pointer calculation for clustered reads. Improve readability of certain parts of the code. Remove redundant spl(s). In vfs_subr, correct usage of vfs_bio_awrite (From Andrew Gallatin <gallatin@cs.duke.edu>). More vtruncbuf problems fixed.
1998-03-19 22:48:16 +00:00
Add barrier write capability to the VFS buffer interface. A barrier write is a disk write request that tells the disk that the buffer being written must be committed to the media along with any writes that preceeded it before any future blocks may be written to the drive. Barrier writes are provided by adding the functions bbarrierwrite (bwrite with barrier) and babarrierwrite (bawrite with barrier). Following a bbarrierwrite the client knows that the requested buffer is on the media. It does not ensure that buffers written before that buffer are on the media. It only ensure that buffers written before that buffer will get to the media before any buffers written after that buffer. A flush command must be sent to the disk to ensure that all earlier written buffers are on the media. Reviewed by: kib Tested by: Peter Holm
2013-02-16 14:51:30 +00:00
if (bp->b_flags & B_BARRIER)
Correct accuracy of the barrier writes accounting. Discussed with: mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation
2018-12-02 12:53:39 +00:00
atomic_add_long(&barrierwrites, 1);
Add barrier write capability to the VFS buffer interface. A barrier write is a disk write request that tells the disk that the buffer being written must be committed to the media along with any writes that preceeded it before any future blocks may be written to the drive. Barrier writes are provided by adding the functions bbarrierwrite (bwrite with barrier) and babarrierwrite (bawrite with barrier). Following a bbarrierwrite the client knows that the requested buffer is on the media. It does not ensure that buffers written before that buffer are on the media. It only ensure that buffers written before that buffer will get to the media before any buffers written after that buffer. A flush command must be sent to the disk to ensure that all earlier written buffers are on the media. Reviewed by: kib Tested by: Peter Holm
2013-02-16 14:51:30 +00:00
In kern_physio.c fix tsleep priority messup. In vfs_bio.c, remove b_generation count usage, remove redundant reassignbuf, remove redundant spl(s), manage page PG_ZERO flags more correctly, utilize in invalid value for b_offset until it is properly initialized. Add asserts for #ifdef DIAGNOSTIC, when b_offset is improperly used. when a process is not performing I/O, and just waiting on a buffer generally, make the sleep priority low. only check page validity in getblk for B_VMIO buffers. In vfs_cluster, add b_offset asserts, correct pointer calculation for clustered reads. Improve readability of certain parts of the code. Remove redundant spl(s). In vfs_subr, correct usage of vfs_bio_awrite (From Andrew Gallatin <gallatin@cs.duke.edu>). More vtruncbuf problems fixed.
1998-03-19 22:48:16 +00:00
oldflags = bp->b_flags;
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
Background writes are entirely an FFS/Softupdates thing. Give FFS vnodes a specific bufwrite method which contains all the background write stuff and then calls into the default bufwrite() for the rest of the job. Remove all the background write related stuff from the normal bufwrite. This drags the softdep_move_dependencies() back into FFS. Long term, it is worth looking at simply copying the data into allocated memory and issuing the bio directly and not create the "shadow buf" in the first place (just like copy-on-write is done in snapshots for instance). I don't think we really gain anything but complexity from doing this with a buf.
2005-02-08 20:29:10 +00:00
KASSERT(!(bp->b_vflags & BV_BKGRDINPROG),
("FFS background buffer should not get here %p", bp));
Several performance improvements for soft updates have been added: 1) Fastpath deletions. When a file is being deleted, check to see if it was so recently created that its inode has not yet been written to disk. If so, the delete can proceed to immediately free the inode. 2) Background writes: No file or block allocations can be done while the bitmap is being written to disk. To avoid these stalls, the bitmap is copied to another buffer which is written thus leaving the original available for futher allocations. 3) Link count tracking. Constantly track the difference in i_effnlink and i_nlink so that inodes that have had no change other than i_effnlink need not be written. 4) Identify buffers with rollback dependencies so that the buffer flushing daemon can choose to skip over them.
2000-01-10 00:24:24 +00:00
In rev. 1.514, iodone on async buffer may happen before code checks the vnode v_flag. For cluster buffers this would result in dereferencing NULL b_vp. To prevent the panic, cache relevant vnode flag before calling bstrategy. Reported by: Peter Holm, kris Tested by: Peter Holm Reviewed by: tegge Pointy hat to: kib
2006-12-20 09:22:31 +00:00
vp = bp->b_vp;
if (vp)
vp_md = vp->v_vflag & VV_MD;
else
vp_md = 0;
In bufwrite(), a dirty buffer is moved to the clean queue before the bufobj counter of the writes in progress is incremented. Other thread inspecting the bufobj would consider it clean. For the regular vnodes, the vnode lock is typically held both by the thread performing the bufwrite() and an other thread doing syncing, which prevents the situation. On the other hand, writes to the VCHR vnodes are done without holding vnode lock. Increment the write ref counter for the buffer object before calling bundirty(). Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-20 21:08:00 +00:00
/*
* Mark the buffer clean. Increment the bufobj write count
* before bundirty() call, to prevent other thread from seeing
* empty dirty list and zero counter for writes in progress,
* falsely indicating that the bufobj is clean.
*/
bufobj_wref(bp->b_bufobj);
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
bundirty(bp);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
bp->b_flags &= ~B_DONE;
bp->b_ioflags &= ~BIO_ERROR;
Remove unused B_WRITEINPROG flag
2004-09-15 21:49:22 +00:00
bp->b_flags |= B_CACHE;
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
bp->b_iocmd = BIO_WRITE;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
vfs_busy_pages(bp, 1);
Fix lockup for loopback NFS mounts. The pipelined I/O limitations could be hit on the client side and prevent the server side from retiring writes. Pipeline operations turned off for all READs (no big loss since reads are usually synchronous) and for NFS writes, and left on for the default bwrite(). (MFC expected prior to 4.3 freeze) Testing by: mjacob, dillon
2001-02-28 04:13:11 +00:00
/*
* Normal bwrites pipeline writes
*/
bp->b_runningbufspace = bp->b_bufsize;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
space = atomic_fetchadd_long(&runningbufspace, bp->b_runningbufspace);
Fix lockup for loopback NFS mounts. The pipelined I/O limitations could be hit on the client side and prevent the server side from retiring writes. Pipeline operations turned off for all READs (no big loss since reads are usually synchronous) and for NFS writes, and left on for the default bwrite(). (MFC expected prior to 4.3 freeze) Testing by: mjacob, dillon
2001-02-28 04:13:11 +00:00
Add four new RCTL resources - readbps, readiops, writebps and writeiops, for limiting disk (actually filesystem) IO. Note that in some cases these limits are not quite precise. It's ok, as long as it's within some reasonable bounds. Testing - and review of the code, in particular the VFS and VM parts - is very welcome. MFC after: 1 month Relnotes: yes Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D5080
2016-04-07 04:23:25 +00:00
#ifdef RACCT
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
if (racct_enable) {
PROC_LOCK(curproc);
racct_add_buf(curproc, bp, 1);
PROC_UNLOCK(curproc);
Add four new RCTL resources - readbps, readiops, writebps and writeiops, for limiting disk (actually filesystem) IO. Note that in some cases these limits are not quite precise. It's ok, as long as it's within some reasonable bounds. Testing - and review of the code, in particular the VFS and VM parts - is very welcome. MFC after: 1 month Relnotes: yes Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D5080
2016-04-07 04:23:25 +00:00
}
Stop checking TD_IDLETHREAD() in buffer cache routines. These predicates are vestigal and cannot be true today. For example, idle threads are not allowed to acquire locks. Also cache curthread in breada(). No functional change intended. Reviewed by: kib, mckusick MFC after: 1 week Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D20066
2019-04-29 13:23:32 +00:00
#endif /* RACCT */
curthread->td_ru.ru_oublock++;
Hopefully fix the remaining glitches with the BUF_*() changes. This should (really this time) fix pageout to swap and a couple of clustering cases. This simplifies BUF_KERNPROC() so that it unconditionally reassigns the lock owner rather than testing B_ASYNC and having the caller decide when to do the reassign. At present this is required because some places use B_CALL/b_iodone to free the buffers without B_ASYNC being set. Also, vfs_cluster.c explicitly calls BUF_KERNPROC() when attaching the buffers rather than the parent walking the cluster_head tailq. Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-06-29 05:59:47 +00:00
if (oldflags & B_ASYNC)
BUF_KERNPROC(bp);
Initialize b_iooffset before calling VOP_[SPEC]STRATEGY
2003-10-18 19:49:46 +00:00
bp->b_iooffset = dbtob(bp->b_blkno);
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
buf_track(bp, __func__);
Move the buffer method vector (buf->b_op) to the bufobj. Extend it with a strategy method. Add bufstrategy() which do the usual VOP_SPECSTRATEGY/VOP_STRATEGY song and dance. Rename ibwrite to bufwrite(). Move the two NFS buf_ops to more sensible places, add bufstrategy to them. Add inlines for bwrite() and bstrategy() which calls through buf->b_bufobj->b_ops->b_{write,strategy}(). Replace almost all VOP_STRATEGY()/VOP_SPECSTRATEGY() calls with bstrategy().
2004-10-24 20:03:41 +00:00
bstrategy(bp);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
A modification that allows the driver strategy to modify the B_ASYNC flag broke things pretty bad (freeing buffer already on queue or other wierd buffer queue errors.) The broken code is left in commented out, but this makes the problem go away for now.
1996-09-13 03:15:45 +00:00
if ((oldflags & B_ASYNC) == 0) {
s/biowait/bufwait/g Prodded by: several.
2000-04-29 16:25:22 +00:00
int rtval = bufwait(bp);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
brelse(bp);
return (rtval);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
} else if (space > hirunningspace) {
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
/*
* don't allow the async write to saturate the I/O
The IO_NOWDRAIN and B_NOWDRAIN hacks are no longer needed to prevent deadlocks with vnode backed md(4) devices because md now uses a kthread to run the bio requests instead of doing it directly from the bio down path.
2003-05-31 16:42:45 +00:00
* system. We will not deadlock here because
Implement IO_NOWDRAIN and B_NOWDRAIN - prevents the buffer cache from blocking in wdrain during a write. This flag needs to be used in devices whos strategy routines turn-around and issue another high level I/O, such as when MD turns around and issues a VOP_WRITE to vnode backing store, in order to avoid deadlocking the dirty buffer draining code. Remove a vprintf() warning from MD when the backing vnode is found to be in-use. The syncer of buf_daemon could be flushing the backing vnode at the time of an MD operation so the warning is not correct. MFC after: 1 week
2001-11-05 18:48:54 +00:00
* we are blocking waiting for I/O that is already in-progress
Allow the bufdaemon and update daemon processes to skip the waitrunningbufspace() calls so that they are always able to proceed and clean up buffer space. Submitted by: Brian Fundakowski Feldman <green@freebsd.org>
2003-11-04 06:30:00 +00:00
* to complete. We do not block here if it is the update
* or syncer daemon trying to clean up as that can lead
* to deadlock.
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
*/
In rev. 1.514, iodone on async buffer may happen before code checks the vnode v_flag. For cluster buffers this would result in dereferencing NULL b_vp. To prevent the panic, cache relevant vnode flag before calling bstrategy. Reported by: Peter Holm, kris Tested by: Peter Holm Reviewed by: tegge Pointy hat to: kib
2006-12-20 09:22:31 +00:00
if ((curthread->td_pflags & TDP_NORUNNINGBUF) == 0 && !vp_md)
Allow the bufdaemon and update daemon processes to skip the waitrunningbufspace() calls so that they are always able to proceed and clean up buffer space. Submitted by: Brian Fundakowski Feldman <green@freebsd.org>
2003-11-04 06:30:00 +00:00
waitrunningbufspace();
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
return (0);
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
}
Cylinder group bitmaps and blocks containing inode for a snapshot file are after snaplock, while other ffs device buffers are before snaplock in global lock order. By itself, this could cause deadlock when bdwrite() tries to flush dirty buffers on snapshotted ffs. If, during the flush, COW activity for snapshot needs to allocate block and ffs_alloccg() selects the cylinder group that is being written by bdwrite(), then kernel would panic due to recursive buffer lock acquision. Avoid dealing with buffers in bdwrite() that are from other side of snaplock divisor in the lock order then the buffer being written. Add new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in the bdwrite(). Default implementation, bufbdflush(), refactors the code from bdwrite(). For ffs device buffers, specialized implementation is used. Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes) Tested by: Peter Holm X-MFC after: 3 weeks (if ever: it changes ABI)
2007-01-23 10:01:19 +00:00
void
bufbdflush(struct bufobj *bo, struct buf *bp)
{
struct buf *nbp;
if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10) {
(void) VOP_FSYNC(bp->b_vp, MNT_NOWAIT, curthread);
altbufferflushes++;
} else if (bo->bo_dirty.bv_cnt > dirtybufthresh) {
BO_LOCK(bo);
/*
* Try to find a buffer to flush.
*/
TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) {
if ((nbp->b_vflags & BV_BKGRDINPROG) ||
BUF_LOCK(nbp,
LK_EXCLUSIVE | LK_NOWAIT, NULL))
continue;
if (bp == nbp)
panic("bdwrite: found ourselves");
BO_UNLOCK(bo);
/* Don't countdeps with the bo lock held. */
if (buf_countdeps(nbp, 0)) {
BO_LOCK(bo);
BUF_UNLOCK(nbp);
continue;
}
if (nbp->b_flags & B_CLUSTEROK) {
vfs_bio_awrite(nbp);
} else {
bremfree(nbp);
bawrite(nbp);
}
dirtybufferflushes++;
break;
}
if (nbp == NULL)
BO_UNLOCK(bo);
}
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Delayed write. (Buffer is marked dirty). Do not bother writing
* anything if the buffer is marked invalid.
*
* Note that since the buffer must be completely valid, we can safely
* set B_CACHE. In fact, we have to set B_CACHE here rather then in
* biodone() in order to prevent getblk from writing the buffer
* out synchronously.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
bdwrite(struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
struct thread *td = curthread;
struct vnode *vp;
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
struct bufobj *bo;
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bdwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
Add barrier write capability to the VFS buffer interface. A barrier write is a disk write request that tells the disk that the buffer being written must be committed to the media along with any writes that preceeded it before any future blocks may be written to the drive. Barrier writes are provided by adding the functions bbarrierwrite (bwrite with barrier) and babarrierwrite (bawrite with barrier). Following a bbarrierwrite the client knows that the requested buffer is on the media. It does not ensure that buffers written before that buffer are on the media. It only ensure that buffers written before that buffer will get to the media before any buffers written after that buffer. A flush command must be sent to the disk to ensure that all earlier written buffers are on the media. Reviewed by: kib Tested by: Peter Holm
2013-02-16 14:51:30 +00:00
KASSERT((bp->b_flags & B_BARRIER) == 0,
("Barrier request in delayed write %p", bp));
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (bp->b_flags & B_INVAL) {
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
brelse(bp);
return;
}
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
/*
* If we have too many dirty buffers, don't create any more.
* If we are wildly over our limit, then force a complete
* cleanup. Otherwise, just keep the situation from getting
When doing cleanup of excessive buffers in bdwrite (see kern/vfs_bio.c delta 1.371) we must ensure that we do not get ourselves into a recursive trap endlessly trying to clean up after ourselves. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 23:59:09 +00:00
* out of control. Note that we have to avoid a recursive
* disaster and not try to clean up after our own cleanup!
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
*/
vp = bp->b_vp;
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
bo = bp->b_bufobj;
Disable nesting of BOP_BDFLUSH(). VOP_FSYNC() call in bdwrite() could result in bdwrite() being reentered, thus causing infinite recursion. Reported and tested by: Peter Holm Reviewed by: tegge MFC after: 2 weeks
2007-04-24 10:59:21 +00:00
if ((td->td_pflags & (TDP_COWINPROGRESS|TDP_INBDFLUSH)) == 0) {
td->td_pflags |= TDP_INBDFLUSH;
Cylinder group bitmaps and blocks containing inode for a snapshot file are after snaplock, while other ffs device buffers are before snaplock in global lock order. By itself, this could cause deadlock when bdwrite() tries to flush dirty buffers on snapshotted ffs. If, during the flush, COW activity for snapshot needs to allocate block and ffs_alloccg() selects the cylinder group that is being written by bdwrite(), then kernel would panic due to recursive buffer lock acquision. Avoid dealing with buffers in bdwrite() that are from other side of snaplock divisor in the lock order then the buffer being written. Add new BOP, bop_bdwrite(), to do dirty buffer flushing for same vnode in the bdwrite(). Default implementation, bufbdflush(), refactors the code from bdwrite(). For ffs device buffers, specialized implementation is used. Reviewed by: tegge, jeff, Russell Cattelan (cattelan xfs org, xfs changes) Tested by: Peter Holm X-MFC after: 3 weeks (if ever: it changes ABI)
2007-01-23 10:01:19 +00:00
BO_BDFLUSH(bo, bp);
Disable nesting of BOP_BDFLUSH(). VOP_FSYNC() call in bdwrite() could result in bdwrite() being reentered, thus causing infinite recursion. Reported and tested by: Peter Holm Reviewed by: tegge MFC after: 2 weeks
2007-04-24 10:59:21 +00:00
td->td_pflags &= ~TDP_INBDFLUSH;
} else
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
recursiveflushes++;
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
bdirty(bp);
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* Set B_CACHE, indicating that the buffer is fully valid. This is
* true even of NFS now.
*/
bp->b_flags |= B_CACHE;
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
/*
* This bmap keeps the system from needing to do the bmap later,
* perhaps when the system is attempting to do a sync. Since it
* is likely that the indirect block -- or whatever other datastructure
* that the filesystem needs is still in memory now, it is a good
* thing to do this. Note also, that if the pageout daemon is
* requesting a sync -- there might not be enough memory to do
* the bmap then... So, this is important to do.
*/
Prevent large files from monopolizing the system buffers. Keep track of the number of dirty buffers held by a vnode. When a bdwrite is done on a buffer, check the existing number of dirty buffers associated with its vnode. If the number rises above vfs.dirtybufthresh (currently 90% of vfs.hidirtybuffers), one of the other (hopefully older) dirty buffers associated with the vnode is written (using bawrite). In the event that this approach fails to curb the growth in it the vnode's number of dirty buffers (due to soft updates rollback dependencies), the more drastic approach of doing a VOP_FSYNC on the vnode is used. This code primarily affects very large and actively written files such as snapshots. This change should eliminate hanging when taking snapshots or doing background fsck on very large filesystems. Hopefully, one day it will be possible to cache filesystem metadata in the VM cache as is done with file data. As it stands, only the buffer cache can be used which limits total metadata storage to about 20Mb no matter how much memory is available on the system. This rather small memory gets badly thrashed causing a lot of extra I/O. For example, taking a snapshot of a 1Tb filesystem minimally requires about 35,000 write operations, but because of the cache thrashing (we only have about 350 buffers at our disposal) ends up doing about 237,540 I/O's thus taking twenty-five minutes instead of four if it could run entirely in the cache. Reported by: Attila Nagy <bra@fsn.hu> Sponsored by: DARPA & NAI Labs.
2003-02-25 06:44:42 +00:00
if (vp->v_type != VCHR && bp->b_lblkno == bp->b_blkno) {
VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL);
Fixes from John Dyson to work around vnode lock hang. Basically, remove the VOP_BMAP calls, and add one to bdwrite. Submitted by: John Dyson
1995-03-03 22:13:16 +00:00
}
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
buf_track(bp, __func__);
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
/*
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
* Set the *dirty* buffer range based upon the VM system dirty
* pages.
*
* Mark the buffer pages as clean. We need to do this here to
* satisfy the vnode_pager and the pageout daemon, so that it
* thinks that the pages have been "cleaned". Note that since
* the pages are in a delayed write buffer -- the VFS layer
* "will" see that the pages get written out on the next sync,
* or perhaps the cluster will be completed.
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
*/
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
vfs_clean_pages_dirty_buf(bp);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
bqrelse(bp);
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
/*
* note: we cannot initiate I/O from a bdwrite even if we wanted to,
* due to the softdep code.
*/
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
/*
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
* bdirty:
*
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
* Turn buffer into delayed write request. We must clear BIO_READ and
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
* B_RELBUF, and we must set B_DELWRI. We reassign the buffer to
* itself to properly update it in the dirty/clean lists. We mark it
* B_DONE to ensure that any asynchronization of the buffer properly
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* clears B_DONE ( else a panic will occur later ).
*
* bdirty() is kinda like bdwrite() - we have to clear B_INVAL which
* might have been set pre-getblk(). Unlike bwrite/bdwrite, bdirty()
* should only be called if the buffer is known-good.
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*
* Since the buffer is not on a queue, we do not update the numfreebuffers
* count.
*
* The buffer must be on QUEUE_NONE.
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
bdirty(struct buf *bp)
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bdirty(%p) vp %p flags %X",
bp, bp->b_vp, bp->b_flags);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE,
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
("bdirty: buffer %p still on queue %d", bp, bp->b_qindex));
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
bp->b_flags &= ~(B_RELBUF);
bp->b_iocmd = BIO_WRITE;
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
if ((bp->b_flags & B_DELWRI) == 0) {
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
bp->b_flags |= /* XXX B_DONE | */ B_DELWRI;
Eliminate unused second argument to reassignbuf() and simplify it accordingly.
2004-07-25 21:24:23 +00:00
reassignbuf(bp);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bdirtyadd(bp);
Reviewed by: dyson@freebsd.org (john Dyson), dg@root.com (david greenman) Submitted by: Kirk McKusick (mcKusick@mckusick.com) Obtained from: WHistle development tree
1998-03-08 09:59:44 +00:00
}
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
* bundirty:
*
* Clear B_DELWRI for buffer.
*
* Since the buffer is not on a queue, we do not update the numfreebuffers
* count.
*
* The buffer must be on QUEUE_NONE.
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
bundirty(struct buf *bp)
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bundirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
- Eliminate the acquisition and release of the bqlock in bremfree() by setting the B_REMFREE flag in the buf. This is done to prevent lock order reversals with code that must call bremfree() with a local lock held. This also reduces overhead by removing two lock operations per buf for fsync() and similar. - Check for the B_REMFREE flag in brelse() and bqrelse() after the bqlock has been acquired so that we may remove ourself from the free-list. - Provide a bremfreef() function to immediately remove a buf from a free-list for use only by NFS. This is done because the nfsclient code overloads the b_freelist queue for its own async. io queue. - Simplify the numfreebuffers accounting by removing a switch statement that executed the same code in every possible case. - getnewbuf() can encounter locked bufs on free-lists once Giant is removed. Remove a panic associated with this condition and delay asserts that inspect the buf until after it is locked. Reviewed by: phk Sponsored by: Isilon Systems, Inc.
2004-11-18 08:44:09 +00:00
KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE,
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
("bundirty: buffer %p still on queue %d", bp, bp->b_qindex));
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
if (bp->b_flags & B_DELWRI) {
bp->b_flags &= ~B_DELWRI;
Eliminate unused second argument to reassignbuf() and simplify it accordingly.
2004-07-25 21:24:23 +00:00
reassignbuf(bp);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bdirtysub(bp);
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
}
Several performance improvements for soft updates have been added: 1) Fastpath deletions. When a file is being deleted, check to see if it was so recently created that its inode has not yet been written to disk. If so, the delete can proceed to immediately free the inode. 2) Background writes: No file or block allocations can be done while the bitmap is being written to disk. To avoid these stalls, the bitmap is copied to another buffer which is written thus leaving the original available for futher allocations. 3) Link count tracking. Constantly track the difference in i_effnlink and i_nlink so that inodes that have had no change other than i_effnlink need not be written. 4) Identify buffers with rollback dependencies so that the buffer flushing daemon can choose to skip over them.
2000-01-10 00:24:24 +00:00
/*
* Since it is now being written, we can clear its deferred write flag.
*/
bp->b_flags &= ~B_DEFERRED;
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
}
/*
* bawrite:
*
* Asynchronous write. Start output on a buffer, but do not wait for
* it to complete. The buffer is released when the output completes.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* bwrite() ( or the VOP routine anyway ) is responsible for handling
* B_INVAL buffers. Not us.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
bawrite(struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
Added some code to make sure that buffers associated with directories and metadata aren't thrashed by regular file I/O. Added mechanism to limit the amount of outstanding I/O on a given vnode. Pagedaemon wakeup policy changed to skew priority a little in favor of file caching. Slight code reorganization to improve clarity. Added a few more comments. Submitted by: John Dyson
1995-02-22 09:30:13 +00:00
bp->b_flags |= B_ASYNC;
Properly vector all bwrite() and BUF_WRITE() calls through the same path and s/BUF_WRITE()/bwrite()/ since it now does the same as bwrite().
2004-03-11 18:02:36 +00:00
(void) bwrite(bp);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Add barrier write capability to the VFS buffer interface. A barrier write is a disk write request that tells the disk that the buffer being written must be committed to the media along with any writes that preceeded it before any future blocks may be written to the drive. Barrier writes are provided by adding the functions bbarrierwrite (bwrite with barrier) and babarrierwrite (bawrite with barrier). Following a bbarrierwrite the client knows that the requested buffer is on the media. It does not ensure that buffers written before that buffer are on the media. It only ensure that buffers written before that buffer will get to the media before any buffers written after that buffer. A flush command must be sent to the disk to ensure that all earlier written buffers are on the media. Reviewed by: kib Tested by: Peter Holm
2013-02-16 14:51:30 +00:00
/*
* babarrierwrite:
*
* Asynchronous barrier write. Start output on a buffer, but do not
* wait for it to complete. Place a write barrier after this write so
* that this buffer and all buffers written before it are committed to
* the disk before any buffers written after this write are committed
* to the disk. The buffer is released when the output completes.
*/
void
babarrierwrite(struct buf *bp)
{
bp->b_flags |= B_ASYNC | B_BARRIER;
(void) bwrite(bp);
}
/*
* bbarrierwrite:
*
* Synchronous barrier write. Start output on a buffer and wait for
* it to complete. Place a write barrier after this write so that
* this buffer and all buffers written before it are committed to
* the disk before any buffers written after this write are committed
* to the disk. The buffer is released when the output completes.
*/
int
bbarrierwrite(struct buf *bp)
{
bp->b_flags |= B_BARRIER;
return (bwrite(bp));
}
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
/*
* bwillwrite:
*
* Called prior to the locking of any vnodes when we are expecting to
* write. We do not want to starve the buffer cache with too many
* dirty buffers so we block here. By blocking prior to the locking
* of any vnodes we attempt to avoid the situation where a locked vnode
* prevents the various system daemons from flushing related buffers.
*/
void
bwillwrite(void)
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
if (buf_dirty_count_severe()) {
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
mtx_lock(&bdirtylock);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
while (buf_dirty_count_severe()) {
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bdirtywait = 1;
msleep(&bdirtywait, &bdirtylock, (PRIBIO + 4),
"flswai", 0);
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
mtx_unlock(&bdirtylock);
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
}
}
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
/*
* Return true if we have too many dirty buffers.
*/
int
buf_dirty_count_severe(void)
{
stylistic polishing.
2004-09-15 20:54:23 +00:00
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
return (!BIT_EMPTY(BUF_DOMAINS, &bdhidirty));
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
* brelse:
*
* Release a busy buffer and, if requested, free its resources. The
* buffer will be stashed in the appropriate bufqueue[] allowing it
* to be accessed later as a cache entity or reused for other purposes.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
brelse(struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
vfs_bio.c: Apply cleanups motivated by Coverity analysis It is believed that the conditions Coverity indicated were actually impossible to hit. So this patch just adds a cleanup to only compute v_mount once in brelse(), and in vfs_bio_getpages() always initializes error to zero to appease the static analyzer. No functional change intended. Submitted by: Darrick Lew <darrick.freebsd AT gmail.com> Reviewed by: kib Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D14613
2018-03-14 22:11:45 +00:00
struct mount *v_mnt;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
int qindex;
The bread() function was inconsistent about whether it would return a buffer pointer in the event of an error (for some errors it would return a buffer pointer and for other errors it would not return a buffer pointer). The cluster_read() function was similarly inconsistent. Clients of these functions were inconsistent in handling errors. Some would assume that no buffer was returned after an error and would thus lose buffers under certain error conditions. Others would assume that brelse() should always be called after an error and would thus panic the system under certain error conditions. To correct both of these problems with minimal code churn, bread() and cluster_write() now always free the buffer when returning an error thus ensuring that buffers will never be lost. The brelse() routine checks for being passed a NULL buffer pointer and silently returns to avoid panics. Thus both approaches to handling error returns from bread() and cluster_read() will work correctly. Future code should be written assuming that bread() and cluster_read() will never return a buffer with an error, so should not attempt to brelse() the buffer when an error is returned. Reviewed by: kib
2016-01-27 21:23:01 +00:00
/*
Minor grammar fix in comment.
2016-02-07 16:18:12 +00:00
* Many functions erroneously call brelse with a NULL bp under rare
The bread() function was inconsistent about whether it would return a buffer pointer in the event of an error (for some errors it would return a buffer pointer and for other errors it would not return a buffer pointer). The cluster_read() function was similarly inconsistent. Clients of these functions were inconsistent in handling errors. Some would assume that no buffer was returned after an error and would thus lose buffers under certain error conditions. Others would assume that brelse() should always be called after an error and would thus panic the system under certain error conditions. To correct both of these problems with minimal code churn, bread() and cluster_write() now always free the buffer when returning an error thus ensuring that buffers will never be lost. The brelse() routine checks for being passed a NULL buffer pointer and silently returns to avoid panics. Thus both approaches to handling error returns from bread() and cluster_read() will work correctly. Future code should be written assuming that bread() and cluster_read() will never return a buffer with an error, so should not attempt to brelse() the buffer when an error is returned. Reviewed by: kib
2016-01-27 21:23:01 +00:00
* error conditions. Simply return when called with a NULL bp.
*/
if (bp == NULL)
return;
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "brelse(%p) vp %p flags %X",
bp, bp->b_vp, bp->b_flags);
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)),
("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp));
As a step towards the elimination of PG_CACHED pages, rework the handling of POSIX_FADV_DONTNEED so that it causes the backing pages to be moved to the head of the inactive queue instead of being cached. This affects the implementation of POSIX_FADV_NOREUSE as well, since it works by applying POSIX_FADV_DONTNEED to file ranges after they have been read or written. At that point the corresponding buffers may still be dirty, so the previous implementation would coalesce successive ranges and apply POSIX_FADV_DONTNEED to the result, ensuring that pages backing the dirty buffers would eventually be cached. To preserve this behaviour in an efficient manner, this change adds a new buf flag, B_NOREUSE, which causes the pages backing a VMIO buf to be placed at the head of the inactive queue when the buf is released. POSIX_FADV_NOREUSE then works by setting this flag in bufs that underlie the specified range. Reviewed by: alc, kib Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D3726
2015-09-30 23:06:29 +00:00
KASSERT((bp->b_flags & B_VMIO) != 0 || (bp->b_flags & B_NOREUSE) == 0,
("brelse: non-VMIO buffer marked NOREUSE"));
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Make recursive getblk() slightly more useful. Keep the buffer state intact if getblk() is done on the already owned buffer. Exit from brelse() early when the lock recursion is detected, otherwise brelse() might prematurely destroy the buffer under some circumstances. Sponsored by: The FreeBSD Foundation Noted by: mckusick Tested by: pho MFC after: 2 weeks
2013-02-27 07:34:09 +00:00
if (BUF_LOCKRECURSED(bp)) {
/*
* Do not process, in particular, do not handle the
* B_INVAL/B_RELBUF and do not release to free list.
*/
BUF_UNLOCK(bp);
return;
}
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
if (bp->b_flags & B_MANAGED) {
bqrelse(bp);
return;
}
Handle errors from background write of the cylinder group blocks. First, on the write error, bufdone() call from ffs_backgroundwrite() panics because pbrelvp() cleared bp->b_bufobj, while brelse() would try to re-dirty the copy of the cg buffer. Handle this by setting B_INVAL for the case of BIO_ERROR. Second, we must re-dirty the real buffer containing the cylinder group block data when background write failed. Real cg buffer was already marked clean in ffs_bufwrite(). After the BV_BKGRDINPROG flag is cleared on the real cg buffer in ffs_backgroundwrite(), buffer scan may reuse the buffer at any moment. The result is lost write, and if the write error was only transient, we get corrupted bitmaps. We cannot re-dirty the original cg buffer in the ffs_backgroundwritedone(), since the context is not sleepable, preventing us from sleeping for origbp' lock. Add BV_BKGDERR flag (protected by the buffer object lock), which is converted into delayed write by brelse(), bqrelse() and buffer scan. In collaboration with: Conrad Meyer <cse.cem@gmail.com> Reviewed by: mckusick Sponsored by: The FreeBSD Foundation (kib), EMC/Isilon storage division (Conrad) MFC after: 2 weeks
2015-06-27 09:44:14 +00:00
if ((bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) {
BO_LOCK(bp->b_bufobj);
bp->b_vflags &= ~BV_BKGRDERR;
BO_UNLOCK(bp->b_bufobj);
bdirty(bp);
}
Rather than not redirting the bp when we get ENXIO, only redirty it when the error is EIO. This catches a much larger class of errors that are unlikely to succeed if retried. Submitted by: bde
2007-12-30 05:53:45 +00:00
if (bp->b_iocmd == BIO_WRITE && (bp->b_ioflags & BIO_ERROR) &&
Don't try to write out bufs that have already failed with ENXIO. This fixes some panics after disconnecting mounted disks. Submitted by: imp (slightly different version, which I've then lost) Reviewed by: kib, imp, mckusick MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D9674
2017-04-14 20:15:34 +00:00
(bp->b_error != ENXIO || !LIST_EMPTY(&bp->b_dep)) &&
The error classification from lower layers is a poor indicator of whether an error is recoverable. Always re-dirty the buffer on errors from write requests. The invalidation we used to do for errors not EIO doesn't need to be done for a device that's really gone, since that's done in a different path. Reviewed by: mckusick@, kib@
2015-10-31 04:53:07 +00:00
!(bp->b_flags & B_INVAL)) {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
Don't try to write out bufs that have already failed with ENXIO. This fixes some panics after disconnecting mounted disks. Submitted by: imp (slightly different version, which I've then lost) Reviewed by: kib, imp, mckusick MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D9674
2017-04-14 20:15:34 +00:00
* Failed write, redirty. All errors except ENXIO (which
Update a comment in brelse() to match reality.
2017-11-20 20:53:03 +00:00
* means the device is gone) are treated as being
* transient.
*
* XXX Treating EIO as transient is not correct; the
* contract with the local storage device drivers is that
* they will only return EIO once the I/O is no longer
* retriable. Network I/O also respects this through the
* guarantees of TCP and/or the internal retries of NFS.
* ENOMEM might be transient, but we also have no way of
* knowing when its ok to retry/reschedule. In general,
* this entire case should be made obsolete through better
* error handling/recovery and resource scheduling.
Don't try to write out bufs that have already failed with ENXIO. This fixes some panics after disconnecting mounted disks. Submitted by: imp (slightly different version, which I've then lost) Reviewed by: kib, imp, mckusick MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D9674
2017-04-14 20:15:34 +00:00
*
* Do this also for buffers that failed with ENXIO, but have
* non-empty dependencies - the soft updates code might need
* to access the buffer to untangle them.
*
* Must clear BIO_ERROR to prevent pages from being scrapped.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*/
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
bp->b_ioflags &= ~BIO_ERROR;
Don't throw away the buffer contents on a fatal write error; just mark the buffer as still being dirty. This isn't a perfect solution, but throwing away the buffer contents will often result in filesystem corruption and this solution will at least correctly deal with transient errors. Submitted by: Kirk McKusick <mckusick@mckusick.com>
1999-01-22 08:59:05 +00:00
bdirty(bp);
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
} else if ((bp->b_flags & (B_NOCACHE | B_INVAL)) ||
Remove the buffercache/vnode side of BIO_DELETE processing in preparation for integration of p4::phk_bufwork. In the future, local filesystems will talk to GEOM directly and they will consequently be able to issue BIO_DELETE directly. Since the removal of the fla driver, BIO_DELETE has effectively been a no-op anyway.
2004-09-13 06:50:42 +00:00
(bp->b_ioflags & BIO_ERROR) || (bp->b_bufsize <= 0)) {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
Don't try to write out bufs that have already failed with ENXIO. This fixes some panics after disconnecting mounted disks. Submitted by: imp (slightly different version, which I've then lost) Reviewed by: kib, imp, mckusick MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D9674
2017-04-14 20:15:34 +00:00
* Either a failed read I/O, or we were asked to free or not
* cache the buffer, or we failed to write to a device that's
* no longer present.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*/
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
bp->b_flags |= B_INVAL;
Use LIST_EMPTY() instead of unrolled version (LIST_FIRST() [!=]= NULL)
2007-02-22 14:52:59 +00:00
if (!LIST_EMPTY(&bp->b_dep))
Virtualizes & untangles the bioops operations vector. Ref: Message-ID: <18317.961014572@critter.freebsd.dk> To: current@
2000-06-16 08:48:51 +00:00
buf_deallocate(bp);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
if (bp->b_flags & B_DELWRI)
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bdirtysub(bp);
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
bp->b_flags &= ~(B_DELWRI | B_CACHE);
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
if ((bp->b_flags & B_VMIO) == 0) {
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
allocbuf(bp, 0);
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
if (bp->b_vp)
b_waiters cannot be adequately protected by the interlock because it is dropped after the call to lockmgr() so just revert this approach using something similar to the precedent one: BUF_LOCKWAITERS() just checks if there are waiters (not the actual number of them) and it is based on newly introduced lockmgr_waiters() which returns if the lockmgr has waiters or not. The name has been choosen differently by old lockwaiters() in order to not confuse them. KPI results enriched by this commit so __FreeBSD_version bumping and manpage update will be happening soon. 'struct buf' also changes, so kernel ABI is disturbed. Bug found by: jeff Approved by: jeff, kib
2008-03-28 12:30:12 +00:00
brelvp(bp);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
}
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
PR: kern/7418 Reviewed by: Luoqi Chen <luoqi@watermarkgroup.com> Fixed problem where write()s can get lost due to buffers flagged B_DELWRI being improperly released in brelse().
1998-09-26 00:12:35 +00:00
/*
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
* We must clear B_RELBUF if B_DELWRI is set. If vfs_vmio_truncate()
PR: kern/7418 Reviewed by: Luoqi Chen <luoqi@watermarkgroup.com> Fixed problem where write()s can get lost due to buffers flagged B_DELWRI being improperly released in brelse().
1998-09-26 00:12:35 +00:00
* is called with B_DELWRI set, the underlying pages may wind up
* getting freed causing a previous write (bdwrite()) to get 'lost'
* because pages associated with a B_DELWRI bp are marked clean.
*
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
* We still allow the B_INVAL case to call vfs_vmio_truncate(), even
PR: kern/7418 Reviewed by: Luoqi Chen <luoqi@watermarkgroup.com> Fixed problem where write()s can get lost due to buffers flagged B_DELWRI being improperly released in brelse().
1998-09-26 00:12:35 +00:00
* if B_DELWRI is set.
*/
if (bp->b_flags & B_DELWRI)
bp->b_flags &= ~B_RELBUF;
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
/*
* VMIO buffer rundown. It is not very necessary to keep a VMIO buffer
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* constituted, not even NFS buffers now. Two flags effect this. If
* B_INVAL, the struct buf is invalidated but the VM object is kept
* around ( i.e. so it is trivial to reconstitute the buffer later ).
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
*
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
* If BIO_ERROR or B_NOCACHE is set, pages in the VM object will be
* invalidated. BIO_ERROR cannot be set for a failed write unless the
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* buffer is also B_INVAL because it hits the re-dirtying code above.
Adjust some comments to prevent future confusion on the implementation. Also add a reference to the buf(9) manual page.
1998-12-22 18:57:30 +00:00
*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Normally we can do this whether a buffer is B_DELWRI or not. If
* the buffer is an NFS buffer, it is tracking piecemeal writes or
Several performance improvements for soft updates have been added: 1) Fastpath deletions. When a file is being deleted, check to see if it was so recently created that its inode has not yet been written to disk. If so, the delete can proceed to immediately free the inode. 2) Background writes: No file or block allocations can be done while the bitmap is being written to disk. To avoid these stalls, the bitmap is copied to another buffer which is written thus leaving the original available for futher allocations. 3) Link count tracking. Constantly track the difference in i_effnlink and i_nlink so that inodes that have had no change other than i_effnlink need not be written. 4) Identify buffers with rollback dependencies so that the buffer flushing daemon can choose to skip over them.
2000-01-10 00:24:24 +00:00
* the commit state and we cannot afford to lose the buffer. If the
* buffer has a background write in progress, we need to keep it
* around to prevent it from being reconstituted and starting a second
* background write.
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
*/
vfs_bio.c: Apply cleanups motivated by Coverity analysis It is believed that the conditions Coverity indicated were actually impossible to hit. So this patch just adds a cleanup to only compute v_mount once in brelse(), and in vfs_bio_getpages() always initializes error to zero to appease the static analyzer. No functional change intended. Submitted by: Darrick Lew <darrick.freebsd AT gmail.com> Reviewed by: kib Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D14613
2018-03-14 22:11:45 +00:00
v_mnt = bp->b_vp != NULL ? bp->b_vp->v_mount : NULL;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
if ((bp->b_flags & B_VMIO) && (bp->b_flags & B_NOCACHE ||
(bp->b_ioflags & BIO_ERROR && bp->b_iocmd == BIO_READ)) &&
vfs_bio.c: Apply cleanups motivated by Coverity analysis It is believed that the conditions Coverity indicated were actually impossible to hit. So this patch just adds a cleanup to only compute v_mount once in brelse(), and in vfs_bio_getpages() always initializes error to zero to appease the static analyzer. No functional change intended. Submitted by: Darrick Lew <darrick.freebsd AT gmail.com> Reviewed by: kib Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D14613
2018-03-14 22:11:45 +00:00
(v_mnt == NULL || (v_mnt->mnt_vfc->vfc_flags & VFCF_NETWORK) == 0 ||
vn_isdisk(bp->b_vp, NULL) || (bp->b_flags & B_DELWRI) == 0)) {
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
vfs_vmio_invalidate(bp);
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
allocbuf(bp, 0);
}
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
As a step towards the elimination of PG_CACHED pages, rework the handling of POSIX_FADV_DONTNEED so that it causes the backing pages to be moved to the head of the inactive queue instead of being cached. This affects the implementation of POSIX_FADV_NOREUSE as well, since it works by applying POSIX_FADV_DONTNEED to file ranges after they have been read or written. At that point the corresponding buffers may still be dirty, so the previous implementation would coalesce successive ranges and apply POSIX_FADV_DONTNEED to the result, ensuring that pages backing the dirty buffers would eventually be cached. To preserve this behaviour in an efficient manner, this change adds a new buf flag, B_NOREUSE, which causes the pages backing a VMIO buf to be placed at the head of the inactive queue when the buf is released. POSIX_FADV_NOREUSE then works by setting this flag in bufs that underlie the specified range. Reviewed by: alc, kib Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D3726
2015-09-30 23:06:29 +00:00
if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0 ||
(bp->b_flags & (B_DELWRI | B_NOREUSE)) == B_NOREUSE) {
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
allocbuf(bp, 0);
As a step towards the elimination of PG_CACHED pages, rework the handling of POSIX_FADV_DONTNEED so that it causes the backing pages to be moved to the head of the inactive queue instead of being cached. This affects the implementation of POSIX_FADV_NOREUSE as well, since it works by applying POSIX_FADV_DONTNEED to file ranges after they have been read or written. At that point the corresponding buffers may still be dirty, so the previous implementation would coalesce successive ranges and apply POSIX_FADV_DONTNEED to the result, ensuring that pages backing the dirty buffers would eventually be cached. To preserve this behaviour in an efficient manner, this change adds a new buf flag, B_NOREUSE, which causes the pages backing a VMIO buf to be placed at the head of the inactive queue when the buf is released. POSIX_FADV_NOREUSE then works by setting this flag in bufs that underlie the specified range. Reviewed by: alc, kib Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D3726
2015-09-30 23:06:29 +00:00
bp->b_flags &= ~B_NOREUSE;
For low memory situations, non-VMIO buffers didnt't release pages back to the system when brelse() was called with B_RELBUF set on the buffer. This could be a problem when the system was low on memory, had many buffers on QUEUE_EMPTYKVA and started to traverse directories. For each getnewbuf(), pages were allocated from the system, driving the free reserve downwards. For each brelse(), the system put the buffer on QUEUE_CLEAN, with B_INVAL set. This commit changes the semantics of B_RELBUF to also free pages from non-VMIO buffers. Reviewed by: alc
2006-02-02 21:37:39 +00:00
if (bp->b_vp != NULL)
b_waiters cannot be adequately protected by the interlock because it is dropped after the call to lockmgr() so just revert this approach using something similar to the precedent one: BUF_LOCKWAITERS() just checks if there are waiters (not the actual number of them) and it is based on newly introduced lockmgr_waiters() which returns if the lockmgr has waiters or not. The name has been choosen differently by old lockwaiters() in order to not confuse them. KPI results enriched by this commit so __FreeBSD_version bumping and manpage update will be happening soon. 'struct buf' also changes, so kernel ABI is disturbed. Bug found by: jeff Approved by: jeff, kib
2008-03-28 12:30:12 +00:00
brelvp(bp);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Various NFS fixes: Make vfs_bio buffer mgmt work better. Buffers were being used after brelse. Make nfs_getpages work independently of other NFS interfaces. This eliminates some difficult recursion problems and decreases pagefault overhead. Remove an erroneous vfs_unbusy_pages. Fix a reentrancy problem, with nfs_vinvalbuf when vnode is already being rundown. Reassignbuf wasn't being called when needed under certain circumstances. (Thanks to Bill Paul for help.)
1998-01-25 06:24:09 +00:00
Fix an old-standing bug that crept in along the several revisions: B_DELWRI cleanup and vnode disassociation should happen just before to assign the buffer to a queue. Reported by: miwi, Volker <volker at vwsoft dot com>, Ben Kaduk <minimarmot at gmail dot com>, Christopher Mallon <christoph dot mallon at gmx dot de> Tested by: lulf, miwi
2009-03-17 16:30:49 +00:00
/*
* If the buffer has junk contents signal it and eventually
* clean up B_DELWRI and diassociate the vnode so that gbincore()
* doesn't find it.
*/
if (bp->b_bufsize == 0 || (bp->b_ioflags & BIO_ERROR) != 0 ||
(bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) != 0)
bp->b_flags |= B_INVAL;
if (bp->b_flags & B_INVAL) {
if (bp->b_flags & B_DELWRI)
bundirty(bp);
if (bp->b_vp)
brelvp(bp);
}
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
buf_track(bp, __func__);
Incorporated post 1.1.5 work from John Dyson. This includes performance improvements via the new routines pmap_qenter/pmap_qremove and pmap_kenter/ pmap_kremove. These routine allow fast mapping of pages for those architectures that have "normal" MMUs. Also included is a fix to the pageout daemon to properly check a queue end condition. Submitted by: John Dyson
1994-08-06 09:15:42 +00:00
/* buffers with no memory */
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (bp->b_bufsize == 0) {
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
buf_free(bp);
return;
}
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
/* buffers with junk contents */
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF) ||
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
(bp->b_ioflags & BIO_ERROR)) {
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA);
- Move BX_BKGRDWAIT and BX_BKGRDINPROG to BV_ and the b_vflags field. - Surround all accesses of the BKGRD{WAIT,INPROG} flags with the vnode interlock. - Don't use the B_LOCKED flag and QUEUE_LOCKED for background write buffers. Check for the BKGRDINPROG flag before recycling or throwing away a buffer. We do this instead because it is not safe for us to move the original buffer to a new queue from the callback on the background write buffer. - Remove the B_LOCKED flag and the locked buffer queue. They are no longer used. - The vnode interlock is used around checks for BKGRDINPROG where it may not be strictly necessary. If we hold the buf lock the a back-ground write will not be started without our knowledge, one may only be completed while we're not looking. Rather than remove the code, Document two of the places where this extra locking is done. A pass should be done to verify and minimize the locking later.
2003-08-28 06:55:18 +00:00
if (bp->b_vflags & BV_BKGRDINPROG)
Several performance improvements for soft updates have been added: 1) Fastpath deletions. When a file is being deleted, check to see if it was so recently created that its inode has not yet been written to disk. If so, the delete can proceed to immediately free the inode. 2) Background writes: No file or block allocations can be done while the bitmap is being written to disk. To avoid these stalls, the bitmap is copied to another buffer which is written thus leaving the original available for futher allocations. 3) Link count tracking. Constantly track the difference in i_effnlink and i_nlink so that inodes that have had no change other than i_effnlink need not be written. 4) Identify buffers with rollback dependencies so that the buffer flushing daemon can choose to skip over them.
2000-01-10 00:24:24 +00:00
panic("losing buffer 2");
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
qindex = QUEUE_CLEAN;
bp->b_flags |= B_AGE;
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
/* remaining buffers */
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
} else if (bp->b_flags & B_DELWRI)
qindex = QUEUE_DIRTY;
else
qindex = QUEUE_CLEAN;
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Add snapshots to the fast filesystem. Most of the changes support the gating of system calls that cause modifications to the underlying filesystem. The gating can be enabled by any filesystem that needs to consistently suspend operations by adding the vop_stdgetwritemount to their set of vnops. Once gating is enabled, the function vfs_write_suspend stops all new write operations to a filesystem, allows any filesystem modifying system calls already in progress to complete, then sync's the filesystem to disk and returns. The function vfs_write_resume allows the suspended write operations to begin again. Gating is not added by default for all filesystems as for SMP systems it adds two extra locks to such critical kernel paths as the write system call. Thus, gating should only be added as needed. Details on the use and current status of snapshots in FFS can be found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness is not included here. Unless and until you create a snapshot file, these changes should have no effect on your system (famous last words).
2000-07-11 22:07:57 +00:00
if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY))
panic("brelse: not dirty");
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_RELBUF | B_DIRECT);
/* binsfree unlocks bp. */
binsfree(bp, qindex);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
/*
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
* Release a buffer back to the appropriate queue but do not try to free
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
* it. The buffer is expected to be used again soon.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* bqrelse() is used by bdwrite() to requeue a delayed write, and used by
* biodone() to requeue an async I/O on completion. It is also used when
* known good buffers need to be requeued but we think we may need the data
* again soon.
This patch implements O_DIRECT about 80% of the way. It takes a patchset Tor created a while ago, removes the raw I/O piece (that has cache coherency problems), and adds a buffer cache / VM freeing piece. Essentially this patch causes O_DIRECT I/O to not be left in the cache, but does not prevent it from going through the cache, hence the 80%. For the last 20% we need a method by which the I/O can be issued directly to buffer supplied by the user process and bypass the buffer cache entirely, but still maintain cache coherency. I also have the code working under -stable but the changes made to sys/file.h may not be MFCable, so an MFC is not on the table yet. Submitted by: tegge, dillon
2001-05-24 07:22:27 +00:00
*
* XXX we should be able to leave the B_RELBUF hint set on completion.
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
*/
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
bqrelse(struct buf *bp)
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
{
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
int qindex;
The buffers b_vflags field is not always properly protected by bufobj lock. If b_bufobj is not NULL, then bufobj lock should be held when manipulating the flags. Not doing this sometimes leaves BV_BKGRDINPROG to be erronously set, causing softdep' getdirtybuf() to stuck indefinitely in "getbuf" sleep, waiting for background write to finish which is not actually performed. Add BO_LOCK() in the cases where it was missed. In collaboration with: pho Tested by: bz Reviewed by: jeff MFC after: 1 month
2010-08-12 08:36:23 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bqrelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
stylistic polishing.
2004-09-15 20:54:23 +00:00
KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)),
("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp));
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
qindex = QUEUE_NONE;
- Introduce the function lockmgr_recursed() which returns true if the lockmgr lkp, when held in exclusive mode, is recursed - Introduce the function BUF_RECURSED() which does the same for bufobj locks based on the top of lockmgr_recursed() - Introduce the function BUF_ISLOCKED() which works like the counterpart VOP_ISLOCKED(9), showing the state of lockmgr linked with the bufobj BUF_RECURSED() and BUF_ISLOCKED() entirely replace the usage of bogus BUF_REFCNT() in a more explicative and SMP-compliant way. This allows us to axe out BUF_REFCNT() and leaving the function lockcount() totally unused in our stock kernel. Further commits will axe lockcount() as well as part of lockmgr() cleanup. KPI results, obviously, broken so further commits will update manpages and freebsd version. Tested by: kris (on UFS and NFS)
2008-01-19 17:36:23 +00:00
if (BUF_LOCKRECURSED(bp)) {
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
/* do not release to free list */
BUF_UNLOCK(bp);
return;
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF);
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
if (bp->b_flags & B_MANAGED) {
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
if (bp->b_flags & B_REMFREE)
bremfreef(bp);
goto out;
Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@
2005-12-07 03:39:08 +00:00
}
- Move BX_BKGRDWAIT and BX_BKGRDINPROG to BV_ and the b_vflags field. - Surround all accesses of the BKGRD{WAIT,INPROG} flags with the vnode interlock. - Don't use the B_LOCKED flag and QUEUE_LOCKED for background write buffers. Check for the BKGRDINPROG flag before recycling or throwing away a buffer. We do this instead because it is not safe for us to move the original buffer to a new queue from the callback on the background write buffer. - Remove the B_LOCKED flag and the locked buffer queue. They are no longer used. - The vnode interlock is used around checks for BKGRDINPROG where it may not be strictly necessary. If we hold the buf lock the a back-ground write will not be started without our knowledge, one may only be completed while we're not looking. Rather than remove the code, Document two of the places where this extra locking is done. A pass should be done to verify and minimize the locking later.
2003-08-28 06:55:18 +00:00
/* buffers with stale but valid contents */
Handle errors from background write of the cylinder group blocks. First, on the write error, bufdone() call from ffs_backgroundwrite() panics because pbrelvp() cleared bp->b_bufobj, while brelse() would try to re-dirty the copy of the cg buffer. Handle this by setting B_INVAL for the case of BIO_ERROR. Second, we must re-dirty the real buffer containing the cylinder group block data when background write failed. Real cg buffer was already marked clean in ffs_bufwrite(). After the BV_BKGRDINPROG flag is cleared on the real cg buffer in ffs_backgroundwrite(), buffer scan may reuse the buffer at any moment. The result is lost write, and if the write error was only transient, we get corrupted bitmaps. We cannot re-dirty the original cg buffer in the ffs_backgroundwritedone(), since the context is not sleepable, preventing us from sleeping for origbp' lock. Add BV_BKGDERR flag (protected by the buffer object lock), which is converted into delayed write by brelse(), bqrelse() and buffer scan. In collaboration with: Conrad Meyer <cse.cem@gmail.com> Reviewed by: mckusick Sponsored by: The FreeBSD Foundation (kib), EMC/Isilon storage division (Conrad) MFC after: 2 weeks
2015-06-27 09:44:14 +00:00
if ((bp->b_flags & B_DELWRI) != 0 || (bp->b_vflags & (BV_BKGRDINPROG |
BV_BKGRDERR)) == BV_BKGRDERR) {
BO_LOCK(bp->b_bufobj);
bp->b_vflags &= ~BV_BKGRDERR;
BO_UNLOCK(bp->b_bufobj);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
qindex = QUEUE_DIRTY;
- Move BX_BKGRDWAIT and BX_BKGRDINPROG to BV_ and the b_vflags field. - Surround all accesses of the BKGRD{WAIT,INPROG} flags with the vnode interlock. - Don't use the B_LOCKED flag and QUEUE_LOCKED for background write buffers. Check for the BKGRDINPROG flag before recycling or throwing away a buffer. We do this instead because it is not safe for us to move the original buffer to a new queue from the callback on the background write buffer. - Remove the B_LOCKED flag and the locked buffer queue. They are no longer used. - The vnode interlock is used around checks for BKGRDINPROG where it may not be strictly necessary. If we hold the buf lock the a back-ground write will not be started without our knowledge, one may only be completed while we're not looking. Rather than remove the code, Document two of the places where this extra locking is done. A pass should be done to verify and minimize the locking later.
2003-08-28 06:55:18 +00:00
} else {
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
if ((bp->b_flags & B_DELWRI) == 0 &&
(bp->b_xflags & BX_VNDIRTY))
panic("bqrelse: not dirty");
As a step towards the elimination of PG_CACHED pages, rework the handling of POSIX_FADV_DONTNEED so that it causes the backing pages to be moved to the head of the inactive queue instead of being cached. This affects the implementation of POSIX_FADV_NOREUSE as well, since it works by applying POSIX_FADV_DONTNEED to file ranges after they have been read or written. At that point the corresponding buffers may still be dirty, so the previous implementation would coalesce successive ranges and apply POSIX_FADV_DONTNEED to the result, ensuring that pages backing the dirty buffers would eventually be cached. To preserve this behaviour in an efficient manner, this change adds a new buf flag, B_NOREUSE, which causes the pages backing a VMIO buf to be placed at the head of the inactive queue when the buf is released. POSIX_FADV_NOREUSE then works by setting this flag in bufs that underlie the specified range. Reviewed by: alc, kib Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D3726
2015-09-30 23:06:29 +00:00
if ((bp->b_flags & B_NOREUSE) != 0) {
brelse(bp);
return;
}
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
qindex = QUEUE_CLEAN;
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
buf_track(bp, __func__);
/* binsfree unlocks bp. */
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
binsfree(bp, qindex);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
return;
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
out:
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
buf_track(bp, __func__);
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
/* unlock */
BUF_UNLOCK(bp);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
}
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
/*
* Complete I/O to a VMIO backed page. Validate the pages as appropriate,
* restore bogus pages.
*/
static void
vfs_vmio_iodone(struct buf *bp)
{
vm_ooffset_t foff;
vm_page_t m;
vm_object_t obj;
vfs: annotate variables only used by debug builds as __unused
2018-05-19 04:59:39 +00:00
struct vnode *vp __unused;
Style fixes. In particular, the variable "bogus" is used like a Boolean. Define it as such. Reviewed by: kib MFC after: 1 week
2017-03-19 23:06:11 +00:00
int i, iosize, resid;
bool bogus;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
obj = bp->b_bufobj->bo_object;
KASSERT(obj->paging_in_progress >= bp->b_npages,
("vfs_vmio_iodone: paging in progress(%d) < b_npages(%d)",
obj->paging_in_progress, bp->b_npages));
vp = bp->b_vp;
KASSERT(vp->v_holdcnt > 0,
("vfs_vmio_iodone: vnode %p has zero hold count", vp));
KASSERT(vp->v_object != NULL,
("vfs_vmio_iodone: vnode %p has no vm_object", vp));
foff = bp->b_offset;
KASSERT(bp->b_offset != NOOFFSET,
("vfs_vmio_iodone: bp %p has no buffer offset", bp));
Style fixes. In particular, the variable "bogus" is used like a Boolean. Define it as such. Reviewed by: kib MFC after: 1 week
2017-03-19 23:06:11 +00:00
bogus = false;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
iosize = bp->b_bcount - bp->b_resid;
VM_OBJECT_WLOCK(obj);
for (i = 0; i < bp->b_npages; i++) {
resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff;
if (resid > iosize)
resid = iosize;
/*
* cleanup bogus pages, restoring the originals
*/
m = bp->b_pages[i];
if (m == bogus_page) {
Style fixes. In particular, the variable "bogus" is used like a Boolean. Define it as such. Reviewed by: kib MFC after: 1 week
2017-03-19 23:06:11 +00:00
bogus = true;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
m = vm_page_lookup(obj, OFF_TO_IDX(foff));
if (m == NULL)
panic("biodone: page disappeared!");
bp->b_pages[i] = m;
} else if ((bp->b_iocmd == BIO_READ) && resid > 0) {
/*
* In the write case, the valid and clean bits are
* already changed correctly ( see bdwrite() ), so we
* only need to do this here in the read case.
*/
KASSERT((m->dirty & vm_page_bits(foff & PAGE_MASK,
resid)) == 0, ("vfs_vmio_iodone: page %p "
"has unexpected dirty bits", m));
vfs_page_set_valid(bp, foff, m);
}
KASSERT(OFF_TO_IDX(foff) == m->pindex,
("vfs_vmio_iodone: foff(%jd)/pindex(%ju) mismatch",
(intmax_t)foff, (uintmax_t)m->pindex));
vm_page_sunbusy(m);
foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK;
iosize -= resid;
}
Perform a single batched update to the object's paging-in-progress count rather than updating it for each page.
2015-10-03 17:04:52 +00:00
vm_object_pip_wakeupn(obj, bp->b_npages);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
VM_OBJECT_WUNLOCK(obj);
if (bogus && buf_mapped(bp)) {
BUF_CHECK_MAPPED(bp);
pmap_qenter(trunc_page((vm_offset_t)bp->b_data),
bp->b_pages, bp->b_npages);
}
}
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
/*
Elide the object lock in the common case in vfs_vmio_unwire(). The object lock was only needed when attempting to free B_DIRECT buffer pages, and for testing for invalid pages (and freeing them if so). Handle the latter by instead moving invalid pages near the head of the inactive queue, where they will be reclaimed quickly. Reviewed by: alc, kib, jeff MFC after: 3 weeks Differential Revision: https://reviews.freebsd.org/D14778
2018-03-21 21:15:43 +00:00
* Unwire a page held by a buf and either free it or update the page queues to
* reflect its recent use.
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
*/
static void
vfs_vmio_unwire(struct buf *bp, vm_page_t m)
{
bool freed;
vm_page_lock(m);
Dequeue wired pages lazily. Previously, wiring a page would cause it to be removed from its page queue. In the common case, unwiring causes it to be enqueued at the tail of that page queue. This change modifies vm_page_wire() to not dequeue the page, thus avoiding the highly contended page queue locks. Instead, vm_page_unwire() takes care of requeuing the page as a single operation, and the page daemon dequeues wired pages as they are encountered during a queue scan to avoid needlessly revisiting them later. For pages in PQ_ACTIVE we do even better, since a requeue is unnecessary. The change improves scalability for some common workloads. For instance, threads wiring pages into the buffer cache no longer need to modify global page queues, and unwiring is usually done by the bufspace thread, so concurrency is not as much of an issue. As another example, many sysctl handlers wire the output buffer to avoid faults on copyout, and since the buffer is likely to be in PQ_ACTIVE, we now entirely avoid modifying the page queue in this case. The change also adds a block comment describing some properties of struct vm_page's reference counters, and the busy lock. Reviewed by: jeff Discussed with: alc, kib MFC after: 1 month Differential Revision: https://reviews.freebsd.org/D11943
2018-02-07 16:57:10 +00:00
if (vm_page_unwire_noq(m)) {
Elide the object lock in the common case in vfs_vmio_unwire(). The object lock was only needed when attempting to free B_DIRECT buffer pages, and for testing for invalid pages (and freeing them if so). Handle the latter by instead moving invalid pages near the head of the inactive queue, where they will be reclaimed quickly. Reviewed by: alc, kib, jeff MFC after: 3 weeks Differential Revision: https://reviews.freebsd.org/D14778
2018-03-21 21:15:43 +00:00
if ((bp->b_flags & B_DIRECT) != 0)
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
freed = vm_page_try_to_free(m);
else
freed = false;
if (!freed) {
/*
Elide the object lock in the common case in vfs_vmio_unwire(). The object lock was only needed when attempting to free B_DIRECT buffer pages, and for testing for invalid pages (and freeing them if so). Handle the latter by instead moving invalid pages near the head of the inactive queue, where they will be reclaimed quickly. Reviewed by: alc, kib, jeff MFC after: 3 weeks Differential Revision: https://reviews.freebsd.org/D14778
2018-03-21 21:15:43 +00:00
* Use a racy check of the valid bits to determine
* whether we can accelerate reclamation of the page.
* The valid bits will be stable unless the page is
* being mapped or is referenced by multiple buffers,
* and in those cases we expect races to be rare. At
* worst we will either accelerate reclamation of a
* valid page and violate LRU, or unnecessarily defer
* reclamation of an invalid page.
*
* The B_NOREUSE flag marks data that is not expected to
* be reused, so accelerate reclamation in that case
* too. Otherwise, maintain LRU.
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
*/
Elide the object lock in the common case in vfs_vmio_unwire(). The object lock was only needed when attempting to free B_DIRECT buffer pages, and for testing for invalid pages (and freeing them if so). Handle the latter by instead moving invalid pages near the head of the inactive queue, where they will be reclaimed quickly. Reviewed by: alc, kib, jeff MFC after: 3 weeks Differential Revision: https://reviews.freebsd.org/D14778
2018-03-21 21:15:43 +00:00
if (m->valid == 0 || (bp->b_flags & B_NOREUSE) != 0)
As a step towards the elimination of PG_CACHED pages, rework the handling of POSIX_FADV_DONTNEED so that it causes the backing pages to be moved to the head of the inactive queue instead of being cached. This affects the implementation of POSIX_FADV_NOREUSE as well, since it works by applying POSIX_FADV_DONTNEED to file ranges after they have been read or written. At that point the corresponding buffers may still be dirty, so the previous implementation would coalesce successive ranges and apply POSIX_FADV_DONTNEED to the result, ensuring that pages backing the dirty buffers would eventually be cached. To preserve this behaviour in an efficient manner, this change adds a new buf flag, B_NOREUSE, which causes the pages backing a VMIO buf to be placed at the head of the inactive queue when the buf is released. POSIX_FADV_NOREUSE then works by setting this flag in bufs that underlie the specified range. Reviewed by: alc, kib Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D3726
2015-09-30 23:06:29 +00:00
vm_page_deactivate_noreuse(m);
Fix some races introduced in r332974. With r332974, when performing a synchronized access of a page's "queue" field, one must first check whether the page is logically dequeued. If so, then the page lock does not prevent the page from being removed from its page queue. Intoduce vm_page_queue(), which returns the page's logical queue index. In some cases, direct access to the "queue" field is still required, but such accesses should be confined to sys/vm. Reported and tested by: pho Reviewed by: kib Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D15280
2018-05-04 17:17:30 +00:00
else if (vm_page_active(m))
Dequeue wired pages lazily. Previously, wiring a page would cause it to be removed from its page queue. In the common case, unwiring causes it to be enqueued at the tail of that page queue. This change modifies vm_page_wire() to not dequeue the page, thus avoiding the highly contended page queue locks. Instead, vm_page_unwire() takes care of requeuing the page as a single operation, and the page daemon dequeues wired pages as they are encountered during a queue scan to avoid needlessly revisiting them later. For pages in PQ_ACTIVE we do even better, since a requeue is unnecessary. The change improves scalability for some common workloads. For instance, threads wiring pages into the buffer cache no longer need to modify global page queues, and unwiring is usually done by the bufspace thread, so concurrency is not as much of an issue. As another example, many sysctl handlers wire the output buffer to avoid faults on copyout, and since the buffer is likely to be in PQ_ACTIVE, we now entirely avoid modifying the page queue in this case. The change also adds a block comment describing some properties of struct vm_page's reference counters, and the busy lock. Reviewed by: jeff Discussed with: alc, kib MFC after: 1 month Differential Revision: https://reviews.freebsd.org/D11943
2018-02-07 16:57:10 +00:00
vm_page_reference(m);
else
Have vm_page_{deactivate,launder}() requeue already-queued pages. In many cases the page is not enqueued so the change will have no effect. However, the change is needed to support an optimization in the fault handler and in some cases (sendfile, the buffer cache) it was being emulated by the caller anyway. Reviewed by: alc Tested by: pho MFC after: 2 weeks X-Differential Revision: https://reviews.freebsd.org/D14625
2018-03-18 16:40:56 +00:00
vm_page_deactivate(m);
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
}
}
vm_page_unlock(m);
}
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
/*
* Perform page invalidation when a buffer is released. The fully invalid
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
* pages will be reclaimed later in vfs_vmio_truncate().
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
*/
static void
vfs_vmio_invalidate(struct buf *bp)
{
vm_object_t obj;
vm_page_t m;
int i, resid, poffset, presid;
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
if (buf_mapped(bp)) {
BUF_CHECK_MAPPED(bp);
pmap_qremove(trunc_page((vm_offset_t)bp->b_data), bp->b_npages);
} else
BUF_CHECK_UNMAPPED(bp);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
/*
* Get the base offset and length of the buffer. Note that
* in the VMIO case if the buffer block size is not
* page-aligned then b_data pointer may not be page-aligned.
* But our b_pages[] array *IS* page aligned.
*
* block sizes less then DEV_BSIZE (usually 512) are not
* supported due to the page granularity bits (m->valid,
* m->dirty, etc...).
*
* See man buf(9) for more information
*/
obj = bp->b_bufobj->bo_object;
resid = bp->b_bufsize;
poffset = bp->b_offset & PAGE_MASK;
VM_OBJECT_WLOCK(obj);
for (i = 0; i < bp->b_npages; i++) {
m = bp->b_pages[i];
if (m == bogus_page)
panic("vfs_vmio_invalidate: Unexpected bogus page.");
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
bp->b_pages[i] = NULL;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
- Fix a nonsense reordering that somehow slipped into my last diff. Reported by: pho
2015-09-23 07:44:07 +00:00
presid = resid > (PAGE_SIZE - poffset) ?
(PAGE_SIZE - poffset) : resid;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
KASSERT(presid >= 0, ("brelse: extra page"));
while (vm_page_xbusied(m)) {
vm_page_lock(m);
VM_OBJECT_WUNLOCK(obj);
Fix a race in vm_page_busy_sleep(9). Suppose that we have an exclusively busy page, and a thread which can accept shared-busy page. In this case, typical code waiting for the page xbusy state to pass is again: VM_OBJECT_WLOCK(object); ... if (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(object); <---1 vm_page_busy_sleep(p, "vmopax"); goto again; } Suppose that the xbusy state owner locked the object, unbusied the page and unlocked the object after we are at the line [1], but before we executed the load of the busy_lock word in vm_page_busy_sleep(). If it happens that there is still no waiters recorded for the busy state, the xbusy owner did not acquired the page lock, so it proceeded. More, suppose that some other thread happen to share-busy the page after xbusy state was relinquished but before the m->busy_lock is read in vm_page_busy_sleep(). Again, that thread only needs vm_object lock to proceed. Then, vm_page_busy_sleep() reads busy_lock value equal to the VPB_SHARERS_WORD(1). In this case, all tests in vm_page_busy_sleep(9) pass and we are going to sleep, despite the page being share-busied. Update check for m->busy_lock == VPB_UNBUSIED in vm_page_busy_sleep(9) to also accept shared-busy state if we only wait for the xbusy state to pass. Merge sequential if()s with the same 'then' clause in vm_page_busy_sleep(). Note that the current code does not share-busy pages from parallel threads, the only way to have more that one sbusy owner is right now is to recurse. Reported and tested by: pho (previous version) Reviewed by: alc, markj Sponsored by: The FreeBSD Foundation MFC after: 1 week Differential revision: https://reviews.freebsd.org/D8196
2016-10-13 14:41:05 +00:00
vm_page_busy_sleep(m, "mbncsh", true);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
VM_OBJECT_WLOCK(obj);
}
if (pmap_page_wired_mappings(m) == 0)
vm_page_set_invalid(m, poffset, presid);
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
vfs_vmio_unwire(bp, m);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
resid -= presid;
poffset = 0;
}
VM_OBJECT_WUNLOCK(obj);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
bp->b_npages = 0;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
}
/*
* Page-granular truncation of an existing VMIO buffer.
*/
static void
vfs_vmio_truncate(struct buf *bp, int desiredpages)
{
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
vm_object_t obj;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
vm_page_t m;
int i;
if (bp->b_npages == desiredpages)
return;
if (buf_mapped(bp)) {
BUF_CHECK_MAPPED(bp);
pmap_qremove((vm_offset_t)trunc_page((vm_offset_t)bp->b_data) +
(desiredpages << PAGE_SHIFT), bp->b_npages - desiredpages);
} else
BUF_CHECK_UNMAPPED(bp);
Elide the object lock in the common case in vfs_vmio_unwire(). The object lock was only needed when attempting to free B_DIRECT buffer pages, and for testing for invalid pages (and freeing them if so). Handle the latter by instead moving invalid pages near the head of the inactive queue, where they will be reclaimed quickly. Reviewed by: alc, kib, jeff MFC after: 3 weeks Differential Revision: https://reviews.freebsd.org/D14778
2018-03-21 21:15:43 +00:00
/*
* The object lock is needed only if we will attempt to free pages.
*/
obj = (bp->b_flags & B_DIRECT) != 0 ? bp->b_bufobj->bo_object : NULL;
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
if (obj != NULL)
VM_OBJECT_WLOCK(obj);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
for (i = desiredpages; i < bp->b_npages; i++) {
m = bp->b_pages[i];
KASSERT(m != bogus_page, ("allocbuf: bogus page found"));
bp->b_pages[i] = NULL;
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
vfs_vmio_unwire(bp, m);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
}
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
if (obj != NULL)
VM_OBJECT_WUNLOCK(obj);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
bp->b_npages = desiredpages;
}
/*
* Byte granular extension of VMIO buffers.
*/
static void
vfs_vmio_extend(struct buf *bp, int desiredpages, int size)
{
/*
* We are growing the buffer, possibly in a
* byte-granular fashion.
*/
vm_object_t obj;
vm_offset_t toff;
vm_offset_t tinc;
vm_page_t m;
/*
* Step 1, bring in the VM pages from the object, allocating
* them if necessary. We must clear B_CACHE if these pages
* are not valid for the range covered by the buffer.
*/
obj = bp->b_bufobj->bo_object;
VM_OBJECT_WLOCK(obj);
Introduce vm_page_grab_pages(), which is intended to replace loops calling vm_page_grab() on consecutive page indices. Besides simplifying the code in the caller, vm_page_grab_pages() allows for batching optimizations. For example, the current implementation replaces calls to vm_page_lookup() on consecutive page indices by cheaper calls to vm_page_next(). Reviewed by: kib, markj Tested by: pho (an earlier version) MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D11926
2017-08-09 04:23:04 +00:00
if (bp->b_npages < desiredpages) {
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
/*
* We must allocate system pages since blocking
* here could interfere with paging I/O, no
* matter which process we are.
*
* Only exclusive busy can be tested here.
* Blocking on shared busy might lead to
* deadlocks once allocbuf() is called after
* pages are vfs_busy_pages().
*/
Modify vm_page_grab_pages() to handle VM_ALLOC_NOWAIT. This will allow its use in sendfile_swapin(). Reviewed by: alc, kib MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D11942
2017-08-11 16:29:22 +00:00
(void)vm_page_grab_pages(obj,
Introduce vm_page_grab_pages(), which is intended to replace loops calling vm_page_grab() on consecutive page indices. Besides simplifying the code in the caller, vm_page_grab_pages() allows for batching optimizations. For example, the current implementation replaces calls to vm_page_lookup() on consecutive page indices by cheaper calls to vm_page_next(). Reviewed by: kib, markj Tested by: pho (an earlier version) MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D11926
2017-08-09 04:23:04 +00:00
OFF_TO_IDX(bp->b_offset) + bp->b_npages,
VM_ALLOC_SYSTEM | VM_ALLOC_IGN_SBUSY |
VM_ALLOC_NOBUSY | VM_ALLOC_WIRED,
&bp->b_pages[bp->b_npages], desiredpages - bp->b_npages);
bp->b_npages = desiredpages;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
}
/*
* Step 2. We've loaded the pages into the buffer,
* we have to figure out if we can still have B_CACHE
* set. Note that B_CACHE is set according to the
* byte-granular range ( bcount and size ), not the
* aligned range ( newbsize ).
*
* The VM test is against m->valid, which is DEV_BSIZE
* aligned. Needless to say, the validity of the data
* needs to also be DEV_BSIZE aligned. Note that this
* fails with NFS if the server or some other client
* extends the file's EOF. If our buffer is resized,
* B_CACHE may remain set! XXX
*/
toff = bp->b_bcount;
tinc = PAGE_SIZE - ((bp->b_offset + toff) & PAGE_MASK);
while ((bp->b_flags & B_CACHE) && toff < size) {
vm_pindex_t pi;
if (tinc > (size - toff))
tinc = size - toff;
pi = ((bp->b_offset & PAGE_MASK) + toff) >> PAGE_SHIFT;
m = bp->b_pages[pi];
vfs_buf_test_cache(bp, bp->b_offset, toff, tinc, m);
toff += tinc;
tinc = PAGE_SIZE;
}
VM_OBJECT_WUNLOCK(obj);
/*
* Step 3, fixup the KVA pmap.
*/
if (buf_mapped(bp))
bpmap_qenter(bp);
else
BUF_CHECK_UNMAPPED(bp);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
}
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
/*
* Check to see if a block at a particular lbn is available for a clustered
* write.
*/
static int
vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno)
{
struct buf *bpa;
int match;
match = 0;
/* If the buf isn't in core skip it */
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
if ((bpa = gbincore(&vp->v_bufobj, lblkno)) == NULL)
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
return (0);
/* If the buf is busy we don't want to wait for it */
- Add an interlock argument to BUF_LOCK and BUF_TIMELOCK. - Remove the buftimelock mutex and acquire the buf's interlock to protect these fields instead. - Hold the vnode interlock while locking bufs on the clean/dirty queues. This reduces some cases from one BUF_LOCK with a LK_NOWAIT and another BUF_LOCK with a LK_TIMEFAIL to a single lock. Reviewed by: arch, mckusick
2003-02-25 03:37:48 +00:00
if (BUF_LOCK(bpa, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
return (0);
/* Only cluster with valid clusterable delayed write buffers */
if ((bpa->b_flags & (B_DELWRI | B_CLUSTEROK | B_INVAL)) !=
(B_DELWRI | B_CLUSTEROK))
goto done;
if (bpa->b_bufsize != size)
goto done;
/*
* Check to see if it is in the expected place on disk and that the
* block has been mapped.
*/
if ((bpa->b_blkno != bpa->b_lblkno) && (bpa->b_blkno == blkno))
match = 1;
done:
BUF_UNLOCK(bpa);
return (match);
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
/*
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
* vfs_bio_awrite:
*
* Implement clustered async writes for clearing out B_DELWRI buffers.
* This is much better then the old way of writing only one buffer at
* a time. Note that we may not be presented with the buffers in the
* correct order, so we search for the cluster in both directions.
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
*/
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
int
stylistic polishing.
2004-09-15 20:54:23 +00:00
vfs_bio_awrite(struct buf *bp)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
{
- Complete part of the unfinished bufobj work by consistently using BO_LOCK/UNLOCK/MTX when manipulating the bufobj. - Create a new lock in the bufobj to lock bufobj fields independently. This leaves the vnode interlock as an 'identity' lock while the bufobj is an io lock. The bufobj lock is ordered before the vnode interlock and also before the mnt ilock. - Exploit this new lock order to simplify softdep_check_suspend(). - A few sync related functions are marked with a new XXX to note that we may not properly interlock against a non-zero bv_cnt when attempting to sync all vnodes on a mountlist. I do not believe this race is important. If I'm wrong this will make these locations easier to find. Reviewed by: kib (earlier diff) Tested by: kris, pho (earlier diff)
2008-03-22 09:15:16 +00:00
struct bufobj *bo;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
int i;
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
int j;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
daddr_t lblkno = bp->b_lblkno;
struct vnode *vp = bp->b_vp;
int ncl;
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
int nwritten;
Make our v_usecount vnode reference count work identically to the original BSD code. The association between the vnode and the vm_object no longer includes reference counts. The major difference is that vm_object's are no longer freed gratuitiously from the vnode, and so once an object is created for the vnode, it will last as long as the vnode does. When a vnode object reference count is incremented, then the underlying vnode reference count is incremented also. The two "objects" are now more intimately related, and so the interactions are now much less complex. When vnodes are now normally placed onto the free queue with an object still attached. The rundown of the object happens at vnode rundown time, and happens with exactly the same filesystem semantics of the original VFS code. There is absolutely no need for vnode_pager_uncache and other travesties like that anymore. A side-effect of these changes is that SMP locking should be much simpler, the I/O copyin/copyout optimizations work, NFS should be more ponderable, and further work on layered filesystems should be less frustrating, because of the totally coherent management of the vnode objects and vnodes. Please be careful with your system while running this code, but I would greatly appreciate feedback as soon a reasonably possible.
1998-01-06 05:26:17 +00:00
int size;
int maxcl;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
int gbflags;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
- Complete part of the unfinished bufobj work by consistently using BO_LOCK/UNLOCK/MTX when manipulating the bufobj. - Create a new lock in the bufobj to lock bufobj fields independently. This leaves the vnode interlock as an 'identity' lock while the bufobj is an io lock. The bufobj lock is ordered before the vnode interlock and also before the mnt ilock. - Exploit this new lock order to simplify softdep_check_suspend(). - A few sync related functions are marked with a new XXX to note that we may not properly interlock against a non-zero bv_cnt when attempting to sync all vnodes on a mountlist. I do not believe this race is important. If I'm wrong this will make these locations easier to find. Reviewed by: kib (earlier diff) Tested by: kris, pho (earlier diff)
2008-03-22 09:15:16 +00:00
bo = &vp->v_bufobj;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
gbflags = (bp->b_data == unmapped_buf) ? GB_UNMAPPED : 0;
Fix a problem that was caused by new (partial) support for merged cache metadata and VBLK type devices. The code is currently mostly disabled, and a work-around has been added to disabled attempted clustered writes for VBLK type device buffers. Clustered write of meta-data is currently a work in progress.
1995-12-13 03:47:01 +00:00
/*
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
* right now we support clustered writing only to regular files. If
* we find a clusterable block we could be in the middle of a cluster
* rather then at the beginning.
Fix a problem that was caused by new (partial) support for merged cache metadata and VBLK type devices. The code is currently mostly disabled, and a work-around has been added to disabled attempted clustered writes for VBLK type device buffers. Clustered write of meta-data is currently a work in progress.
1995-12-13 03:47:01 +00:00
*/
if ((vp->v_type == VREG) &&
(vp->v_mount != 0) && /* Only on nodes that have the size info */
Minor optimization.
1995-08-24 13:28:16 +00:00
(bp->b_flags & (B_CLUSTEROK | B_INVAL)) == B_CLUSTEROK) {
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
size = vp->v_mount->mnt_stat.f_iosize;
maxcl = MAXPHYS / size;
Fixed a variety of deadlock and panic bugs, removed the bypass code, and implemented the ability to limit bufferspace by memory consumed. (vfs_bio.c) Fixed recently introduced bugs that caused extra I/O to happen in some cases. (vfs_cluster.c) Submitted by: John Dyson
1995-01-24 10:00:46 +00:00
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_RLOCK(bo);
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
for (i = 1; i < maxcl; i++)
if (vfs_bio_clcheck(vp, size, lblkno + i,
bp->b_blkno + ((i * size) >> DEV_BSHIFT)) == 0)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
break;
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
for (j = 1; i + j <= maxcl && j <= lblkno; j++)
if (vfs_bio_clcheck(vp, size, lblkno - j,
bp->b_blkno - ((j * size) >> DEV_BSHIFT)) == 0)
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
break;
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_RUNLOCK(bo);
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
--j;
ncl = i + j;
MFS doesn't bother to associate a struct mount with the vnode...so work around this by not trying to cluster this type of I/O. Submitted by: John Dyson
1995-01-11 01:53:18 +00:00
/*
* this is a possible cluster write
*/
if (ncl != 1) {
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
BUF_UNLOCK(bp);
Add currently unused flag argument to the cluster_read(), cluster_write() and cluster_wbuild() functions. The flags to be allowed are a subset of the GB_* flags for getblk(). Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-14 20:28:26 +00:00
nwritten = cluster_wbuild(vp, size, lblkno - j, ncl,
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
gbflags);
Add currently unused flag argument to the cluster_read(), cluster_write() and cluster_wbuild() functions. The flags to be allowed are a subset of the GB_* flags for getblk(). Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-14 20:28:26 +00:00
return (nwritten);
MFS doesn't bother to associate a struct mount with the vnode...so work around this by not trying to cluster this type of I/O. Submitted by: John Dyson
1995-01-11 01:53:18 +00:00
}
Tie up some loose ends in vnode/object management. Remove an unneeded config option in pmap. Fix a problem with faulting in pages. Clean-up some loose ends in swap pager memory management. The system should be much more stable, but all subtile bugs aren't fixed yet.
1998-01-17 09:17:02 +00:00
}
General fixes to the vfs clustring code: 1) Make cluster buffer list be a non-malloced chain. This eliminates yet another 'evil' M_WAITOK and generally cleans up the code. 2) Fix write clustering for ext2fs. It was just broken. Also, ffs clustering had an efficiency problem that more bawrites were happening than should have been. 3) Make changes to buf.h to support the above, plus remove b_pfcent at the request of David Greenman. Reviewed by: davidg (partially)
1995-11-19 19:54:31 +00:00
bremfree(bp);
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
bp->b_flags |= B_ASYNC;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
/*
MFS doesn't bother to associate a struct mount with the vnode...so work around this by not trying to cluster this type of I/O. Submitted by: John Dyson
1995-01-11 01:53:18 +00:00
* default (old) behavior, writing out only one block
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* XXX returns b_bufsize instead of b_bcount for nwritten?
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
*/
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
nwritten = bp->b_bufsize;
Properly vector all bwrite() and BUF_WRITE() calls through the same path and s/BUF_WRITE()/bwrite()/ since it now does the same as bwrite().
2004-03-11 18:02:36 +00:00
(void) bwrite(bp);
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Some style fixes. Sponsored by: The FreeBSD Foundation
2013-03-14 20:31:39 +00:00
return (nwritten);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* getnewbuf_kva:
*
* Allocate KVA for an empty buf header according to gbflags.
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
*/
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
static int
getnewbuf_kva(struct buf *bp, int gbflags, int maxsize)
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
{
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_UNMAPPED) {
If the buffer lock has waiters after the buffer has changed identity then getnewbuf() needs to drop the buffer in order to wake waiters that might sleep on the buffer in the context of the old identity.
2006-10-02 02:06:27 +00:00
/*
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
* In order to keep fragmentation sane we only allocate kva
* in BKVASIZE chunks. XXX with vmem we can do page size.
If the buffer lock has waiters after the buffer has changed identity then getnewbuf() needs to drop the buffer in order to wake waiters that might sleep on the buffer in the context of the old identity.
2006-10-02 02:06:27 +00:00
*/
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
maxsize = (maxsize + BKVAMASK) & ~BKVAMASK;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (maxsize != bp->b_kvasize &&
bufkva_alloc(bp, maxsize, gbflags))
return (ENOSPC);
Implement a new totally dynamic (up to MAXPHYS) buffer kva allocation scheme. Additionally, add the capability for checking for unexpected kernel page faults. The maximum amount of kva space for buffers hasn't been decreased from where it is, but it will now be possible to do so. This scheme manages the kva space similar to the buffers themselves. If there isn't enough kva space because of usage or fragementation, buffers will be reclaimed until a buffer allocation is successful. This scheme should be very resistant to fragmentation problems until/if the LFS code is fixed and uses the bogus buffer locking scheme -- but a 'fixed' LFS is not likely to use such a scheme. Now there should be NO problem allocating buffers up to MAXPHYS.
1996-11-30 22:41:49 +00:00
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return (0);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
/*
* getnewbuf:
*
* Find and initialize a new buffer header, freeing up existing buffers
* in the bufqueues as necessary. The new buffer is returned locked.
*
* We block if:
* We have insufficient buffer headers
* We have insufficient buffer space
- Add a general purpose resource allocator, vmem, from NetBSD. It was originally inspired by the Solaris vmem detailed in the proceedings of usenix 2001. The NetBSD version was heavily refactored for bugs and simplicity. - Use this resource allocator to allocate the buffer and transient maps. Buffer cache defrags are reduced by 25% when used by filesystems with mixed block sizes. Ultimately this may permit dynamic buffer cache sizing on low KVA machines. Discussed with: alc, kib, attilio Tested by: pho Sponsored by: EMC / Isilon Storage Division
2013-06-28 03:51:20 +00:00
* buffer_arena is too fragmented ( space reservation fails )
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
* If we have to flush dirty buffers ( but we try to avoid this )
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
*
* The caller is responsible for releasing the reserved bufspace after
* allocbuf() is called.
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
*/
static struct buf *
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int maxsize, int gbflags)
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufdomain *bd;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
struct buf *bp;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bool metadata, reserved;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
getnewbuf: Initialize bp to avoid uninitialized pointer dereference and brelse(). This came in recently in r289279. Coverity CID: 1331561
2015-10-29 19:02:24 +00:00
bp = NULL;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
KASSERT((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC,
("GB_KVAALLOC only makes sense with GB_UNMAPPED"));
if (!unmapped_buf_allowed)
gbflags &= ~(GB_UNMAPPED | GB_KVAALLOC);
if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 ||
vp->v_type == VCHR)
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
metadata = true;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
else
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
metadata = false;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (vp == NULL)
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = &bdomain[0];
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
else
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bd = &bdomain[vp->v_bufobj.bo_domain];
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
counter_u64_add(getnewbufcalls, 1);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
reserved = false;
do {
if (reserved == false &&
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_reserve(bd, maxsize, metadata) != 0) {
counter_u64_add(getnewbufrestarts, 1);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
continue;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
reserved = true;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if ((bp = buf_alloc(bd)) == NULL) {
counter_u64_add(getnewbufrestarts, 1);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
continue;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (getnewbuf_kva(bp, gbflags, maxsize) == 0)
return (bp);
break;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
} while (buf_recycle(bd, false) == 0);
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (reserved)
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_release(bd, maxsize);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (bp != NULL) {
bp->b_flags |= B_INVAL;
brelse(bp);
Fix a problem with the new buffer_map management code. Additionally, decrease the size of buffer_map to approx 2/3 of what it used to be (buffer_map can be smaller now.) The original commit of these changes increased the size of buffer_map to the point where the system would not boot on large systems -- now large systems with large caches will have even less problems than before.
1996-12-01 15:46:40 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
bufspace_wait(bd, vp, gbflags, slpflag, slptimeo);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
return (NULL);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
/*
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
* buf_daemon:
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
* buffer flushing daemon. Buffers are normally flushed by the
* update daemon but if it cannot keep up this process starts to
* take the load in an attempt to prevent getnewbuf() from blocking.
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
*/
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
static struct kproc_desc buf_kp = {
"bufdaemon",
buf_daemon,
&bufdaemonproc
};
In keeping with style(9)'s recommendations on macros, use a ';' after each SYSINIT() macro invocation. This makes a number of lightweight C parsers much happier with the FreeBSD kernel source, including cflow's prcc and lxr. MFC after: 1 month Discussed with: imp, rink
2008-03-16 10:58:09 +00:00
SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &buf_kp);
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
static int
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
buf_flush(struct vnode *vp, struct bufdomain *bd, int target)
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
{
int flushed;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
flushed = flushbufqueues(vp, bd, target, 0);
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
if (flushed == 0) {
/*
* Could not find any buffers without rollback
* dependencies, so just write the first one
* in the hopes of eventually making progress.
*/
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
if (vp != NULL && target > 2)
target /= 2;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
flushbufqueues(vp, bd, target, 1);
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
}
return (flushed);
}
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
static void
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
buf_daemon()
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
{
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
struct bufdomain *bd;
int speedupreq;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
int lodirty;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
int i;
Major update to the way synchronization is done in the kernel. Highlights include: * Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The alpha port is still in transition and currently uses both.) * Per-CPU idle processes. * Interrupts are run in their own separate kernel threads and can be preempted (i386 only). Partially contributed by: BSDi (BSD/OS) Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
2000-09-07 01:33:02 +00:00
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
/*
* This process needs to be suspended prior to shutdown sync.
*/
Make bufdaemon and bufspacedaemon use kthread_suspend_check instead of kproc_suspend_check. In r329612 bufspacedaemon was turned into a thread of the bufdaemon process causing both to call kproc_suspend_check with the same proc argument and that function contains the following while loop: while (SIGISMEMBER(p->p_siglist, SIGSTOP)) { wakeup(&p->p_siglist); msleep(&p->p_siglist, &p->p_mtx, PPAUSE, "kpsusp", 0); } So one thread wakes up the other and the other wakes up the first again, locking up UP machines on shutdown. Also register the shutdown handlers with SHUTDOWN_PRI_LAST + 100 so they run after the syncer has shutdown, because the syncer can cause a situation where bufdaemon help is needed to proceed. PR: 227404 Reviewed by: kib Tested by: cy, rmacklem
2018-04-22 16:05:29 +00:00
EVENTHANDLER_REGISTER(shutdown_pre_sync, kthread_shutdown, curthread,
SHUTDOWN_PRI_LAST + 100);
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
/*
* Start the buf clean daemons as children threads.
*/
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
for (i = 0 ; i < buf_domains; i++) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
int error;
error = kthread_add((void (*)(void *))bufspace_daemon,
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
&bdomain[i], curproc, NULL, 0, 0, "bufspacedaemon-%d", i);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
if (error)
panic("error %d spawning bufspace daemon", error);
}
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
/*
* This process is allowed to take the buffer cache to the limit
*/
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
curthread->td_pflags |= TDP_NORUNNINGBUF | TDP_BUFNEED;
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_lock(&bdlock);
Introduce a mechanism to suspend/resume system processes. Suspend syncer and bufdaemon prior to disk sync during system shutdown.
2000-01-07 08:36:44 +00:00
for (;;) {
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
bd_request = 0;
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_unlock(&bdlock);
Make bufdaemon and bufspacedaemon use kthread_suspend_check instead of kproc_suspend_check. In r329612 bufspacedaemon was turned into a thread of the bufdaemon process causing both to call kproc_suspend_check with the same proc argument and that function contains the following while loop: while (SIGISMEMBER(p->p_siglist, SIGSTOP)) { wakeup(&p->p_siglist); msleep(&p->p_siglist, &p->p_mtx, PPAUSE, "kpsusp", 0); } So one thread wakes up the other and the other wakes up the first again, locking up UP machines on shutdown. Also register the shutdown handlers with SHUTDOWN_PRI_LAST + 100 so they run after the syncer has shutdown, because the syncer can cause a situation where bufdaemon help is needed to proceed. PR: 227404 Reviewed by: kib Tested by: cy, rmacklem
2018-04-22 16:05:29 +00:00
kthread_suspend_check();
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
/*
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
* Save speedupreq for this pass and reset to capture new
* requests.
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
*/
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
speedupreq = bd_speedupreq;
bd_speedupreq = 0;
/*
* Flush each domain sequentially according to its level and
* the speedup request.
*/
for (i = 0; i < buf_domains; i++) {
bd = &bdomain[i];
if (speedupreq)
lodirty = bd->bd_numdirtybuffers / 2;
else
lodirty = bd->bd_lodirtybuffers;
while (bd->bd_numdirtybuffers > lodirty) {
if (buf_flush(NULL, bd,
bd->bd_numdirtybuffers - lodirty) == 0)
break;
kern_yield(PRI_USER);
}
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
}
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
/*
* Only clear bd_request if we have reached our low water
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
* mark. The buf_daemon normally waits 1 second and
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
* then incrementally flushes any dirty buffers that have
* built up, within reason.
*
* If we were unable to hit our low water mark and couldn't
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
* find any flushable buffers, we sleep for a short period
* to avoid endless loops on unlockable buffers.
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
*/
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
mtx_lock(&bdlock);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
if (!BIT_EMPTY(BUF_DOMAINS, &bdlodirty)) {
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
/*
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
* We reached our low water mark, reset the
* request and sleep until we are needed again.
* The sleep is just so the suspend code works.
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
*/
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
bd_request = 0;
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
/*
* Do an extra wakeup in case dirty threshold
* changed via sysctl and the explicit transition
* out of shortfall was missed.
*/
bdirtywakeup();
if (runningbufspace <= lorunningspace)
runningwakeup();
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
msleep(&bd_request, &bdlock, PVM, "psleep", hz);
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
} else {
/*
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
* We couldn't find any flushable dirty buffers but
* still have too many dirty buffers, we
* have to sleep and try again. (rare)
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
*/
- Lock down the buffer cache's infrastructure code. This includes locks on buf lists, synchronization variables, and atomic ops for the counters. This change does not remove giant from any code although some pushdown may be possible. - In vfs_bio_awrite() don't access buf fields without the buf lock.
2003-02-09 09:47:31 +00:00
msleep(&bd_request, &bdlock, PVM, "qsleep", hz / 10);
Reimplement buf_daemon / getnewbuf() interaction for dealing with stressful situations. buf_daemon now makes a distinction between being woken up and its sleep timing out, and as a consequence is now much better able to dynamically tune itself to its environment. Reviewed by: Alfred Perlstein <bright@wintelcom.net>
1999-12-20 20:28:40 +00:00
}
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
}
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
}
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
/*
* flushbufqueues:
*
* Try to flush a buffer in the dirty queue. We must be careful to
* free up B_INVAL buffers instead of write them, which NFS is
* particularly sensitive to.
*/
Make various vnode related functions static
2005-02-10 12:28:58 +00:00
static int flushwithdeps = 0;
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
SYSCTL_INT(_vfs, OID_AUTO, flushwithdeps, CTLFLAG_RW, &flushwithdeps,
0, "Number of buffers flushed with dependecies that require rollbacks");
stylistic polishing.
2004-09-15 20:54:23 +00:00
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
static int
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
flushbufqueues(struct vnode *lvp, struct bufdomain *bd, int target,
int flushdeps)
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
{
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
struct bufqueue *bq;
In flushbufqueues(), do not allocate sentinel buffer on the stack, struct buf is large. Use sleeping malloc(9) call, and zero the allocated buf as a debugging feature.
2009-04-16 09:37:48 +00:00
struct buf *sentinel;
When the number of dirty buffers rises too high, the buf_daemon runs to help clean up. After selecting a potential buffer to write, this patch has it acquire a lock on the vnode that owns the buffer before trying to write it. The vnode lock is necessary to avoid a race with some other process holding the vnode locked and trying to flush its dirty buffers. In particular, if the vnode in question is a snapshot file, then the race can lead to a deadlock. To avoid slowing down the buf_daemon, it does a non-blocking lock request when trying to lock the vnode. If it fails to get the lock it skips over the buffer and continues down its queue looking for buffers to flush. Sponsored by: DARPA & NAI Labs.
2002-10-18 01:29:59 +00:00
struct vnode *vp;
- Add a missing vn_start_write() to flushbufqueues(). This could have caused snapshot related problems. - The vp can not be NULL here or we would panic in vfs_bio_awrite(). Stop confusing the logic by checking for it in several places. Submitted by: kirk and then rototilled by me to remove vp == NULL checks.
2003-10-05 22:16:08 +00:00
struct mount *mp;
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
struct buf *bp;
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
int hasdeps;
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
int flushed;
Reimplement r255797 using LK_TRYUPGRADE. The r255797 was: Increase the chance of the buffer write from the bufdaemon helper context to succeed. If the locked vnode which owns the buffer to be written is shared locked, try the non-blocking upgrade of the lock to exclusive. PR: kern/178997 Reported and tested by: Klaus Weber <fbsd-bugs-2013-1@unix-admin.de> Sponsored by: The FreeBSD Foundation MFC after: 1 week Approved by: re (glebius)
2013-09-29 18:04:57 +00:00
int error;
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
bool unlock;
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
flushed = 0;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bq = &bd->bd_dirtyq;
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
bp = NULL;
In flushbufqueues(), do not allocate sentinel buffer on the stack, struct buf is large. Use sleeping malloc(9) call, and zero the allocated buf as a debugging feature.
2009-04-16 09:37:48 +00:00
sentinel = malloc(sizeof(struct buf), M_TEMP, M_WAITOK | M_ZERO);
sentinel->b_qindex = QUEUE_SENTINEL;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_LOCK(bq);
TAILQ_INSERT_HEAD(&bq->bq_queue, sentinel, b_freelist);
BQ_UNLOCK(bq);
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
while (flushed != target) {
When helping the bufdaemon from the buffer allocation context, there is no sense to walk the whole dirty buffer queue. We are only interested in, and can operate on, the buffers owned by the current vnode [1]. Instead of calling generic queue flush routine, do VOP_FSYNC() if possible. Holding the dirty buffer queue lock in the bufdaemon, without dropping it, can cause starvation of buffer writes from other threads. This is esp. easy to reproduce on the big memory machines, where large files are written, causing almost all dirty buffers accumulating in several big files, which vnodes are locked by writers. Bufdaemon cannot flush any buffer, but is iterating over the whole dirty queue continuously. Since dirty queue mutex is not dropped, bufdone() in g_up thread is starved, usually deadlocking the machine [2]. Mitigate this by dropping the queue lock after the vnode is locked, allowing other queue lock contenders to make a progress. Discussed with: Jeff [1] Reported by: pho [2] Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Approved by: re (hrs)
2013-10-02 06:00:34 +00:00
maybe_yield();
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_LOCK(bq);
In flushbufqueues(), do not allocate sentinel buffer on the stack, struct buf is large. Use sleeping malloc(9) call, and zero the allocated buf as a debugging feature.
2009-04-16 09:37:48 +00:00
bp = TAILQ_NEXT(sentinel, b_freelist);
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
if (bp != NULL) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
TAILQ_REMOVE(&bq->bq_queue, sentinel, b_freelist);
TAILQ_INSERT_AFTER(&bq->bq_queue, bp, sentinel,
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
b_freelist);
When helping the bufdaemon from the buffer allocation context, there is no sense to walk the whole dirty buffer queue. We are only interested in, and can operate on, the buffers owned by the current vnode [1]. Instead of calling generic queue flush routine, do VOP_FSYNC() if possible. Holding the dirty buffer queue lock in the bufdaemon, without dropping it, can cause starvation of buffer writes from other threads. This is esp. easy to reproduce on the big memory machines, where large files are written, causing almost all dirty buffers accumulating in several big files, which vnodes are locked by writers. Bufdaemon cannot flush any buffer, but is iterating over the whole dirty queue continuously. Since dirty queue mutex is not dropped, bufdone() in g_up thread is starved, usually deadlocking the machine [2]. Mitigate this by dropping the queue lock after the vnode is locked, allowing other queue lock contenders to make a progress. Discussed with: Jeff [1] Reported by: pho [2] Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Approved by: re (hrs)
2013-10-02 06:00:34 +00:00
} else {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_UNLOCK(bq);
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
break;
When helping the bufdaemon from the buffer allocation context, there is no sense to walk the whole dirty buffer queue. We are only interested in, and can operate on, the buffers owned by the current vnode [1]. Instead of calling generic queue flush routine, do VOP_FSYNC() if possible. Holding the dirty buffer queue lock in the bufdaemon, without dropping it, can cause starvation of buffer writes from other threads. This is esp. easy to reproduce on the big memory machines, where large files are written, causing almost all dirty buffers accumulating in several big files, which vnodes are locked by writers. Bufdaemon cannot flush any buffer, but is iterating over the whole dirty queue continuously. Since dirty queue mutex is not dropped, bufdone() in g_up thread is starved, usually deadlocking the machine [2]. Mitigate this by dropping the queue lock after the vnode is locked, allowing other queue lock contenders to make a progress. Discussed with: Jeff [1] Reported by: pho [2] Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Approved by: re (hrs)
2013-10-02 06:00:34 +00:00
}
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
/*
* Skip sentinels inserted by other invocations of the
* flushbufqueues(), taking care to not reorder them.
*
* Only flush the buffers that belong to the
* vnode locked by the curthread.
*/
if (bp->b_qindex == QUEUE_SENTINEL || (lvp != NULL &&
bp->b_vp != lvp)) {
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_UNLOCK(bq);
vfs_bio: Remove a leading space (style) Introduced in r282085. Sponsored by: Dell EMC Isilon
2016-10-05 23:42:02 +00:00
continue;
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
}
When helping the bufdaemon from the buffer allocation context, there is no sense to walk the whole dirty buffer queue. We are only interested in, and can operate on, the buffers owned by the current vnode [1]. Instead of calling generic queue flush routine, do VOP_FSYNC() if possible. Holding the dirty buffer queue lock in the bufdaemon, without dropping it, can cause starvation of buffer writes from other threads. This is esp. easy to reproduce on the big memory machines, where large files are written, causing almost all dirty buffers accumulating in several big files, which vnodes are locked by writers. Bufdaemon cannot flush any buffer, but is iterating over the whole dirty queue continuously. Since dirty queue mutex is not dropped, bufdone() in g_up thread is starved, usually deadlocking the machine [2]. Mitigate this by dropping the queue lock after the vnode is locked, allowing other queue lock contenders to make a progress. Discussed with: Jeff [1] Reported by: pho [2] Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Approved by: re (hrs)
2013-10-02 06:00:34 +00:00
error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_UNLOCK(bq);
When helping the bufdaemon from the buffer allocation context, there is no sense to walk the whole dirty buffer queue. We are only interested in, and can operate on, the buffers owned by the current vnode [1]. Instead of calling generic queue flush routine, do VOP_FSYNC() if possible. Holding the dirty buffer queue lock in the bufdaemon, without dropping it, can cause starvation of buffer writes from other threads. This is esp. easy to reproduce on the big memory machines, where large files are written, causing almost all dirty buffers accumulating in several big files, which vnodes are locked by writers. Bufdaemon cannot flush any buffer, but is iterating over the whole dirty queue continuously. Since dirty queue mutex is not dropped, bufdone() in g_up thread is starved, usually deadlocking the machine [2]. Mitigate this by dropping the queue lock after the vnode is locked, allowing other queue lock contenders to make a progress. Discussed with: Jeff [1] Reported by: pho [2] Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Approved by: re (hrs)
2013-10-02 06:00:34 +00:00
if (error != 0)
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
continue;
Remove b_pin_count from struct buf. It was added in r153192 for XFS and doesn't appear to have been used for anything else. XFS was disconnected in r241607 and removed entirely in r247631. Reported by: mlaier Reviewed by: imp, kib Differential Revision: https://reviews.freebsd.org/D7468
2016-08-11 07:58:23 +00:00
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
/*
* BKGRDINPROG can only be set with the buf and bufobj
* locks both held. We tolerate a race to clear it here.
*/
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
if ((bp->b_vflags & BV_BKGRDINPROG) != 0 ||
(bp->b_flags & B_DELWRI) == 0) {
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
BUF_UNLOCK(bp);
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
continue;
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
}
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
if (bp->b_flags & B_INVAL) {
When helping the bufdaemon from the buffer allocation context, there is no sense to walk the whole dirty buffer queue. We are only interested in, and can operate on, the buffers owned by the current vnode [1]. Instead of calling generic queue flush routine, do VOP_FSYNC() if possible. Holding the dirty buffer queue lock in the bufdaemon, without dropping it, can cause starvation of buffer writes from other threads. This is esp. easy to reproduce on the big memory machines, where large files are written, causing almost all dirty buffers accumulating in several big files, which vnodes are locked by writers. Bufdaemon cannot flush any buffer, but is iterating over the whole dirty queue continuously. Since dirty queue mutex is not dropped, bufdone() in g_up thread is starved, usually deadlocking the machine [2]. Mitigate this by dropping the queue lock after the vnode is locked, allowing other queue lock contenders to make a progress. Discussed with: Jeff [1] Reported by: pho [2] Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Approved by: re (hrs)
2013-10-02 06:00:34 +00:00
bremfreef(bp);
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
brelse(bp);
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
flushed++;
continue;
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
}
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
Use LIST_EMPTY() instead of unrolled version (LIST_FIRST() [!=]= NULL)
2007-02-22 14:52:59 +00:00
if (!LIST_EMPTY(&bp->b_dep) && buf_countdeps(bp, 0)) {
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
if (flushdeps == 0) {
BUF_UNLOCK(bp);
continue;
}
hasdeps = 1;
} else
hasdeps = 0;
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
/*
* We must hold the lock on a vnode before writing
* one of its buffers. Otherwise we may confuse, or
* in the case of a snapshot vnode, deadlock the
* system.
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
*
* The lock order here is the reverse of the normal
* of vnode followed by buf lock. This is ok because
* the NOWAIT will prevent deadlock.
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
*/
- Add a missing vn_start_write() to flushbufqueues(). This could have caused snapshot related problems. - The vp can not be NULL here or we would panic in vfs_bio_awrite(). Stop confusing the logic by checking for it in several places. Submitted by: kirk and then rototilled by me to remove vp == NULL checks.
2003-10-05 22:16:08 +00:00
vp = bp->b_vp;
if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
BUF_UNLOCK(bp);
continue;
}
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
if (lvp == NULL) {
unlock = true;
error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT);
} else {
ASSERT_VOP_LOCKED(vp, "getbuf");
unlock = false;
error = VOP_ISLOCKED(vp) == LK_EXCLUSIVE ? 0 :
vn_lock(vp, LK_TRYUPGRADE);
}
Reimplement r255797 using LK_TRYUPGRADE. The r255797 was: Increase the chance of the buffer write from the bufdaemon helper context to succeed. If the locked vnode which owns the buffer to be written is shared locked, try the non-blocking upgrade of the lock to exclusive. PR: kern/178997 Reported and tested by: Klaus Weber <fbsd-bugs-2013-1@unix-admin.de> Sponsored by: The FreeBSD Foundation MFC after: 1 week Approved by: re (glebius)
2013-09-29 18:04:57 +00:00
if (error == 0) {
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "flushbufqueue(%p) vp %p flags %X",
bp, bp->b_vp, bp->b_flags);
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
if (curproc == bufdaemonproc) {
vfs_bio_awrite(bp);
} else {
bremfree(bp);
bwrite(bp);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
counter_u64_add(notbufdflushes, 1);
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
}
- Add a missing vn_start_write() to flushbufqueues(). This could have caused snapshot related problems. - The vp can not be NULL here or we would panic in vfs_bio_awrite(). Stop confusing the logic by checking for it in several places. Submitted by: kirk and then rototilled by me to remove vp == NULL checks.
2003-10-05 22:16:08 +00:00
vn_finished_write(mp);
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
if (unlock)
VOP_UNLOCK(vp, 0);
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
flushwithdeps += hasdeps;
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
flushed++;
Partially revert r255986: do not call VOP_FSYNC() when helping bufdaemon in getnewbuf(), do use buf_flush(). The difference is that bufdaemon uses TRYLOCK to get buffer locks, which allows calls to getnewbuf() while another buffer is locked. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2015-04-27 11:13:19 +00:00
/*
* Sleeping on runningbufspace while holding
* vnode lock leads to deadlock.
*/
if (curproc == bufdaemonproc &&
runningbufspace > hirunningspace)
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
waitrunningbufspace();
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
continue;
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
}
- Add a mising vn_finished_write() Pointy hat: jeff Found by: robert Obtained from: kirk
2003-10-14 00:38:34 +00:00
vn_finished_write(mp);
- Remove a race between fsync like functions and flushbufqueues() by requiring locked bufs in vfs_bio_awrite(). Previously the buf could have been written out by fsync before we acquired the buf lock if it weren't for giant. The cluster_wbuild() handles this race properly but the single write at the end of vfs_bio_awrite() would not. - Modify flushbufqueues() so there is only one copy of the loop. Pass a parameter in that says whether or not we should sync bufs with deps. - Call flushbufqueues() a second time and then break if we couldn't find any bufs without deps.
2003-03-13 07:19:23 +00:00
BUF_UNLOCK(bp);
The buffer daemon cannot skip over buffers owned by locked inodes as they may be the only viable ones to flush. Thus it will now wait for an inode lock if the other alternatives will result in rollbacks (and immediate redirtying of the buffer). If only buffers with rollbacks are available, one will be flushed, but then the buffer daemon will wait briefly before proceeding. Failing to wait briefly effectively deadlocks a uniprocessor since every other process writing to that filesystem will wait for the buffer daemon to clean up which takes close enough to forever to feel like a deadlock. Reported by: Archie Cobbs <archie@dellroad.org> Sponsored by: DARPA & NAI Labs. Approved by: re
2002-12-14 01:35:30 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
BQ_LOCK(bq);
TAILQ_REMOVE(&bq->bq_queue, sentinel, b_freelist);
BQ_UNLOCK(bq);
In flushbufqueues(), do not allocate sentinel buffer on the stack, struct buf is large. Use sleeping malloc(9) call, and zero the allocated buf as a debugging feature.
2009-04-16 09:37:48 +00:00
free(sentinel, M_TEMP);
- Under heavy IO load the buf daemon can run for many hundereds of milliseconds due to what is essentially n^2 algorithmic complexity. This change makes the algorithm N*2 instead. This heavy processing manifested itself as skipping in audio and video playback due to the long scheduling latencies and contention on giant by pcm. - flushbufqueues() is now responsible for flushing multiple buffers rather than one at a time. This allows us to save our progress in the list by using a sentinal. We must do the numdirtywakeup() and waitrunningbufspace() here now rather than in buf_daemon(). - Also add a uio_yield() after we have processed the list once for bufs without deps and again for bufs with deps. This is to release Giant and allow any other giant locked code to proceed. Tested by: Many users on current@ Revealed by: schedgraph traces sent by Emil Mikulic & Anthony Ginepro
2005-06-08 20:26:05 +00:00
return (flushed);
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
* Check to see if a block is currently memory resident.
*/
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
struct buf *
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
incore(struct bufobj *bo, daddr_t blkno)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
struct buf *bp;
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_RLOCK(bo);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
bp = gbincore(bo, blkno);
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_RUNLOCK(bo);
Improvements to the cluster code, minor vfs_bio efficiency: Better performance -- more aggressive read-ahead under certain circumstanses. Mods to support clustering on small ( < PAGE_SIZE) block size filesystems (e.g. ext2fs, msdosfs.)
1995-09-03 19:56:15 +00:00
return (bp);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
/*
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
* Returns true if no I/O is needed to access the
* associated VM object. This is like incore except
* it also hunts around in the VM system for the data.
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
*/
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
static int
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
inmem(struct vnode * vp, daddr_t blkno)
{
vm_object_t obj;
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
vm_offset_t toff, tinc, size;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
vm_page_t m;
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
vm_ooffset_t off;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
- Lock accesses to v_numoutput. - Lock calls to gbincore.
2002-09-25 02:11:37 +00:00
ASSERT_VOP_LOCKED(vp, "inmem");
With Alfred's permission, remove vm_mtx in favor of a fine-grained approach (this commit is just the first stage). Also add various GIANT_ macros to formalize the removal of Giant, making it easy to test in a more piecemeal fashion. These macros will allow us to test fine-grained locks to a degree before removing Giant, and also after, and to remove Giant in a piecemeal fashion via sysctl's on those subsystems which the authors believe can operate without Giant.
2001-07-04 16:20:28 +00:00
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
if (incore(&vp->v_bufobj, blkno))
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
return 1;
Another minor optimization, this time to incore().
1995-08-24 13:59:14 +00:00
if (vp->v_mount == NULL)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
return 0;
Don't use VOP_GETVOBJECT, use vp->v_object directly.
2005-01-25 00:40:01 +00:00
obj = vp->v_object;
if (obj == NULL)
return (0);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
size = PAGE_SIZE;
if (size > vp->v_mount->mnt_stat.f_iosize)
size = vp->v_mount->mnt_stat.f_iosize;
fix intermediate overflow in 'quad = int * int' situation by casting the arguments to the multiply to a quad equivalent. In this case, vm_ooffset_t. Reviewed by: Archie Cobbs <archie@whistle.com>
1998-12-14 21:17:37 +00:00
off = (vm_ooffset_t)blkno * (vm_ooffset_t)vp->v_mount->mnt_stat.f_iosize;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Use readlocking now that assertions on vm_page_lookup() are relaxed. Sponsored by: EMC / Isilon storage division Reviewed by: alc Tested by: flo, pho
2013-05-17 20:03:55 +00:00
VM_OBJECT_RLOCK(obj);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) {
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
m = vm_page_lookup(obj, OFF_TO_IDX(off + toff));
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (!m)
Introduce a global lock for the vm subsystem (vm_mtx). vm_mtx does not recurse and is required for most low level vm operations. faults can not be taken without holding Giant. Memory subsystems can now call the base page allocators safely. Almost all atomic ops were removed as they are covered under the vm mutex. Alpha and ia64 now need to catch up to i386's trap handlers. FFS and NFS have been tested, other filesystems will need minor changes (grabbing the vm lock when twiddling page properties). Reviewed (partially) by: jake, jhb
2001-05-19 01:28:09 +00:00
goto notinmem;
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
tinc = size;
if (tinc > PAGE_SIZE - ((toff + off) & PAGE_MASK))
tinc = PAGE_SIZE - ((toff + off) & PAGE_MASK);
vm_page_is_valid() wasn't expecting a large offset argument, it's expecting a sub-page offset. We were passing the file position, and vm_page_bits() could do some interesting things when base was larger PAGE_SIZE. if (size > PAGE_SIZE - base) size = PAGE_SIZE - base; is interesting when (PAGE_SIZE - base) is negative. I could imagine that this could have interesting consequences for memory page -> device block bit validation.
1998-05-01 15:10:59 +00:00
if (vm_page_is_valid(m,
(vm_offset_t) ((toff + off) & PAGE_MASK), tinc) == 0)
Introduce a global lock for the vm subsystem (vm_mtx). vm_mtx does not recurse and is required for most low level vm operations. faults can not be taken without holding Giant. Memory subsystems can now call the base page allocators safely. Almost all atomic ops were removed as they are covered under the vm mutex. Alpha and ia64 now need to catch up to i386's trap handlers. FFS and NFS have been tested, other filesystems will need minor changes (grabbing the vm lock when twiddling page properties). Reviewed (partially) by: jake, jhb
2001-05-19 01:28:09 +00:00
goto notinmem;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Use readlocking now that assertions on vm_page_lookup() are relaxed. Sponsored by: EMC / Isilon storage division Reviewed by: alc Tested by: flo, pho
2013-05-17 20:03:55 +00:00
VM_OBJECT_RUNLOCK(obj);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
return 1;
Introduce a global lock for the vm subsystem (vm_mtx). vm_mtx does not recurse and is required for most low level vm operations. faults can not be taken without holding Giant. Memory subsystems can now call the base page allocators safely. Almost all atomic ops were removed as they are covered under the vm mutex. Alpha and ia64 now need to catch up to i386's trap handlers. FFS and NFS have been tested, other filesystems will need minor changes (grabbing the vm lock when twiddling page properties). Reviewed (partially) by: jake, jhb
2001-05-19 01:28:09 +00:00
notinmem:
Use readlocking now that assertions on vm_page_lookup() are relaxed. Sponsored by: EMC / Isilon storage division Reviewed by: alc Tested by: flo, pho
2013-05-17 20:03:55 +00:00
VM_OBJECT_RUNLOCK(obj);
Introduce a global lock for the vm subsystem (vm_mtx). vm_mtx does not recurse and is required for most low level vm operations. faults can not be taken without holding Giant. Memory subsystems can now call the base page allocators safely. Almost all atomic ops were removed as they are covered under the vm mutex. Alpha and ia64 now need to catch up to i386's trap handlers. FFS and NFS have been tested, other filesystems will need minor changes (grabbing the vm lock when twiddling page properties). Reviewed (partially) by: jake, jhb
2001-05-19 01:28:09 +00:00
return (0);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
/*
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
* Set the dirty range for a buffer based on the status of the dirty
* bits in the pages comprising the buffer. The range is limited
* to the size of the buffer.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
* Tell the VM system that the pages associated with this buffer
* are clean. This is used for delayed writes where the data is
* going to go to disk eventually without additional VM intevention.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
* Note that while we only really need to clean through to b_bcount, we
* just go ahead and clean through to b_bufsize.
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
*/
static void
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
vfs_clean_pages_dirty_buf(struct buf *bp)
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
{
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
vm_ooffset_t foff, noff, eoff;
vm_page_t m;
int i;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0)
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
return;
Silence warnings.
1999-01-12 11:59:34 +00:00
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
foff = bp->b_offset;
KASSERT(bp->b_offset != NOOFFSET,
("vfs_clean_pages_dirty_buf: no buffer offset"));
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(bp->b_bufobj->bo_object);
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
vfs_drain_busy_pages(bp);
Refactor vfs_setdirty(), creating vfs_setdirty_locked_object(). Call vfs_setdirty_locked_object() from vfs_busy_pages() instead of vfs_setdirty(), thereby eliminating a second acquisition and release of the same vm object lock.
2006-10-29 00:04:39 +00:00
vfs_setdirty_locked_object(bp);
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
for (i = 0; i < bp->b_npages; i++) {
noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK;
eoff = noff;
if (eoff > bp->b_offset + bp->b_bufsize)
eoff = bp->b_offset + bp->b_bufsize;
m = bp->b_pages[i];
vfs_page_set_validclean(bp, foff, m);
/* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */
foff = noff;
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object);
Refactor vfs_setdirty(), creating vfs_setdirty_locked_object(). Call vfs_setdirty_locked_object() from vfs_busy_pages() instead of vfs_setdirty(), thereby eliminating a second acquisition and release of the same vm object lock.
2006-10-29 00:04:39 +00:00
}
static void
vfs_setdirty_locked_object(struct buf *bp)
{
vm_object_t object;
int i;
object = bp->b_bufobj->bo_object;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(object);
Eliminate a comment describing code that was deleted over eight years ago. Move another comment to its proper place. Fix a typo in a third comment.
2009-06-01 06:12:08 +00:00
/*
* We qualify the scan for modified pages on whether the
* object has been flushed yet.
*/
Remove OBJ_CLEANING flag. The vfs_setdirty_locked_object() is the only consumer of the flag, and it used the flag because OBJ_MIGHTBEDIRTY was cleared early in vm_object_page_clean, before the cleaning pass was done. This is no longer true after r216799. Moreover, since OBJ_CLEANING is a flag, and not the counter, it could be reset too prematurely when parallel vm_object_page_clean() are performed. Reviewed by: alc (as a part of the bigger patch) MFC after: 1 month (after r216799 is merged)
2010-12-29 22:26:49 +00:00
if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
vm_offset_t boffset;
vm_offset_t eoffset;
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
/*
* test the pages to see if they have been modified directly
* by users through the VM system.
*/
Make vm_page's PG_ZERO flag immutable between the time of the page's allocation and deallocation. This flag's principal use is shortly after allocation. For such cases, clearing the flag is pointless. The only unusual use of PG_ZERO is in vfs_bio_clrbuf(). However, allocbuf() never requests a prezeroed page. So, vfs_bio_clrbuf() never sees a prezeroed page. Reviewed by: tegge@
2004-05-06 05:03:23 +00:00
for (i = 0; i < bp->b_npages; i++)
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
vm_page_test_dirty(bp->b_pages[i]);
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Calculate the encompassing dirty range, boffset and eoffset,
* (eoffset - boffset) bytes.
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
*/
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
for (i = 0; i < bp->b_npages; i++) {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
if (bp->b_pages[i]->dirty)
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
break;
}
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
boffset = (i << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK);
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
for (i = bp->b_npages - 1; i >= 0; --i) {
if (bp->b_pages[i]->dirty) {
break;
}
}
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
eoffset = ((i + 1) << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK);
/*
* Fit it to the buffer.
*/
if (eoffset > bp->b_bcount)
eoffset = bp->b_bcount;
/*
* If we have a good dirty range, merge with the existing
* dirty range.
*/
if (boffset < eoffset) {
if (bp->b_dirtyoff > boffset)
bp->b_dirtyoff = boffset;
if (bp->b_dirtyend < eoffset)
bp->b_dirtyend = eoffset;
}
Changes to fix the following bugs: 1) Files weren't properly synced on filesystems other than UFS. In some cases, this lead to lost data. Most likely would be noticed on NFS. The fix is to make the VM page sync/object_clean general rather than in each filesystem. 2) Mixing regular and mmaped file I/O on NFS was very broken. It caused chunks of files to end up as zeroes rather than the intended contents. The fix was to fix several race conditions and to kludge up the "b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention to page modifications that occurred via the mmapping. Reviewed by: David Greenman Submitted by: John Dyson
1995-05-21 21:39:31 +00:00
}
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
* Allocate the KVA mapping for an existing buffer.
* If an unmapped buffer is provided but a mapped buffer is requested, take
* also care to properly setup mappings between pages and KVA.
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
*/
static void
bp_unmapped_get_kva(struct buf *bp, daddr_t blkno, int size, int gbflags)
{
int bsize, maxsize, need_mapping, need_kva;
off_t offset;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
need_mapping = bp->b_data == unmapped_buf &&
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
(gbflags & GB_UNMAPPED) == 0;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
need_kva = bp->b_kvabase == unmapped_buf &&
bp->b_data == unmapped_buf &&
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
(gbflags & GB_KVAALLOC) != 0;
if (!need_mapping && !need_kva)
return;
BUF_CHECK_UNMAPPED(bp);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (need_mapping && bp->b_kvabase != unmapped_buf) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
* Buffer is not mapped, but the KVA was already
* reserved at the time of the instantiation. Use the
* allocated space.
*/
goto has_addr;
}
/*
* Calculate the amount of the address space we would reserve
* if the buffer was mapped.
*/
bsize = vn_isdisk(bp->b_vp, NULL) ? DEV_BSIZE : bp->b_bufobj->bo_bsize;
Add KASSERTs to catch the case where a developer may have forgotten to set bo_bsize on a bufobj. This is a slight modification of the patch provided. PR: 193146 Submitted by: Conrad Meyer <conrad.meyer@isilon.com> Sponsored by: EMC Isilon Storage Division
2014-09-04 00:10:06 +00:00
KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize"));
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
offset = blkno * bsize;
maxsize = size + (offset & PAGE_MASK);
maxsize = imax(maxsize, bsize);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
while (bufkva_alloc(bp, maxsize, gbflags) != 0) {
if ((gbflags & GB_NOWAIT_BD) != 0) {
/*
* XXXKIB: defragmentation cannot
* succeed, not sure what else to do.
*/
panic("GB_NOWAIT_BD and GB_UNMAPPED %p", bp);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
counter_u64_add(mappingrestarts, 1);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bufspace_wait(bufdomain(bp), bp->b_vp, gbflags, 0, 0);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
has_addr:
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (need_mapping) {
/* b_offset is handled by bpmap_qenter. */
bp->b_data = bp->b_kvabase;
BUF_CHECK_MAPPED(bp);
bpmap_qenter(bp);
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
struct buf *
getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo,
int flags)
{
struct buf *bp;
int error;
error = getblkx(vp, blkno, size, slpflag, slptimeo, flags, &bp);
if (error != 0)
return (NULL);
return (bp);
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
* getblkx:
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* Get a block given a specified block and offset into a file/device.
* The buffers B_DONE bit will be cleared on return, making it almost
* ready for an I/O initiation. B_INVAL may or may not be set on
* return. The caller should clear B_INVAL prior to initiating a
* READ.
*
* For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for
* an existing buffer.
*
* For a VMIO buffer, B_CACHE is modified according to the backing VM.
* If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set
* and then cleared based on the backing VM. If the previous buffer is
* non-0-sized but invalid, B_CACHE will be cleared.
*
* If getblk() must create a new buffer, the new buffer is returned with
* both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which
* case it is returned with B_INVAL clear and B_CACHE set based on the
* backing VM.
*
Properly vector all bwrite() and BUF_WRITE() calls through the same path and s/BUF_WRITE()/bwrite()/ since it now does the same as bwrite().
2004-03-11 18:02:36 +00:00
* getblk() also forces a bwrite() for any B_DELWRI buffer whos
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* B_CACHE bit is clear.
*
* What this means, basically, is that the caller should use B_CACHE to
* determine whether the buffer is fully valid or not and should clear
* B_INVAL prior to issuing a read. If the caller intends to validate
* the buffer by loading its data area with something, the caller needs
* to clear B_INVAL. If the caller does this without issuing an I/O,
* the caller should set B_CACHE ( as an optimization ), else the caller
* should issue the I/O and biodone() will set B_CACHE if the I/O was
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* a write attempt or if it was a successful read. If the caller
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
* intends to issue a READ, the caller must clear B_INVAL and BIO_ERROR
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* prior to issuing the READ. biodone() will *not* clear B_INVAL.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
int
getblkx(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo,
int flags, struct buf **bpp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
struct buf *bp;
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
struct bufobj *bo;
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
daddr_t d_blkno;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
int bsize, error, maxsize, vmio;
off_t offset;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "getblk(%p, %ld, %d)", vp, (long)blkno, size);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
KASSERT((flags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC,
("GB_KVAALLOC only makes sense with GB_UNMAPPED"));
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
ASSERT_VOP_LOCKED(vp, "getblk");
Make MAXBCACHEBUF a tunable called vfs.maxbcachebuf. By making MAXBCACHEBUF a tunable, it can be increased to allow for larger read/write data sizes for the NFS client. The tunable is limited to MAXPHYS, which is currently 128K. Making MAXPHYS a tunable or increasing its value is being discussed, since it would be nice to support a read/write data size of 1Mbyte for the NFS client when mounting the AmazonEFS file service. Reviewed by: kib MFC after: 2 weeks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D10991
2017-06-17 22:24:19 +00:00
if (size > maxbcachebuf)
panic("getblk: size(%d) > maxbcachebuf(%d)\n", size,
maxbcachebuf);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if (!unmapped_buf_allowed)
flags &= ~(GB_UNMAPPED | GB_KVAALLOC);
Potentially fix a problem, whereby MSDOSFS can request buffers larger than the vfs layer can provide. We now automatically support 32K clusters if MSDOSFS is installed, and panic if a filesystem tries to allocate a buffer larger than MAXBSIZE. This commit is a result of some "prodding" by BDE.
1996-11-28 04:26:04 +00:00
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
bo = &vp->v_bufobj;
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
d_blkno = blkno;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
loop:
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_RLOCK(bo);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
bp = gbincore(bo, blkno);
if (bp != NULL) {
- Add a new 'flags' parameter to getblk(). - Define one flag GB_LOCK_NOWAIT that tells getblk() to pass the LK_NOWAIT flag to the initial BUF_LOCK(). This will eventually be used in cases were we want to use a buffer only if it is not currently in use. - Convert all consumers of the getblk() api to use this extra parameter. Reviwed by: arch Not objected to by: mckusick
2003-03-04 00:04:44 +00:00
int lockflags;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
Do not call bremfree for managed buffers. Calling bremfree for these buffers results in panic: "bremfree: buffer %p not on a queue." Approved by: kib
2012-05-15 09:55:15 +00:00
* Buffer is in-core. If the buffer is not busy nor managed,
* it must be on a queue.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*/
- Add a new 'flags' parameter to getblk(). - Define one flag GB_LOCK_NOWAIT that tells getblk() to pass the LK_NOWAIT flag to the initial BUF_LOCK(). This will eventually be used in cases were we want to use a buffer only if it is not currently in use. - Convert all consumers of the getblk() api to use this extra parameter. Reviwed by: arch Not objected to by: mckusick
2003-03-04 00:04:44 +00:00
lockflags = LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
if ((flags & GB_LOCK_NOWAIT) != 0)
- Add a new 'flags' parameter to getblk(). - Define one flag GB_LOCK_NOWAIT that tells getblk() to pass the LK_NOWAIT flag to the initial BUF_LOCK(). This will eventually be used in cases were we want to use a buffer only if it is not currently in use. - Convert all consumers of the getblk() api to use this extra parameter. Reviwed by: arch Not objected to by: mckusick
2003-03-04 00:04:44 +00:00
lockflags |= LK_NOWAIT;
error = BUF_TIMELOCK(bp, lockflags,
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_LOCKPTR(bo), "getblk", slpflag, slptimeo);
- Add an interlock argument to BUF_LOCK and BUF_TIMELOCK. - Remove the buftimelock mutex and acquire the buf's interlock to protect these fields instead. - Hold the vnode interlock while locking bufs on the clean/dirty queues. This reduces some cases from one BUF_LOCK with a LK_NOWAIT and another BUF_LOCK with a LK_TIMEFAIL to a single lock. Reviewed by: arch, mckusick
2003-02-25 03:37:48 +00:00
/*
* If we slept and got the lock we have to restart in case
* the buffer changed identities.
*/
if (error == ENOLCK)
goto loop;
/* We timed out or were interrupted. */
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
else if (error != 0)
return (error);
Make recursive getblk() slightly more useful. Keep the buffer state intact if getblk() is done on the already owned buffer. Exit from brelse() early when the lock recursion is detected, otherwise brelse() might prematurely destroy the buffer under some circumstances. Sponsored by: The FreeBSD Foundation Noted by: mckusick Tested by: pho MFC after: 2 weeks
2013-02-27 07:34:09 +00:00
/* If recursed, assume caller knows the rules. */
else if (BUF_LOCKRECURSED(bp))
goto end;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
* The buffer is locked. B_CACHE is cleared if the buffer is
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
* invalid. Otherwise, for a non-VMIO buffer, B_CACHE is set
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* and for a VMIO buffer B_CACHE is adjusted according to the
* backing VM cache.
*/
if (bp->b_flags & B_INVAL)
bp->b_flags &= ~B_CACHE;
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
else if ((bp->b_flags & (B_VMIO | B_INVAL)) == 0)
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
bp->b_flags |= B_CACHE;
Do not call bremfree for managed buffers. Calling bremfree for these buffers results in panic: "bremfree: buffer %p not on a queue." Approved by: kib
2012-05-15 09:55:15 +00:00
if (bp->b_flags & B_MANAGED)
MPASS(bp->b_qindex == QUEUE_NONE);
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
else
Do not call bremfree for managed buffers. Calling bremfree for these buffers results in panic: "bremfree: buffer %p not on a queue." Approved by: kib
2012-05-15 09:55:15 +00:00
bremfree(bp);
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Some style fixes. Sponsored by: The FreeBSD Foundation
2013-03-14 20:31:39 +00:00
* check for size inconsistencies for non-VMIO case.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
if (bp->b_bcount != size) {
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
if ((bp->b_flags & B_VMIO) == 0 ||
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
(size > bp->b_kvasize)) {
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
if (bp->b_flags & B_DELWRI) {
Correct a problem where data OR metadata could be thrown away if a buffer is grown.
1998-03-17 17:36:05 +00:00
bp->b_flags |= B_NOCACHE;
Properly vector all bwrite() and BUF_WRITE() calls through the same path and s/BUF_WRITE()/bwrite()/ since it now does the same as bwrite().
2004-03-11 18:02:36 +00:00
bwrite(bp);
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
} else {
Use LIST_EMPTY() instead of unrolled version (LIST_FIRST() [!=]= NULL)
2007-02-22 14:52:59 +00:00
if (LIST_EMPTY(&bp->b_dep)) {
Correct a problem where data OR metadata could be thrown away if a buffer is grown.
1998-03-17 17:36:05 +00:00
bp->b_flags |= B_RELBUF;
brelse(bp);
} else {
bp->b_flags |= B_NOCACHE;
Properly vector all bwrite() and BUF_WRITE() calls through the same path and s/BUF_WRITE()/bwrite()/ since it now does the same as bwrite().
2004-03-11 18:02:36 +00:00
bwrite(bp);
Correct a problem where data OR metadata could be thrown away if a buffer is grown.
1998-03-17 17:36:05 +00:00
}
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
}
These changes fix a bug in the clustering code that I made worse when adding support for EXT2FS. Note that the Sig-11 problems appear to be caused by this, but there is still probably an underlying VM problem that let this clustering bug cause vnode objects to appear to be corrupted. The direct manifestation of this bug would have been severely mis-read files. It is possible that processes would Sig-11 on very damaged input files and might explain the mysterious differences in system behaviour when phk's malloc is being used.
1995-09-23 21:12:45 +00:00
goto loop;
}
}
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
* Handle the case of unmapped buffer which should
* become mapped, or the buffer for which KVA
* reservation is requested.
*/
bp_unmapped_get_kva(bp, blkno, size, flags);
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
/*
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* If the size is inconsistent in the VMIO case, we can resize
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* the buffer. This might lead to B_CACHE getting set or
* cleared. If the size has not changed, B_CACHE remains
* unchanged from its previous state.
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
*/
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
allocbuf(bp, size);
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
Split DIAGNOSTIC -> DIAGNOSTIC, INVARIANTS, and INVARIANT_SUPPORT as discussed on -hackers. Introduce 'KASSERT(assertion, ("panic message", args))' for simple check + panic. Reviewed by: msmith
1999-01-08 17:31:30 +00:00
KASSERT(bp->b_offset != NOOFFSET,
KNFize, by bde.
1999-01-10 01:58:29 +00:00
("getblk: no buffer offset"));
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* A buffer with B_DELWRI set and B_CACHE clear must
* be committed before we can return the buffer in
* order to prevent the caller from issuing a read
* ( due to B_CACHE not being set ) and overwriting
* it.
Fix an expression parenthesization typo in a conditional. It should not have any operational effects other then to make the code in question a little faster. Also added a more involved comment.
1999-01-23 06:36:15 +00:00
*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Most callers, including NFS and FFS, need this to
* operate properly either because they assume they
* can issue a read if B_CACHE is not set, or because
* ( for example ) an uncached B_DELWRI might loop due
* to softupdates re-dirtying the buffer. In the latter
* case, B_CACHE is set after the first write completes,
* preventing further loops.
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
* NOTE! b*write() sets B_CACHE. If we cleared B_CACHE
* above while extending the buffer, we cannot allow the
* buffer to remain with B_CACHE set after the write
* completes or it will represent a corrupt state. To
* deal with this we set B_NOCACHE to scrap the buffer
* after the write.
*
* We might be able to do something fancy, like setting
* B_CACHE in bwrite() except if B_DELWRI is already set,
* so the below call doesn't set B_CACHE, but that gets real
* confusing. This is much easier.
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
*/
if ((bp->b_flags & (B_CACHE|B_DELWRI)) == B_DELWRI) {
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
bp->b_flags |= B_NOCACHE;
Properly vector all bwrite() and BUF_WRITE() calls through the same path and s/BUF_WRITE()/bwrite()/ since it now does the same as bwrite().
2004-03-11 18:02:36 +00:00
bwrite(bp);
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
goto loop;
}
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
bp->b_flags &= ~B_DONE;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
} else {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* Buffer is not in-core, create new buffer. The buffer
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
* returned by getnewbuf() is locked. Note that the returned
* buffer is also considered valid (not marked B_INVAL).
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*/
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
BO_RUNLOCK(bo);
- Define a new flag for getblk(): GB_NOCREAT. This flag causes getblk() to bail out if the buffer is not already present. - The buffer returned by incore() is not locked and should not be sent to brelse(). Use getblk() with the new GB_NOCREAT flag to preserve the desired semantics.
2003-08-31 08:50:11 +00:00
/*
* If the user does not want us to create the buffer, bail out
* here.
*/
- Remove long dead splbio() calls and comments relating to the old synchronization mechanism.
2005-04-30 12:18:50 +00:00
if (flags & GB_NOCREAT)
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
return (EEXIST);
- Consolidate duplicate code into support functions. - Split the bqlock into bqclean and bqdirty locks. - Only acquire the wakeup synchronization locks when we cross a threshold requiring them. - Restructure the way flushbufqueues() targets work so they are more smp friendly and sane. Reviewed by: kib Discussed with: mckusick, attilio Sponsored by: EMC / Isilon Storage Division M vfs_bio.c
2013-06-05 23:53:00 +00:00
bo_bsize: revert r205860 and take an alternative approch in getblk In r205860 I missed the fact that there is code that strongly assumes that devvp bo_bsize is equal to underlying provider's sectorsize. In those places it is hard to obtain the sectorsize in an alternative way if devvp bo_bsize is set to something else. So, I am reverting bo_bsize assigment in g_vfs_open. Instead, in getblk I use DEV_BSIZE block size for b_offset calculation if vp is a disk vp as reported by vn_isdisk. This should coinside with vp being a devvp. Reported by: Mykola Dzham <i@levsha.me> Tested by: Mykola Dzham <i@levsha.me> Pointyhat to: avg MFC after: 2 weeks X-ToDo: convert bread(devvp) in all fs to use bo_bsize-d blocks
2010-04-02 15:12:31 +00:00
bsize = vn_isdisk(vp, NULL) ? DEV_BSIZE : bo->bo_bsize;
Add KASSERTs to catch the case where a developer may have forgotten to set bo_bsize on a bufobj. This is a slight modification of the patch provided. PR: 193146 Submitted by: Conrad Meyer <conrad.meyer@isilon.com> Sponsored by: EMC Isilon Storage Division
2014-09-04 00:10:06 +00:00
KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize"));
This commit adds basic support for the UFS2 filesystem. The UFS2 filesystem expands the inode to 256 bytes to make space for 64-bit block pointers. It also adds a file-creation time field, an ability to use jumbo blocks per inode to allow extent like pointer density, and space for extended attributes (up to twice the filesystem block size worth of attributes, e.g., on a 16K filesystem, there is space for 32K of attributes). UFS2 fully supports and runs existing UFS1 filesystems. New filesystems built using newfs can be built in either UFS1 or UFS2 format using the -O option. In this commit UFS1 is the default format, so if you want to build UFS2 format filesystems, you must specify -O 2. This default will be changed to UFS2 when UFS2 proves itself to be stable. In this commit the boot code for reading UFS2 filesystems is not compiled (see /sys/boot/common/ufsread.c) as there is insufficient space in the boot block. Once the size of the boot block is increased, this code can be defined. Things to note: the definition of SBSIZE has changed to SBLOCKSIZE. The header file <ufs/ufs/dinode.h> must be included before <ufs/ffs/fs.h> so as to get the definitions of ufs2_daddr_t and ufs_lbn_t. Still TODO: Verify that the first level bootstraps work for all the architectures. Convert the utility ffsinfo to understand UFS2 and test growfs. Add support for the extended attribute storage. Update soft updates to ensure integrity of extended attribute storage. Switch the current extended attribute interfaces to use the extended attribute storage. Add the extent like functionality (framework is there, but is currently never used). Sponsored by: DARPA & NAI Labs. Reviewed by: Poul-Henning Kamp <phk@freebsd.org>
2002-06-21 06:18:05 +00:00
offset = blkno * bsize;
Don't use VOP_GETVOBJECT, use vp->v_object directly.
2005-01-25 00:40:01 +00:00
vmio = vp->v_object != NULL;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if (vmio) {
maxsize = size + (offset & PAGE_MASK);
} else {
maxsize = size;
/* Do not allow non-VMIO notmapped buffers. */
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
flags &= ~(GB_UNMAPPED | GB_KVAALLOC);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
maxsize = imax(maxsize, bsize);
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
if ((flags & GB_NOSPARSE) != 0 && vmio &&
!vn_isdisk(vp, NULL)) {
error = VOP_BMAP(vp, blkno, NULL, &d_blkno, 0, 0);
KASSERT(error != EOPNOTSUPP,
("GB_NOSPARSE from fs not supporting bmap, vp %p",
vp));
if (error != 0)
return (error);
if (d_blkno == -1)
return (EJUSTRETURN);
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bp = getnewbuf(vp, slpflag, slptimeo, maxsize, flags);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
if (bp == NULL) {
- Remove long dead splbio() calls and comments relating to the old synchronization mechanism.
2005-04-30 12:18:50 +00:00
if (slpflag || slptimeo)
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
return (ETIMEDOUT);
Add a sched_yield() to work around low memory conditions in the current code. Things seem to get stuck in low memory conditions where no bufs are available, the reclamation path is called to wakeup the daemon, but no sleeping is done. Because of this, we are stuck in a tight loop in the current process and never run said reclamation path. This was introduced in r289279 . This is only a temporary workaround to restore system usefulness until the more permanent solutions can be found. Tested: * Carambola2, 64MB (and 32MB by manual config.)
2015-11-07 04:04:00 +00:00
/*
* XXX This is here until the sleep path is diagnosed
* enough to work under very low memory conditions.
*
* There's an issue on low memory, 4BSD+non-preempt
* systems (eg MIPS routers with 32MB RAM) where buffer
* exhaustion occurs without sleeping for buffer
* reclaimation. This just sticks in a loop and
* constantly attempts to allocate a buffer, which
* hits exhaustion and tries to wakeup bufdaemon.
* This never happens because we never yield.
*
* The real solution is to identify and fix these cases
* so we aren't effectively busy-waiting in a loop
* until the reclaimation path has cycles to run.
*/
kern_yield(PRI_USER);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
goto loop;
}
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
Added some code to make sure that buffers associated with directories and metadata aren't thrashed by regular file I/O. Added mechanism to limit the amount of outstanding I/O on a given vnode. Pagedaemon wakeup policy changed to skew priority a little in favor of file caching. Slight code reorganization to improve clarity. Added a few more comments. Submitted by: John Dyson
1995-02-22 09:30:13 +00:00
/*
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
* This code is used to make sure that a buffer is not
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
* created while the getnewbuf routine is blocked.
Make comment match code.
1999-03-02 21:23:38 +00:00
* This can be a problem whether the vnode is locked or not.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* If the buffer is created out from under us, we have to
- Remove long dead splbio() calls and comments relating to the old synchronization mechanism.
2005-04-30 12:18:50 +00:00
* throw away the one we just created.
Replace the global buffer hash table with per-vnode splay trees using a methodology similar to the vm_map_entry splay and the VM splay that Alan Cox is working on. Extensive testing has appeared to have shown no increase in overhead. Disadvantages Dirties more cache lines during lookups. Not as fast as a hash table lookup (but still N log N and optimal when there is locality of reference). Advantages vnode->v_dirtyblkhd is now perfectly sorted, making fsync/sync/filesystem syncer operate more efficiently. I get to rip out all the old hacks (some of which were mine) that tried to keep the v_dirtyblkhd tailq sorted. The per-vnode splay tree should be easier to lock / SMPng pushdown on vnodes will be easier. This commit along with another that Alan is working on for the VM page global hash table will allow me to implement ranged fsync(), optimize server-side nfs commit rpcs, and implement partial syncs by the filesystem syncer (aka filesystem syncer would detect that someone is trying to get the vnode lock, remembers its place, and skip to the next vnode). Note that the buffer cache splay is somewhat more complex then other splays due to special handling of background bitmap writes (multiple buffers with the same lblkno in the same vnode), and B_INVAL discontinuities between the old hash table and the existence of the buffer on the v_cleanblkhd list. Suggested by: alc
2002-07-10 17:02:32 +00:00
*
* Note: this must occur before we associate the buffer
* with the vp especially considering limitations in
* the splay tree implementation when dealing with duplicate
* lblkno's.
Added some code to make sure that buffers associated with directories and metadata aren't thrashed by regular file I/O. Added mechanism to limit the amount of outstanding I/O on a given vnode. Pagedaemon wakeup policy changed to skew priority a little in favor of file caching. Slight code reorganization to improve clarity. Added a few more comments. Submitted by: John Dyson
1995-02-22 09:30:13 +00:00
*/
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
BO_LOCK(bo);
if (gbincore(bo, blkno)) {
BO_UNLOCK(bo);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
bp->b_flags |= B_INVAL;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bufspace_release(bufdomain(bp), maxsize);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
brelse(bp);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
goto loop;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
Added some code to make sure that buffers associated with directories and metadata aren't thrashed by regular file I/O. Added mechanism to limit the amount of outstanding I/O on a given vnode. Pagedaemon wakeup policy changed to skew priority a little in favor of file caching. Slight code reorganization to improve clarity. Added a few more comments. Submitted by: John Dyson
1995-02-22 09:30:13 +00:00
/*
* Insert the buffer into the hash, so that it can
* be found by incore.
*/
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
bp->b_lblkno = blkno;
bp->b_blkno = d_blkno;
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
bp->b_offset = offset;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
bgetvp(vp, bp);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
BO_UNLOCK(bo);
Added some code to make sure that buffers associated with directories and metadata aren't thrashed by regular file I/O. Added mechanism to limit the amount of outstanding I/O on a given vnode. Pagedaemon wakeup policy changed to skew priority a little in favor of file caching. Slight code reorganization to improve clarity. Added a few more comments. Submitted by: John Dyson
1995-02-22 09:30:13 +00:00
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* set B_VMIO bit. allocbuf() the buffer bigger. Since the
* buffer size starts out as 0, B_CACHE will be set by
* allocbuf() for the VMIO case prior to it testing the
* backing store for validity.
*/
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
if (vmio) {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
bp->b_flags |= B_VMIO;
Don't use VOP_GETVOBJECT, use vp->v_object directly.
2005-01-25 00:40:01 +00:00
KASSERT(vp->v_object == bp->b_bufobj->bo_object,
Change buf->b_object to buf->b_bufobj->bo_object some whitespace fixes.
2004-11-04 09:06:54 +00:00
("ARGH! different b_bufobj->bo_object %p %p %p\n",
Don't use VOP_GETVOBJECT, use vp->v_object directly.
2005-01-25 00:40:01 +00:00
bp, vp->v_object, bp->b_bufobj->bo_object));
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
} else {
bp->b_flags &= ~B_VMIO;
Change buf->b_object to buf->b_bufobj->bo_object some whitespace fixes.
2004-11-04 09:06:54 +00:00
KASSERT(bp->b_bufobj->bo_object == NULL,
("ARGH! has b_bufobj->bo_object %p %p\n",
bp, bp->b_bufobj->bo_object));
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
BUF_CHECK_MAPPED(bp);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Added some code to make sure that buffers associated with directories and metadata aren't thrashed by regular file I/O. Added mechanism to limit the amount of outstanding I/O on a given vnode. Pagedaemon wakeup policy changed to skew priority a little in favor of file caching. Slight code reorganization to improve clarity. Added a few more comments. Submitted by: John Dyson
1995-02-22 09:30:13 +00:00
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
allocbuf(bp, size);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bufspace_release(bufdomain(bp), maxsize);
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
bp->b_flags &= ~B_DONE;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
}
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp);
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Make recursive getblk() slightly more useful. Keep the buffer state intact if getblk() is done on the already owned buffer. Exit from brelse() early when the lock recursion is detected, otherwise brelse() might prematurely destroy the buffer under some circumstances. Sponsored by: The FreeBSD Foundation Noted by: mckusick Tested by: pho MFC after: 2 weeks
2013-02-27 07:34:09 +00:00
end:
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
buf_track(bp, __func__);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
KASSERT(bp->b_bufobj == bo,
- Add and enhance asserts related to the wrong bufobj panic. Sponsored by: Isilon Systems, Inc. Approved by: re (blanket vfs)
2005-06-14 20:32:27 +00:00
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
Detect and optimize reads from the hole on UFS. - Create getblkx(9) variant of getblk(9) which can return error. - Add GB_NOSPARSE flag for getblk()/getblkx() which requests that BMAP was performed before the buffer is created, and EJUSTRETURN returned in case the requested block does not exist. - Make ffs_read() use GB_NOSPARSE to avoid instantiating buffer (and allocating the pages for it), copying from zero_region instead. The end result is less page allocations and buffer recycling when a hole is read, which is important for some benchmarks. Requested and reviewed by: jeff Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D14917
2018-05-13 09:47:28 +00:00
*bpp = bp;
return (0);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Get an empty, disassociated buffer of given size. The buffer is initially
* set to B_INVAL.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
struct buf *
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
geteblk(int size, int flags)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
struct buf *bp;
Commit the buffer cache cleanup patch to 4.x and 5.x. This patch fixes a fragmentation problem due to geteblk() reserving too much space for the buffer and imposes a larger granularity (16K) on KVA reservations for the buffer cache to avoid fragmentation issues. The buffer cache size calculations have been redone to simplify them (fewer defines, better comments, less chance of running out of KVA). The geteblk() fix solves a performance problem that DG was able reproduce. This patch does not completely fix the KVA fragmentation problems, but it goes a long way Mostly Reviewed by: bde and others Approved by: jkh
2000-03-27 21:29:33 +00:00
int maxsize;
maxsize = (size + BKVAMASK) & ~BKVAMASK;
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
while ((bp = getnewbuf(NULL, 0, 0, maxsize, flags)) == NULL) {
Fix two issues with bufdaemon, often causing the processes to hang in the "nbufkv" sleep. First, ffs background cg group block write requests a new buffer for the shadow copy. When ffs_bufwrite() is called from the bufdaemon due to buffers shortage, requesting the buffer deadlock bufdaemon. Introduce a new flag for getnewbuf(), GB_NOWAIT_BD, to request getblk to not block while allocating the buffer, and return failure instead. Add a flag argument to the geteblk to allow to pass the flags to getblk(). Do not repeat the getnewbuf() call from geteblk if buffer allocation failed and either GB_NOWAIT_BD is specified, or geteblk() is called from bufdaemon (or its helper, see below). In ffs_bufwrite(), fall back to synchronous cg block write if shadow block allocation failed. Since r107847, buffer write assumes that vnode owning the buffer is locked. The second problem is that buffer cache may accumulate many buffers belonging to limited number of vnodes. With such workload, quite often threads that own the mentioned vnodes locks are trying to read another block from the vnodes, and, due to buffer cache exhaustion, are asking bufdaemon for help. Bufdaemon is unable to make any substantial progress because the vnodes are locked. Allow the threads owning vnode locks to help the bufdaemon by doing the flush pass over the buffer cache before getnewbuf() is going to uninterruptible sleep. Move the flushing code from buf_daemon() to new helper function buf_do_flush(), that is called from getnewbuf(). The number of buffers flushed by single call to buf_do_flush() from getnewbuf() is limited by new sysctl vfs.flushbufqtarget. Prevent recursive calls to buf_do_flush() by marking the bufdaemon and threads that temporarily help bufdaemon by TDP_BUFNEED flag. In collaboration with: pho Reviewed by: tegge (previous version) Tested by: glebius, yandex ... MFC after: 3 weeks
2009-03-16 15:39:46 +00:00
if ((flags & GB_NOWAIT_BD) &&
(curthread->td_pflags & TDP_BUFNEED) != 0)
return (NULL);
}
Removed third arg (vmio) to allocbuf() that was added with the original merged cache changes, and figure it out based on the B_VMIO buffer flag. Fixes a problem where delayed write VMIO buffers would sometimes get recopied into kernel-alloced memory. Submitted by: John Dyson
1995-03-26 23:29:13 +00:00
allocbuf(bp, size);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
bufspace_release(bufdomain(bp), maxsize);
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
bp->b_flags |= B_INVAL; /* b_dep cleared by getnewbuf() */
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
return (bp);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
/*
* Truncate the backing store for a non-vmio buffer.
*/
static void
vfs_nonvmio_truncate(struct buf *bp, int newbsize)
{
if (bp->b_flags & B_MALLOC) {
/*
* malloced buffers are not shrunk
*/
if (newbsize == 0) {
bufmallocadjust(bp, 0);
free(bp->b_data, M_BIOBUF);
bp->b_data = bp->b_kvabase;
bp->b_flags &= ~B_MALLOC;
}
return;
}
vm_hold_free_pages(bp, newbsize);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufspace_adjust(bp, newbsize);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
}
/*
* Extend the backing for a non-VMIO buffer.
*/
static void
vfs_nonvmio_extend(struct buf *bp, int newbsize)
{
caddr_t origbuf;
int origbufsize;
/*
* We only use malloced memory on the first allocation.
* and revert to page-allocated memory when the buffer
* grows.
*
* There is a potential smp race here that could lead
* to bufmallocspace slightly passing the max. It
* is probably extremely rare and not worth worrying
* over.
*/
if (bp->b_bufsize == 0 && newbsize <= PAGE_SIZE/2 &&
bufmallocspace < maxbufmallocspace) {
bp->b_data = malloc(newbsize, M_BIOBUF, M_WAITOK);
bp->b_flags |= B_MALLOC;
bufmallocadjust(bp, newbsize);
return;
}
/*
* If the buffer is growing on its other-than-first
* allocation then we revert to the page-allocation
* scheme.
*/
origbuf = NULL;
origbufsize = 0;
if (bp->b_flags & B_MALLOC) {
origbuf = bp->b_data;
origbufsize = bp->b_bufsize;
bp->b_data = bp->b_kvabase;
bufmallocadjust(bp, 0);
bp->b_flags &= ~B_MALLOC;
newbsize = round_page(newbsize);
}
vm_hold_load_pages(bp, (vm_offset_t) bp->b_data + bp->b_bufsize,
(vm_offset_t) bp->b_data + newbsize);
if (origbuf != NULL) {
bcopy(origbuf, bp->b_data, origbufsize);
free(origbuf, M_BIOBUF);
}
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufspace_adjust(bp, newbsize);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
* This code constitutes the buffer memory from either anonymous system
* memory (in the case of non-VMIO operations) or from an associated
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
* VM object (in the case of VMIO operations). This code is able to
* resize a buffer up or down.
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
*
* Note that this code is tricky, and has many complications to resolve
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* deadlock or inconsistent data situations. Tread lightly!!!
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
* There are B_CACHE and B_DELWRI interactions that must be dealt with by
* the caller. Calling this code willy nilly can result in the loss of data.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* allocbuf() only adjusts B_CACHE for VMIO buffers. getblk() deals with
* B_CACHE for the non-VMIO case.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
int
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
allocbuf(struct buf *bp, int size)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
int newbsize;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Fixed broken offset use in vfs_unbusy_pages() which resulted in several different types of panics/inconsistencies with NFS clients. Cleared PG_WANTED where appropriate. Added checks for buffer busy in allocbuf and biodone. Reviewed by: John Dyson
1995-07-25 05:03:06 +00:00
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
if (bp->b_bcount == size)
return (1);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (bp->b_kvasize != 0 && bp->b_kvasize < size)
Implement a new totally dynamic (up to MAXPHYS) buffer kva allocation scheme. Additionally, add the capability for checking for unexpected kernel page faults. The maximum amount of kva space for buffers hasn't been decreased from where it is, but it will now be possible to do so. This scheme manages the kva space similar to the buffers themselves. If there isn't enough kva space because of usage or fragementation, buffers will be reclaimed until a buffer allocation is successful. This scheme should be very resistant to fragmentation problems until/if the LFS code is fixed and uses the bogus buffer locking scheme -- but a 'fixed' LFS is not likely to use such a scheme. Now there should be NO problem allocating buffers up to MAXPHYS.
1996-11-30 22:41:49 +00:00
panic("allocbuf: buffer too small");
sys: use our roundup2/rounddown2() macros when param.h is available. rounddown2 tends to produce longer lines than the original code and when the code has a high indentation level it was not really advantageous to do the replacement. This tries to strike a balance between readability using the macros and flexibility of having the expressions, so not everything is converted.
2016-04-21 19:57:40 +00:00
newbsize = roundup2(size, DEV_BSIZE);
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
if ((bp->b_flags & B_VMIO) == 0) {
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
if ((bp->b_flags & B_MALLOC) == 0)
newbsize = round_page(newbsize);
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Just get anonymous memory from the kernel. Don't
* mess with B_CACHE.
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
*/
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
if (newbsize < bp->b_bufsize)
vfs_nonvmio_truncate(bp, newbsize);
else if (newbsize > bp->b_bufsize)
vfs_nonvmio_extend(bp, newbsize);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
} else {
int desiredpages;
Correctly handle misaligned VMIO buffer (whose start or end offset in the VM object are not page aligned). This should fix the mount_msdos panic after a failed attemp to mount as ffs. Reviewed By: Matthew Dillon <dillon@apollo.backplane.com> Archie Cobbs <archie@whistle.com> Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
1998-12-22 14:43:58 +00:00
desiredpages = (size == 0) ? 0 :
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
num_pages((bp->b_offset & PAGE_MASK) + newbsize);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
1) Fix a bug that a buffer is removed from a queue, but the queue type is not set to QUEUE_NONE. This appears to have caused a hang bug that has been lurking. 2) Fix bugs that brelse'ing locked buffers do not "free" them, but the code assumes so. This can cause hangs when LFS is used. 3) Use malloced memory for directories when applicable. The amount of malloced memory is seriously limited, but should decrease the amount of memory used by an average directory to 1/4 - 1/2 previous. This capability is fully tunable. (Note that there is no config parameter, and might never be.) 4) Bias slightly the buffer cache usage towards non-VMIO buffers. Since the data in VMIO buffers is not lost when the buffer is reclaimed, this will help performance. This is adjustable also.
1996-03-02 04:40:56 +00:00
if (bp->b_flags & B_MALLOC)
panic("allocbuf: VMIO buffer can't be malloced");
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* Set B_CACHE initially if buffer is 0 length or will become
* 0-length.
*/
if (size == 0 || bp->b_bufsize == 0)
bp->b_flags |= B_CACHE;
1) Fix a bug that a buffer is removed from a queue, but the queue type is not set to QUEUE_NONE. This appears to have caused a hang bug that has been lurking. 2) Fix bugs that brelse'ing locked buffers do not "free" them, but the code assumes so. This can cause hangs when LFS is used. 3) Use malloced memory for directories when applicable. The amount of malloced memory is seriously limited, but should decrease the amount of memory used by an average directory to 1/4 - 1/2 previous. This capability is fully tunable. (Note that there is no config parameter, and might never be.) 4) Bias slightly the buffer cache usage towards non-VMIO buffers. Since the data in VMIO buffers is not lost when the buffer is reclaimed, this will help performance. This is adjustable also.
1996-03-02 04:40:56 +00:00
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
if (newbsize < bp->b_bufsize)
vfs_vmio_truncate(bp, desiredpages);
/* XXX This looks as if it should be newbsize > b_bufsize */
else if (size > bp->b_bcount)
vfs_vmio_extend(bp, desiredpages, size);
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
bufspace_adjust(bp, newbsize);
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
}
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
bp->b_bcount = size; /* requested buffer size. */
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function. - Allow vfs_vmio_invalidate() to free the pages, leaving us with a single loop and bufobj lock when B_NOCACHE/B_INVAL is used. - Eliminate the special B_ASYNC handling on free that has not been relevant for some time. - Remove the extraneous page busy from vfs_vmio_truncate(). Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon storage division
2015-09-27 05:16:06 +00:00
return (1);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
extern int inflight_transient_maps;
Avoid completing I/O when dumping core after a panic. Filesystem or pager completion callbacks are generally non-functional after a panic and may trigger deadlocks if invoked in this context (e.g., by attempting to destroying a buffer mapping). To avoid this situation, short-circuit I/O completion in biodone(). Reviewed by: imp Discussed with: mav MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D15592
2018-06-01 23:49:32 +00:00
static struct bio_queue nondump_bios;
Oops, broke the build there. Uninline biodone() now that it is non-trivial. Introduce biowait() function. Currently there is a race condition and the mitigation is a timeout/retry. It is not obvious what kind of locking (if any) is suitable for BIO_DONE, since the majority of users take are of this themselves, and only a few places actually rely on the wakeup. Sponsored by: DARPA & NAI Labs.
2002-09-13 11:28:31 +00:00
void
biodone(struct bio *bp)
{
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
struct mtx *mtxp;
Close a race in biodone(), whereby the bio_done field of the passed bio may have been freed and reassigned by the wakeup before being tested after releasing the bdonelock. There's a non-zero chance this is the cause of a few of the crashes knocking around with biodone() sitting in the stack backtrace. Reviewed By: phk@
2005-09-29 10:37:20 +00:00
void (*done)(struct bio *);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
vm_offset_t start, end;
stylistic polishing.
2004-09-15 20:54:23 +00:00
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
biotrack(bp, __func__);
Avoid completing I/O when dumping core after a panic. Filesystem or pager completion callbacks are generally non-functional after a panic and may trigger deadlocks if invoked in this context (e.g., by attempting to destroying a buffer mapping). To avoid this situation, short-circuit I/O completion in biodone(). Reviewed by: imp Discussed with: mav MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D15592
2018-06-01 23:49:32 +00:00
/*
* Avoid completing I/O when dumping after a panic since that may
* result in a deadlock in the filesystem or pager code. Note that
* this doesn't affect dumps that were started manually since we aim
* to keep the system usable after it has been resumed.
*/
if (__predict_false(dumping && SCHEDULER_STOPPED())) {
TAILQ_INSERT_HEAD(&nondump_bios, bp, bio_queue);
return;
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if ((bp->bio_flags & BIO_TRANSIENT_MAPPING) != 0) {
MFprojects/camlock r256619: Restore BIO_UNMAPPED and BIO_TRANSIENT_MAPPING in biodonne() when unmapping temporary mapped buffer. That fixes double unmap if biodone() called twice for the same BIO (but with different done methods). Move mapping removal before calling bio_done() method. I believe that it is very wrong to do anything to BIO after reporting completion. kib@ thinks it was done for some forgotten now case when bio_done() method needed mapped buffer. But 1) if BIO was sent as unmapped, then IMO done() should be called in the same way; 2) IMO there is no guatantee that buffer will be mapped at this point at all, for example, if all underlying stack supports unmapped I/O, so bio_done() handler can not expect that.
2013-10-21 06:44:55 +00:00
bp->bio_flags &= ~BIO_TRANSIENT_MAPPING;
bp->bio_flags |= BIO_UNMAPPED;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
start = trunc_page((vm_offset_t)bp->bio_data);
end = round_page((vm_offset_t)bp->bio_data + bp->bio_length);
Reset bp->bio_done to unmapped_buf when removing a transient map in biodone. Submitted by: Scott Ferris <scott.ferris@isilon.com> Sponsored by: EMC / Isilon Storage Division Reviewed by: kib
2015-03-16 20:00:09 +00:00
bp->bio_data = unmapped_buf;
Use atop() instead of OFF_TO_IDX() for convertion of addresses or addresses offsets, as intended. Suggested and reviewed by: alc Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2017-03-14 19:39:17 +00:00
pmap_qremove(start, atop(end - start));
MFprojects/camlock r256619: Restore BIO_UNMAPPED and BIO_TRANSIENT_MAPPING in biodonne() when unmapping temporary mapped buffer. That fixes double unmap if biodone() called twice for the same BIO (but with different done methods). Move mapping removal before calling bio_done() method. I believe that it is very wrong to do anything to BIO after reporting completion. kib@ thinks it was done for some forgotten now case when bio_done() method needed mapped buffer. But 1) if BIO was sent as unmapped, then IMO done() should be called in the same way; 2) IMO there is no guatantee that buffer will be mapped at this point at all, for example, if all underlying stack supports unmapped I/O, so bio_done() handler can not expect that.
2013-10-21 06:44:55 +00:00
vmem_free(transient_arena, start, end - start);
atomic_add_int(&inflight_transient_maps, -1);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
Close a race in biodone(), whereby the bio_done field of the passed bio may have been freed and reassigned by the wakeup before being tested after releasing the bdonelock. There's a non-zero chance this is the cause of a few of the crashes knocking around with biodone() sitting in the stack backtrace. Reviewed By: phk@
2005-09-29 10:37:20 +00:00
done = bp->bio_done;
MFprojects/camlock r256370: - Take BIO lock in biodone() only when there is no completion callback set and so we should wake up thread waiting in biowait(). - Remove msleep() timeout from biowait(). It was added 11 years ago, when there was no locks used, and it should not be needed any more.
2013-10-16 09:56:40 +00:00
if (done == NULL) {
mtxp = mtx_pool_find(mtxpool_sleep, bp);
mtx_lock(mtxp);
bp->bio_flags |= BIO_DONE;
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
wakeup(bp);
MFprojects/camlock r256370: - Take BIO lock in biodone() only when there is no completion callback set and so we should wake up thread waiting in biowait(). - Remove msleep() timeout from biowait(). It was added 11 years ago, when there was no locks used, and it should not be needed any more.
2013-10-16 09:56:40 +00:00
mtx_unlock(mtxp);
Do not set BIO_DONE if the BIO specifies a completion handler. biowait() will otherwise race with completions of such BIOs. In-tree code only calls biowait() on BIOs that do not specify a handler, so this change should not have any functional impact. Reviewed by: mav MFC after: 1 month Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D9070
2017-01-10 21:41:28 +00:00
} else
Close a race in biodone(), whereby the bio_done field of the passed bio may have been freed and reassigned by the wakeup before being tested after releasing the bdonelock. There's a non-zero chance this is the cause of a few of the crashes knocking around with biodone() sitting in the stack backtrace. Reviewed By: phk@
2005-09-29 10:37:20 +00:00
done(bp);
Oops, broke the build there. Uninline biodone() now that it is non-trivial. Introduce biowait() function. Currently there is a race condition and the mitigation is a timeout/retry. It is not obvious what kind of locking (if any) is suitable for BIO_DONE, since the majority of users take are of this themselves, and only a few places actually rely on the wakeup. Sponsored by: DARPA & NAI Labs.
2002-09-13 11:28:31 +00:00
}
/*
* Wait for a BIO to finish.
*/
int
biowait(struct bio *bp, const char *wchan)
{
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
struct mtx *mtxp;
Oops, broke the build there. Uninline biodone() now that it is non-trivial. Introduce biowait() function. Currently there is a race condition and the mitigation is a timeout/retry. It is not obvious what kind of locking (if any) is suitable for BIO_DONE, since the majority of users take are of this themselves, and only a few places actually rely on the wakeup. Sponsored by: DARPA & NAI Labs.
2002-09-13 11:28:31 +00:00
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
mtxp = mtx_pool_find(mtxpool_sleep, bp);
mtx_lock(mtxp);
Oops, broke the build there. Uninline biodone() now that it is non-trivial. Introduce biowait() function. Currently there is a race condition and the mitigation is a timeout/retry. It is not obvious what kind of locking (if any) is suitable for BIO_DONE, since the majority of users take are of this themselves, and only a few places actually rely on the wakeup. Sponsored by: DARPA & NAI Labs.
2002-09-13 11:28:31 +00:00
while ((bp->bio_flags & BIO_DONE) == 0)
MFprojects/camlock r256370: - Take BIO lock in biodone() only when there is no completion callback set and so we should wake up thread waiting in biowait(). - Remove msleep() timeout from biowait(). It was added 11 years ago, when there was no locks used, and it should not be needed any more.
2013-10-16 09:56:40 +00:00
msleep(bp, mtxp, PRIBIO, wchan, 0);
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
mtx_unlock(mtxp);
s/Danglish/English/ Some style issues. Change the timeout to be hz/10 instead of hz. Brucification by: bde.
2002-09-15 17:52:35 +00:00
if (bp->bio_error != 0)
Oops, broke the build there. Uninline biodone() now that it is non-trivial. Introduce biowait() function. Currently there is a race condition and the mitigation is a timeout/retry. It is not obvious what kind of locking (if any) is suitable for BIO_DONE, since the majority of users take are of this themselves, and only a few places actually rely on the wakeup. Sponsored by: DARPA & NAI Labs.
2002-09-13 11:28:31 +00:00
return (bp->bio_error);
Make biowait() check bio_error before the BIO_ERROR flag, to propery catch internal GEOM use of bio_error. Sponsored by: DARPA & NAI Labs.
2002-09-26 16:32:14 +00:00
if (!(bp->bio_flags & BIO_ERROR))
return (0);
Oops, broke the build there. Uninline biodone() now that it is non-trivial. Introduce biowait() function. Currently there is a race condition and the mitigation is a timeout/retry. It is not obvious what kind of locking (if any) is suitable for BIO_DONE, since the majority of users take are of this themselves, and only a few places actually rely on the wakeup. Sponsored by: DARPA & NAI Labs.
2002-09-13 11:28:31 +00:00
return (EIO);
}
Un-inline the non-trivial "trivial" bio* functions. Untangle devstat_end_transaction_bio()
2002-09-14 19:34:11 +00:00
void
biofinish(struct bio *bp, struct devstat *stat, int error)
{
if (error) {
bp->bio_error = error;
bp->bio_flags |= BIO_ERROR;
}
if (stat != NULL)
devstat_end_transaction_bio(stat, bp);
biodone(bp);
}
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
#if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING)
void
biotrack_buf(struct bio *bp, const char *location)
{
buf_track(bp->bio_track_bp, location);
}
#endif
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Hmm, diff/patch still doesn't like me. Missed one s/biowait/bufwait/g
2000-04-30 06:16:03 +00:00
* bufwait:
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* Wait for buffer I/O completion, returning error status. The buffer
Correct typos, mostly s/ a / an / where appropriate. Some whitespace cleanup, especially in troff files.
2003-01-01 18:49:04 +00:00
* is left locked and B_DONE on return. B_EINTR is converted into an EINTR
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* error and cleared.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
int
stylistic polishing.
2004-09-15 20:54:23 +00:00
bufwait(struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
if (bp->b_iocmd == BIO_READ)
bwait(bp, PRIBIO, "biord");
else
bwait(bp, PRIBIO, "biowr");
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
if (bp->b_flags & B_EINTR) {
bp->b_flags &= ~B_EINTR;
return (EINTR);
}
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
if (bp->b_ioflags & BIO_ERROR) {
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
return (bp->b_error ? bp->b_error : EIO);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
} else {
return (0);
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/*
Give struct bio it's own call back mechanism.
2000-05-01 13:36:25 +00:00
* bufdone:
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* Finish I/O on a buffer, optionally calling a completion function.
* This is usually called from an interrupt so process blocking is
* not allowed.
*
* biodone is also responsible for setting B_CACHE in a B_VMIO bp.
* In a non-VMIO bp, B_CACHE will be set on the next getblk()
* assuming B_INVAL is clear.
*
* For the VMIO case, we set B_CACHE if the op was a read and no
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* read error occurred, or if the op was a write. B_CACHE is never
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* set if the buffer is invalid or otherwise uncacheable.
*
[skip ci] fix typo in comment from r59840 MFC after: 2 weeks Sponsored by: The FreeBSD Foundation
2019-04-26 15:00:59 +00:00
* bufdone does not mess with B_INVAL, allowing the I/O routine or the
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* initiator to leave B_INVAL set to brelse the buffer out of existence
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* in the biodone routine.
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
*/
Complete the bio/buf divorce for all code below devfs::strategy Exceptions: Vinum untouched. This means that it cannot be compiled. Greg Lehey is on the case. CCD not converted yet, casts to struct buf (still safe) atapi-cd casts to struct buf to examine B_PHYS
2000-04-15 05:54:02 +00:00
void
bufdone(struct buf *bp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
- Don't drop the wref on the bufobj until after bufdone() has completed. Without this, threads waiting in bufobj_wwait() may wakeup prior to bufdone() completing. Sponsored by: Isilon Systems, Inc.
2005-01-28 17:48:58 +00:00
struct bufobj *dropobj;
Remove __P.
2002-03-19 21:25:46 +00:00
void (*biodone)(struct buf *);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
buf_track(bp, __func__);
- Add CTR calls to trace the lifecycle of a buffer. - Remove some KASSERTs which are invalid if the appropriate lock is not held. - Slightly restructure bremfree() so that it is more sane. - Change the flush code in bdwrite() to avoid acquiring a mutex whenever possible. - Change the flush code in bdwrite() to avoid holding the bufobj mutex while calling buf_countdeps(). This introduces a lock-order relationship with the softdep lock that can not otherwise be resolved. - Don't set B_DONE until bufdone() is complete, otherwise another processor may believe the buf is done before it is. - Only acquire Giant if the caller has set b_iodone. Don't grab giant around normal bufdone() calls. Sponsored By: Isilon Systems, Inc.
2005-01-24 10:47:04 +00:00
CTR3(KTR_BUF, "bufdone(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
- Don't drop the wref on the bufobj until after bufdone() has completed. Without this, threads waiting in bufobj_wwait() may wakeup prior to bufdone() completing. Sponsored by: Isilon Systems, Inc.
2005-01-28 17:48:58 +00:00
dropobj = NULL;
Fix a problem with the VN device. Specifically, the VN device can cause a problem of spiraling death due to buffer resource limitations. The vfs_bio code in general had little ability to handle buffer resource management, and now it does. Also, there are a lot more knobs for tuning the vfs_bio code now. The knobs came free because of the need that there always be some immediately available buffers (non-delayed or locked) for use. Note that the buffer cache code is much less likely to get bogged down with lots of delayed writes, even more so than before.
1997-06-15 17:56:53 +00:00
Reviewed by: Many at differnt times in differnt parts, including alan, john, me, luoqi, and kirk Submitted by: Matt Dillon <dillon@frebsd.org> This change implements a relatively sophisticated fix to getnewbuf(). There were two problems with getnewbuf(). First, the writerecursion can lead to a system stack overflow when you have NFS and/or VN devices in the system. Second, the free/dirty buffer accounting was completely broken. Not only did the nfs routines blow it trying to manually account for the buffer state, but the accounting that was done did not work well with the purpose of their existance: figuring out when getnewbuf() needs to sleep. The meat of the change is to kern/vfs_bio.c. The remaining diffs are all minor except for NFS, which includes both the fixes for bp interaction AND fixes for a 'biodone(): buffer already done' lockup. Sys/buf.h also contains a chaining structure which is not used by this patchset but is used by other patches that are coming soon. This patch deliniated by tags PRE_MAT_GETBUF and POST_MAT_GETBUF. (sorry for the missing T matt)
1999-03-12 02:24:58 +00:00
KASSERT(!(bp->b_flags & B_DONE), ("biodone: bp %p already done", bp));
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
Fixed broken offset use in vfs_unbusy_pages() which resulted in several different types of panics/inconsistencies with NFS clients. Cleared PG_WANTED where appropriate. Added checks for buffer busy in allocbuf and biodone. Reviewed by: John Dyson
1995-07-25 05:03:06 +00:00
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
runningbufwakeup(bp);
- Don't drop the wref on the bufobj until after bufdone() has completed. Without this, threads waiting in bufobj_wwait() may wakeup prior to bufdone() completing. Sponsored by: Isilon Systems, Inc.
2005-01-28 17:48:58 +00:00
if (bp->b_iocmd == BIO_WRITE)
dropobj = bp->b_bufobj;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
/* call optional completion function if requested */
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
if (bp->b_iodone != NULL) {
Draw the outline of "struct bio". Struct bio is the future carrier of I/O requests for "struct buf".
2000-04-02 09:26:51 +00:00
biodone = bp->b_iodone;
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
bp->b_iodone = NULL;
Draw the outline of "struct bio". Struct bio is the future carrier of I/O requests for "struct buf".
2000-04-02 09:26:51 +00:00
(*biodone) (bp);
- Don't drop the wref on the bufobj until after bufdone() has completed. Without this, threads waiting in bufobj_wwait() may wakeup prior to bufdone() completing. Sponsored by: Isilon Systems, Inc.
2005-01-28 17:48:58 +00:00
if (dropobj)
bufobj_wdrop(dropobj);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
return;
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (bp->b_flags & B_VMIO) {
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* Set B_CACHE if the op was a normal read and no error
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* occurred. B_CACHE is set for writes in the b*write()
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* routines.
*/
Remove B_READ, B_WRITE and B_FREEBUF and replace them with a new field in struct buf: b_iocmd. The b_iocmd is enforced to have exactly one bit set. B_WRITE was bogusly defined as zero giving rise to obvious coding mistakes. Also eliminate the redundant struct buf flag B_CALL, it can just as efficiently be done by comparing b_iodone to NULL. Should you get a panic or drop into the debugger, complaining about "b_iocmd", don't continue. It is likely to write on your disk where it should have been reading. This change is a step in the direction towards a stackable BIO capability. A lot of this patch were machine generated (Thanks to style(9) compliance!) Vinum users: Greg has not had time to test this yet, be careful.
2000-03-20 10:44:49 +00:00
if (bp->b_iocmd == BIO_READ &&
Move B_ERROR flag to b_ioflags and call it BIO_ERROR. (Much of this done by script) Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED. Move b_pblkno and b_iodone_chain to struct bio while we transition, they will be obsoleted once bio structs chain/stack. Add bio_queue field for struct bio aware disksort. Address a lot of stylistic issues brought up by bde.
2000-04-02 15:24:56 +00:00
!(bp->b_flags & (B_INVAL|B_NOCACHE)) &&
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
!(bp->b_ioflags & BIO_ERROR))
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
bp->b_flags |= B_CACHE;
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
vfs_vmio_iodone(bp);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Merge biodone_finish() back into biodone(). The primary purpose is to make the order of operations clearer to avoid the race condition that was fixed in r328914. In particular, this commit corrects a similar race that existed in the soft updates callback. Doing some sleuthing through the SVN repository, it appears that bufdone_finish() was added to support XFS: ------------------------------------------------------------------------ r153192 | rodrigc | 2005-12-06 19:39:08 -0800 (Tue, 06 Dec 2005) | 13 lines Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@ ------------------------------------------------------------------------ It does not appear to ever have been used for anything else. XFS was disconnected in r241607: ------------------------------------------------------------------------ r241607 | attilio | 2012-10-16 03:04:00 -0700 (Tue, 16 Oct 2012) | 5 lines Disconnect non-MPSAFE XFS from the build in preparation for dropping GIANT from VFS. This is not targeted for MFC. ------------------------------------------------------------------------ and removed entirely in r247631: ------------------------------------------------------------------------ r247631 | attilio | 2013-03-02 07:33:54 -0800 (Sat, 02 Mar 2013) | 5 lines Garbage collect XFS bits which are now already completely disconnected from the tree since few months. This is not targeted for MFC. ------------------------------------------------------------------------ Since XFS support is gone, there is no reason to retain biodone_finish(). Suggested by: Warner Losh (imp) Discussed with: cem, kib Tested by: Peter Holm (pho)
2018-02-09 19:50:47 +00:00
if (!LIST_EMPTY(&bp->b_dep))
buf_complete(bp);
Occasional cylinder-group check-hash errors were being reported on systems running with a heavy filesystem load. Tracking down this bug was elusive because there were actually two problems. Sometimes the in-memory check hash was wrong and sometimes the check hash computed when doing the read was wrong. The occurrence of either error caused a check-hash mismatch to be reported. The first error was that the check hash in the in-memory cylinder group was incorrect. This error was caused by the following sequence of events: - We read a cylinder-group buffer and the check hash is valid. - We update its cg_time and cg_old_time which makes the in-memory check-hash value invalid but we do not mark the cylinder group dirty. - We do not make any other changes to the cylinder group, so we never mark it dirty, thus do not write it out, and hence never update the incorrect check hash for the in-memory buffer. - Later, the buffer gets freed, but the page with the old incorrect check hash is still in the VM cache. - Later, we read the cylinder group again, and the first page with the old check hash is still in the VM cache, but some other pages are not, so we have to do a read. - The read does not actually get the first page from disk, but rather from the VM cache, resulting in the old check hash in the buffer. - The value computed after doing the read does not match causing the error to be printed. The fix for this problem is to only set cg_time and cg_old_time as the cylinder group is being written to disk. This keeps the in-memory check-hash valid unless the cylinder group has had other modifications which will require it to be written with a new check hash calculated. It also requires that the check hash be recalculated in the in-memory cylinder group when it is marked clean after doing a background write. The second problem was that the check hash computed at the end of the read was incorrect because the calculation of the check hash on completion of the read was being done too soon. - When a read completes we had the following sequence: - bufdone() -- b_ckhashcalc (calculates check hash) -- bufdone_finish() --- vfs_vmio_iodone() (replaces bogus pages with the cached ones) - When we are reading a buffer where one or more pages are already in memory (but not all pages, or we wouldn't be doing the read), the I/O is done with bogus_page mapped in for the pages that exist in the VM cache. This mapping is done to avoid corrupting the cached pages if there is any I/O overrun. The vfs_vmio_iodone() function is responsible for replacing the bogus_page(s) with the cached ones. But we were calculating the check hash before the bogus_page(s) were replaced. Hence, when we were calculating the check hash, we were partly reading from bogus_page, which means we calculated a bad check hash (e.g., because multiple pages have been mapped to bogus_page, so its contents are indeterminate). The second fix is to move the check-hash calculation from bufdone() to bufdone_finish() after the call to vfs_vmio_iodone() so that it computes the check hash over the correct set of pages. With these two changes, the occasional cylinder-group check-hash errors are gone. Submitted by: David Pfitzner <dpfitzner@netflix.com> Reviewed by: kib Tested by: David Pfitzner
2018-02-06 00:19:46 +00:00
if ((bp->b_flags & B_CKHASH) != 0) {
KASSERT(bp->b_iocmd == BIO_READ,
Merge biodone_finish() back into biodone(). The primary purpose is to make the order of operations clearer to avoid the race condition that was fixed in r328914. In particular, this commit corrects a similar race that existed in the soft updates callback. Doing some sleuthing through the SVN repository, it appears that bufdone_finish() was added to support XFS: ------------------------------------------------------------------------ r153192 | rodrigc | 2005-12-06 19:39:08 -0800 (Tue, 06 Dec 2005) | 13 lines Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@ ------------------------------------------------------------------------ It does not appear to ever have been used for anything else. XFS was disconnected in r241607: ------------------------------------------------------------------------ r241607 | attilio | 2012-10-16 03:04:00 -0700 (Tue, 16 Oct 2012) | 5 lines Disconnect non-MPSAFE XFS from the build in preparation for dropping GIANT from VFS. This is not targeted for MFC. ------------------------------------------------------------------------ and removed entirely in r247631: ------------------------------------------------------------------------ r247631 | attilio | 2013-03-02 07:33:54 -0800 (Sat, 02 Mar 2013) | 5 lines Garbage collect XFS bits which are now already completely disconnected from the tree since few months. This is not targeted for MFC. ------------------------------------------------------------------------ Since XFS support is gone, there is no reason to retain biodone_finish(). Suggested by: Warner Losh (imp) Discussed with: cem, kib Tested by: Peter Holm (pho)
2018-02-09 19:50:47 +00:00
("bufdone: b_iocmd %d not BIO_READ", bp->b_iocmd));
KASSERT(buf_mapped(bp), ("bufdone: bp %p not mapped", bp));
Occasional cylinder-group check-hash errors were being reported on systems running with a heavy filesystem load. Tracking down this bug was elusive because there were actually two problems. Sometimes the in-memory check hash was wrong and sometimes the check hash computed when doing the read was wrong. The occurrence of either error caused a check-hash mismatch to be reported. The first error was that the check hash in the in-memory cylinder group was incorrect. This error was caused by the following sequence of events: - We read a cylinder-group buffer and the check hash is valid. - We update its cg_time and cg_old_time which makes the in-memory check-hash value invalid but we do not mark the cylinder group dirty. - We do not make any other changes to the cylinder group, so we never mark it dirty, thus do not write it out, and hence never update the incorrect check hash for the in-memory buffer. - Later, the buffer gets freed, but the page with the old incorrect check hash is still in the VM cache. - Later, we read the cylinder group again, and the first page with the old check hash is still in the VM cache, but some other pages are not, so we have to do a read. - The read does not actually get the first page from disk, but rather from the VM cache, resulting in the old check hash in the buffer. - The value computed after doing the read does not match causing the error to be printed. The fix for this problem is to only set cg_time and cg_old_time as the cylinder group is being written to disk. This keeps the in-memory check-hash valid unless the cylinder group has had other modifications which will require it to be written with a new check hash calculated. It also requires that the check hash be recalculated in the in-memory cylinder group when it is marked clean after doing a background write. The second problem was that the check hash computed at the end of the read was incorrect because the calculation of the check hash on completion of the read was being done too soon. - When a read completes we had the following sequence: - bufdone() -- b_ckhashcalc (calculates check hash) -- bufdone_finish() --- vfs_vmio_iodone() (replaces bogus pages with the cached ones) - When we are reading a buffer where one or more pages are already in memory (but not all pages, or we wouldn't be doing the read), the I/O is done with bogus_page mapped in for the pages that exist in the VM cache. This mapping is done to avoid corrupting the cached pages if there is any I/O overrun. The vfs_vmio_iodone() function is responsible for replacing the bogus_page(s) with the cached ones. But we were calculating the check hash before the bogus_page(s) were replaced. Hence, when we were calculating the check hash, we were partly reading from bogus_page, which means we calculated a bad check hash (e.g., because multiple pages have been mapped to bogus_page, so its contents are indeterminate). The second fix is to move the check-hash calculation from bufdone() to bufdone_finish() after the call to vfs_vmio_iodone() so that it computes the check hash over the correct set of pages. With these two changes, the occasional cylinder-group check-hash errors are gone. Submitted by: David Pfitzner <dpfitzner@netflix.com> Reviewed by: kib Tested by: David Pfitzner
2018-02-06 00:19:46 +00:00
(*bp->b_ckhashcalc)(bp);
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
/*
* For asynchronous completions, release the buffer now. The brelse
Convert buffer locking from using the B_BUSY and B_WANTED flags to using lockmgr locks. This commit should be functionally equivalent to the old semantics. That is, all buffer locking is done with LK_EXCLUSIVE requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will be done in future commits.
1999-06-26 02:47:16 +00:00
* will do a wakeup there if necessary - so no need to do a wakeup
* here in the async case. The sync case always needs to do a wakeup.
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
*/
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
if (bp->b_flags & B_ASYNC) {
Some refactoring of the buf/vm interface. - Eliminate bogus page replacement that is inconsistently applied in the invalidation loop in brelse. This has been a no-op in modern times as biodone() is responsible for cleaning up after bogus pages. This would've spammed the console with printfs at a minimum. - Allow the compiler and human readers alike to reason about allocbuf() by splitting it into constituent parts. - Separate the VM manipulating and buf manipulating code in brelse() and bufdone() so that the intentions are clear. This makes it evident that there are several duplicated buf pages loops that will be consolidated at a later time. Reviewed by: kib Tested by: pho Sponsored by: EMC / Isilon Storage Division
2015-09-22 23:57:52 +00:00
if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_RELBUF)) ||
(bp->b_ioflags & BIO_ERROR))
init_main.c subr_autoconf.c: Add support for "interrupt driven configuration hooks". A component of the kernel can register a hook, most likely during auto-configuration, and receive a callback once interrupt services are available. This callback will occur before the root and dump devices are configured, so the configuration task can affect the selection of those two devices or complete any tasks that need to be performed prior to launching init. System boot is posponed so long as a hook is registered. The hook owner is responsible for removing the hook once their task is complete or the system boot can continue. kern_acct.c kern_clock.c kern_exit.c kern_synch.c kern_time.c: Change the interface and implementation for the kernel callout service. The new implemntaion is based on the work of Adam M. Costello and George Varghese, published in a technical report entitled "Redesigning the BSD Callout and Timer Facilities". The interface used in FreeBSD is a little different than the one outlined in the paper. The new function prototypes are: struct callout_handle timeout(void (*func)(void *), void *arg, int ticks); void untimeout(void (*func)(void *), void *arg, struct callout_handle handle); If a client wishes to remove a timeout, it must store the callout_handle returned by timeout and pass it to untimeout. The new implementation gives 0(1) insert and removal of callouts making this interface scale well even for applications that keep 100s of callouts outstanding. See the updated timeout.9 man page for more details.
1997-09-21 22:00:25 +00:00
brelse(bp);
else
bqrelse(bp);
- Don't drop the wref on the bufobj until after bufdone() has completed. Without this, threads waiting in bufobj_wwait() may wakeup prior to bufdone() completing. Sponsored by: Isilon Systems, Inc.
2005-01-28 17:48:58 +00:00
} else
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
bdone(bp);
Merge biodone_finish() back into biodone(). The primary purpose is to make the order of operations clearer to avoid the race condition that was fixed in r328914. In particular, this commit corrects a similar race that existed in the soft updates callback. Doing some sleuthing through the SVN repository, it appears that bufdone_finish() was added to support XFS: ------------------------------------------------------------------------ r153192 | rodrigc | 2005-12-06 19:39:08 -0800 (Tue, 06 Dec 2005) | 13 lines Changes imported from XFS for FreeBSD project: - add fields to struct buf (needed by XFS) - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3 - b_pin_count, count of pinned buffer - add new B_MANAGED flag - add breada() function to initiate asynchronous I/O on read-ahead blocks. - add bufdone_finish(), bpin(), bunpin_wait() functions Patches provided by: kan Reviewed by: phk Silence on: arch@ ------------------------------------------------------------------------ It does not appear to ever have been used for anything else. XFS was disconnected in r241607: ------------------------------------------------------------------------ r241607 | attilio | 2012-10-16 03:04:00 -0700 (Tue, 16 Oct 2012) | 5 lines Disconnect non-MPSAFE XFS from the build in preparation for dropping GIANT from VFS. This is not targeted for MFC. ------------------------------------------------------------------------ and removed entirely in r247631: ------------------------------------------------------------------------ r247631 | attilio | 2013-03-02 07:33:54 -0800 (Sat, 02 Mar 2013) | 5 lines Garbage collect XFS bits which are now already completely disconnected from the tree since few months. This is not targeted for MFC. ------------------------------------------------------------------------ Since XFS support is gone, there is no reason to retain biodone_finish(). Suggested by: Warner Losh (imp) Discussed with: cem, kib Tested by: Peter Holm (pho)
2018-02-09 19:50:47 +00:00
if (dropobj)
bufobj_wdrop(dropobj);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
/*
* This routine is called in lieu of iodone in the case of
* incomplete I/O. This keeps the busy status for pages
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* consistent.
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
*/
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
vfs_unbusy_pages(struct buf *bp)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
{
Nitpicking and dusting performed on a train. Removes trivial warnings about unused variables, labels and other lint.
1998-10-25 17:44:59 +00:00
int i;
undent some functions a bit.
2004-09-15 21:08:58 +00:00
vm_object_t obj;
vm_page_t m;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
This implements a better launder limiting solution. There was a solution in 4.2-REL which I ripped out in -stable and -current when implementing the low-memory handling solution. However, maxlaunder turns out to be the saving grace in certain very heavily loaded systems (e.g. newsreader box). The new algorithm limits the number of pages laundered in the first pageout daemon pass. If that is not sufficient then suceessive will be run without any limit. Write I/O is now pipelined using two sysctls, vfs.lorunningspace and vfs.hirunningspace. This prevents excessive buffered writes in the disk queues which cause long (multi-second) delays for reads. It leads to more stable (less jerky) and generally faster I/O streaming to disk by allowing required read ops (e.g. for indirect blocks and such) to occur without interrupting the write stream, amoung other things. NOTE: eventually, filesystem write I/O pipelining needs to be done on a per-device basis. At the moment it is globalized.
2000-12-26 19:41:38 +00:00
runningbufwakeup(bp);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if (!(bp->b_flags & B_VMIO))
return;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Change buf->b_object to buf->b_bufobj->bo_object some whitespace fixes.
2004-11-04 09:06:54 +00:00
obj = bp->b_bufobj->bo_object;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(obj);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
for (i = 0; i < bp->b_npages; i++) {
m = bp->b_pages[i];
if (m == bogus_page) {
m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_offset) + i);
Add b_bufobj to struct buf which eventually will eliminate the need for b_vp. Initialize b_bufobj for all buffers. Make incore() and gbincore() take a bufobj instead of a vnode. Make inmem() local to vfs_bio.c Change a lot of VI_[UN]LOCK(bp->b_vp) to BO_[UN]LOCK(bp->b_bufobj) also VI_MTX() to BO_MTX(), Make buf_vlist_add() take a bufobj instead of a vnode. Eliminate other uses of bp->b_vp where bp->b_bufobj will do. Various minor polishing: remove "register", turn panic into KASSERT, use new function declarations, TAILQ_FOREACH_SAFE() etc.
2004-10-22 08:47:20 +00:00
if (!m)
undent some functions a bit.
2004-09-15 21:08:58 +00:00
panic("vfs_unbusy_pages: page missing\n");
bp->b_pages[i] = m;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (buf_mapped(bp)) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
BUF_CHECK_MAPPED(bp);
pmap_qenter(trunc_page((vm_offset_t)bp->b_data),
bp->b_pages, bp->b_npages);
} else
BUF_CHECK_UNMAPPED(bp);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
vm_page_sunbusy(m);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Perform a single batched update to the object's paging-in-progress count rather than updating it for each page.
2015-10-03 17:04:52 +00:00
vm_object_pip_wakeupn(obj, bp->b_npages);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
/*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* vfs_page_set_valid:
This is a rather large commit that encompasses the new swapper, changes to the VM system to support the new swapper, VM bug fixes, several VM optimizations, and some additional revamping of the VM code. The specific bug fixes will be documented with additional forced commits. This commit is somewhat rough in regards to code cleanup issues. Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
1999-01-21 08:29:12 +00:00
*
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* Set the valid bits in a page based on the supplied offset. The
* range is restricted to the buffer's size.
*
* This routine is typically called after a read completes.
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
*/
static void
Eliminate vfs_page_set_valid()'s unused argument.
2007-12-02 01:28:35 +00:00
vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m)
Eliminate page queues locking from bufdone_finish() through the following changes: Rename vfs_page_set_valid() to vfs_page_set_validclean() to reflect what this function actually does. Suggested by: tegge Introduce a new version of vfs_page_set_valid() that does no more than what the function's name implies. Specifically, it does not update the page's dirty mask, and thus it does not require the page queues lock to be held. Update two of the three callers to the old vfs_page_set_valid() to call vfs_page_set_validclean() instead because they actually require the page's dirty mask to be cleared. Introduce vm_page_set_valid(). Reviewed by: tegge
2009-05-13 05:39:39 +00:00
{
vm_ooffset_t eoff;
/*
* Compute the end offset, eoff, such that [off, eoff) does not span a
* page boundary and eoff is not greater than the end of the buffer.
* The end of the buffer, in this case, is our file EOF, not the
* allocation size of the buffer.
*/
eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK;
if (eoff > bp->b_offset + bp->b_bcount)
eoff = bp->b_offset + bp->b_bcount;
/*
* Set valid range. This is typically the entire buffer and thus the
* entire page.
*/
if (eoff > off)
Rename vm_page_set_valid() to vm_page_set_valid_range(). The vm_page_set_valid() is the most reasonable name for the m->valid accessor. Reviewed by: attilio, alc
2011-11-30 17:39:00 +00:00
vm_page_set_valid_range(m, off & PAGE_MASK, eoff - off);
Eliminate page queues locking from bufdone_finish() through the following changes: Rename vfs_page_set_valid() to vfs_page_set_validclean() to reflect what this function actually does. Suggested by: tegge Introduce a new version of vfs_page_set_valid() that does no more than what the function's name implies. Specifically, it does not update the page's dirty mask, and thus it does not require the page queues lock to be held. Update two of the three callers to the old vfs_page_set_valid() to call vfs_page_set_validclean() instead because they actually require the page's dirty mask to be cleared. Introduce vm_page_set_valid(). Reviewed by: tegge
2009-05-13 05:39:39 +00:00
}
/*
* vfs_page_set_validclean:
*
* Set the valid bits and clear the dirty bits in a page based on the
* supplied offset. The range is restricted to the buffer's size.
*/
static void
vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m)
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
{
The previous fix didn't work properly for small block size filesystems, which caused very slow file access for cd9660 and some ext2fs filesystems. Reviewed by: bde
1997-05-30 22:25:35 +00:00
vm_ooffset_t soff, eoff;
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
Catch a case spotted by Tor where files mmapped could leave garbage in the unallocated parts of the last page when the file ended on a frag but not a page boundary. Delimitted by tags PRE_MATT_MMAP_EOF and POST_MATT_MMAP_EOF, in files alpha/alpha/pmap.c i386/i386/pmap.c nfs/nfs_bio.c vm/pmap.h vm/vm_page.c vm/vm_page.h vm/vnode_pager.c miscfs/specfs/spec_vnops.c ufs/ufs/ufs_readwrite.c kern/vfs_bio.c Submitted by: Matt Dillon <dillon@freebsd.org> Reviewed by: Alan Cox <alc@freebsd.org>
1999-04-05 19:38:30 +00:00
/*
* Start and end offsets in buffer. eoff - soff may not cross a
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* page boundary or cross the end of the buffer. The end of the
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* buffer, in this case, is our file EOF, not the allocation size
* of the buffer.
Catch a case spotted by Tor where files mmapped could leave garbage in the unallocated parts of the last page when the file ended on a frag but not a page boundary. Delimitted by tags PRE_MATT_MMAP_EOF and POST_MATT_MMAP_EOF, in files alpha/alpha/pmap.c i386/i386/pmap.c nfs/nfs_bio.c vm/pmap.h vm/vm_page.c vm/vm_page.h vm/vnode_pager.c miscfs/specfs/spec_vnops.c ufs/ufs/ufs_readwrite.c kern/vfs_bio.c Submitted by: Matt Dillon <dillon@freebsd.org> Reviewed by: Alan Cox <alc@freebsd.org>
1999-04-05 19:38:30 +00:00
*/
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
soff = off;
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
eoff = (off + PAGE_SIZE) & ~(off_t)PAGE_MASK;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
if (eoff > bp->b_offset + bp->b_bcount)
eoff = bp->b_offset + bp->b_bcount;
Catch a case spotted by Tor where files mmapped could leave garbage in the unallocated parts of the last page when the file ended on a frag but not a page boundary. Delimitted by tags PRE_MATT_MMAP_EOF and POST_MATT_MMAP_EOF, in files alpha/alpha/pmap.c i386/i386/pmap.c nfs/nfs_bio.c vm/pmap.h vm/vm_page.c vm/vm_page.h vm/vnode_pager.c miscfs/specfs/spec_vnops.c ufs/ufs/ufs_readwrite.c kern/vfs_bio.c Submitted by: Matt Dillon <dillon@freebsd.org> Reviewed by: Alan Cox <alc@freebsd.org>
1999-04-05 19:38:30 +00:00
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* Set valid range. This is typically the entire buffer and thus the
* entire page.
*/
if (eoff > soff) {
vm_page_set_validclean(
m,
(vm_offset_t) (soff & PAGE_MASK),
(vm_offset_t) (eoff - soff)
);
}
Fix a few bugs with NFS and mmap caused by NFS' use of b_validoff and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully can be tidied up later by a slight redesign. PR: kern/2573, kern/2754, kern/3046 (possibly) Reviewed by: dyson
1997-05-19 14:36:56 +00:00
}
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
/*
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
* Ensure that all buffer pages are not exclusive busied. If any page is
* exclusive busy, drain it.
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
*/
Both cluster_rbuild() and cluster_wbuild() sometimes set the pages shared busy without first draining the hard busy state. Previously it went unnoticed since VPO_BUSY and m->busy fields were distinct, and vm_page_io_start() did not verified that the passed page has VPO_BUSY flag cleared, but such page state is wrong. New implementation is more strict and catched this case. Drain the busy state as needed, before calling vm_page_sbusy(). Tested by: pho, jkim Sponsored by: The FreeBSD Foundation
2013-08-22 18:26:45 +00:00
void
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
vfs_drain_busy_pages(struct buf *bp)
{
vm_page_t m;
int i, last_busied;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(bp->b_bufobj->bo_object);
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
last_busied = 0;
for (i = 0; i < bp->b_npages; i++) {
m = bp->b_pages[i];
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
if (vm_page_xbusied(m)) {
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
for (; last_busied < i; last_busied++)
The vm_pageout_flush() functions sbusies pages in the passed pages run. After that, the pager put method is called, usually translated to VOP_WRITE(). For the filesystems which use buffer cache, bufwrite() sbusies the buffer pages again, waiting for the xbusy state to drain. The later is done in vfs_drain_busy_pages(), which is called with the buffer pages already sbusied (by vm_pageout_flush()). Since vfs_drain_busy_pages() can only wait for one page at the time, and during the wait, the object lock is dropped, previous pages in the buffer must be protected from other threads busying them. Up to the moment, it was done by xbusying the pages, that is incompatible with the sbusy state in the new implementation of busy. Switch to sbusy. Reported and tested by: pho Sponsored by: The FreeBSD Foundation
2013-09-05 12:56:08 +00:00
vm_page_sbusy(bp->b_pages[last_busied]);
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
while (vm_page_xbusied(m)) {
vm_page_lock(m);
VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object);
Fix a race in vm_page_busy_sleep(9). Suppose that we have an exclusively busy page, and a thread which can accept shared-busy page. In this case, typical code waiting for the page xbusy state to pass is again: VM_OBJECT_WLOCK(object); ... if (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(object); <---1 vm_page_busy_sleep(p, "vmopax"); goto again; } Suppose that the xbusy state owner locked the object, unbusied the page and unlocked the object after we are at the line [1], but before we executed the load of the busy_lock word in vm_page_busy_sleep(). If it happens that there is still no waiters recorded for the busy state, the xbusy owner did not acquired the page lock, so it proceeded. More, suppose that some other thread happen to share-busy the page after xbusy state was relinquished but before the m->busy_lock is read in vm_page_busy_sleep(). Again, that thread only needs vm_object lock to proceed. Then, vm_page_busy_sleep() reads busy_lock value equal to the VPB_SHARERS_WORD(1). In this case, all tests in vm_page_busy_sleep(9) pass and we are going to sleep, despite the page being share-busied. Update check for m->busy_lock == VPB_UNBUSIED in vm_page_busy_sleep(9) to also accept shared-busy state if we only wait for the xbusy state to pass. Merge sequential if()s with the same 'then' clause in vm_page_busy_sleep(). Note that the current code does not share-busy pages from parallel threads, the only way to have more that one sbusy owner is right now is to recurse. Reported and tested by: pho (previous version) Reviewed by: alc, markj Sponsored by: The FreeBSD Foundation MFC after: 1 week Differential revision: https://reviews.freebsd.org/D8196
2016-10-13 14:41:05 +00:00
vm_page_busy_sleep(m, "vbpage", true);
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
VM_OBJECT_WLOCK(bp->b_bufobj->bo_object);
}
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
}
}
for (i = 0; i < last_busied; i++)
The vm_pageout_flush() functions sbusies pages in the passed pages run. After that, the pager put method is called, usually translated to VOP_WRITE(). For the filesystems which use buffer cache, bufwrite() sbusies the buffer pages again, waiting for the xbusy state to drain. The later is done in vfs_drain_busy_pages(), which is called with the buffer pages already sbusied (by vm_pageout_flush()). Since vfs_drain_busy_pages() can only wait for one page at the time, and during the wait, the object lock is dropped, previous pages in the buffer must be protected from other threads busying them. Up to the moment, it was done by xbusying the pages, that is incompatible with the sbusy state in the new implementation of busy. Switch to sbusy. Reported and tested by: pho Sponsored by: The FreeBSD Foundation
2013-09-05 12:56:08 +00:00
vm_page_sunbusy(bp->b_pages[i]);
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
}
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
/*
* This routine is called before a device strategy routine.
* It is used to tell the VM system that paging I/O is in
* progress, and treat the pages associated with the buffer
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
* almost as being exclusive busy. Also the object paging_in_progress
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
* flag is handled to make sure that the object doesn't become
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* inconsistent.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* Since I/O has not been initiated yet, certain buffer flags
sys/kern: spelling fixes. Mostly on comments but affects some debug messages. MFC after: 2 weeks
2016-04-29 21:54:28 +00:00
* such as BIO_ERROR or B_INVAL may be in an inconsistent state
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
* and should be ignored.
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
*/
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
void
stylistic polishing.
2004-09-15 20:54:23 +00:00
vfs_busy_pages(struct buf *bp, int clear_modify)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
{
undent some functions a bit.
2004-09-15 21:08:58 +00:00
vm_object_t obj;
vm_ooffset_t foff;
vm_page_t m;
Style fixes. In particular, the variable "bogus" is used like a Boolean. Define it as such. Reviewed by: kib MFC after: 1 week
2017-03-19 23:06:11 +00:00
int i;
bool bogus;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if (!(bp->b_flags & B_VMIO))
return;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Change buf->b_object to buf->b_bufobj->bo_object some whitespace fixes.
2004-11-04 09:06:54 +00:00
obj = bp->b_bufobj->bo_object;
undent some functions a bit.
2004-09-15 21:08:58 +00:00
foff = bp->b_offset;
KASSERT(bp->b_offset != NOOFFSET,
("vfs_busy_pages: no buffer offset"));
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(obj);
Reorganize the code in bdwrite() which handles move of dirtiness from the buffer pages to buffer. Combine the code to set buffer dirty range (previously in vfs_setdirty()) and to clean the pages (vfs_clean_pages()) into new function vfs_clean_pages_dirty_buf(). Now the vm object lock is acquired only once. Drain the VPO_BUSY bit of the buffer pages before setting valid and clean bits in vfs_clean_pages_dirty_buf() with new helper vfs_drain_busy_pages(). pmap_clear_modify() asserts that page is not busy. In vfs_busy_pages(), move the wait for draining of VPO_BUSY before the dirtyness handling, to follow the structure of vfs_clean_pages_dirty_buf(). Reported and tested by: pho Suggested and reviewed by: alc MFC after: 2 weeks
2010-06-08 17:54:28 +00:00
vfs_drain_busy_pages(bp);
Refactor vfs_setdirty(), creating vfs_setdirty_locked_object(). Call vfs_setdirty_locked_object() from vfs_busy_pages() instead of vfs_setdirty(), thereby eliminating a second acquisition and release of the same vm object lock.
2006-10-29 00:04:39 +00:00
if (bp->b_bufsize != 0)
vfs_setdirty_locked_object(bp);
Style fixes. In particular, the variable "bogus" is used like a Boolean. Define it as such. Reviewed by: kib MFC after: 1 week
2017-03-19 23:06:11 +00:00
bogus = false;
undent some functions a bit.
2004-09-15 21:08:58 +00:00
for (i = 0; i < bp->b_npages; i++) {
m = bp->b_pages[i];
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if ((bp->b_flags & B_CLUSTER) == 0) {
vm_object_pip_add(obj, 1);
The soft and hard busy mechanism rely on the vm object lock to work. Unify the 2 concept into a real, minimal, sxlock where the shared acquisition represent the soft busy and the exclusive acquisition represent the hard busy. The old VPO_WANTED mechanism becames the hard-path for this new lock and it becomes per-page rather than per-object. The vm_object lock becames an interlock for this functionality: it can be held in both read or write mode. However, if the vm_object lock is held in read mode while acquiring or releasing the busy state, the thread owner cannot make any assumption on the busy state unless it is also busying it. Also: - Add a new flag to directly shared busy pages while vm_page_alloc and vm_page_grab are being executed. This will be very helpful once these functions happen under a read object lock. - Move the swapping sleep into its own per-object flag The KPI is heavilly changed this is why the version is bumped. It is very likely that some VM ports users will need to change their own code. Sponsored by: EMC / Isilon storage division Discussed with: alc Reviewed by: jeff, kib Tested by: gavin, bapt (older version) Tested by: pho, scottl
2013-08-09 11:11:11 +00:00
vm_page_sbusy(m);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
undent some functions a bit.
2004-09-15 21:08:58 +00:00
/*
* When readying a buffer for a read ( i.e
* clear_modify == 0 ), it is important to do
* bogus_page replacement for valid pages in
* partially instantiated buffers. Partially
* instantiated buffers can, in turn, occur when
* reconstituting a buffer from its VM backing store
* base. We only have to do this if B_CACHE is
* clear ( which causes the I/O to occur in the
* first place ). The replacement prevents the read
* I/O from overwriting potentially dirty VM-backed
* pages. XXX bogus page replacement is, uh, bogus.
* It may not work properly with small-block devices.
* We need to find a better way.
*/
Revert CVS revision 1.94 (svn r16840). Current pmap implementations don't suffer from the race condition that motivated revision 1.94. Consequently, the work-around that was implemented by revision 1.94 is no longer needed. Moreover, reverting this work-around eliminates the need for vfs_busy_pages() to acquire the page queues lock when preparing a buffer for read. Reviewed by: tegge
2009-05-11 05:16:57 +00:00
if (clear_modify) {
pmap_remove_write(m);
Eliminate page queues locking from bufdone_finish() through the following changes: Rename vfs_page_set_valid() to vfs_page_set_validclean() to reflect what this function actually does. Suggested by: tegge Introduce a new version of vfs_page_set_valid() that does no more than what the function's name implies. Specifically, it does not update the page's dirty mask, and thus it does not require the page queues lock to be held. Update two of the three callers to the old vfs_page_set_valid() to call vfs_page_set_validclean() instead because they actually require the page's dirty mask to be cleared. Introduce vm_page_set_valid(). Reviewed by: tegge
2009-05-13 05:39:39 +00:00
vfs_page_set_validclean(bp, foff, m);
Revert CVS revision 1.94 (svn r16840). Current pmap implementations don't suffer from the race condition that motivated revision 1.94. Consequently, the work-around that was implemented by revision 1.94 is no longer needed. Moreover, reverting this work-around eliminates the need for vfs_busy_pages() to acquire the page queues lock when preparing a buffer for read. Reviewed by: tegge
2009-05-11 05:16:57 +00:00
} else if (m->valid == VM_PAGE_BITS_ALL &&
Change buf->b_object to buf->b_bufobj->bo_object some whitespace fixes.
2004-11-04 09:06:54 +00:00
(bp->b_flags & B_CACHE) == 0) {
undent some functions a bit.
2004-09-15 21:08:58 +00:00
bp->b_pages[i] = bogus_page;
Style fixes. In particular, the variable "bogus" is used like a Boolean. Define it as such. Reviewed by: kib MFC after: 1 week
2017-03-19 23:06:11 +00:00
bogus = true;
undent some functions a bit.
2004-09-15 21:08:58 +00:00
}
foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (bogus && buf_mapped(bp)) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
BUF_CHECK_MAPPED(bp);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
pmap_qenter(trunc_page((vm_offset_t)bp->b_data),
bp->b_pages, bp->b_npages);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
Introduce vfs_bio_set_valid() and use it from ffs_realloccg(). This eliminates the misuse of vfs_bio_clrbuf() by ffs_realloccg(). In collaboration with: tegge
2009-05-17 20:26:00 +00:00
/*
* vfs_bio_set_valid:
*
* Set the range within the buffer to valid. The range is
* relative to the beginning of the buffer, b_offset. Note that
* b_offset itself may be offset from the beginning of the first
* page.
*/
void
vfs_bio_set_valid(struct buf *bp, int base, int size)
{
int i, n;
vm_page_t m;
if (!(bp->b_flags & B_VMIO))
return;
/*
* Fixup base to be relative to beginning of first page.
* Set initial n to be the maximum number of bytes in the
* first page that can be validated.
*/
base += (bp->b_offset & PAGE_MASK);
n = PAGE_SIZE - (base & PAGE_MASK);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(bp->b_bufobj->bo_object);
Introduce vfs_bio_set_valid() and use it from ffs_realloccg(). This eliminates the misuse of vfs_bio_clrbuf() by ffs_realloccg(). In collaboration with: tegge
2009-05-17 20:26:00 +00:00
for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) {
m = bp->b_pages[i];
if (n > size)
n = size;
Rename vm_page_set_valid() to vm_page_set_valid_range(). The vm_page_set_valid() is the most reasonable name for the m->valid accessor. Reviewed by: attilio, alc
2011-11-30 17:39:00 +00:00
vm_page_set_valid_range(m, base & PAGE_MASK, n);
Introduce vfs_bio_set_valid() and use it from ffs_realloccg(). This eliminates the misuse of vfs_bio_clrbuf() by ffs_realloccg(). In collaboration with: tegge
2009-05-17 20:26:00 +00:00
base += n;
size -= n;
n = PAGE_SIZE;
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object);
Introduce vfs_bio_set_valid() and use it from ffs_realloccg(). This eliminates the misuse of vfs_bio_clrbuf() by ffs_realloccg(). In collaboration with: tegge
2009-05-17 20:26:00 +00:00
}
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
/*
* vfs_bio_clrbuf:
*
Several changes to vfs_bio_clrbuf(): Provide a more descriptive comment. Eliminate dead code. The page cannot possibly have PG_ZERO set. Eliminate unnecessary blank lines. Reviewed by: tegge
2009-05-17 23:25:53 +00:00
* If the specified buffer is a non-VMIO buffer, clear the entire
* buffer. If the specified buffer is a VMIO buffer, clear and
* validate only the previously invalid portions of the buffer.
* This routine essentially fakes an I/O, so we need to clear
* BIO_ERROR and B_INVAL.
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
*
* Note that while we only theoretically need to clear through b_bcount,
* we go ahead and clear through b_bufsize.
*/
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
void
Fix a bug in vfs_bio_clrbuf(). The single-page-clrbuf optimization was improperly clearing more then just the invalid portions of the page. (This bug is not known to have been triggered by anything). Submitted by: tegge MFC after: 7 days
2002-06-22 19:09:35 +00:00
vfs_bio_clrbuf(struct buf *bp)
{
Rewrite the vfs_bio_clrbuf(9) to not access the b_data for B_VMIO buffers directly, use pmap_zero_page_area(9) for each zeroing page region instead. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-14 19:48:25 +00:00
int i, j, mask, sa, ea, slide;
Introduce a global lock for the vm subsystem (vm_mtx). vm_mtx does not recurse and is required for most low level vm operations. faults can not be taken without holding Giant. Memory subsystems can now call the base page allocators safely. Almost all atomic ops were removed as they are covered under the vm mutex. Alpha and ia64 now need to catch up to i386's trap handlers. FFS and NFS have been tested, other filesystems will need minor changes (grabbing the vm lock when twiddling page properties). Reviewed (partially) by: jake, jhb
2001-05-19 01:28:09 +00:00
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if ((bp->b_flags & (B_VMIO | B_MALLOC)) != B_VMIO) {
clrbuf(bp);
return;
}
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(bp->b_bufobj->bo_object);
Change buf->b_object to buf->b_bufobj->bo_object some whitespace fixes.
2004-11-04 09:06:54 +00:00
if ((bp->b_npages == 1) && (bp->b_bufsize < PAGE_SIZE) &&
undent some functions a bit.
2004-09-15 21:08:58 +00:00
(bp->b_offset & PAGE_MASK) == 0) {
if (bp->b_pages[0] == bogus_page)
goto unlock;
mask = (1 << (bp->b_bufsize / DEV_BSIZE)) - 1;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[0]->object);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if ((bp->b_pages[0]->valid & mask) == mask)
goto unlock;
Several changes to vfs_bio_clrbuf(): Provide a more descriptive comment. Eliminate dead code. The page cannot possibly have PG_ZERO set. Eliminate unnecessary blank lines. Reviewed by: tegge
2009-05-17 23:25:53 +00:00
if ((bp->b_pages[0]->valid & mask) == 0) {
Rewrite the vfs_bio_clrbuf(9) to not access the b_data for B_VMIO buffers directly, use pmap_zero_page_area(9) for each zeroing page region instead. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-14 19:48:25 +00:00
pmap_zero_page_area(bp->b_pages[0], 0, bp->b_bufsize);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
bp->b_pages[0]->valid |= mask;
goto unlock;
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
}
undent some functions a bit.
2004-09-15 21:08:58 +00:00
}
Rewrite the vfs_bio_clrbuf(9) to not access the b_data for B_VMIO buffers directly, use pmap_zero_page_area(9) for each zeroing page region instead. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-14 19:48:25 +00:00
sa = bp->b_offset & PAGE_MASK;
slide = 0;
for (i = 0; i < bp->b_npages; i++, sa = 0) {
slide = imin(slide + PAGE_SIZE, bp->b_offset + bp->b_bufsize);
ea = slide & PAGE_MASK;
if (ea == 0)
ea = PAGE_SIZE;
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if (bp->b_pages[i] == bogus_page)
continue;
Rewrite the vfs_bio_clrbuf(9) to not access the b_data for B_VMIO buffers directly, use pmap_zero_page_area(9) for each zeroing page region instead. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-14 19:48:25 +00:00
j = sa / DEV_BSIZE;
undent some functions a bit.
2004-09-15 21:08:58 +00:00
mask = ((1 << ((ea - sa) / DEV_BSIZE)) - 1) << j;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[i]->object);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
if ((bp->b_pages[i]->valid & mask) == mask)
continue;
Several changes to vfs_bio_clrbuf(): Provide a more descriptive comment. Eliminate dead code. The page cannot possibly have PG_ZERO set. Eliminate unnecessary blank lines. Reviewed by: tegge
2009-05-17 23:25:53 +00:00
if ((bp->b_pages[i]->valid & mask) == 0)
Rewrite the vfs_bio_clrbuf(9) to not access the b_data for B_VMIO buffers directly, use pmap_zero_page_area(9) for each zeroing page region instead. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-14 19:48:25 +00:00
pmap_zero_page_area(bp->b_pages[i], sa, ea - sa);
Several changes to vfs_bio_clrbuf(): Provide a more descriptive comment. Eliminate dead code. The page cannot possibly have PG_ZERO set. Eliminate unnecessary blank lines. Reviewed by: tegge
2009-05-17 23:25:53 +00:00
else {
undent some functions a bit.
2004-09-15 21:08:58 +00:00
for (; sa < ea; sa += DEV_BSIZE, j++) {
Rewrite the vfs_bio_clrbuf(9) to not access the b_data for B_VMIO buffers directly, use pmap_zero_page_area(9) for each zeroing page region instead. Sponsored by: The FreeBSD Foundation Tested by: pho MFC after: 2 weeks
2013-03-14 19:48:25 +00:00
if ((bp->b_pages[i]->valid & (1 << j)) == 0) {
pmap_zero_page_area(bp->b_pages[i],
sa, DEV_BSIZE);
}
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
}
}
undent some functions a bit.
2004-09-15 21:08:58 +00:00
bp->b_pages[i]->valid |= mask;
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
}
undent some functions a bit.
2004-09-15 21:08:58 +00:00
unlock:
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object);
undent some functions a bit.
2004-09-15 21:08:58 +00:00
bp->b_resid = 0;
Changes from John Dyson and myself: Fixed remaining known bugs in the buffer IO and VM system. vfs_bio.c: Fixed some race conditions and locking bugs. Improved performance by removing some (now) unnecessary code and fixing some broken logic. Fixed process accounting of # of FS outputs. Properly handle NFS interrupts (B_EINTR). (various) Replaced calls to clrbuf() with calls to an optimized routine call vfs_bio_clrbuf(). (various FS sync) Sync out modified vnode_pager backed pages. ffs_vnops.c: Do two passes: Sync out file data first, then indirect blocks. vm_fault.c: Fixed deadly embrace caused by acquiring locks in the wrong order. vnode_pager.c: Changed to use buffer I/O system for writing out modified pages. This should fix the problem with the modification date previous not getting updated. Also dramatically simplifies the code. Note that this is going to change in the future and be implemented via VOP_PUTPAGES(). vm_object.c: Fixed a pile of bugs related to cleaning (vnode) objects. The performance of vm_object_page_clean() is terrible when dealing with huge objects, but this will change when we implement a binary tree to keep the object pages sorted. vm_pageout.c: Fixed broken clustering of pageouts. Fixed race conditions and other lockup style bugs in the scanning of pages. Improved performance.
1995-04-09 06:02:46 +00:00
}
Add a helper function vfs_bio_bzero_buf() to zero the portion of the buffer, transparently handling mapped or unmapped buffers. Its intent is to replace the use of bzero(bp->b_data) in cases where the buffer might be unmapped, to avoid unneeded upgrades. Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-19 14:27:14 +00:00
void
vfs_bio_bzero_buf(struct buf *bp, int base, int size)
{
vm_page_t m;
int i, n;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (buf_mapped(bp)) {
Add a helper function vfs_bio_bzero_buf() to zero the portion of the buffer, transparently handling mapped or unmapped buffers. Its intent is to replace the use of bzero(bp->b_data) in cases where the buffer might be unmapped, to avoid unneeded upgrades. Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-19 14:27:14 +00:00
BUF_CHECK_MAPPED(bp);
bzero(bp->b_data + base, size);
} else {
BUF_CHECK_UNMAPPED(bp);
n = PAGE_SIZE - (base & PAGE_MASK);
for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) {
m = bp->b_pages[i];
if (n > size)
n = size;
pmap_zero_page_area(m, base & PAGE_MASK, n);
base += n;
size -= n;
n = PAGE_SIZE;
}
}
}
Release laundered vnode pages to the head of the inactive queue. The swap pager enqueues laundered pages near the head of the inactive queue to avoid another trip through LRU before reclamation. This change adds support for this behaviour to the vnode pager and makes use of it in UFS and ext2fs. Some ioflag handling is consolidated into a common subroutine so that this support can be easily extended to other filesystems which make use of the buffer cache. No changes are needed for ZFS since its putpages routine always undirties the pages before returning, and the laundry thread requeues the pages appropriately in this case. Reviewed by: alc, kib Differential Revision: https://reviews.freebsd.org/D8589
2016-11-23 17:53:07 +00:00
/*
* Update buffer flags based on I/O request parameters, optionally releasing the
* buffer. If it's VMIO or direct I/O, the buffer pages are released to the VM,
* where they may be placed on a page queue (VMIO) or freed immediately (direct
* I/O). Otherwise the buffer is released to the cache.
*/
static void
b_io_dismiss(struct buf *bp, int ioflag, bool release)
{
KASSERT((ioflag & IO_NOREUSE) == 0 || (ioflag & IO_VMIO) != 0,
("buf %p non-VMIO noreuse", bp));
if ((ioflag & IO_DIRECT) != 0)
bp->b_flags |= B_DIRECT;
This is an additional fix for bug report 230962. When using extended attributes, the kernel can panic with either "ffs_truncate3" or with "softdep_deallocate_dependencies: dangling deps". The problem arises because the flushbuflist() function which is called to clear out buffers is passed either the V_NORMAL flag to indicate that it should flush buffer associated with the contents of the file or the V_ALT flag to indicate that it should flush the buffers associated with the extended attribute data. The buffers containing the extended attribute data are identified by having their BX_ALTDATA flag set in the buffer's b_xflags field. The BX_ALTDATA flag is set on the buffer when the extended attribute block is first allocated or when its contents are read in from the disk. On a busy system, a buffer may be reused for another purpose, but the contents of the block that it contained continues to be held in the main page cache. Each physical page is identified as holding the contents of a logical block within a specified file (identified by a vnode). When a request is made to read a file, the kernel first looks for the block in the existing buffers. If it is not found there, it checks the page cache to see if it is still there. If it is found in the page cache, then it is remapped into a new buffer thus avoiding the need to read it in from the disk. The bug is that when a buffer request made for an extended attribute is fulfilled by reconstituting a buffer from the page cache rather than reading it in from disk, the BX_ALTDATA flag was not being set. Thus the flushbuflist() function would never clear it out and the "ffs_truncate3" panic would occur because the vnode being cleared still had buffers on its clean-buffer list. If the extended attribute was being updated, it is first read, then updated, and finally written. If the read is fulfilled by reconstituting the buffer from the page cache the BX_ALTDATA flag was not set and thus the dirty buffer would never be flushed by flushbuflist(). Eventually the buffer would be recycled. Since it was never written it would have an unfinished dependency which would trigger the "softdep_deallocate_dependencies: dangling deps" panic. The fix is to ensure that the BX_ALTDATA flag is set when a buffer has been reconstituted from the page cache. PR: 230962 Reported by: 2t8mr7kx9f@protonmail.com Reviewed by: kib Tested by: Peter Holm MFC after: 1 week Sponsored by: Netflix
2019-03-12 19:08:41 +00:00
if ((ioflag & IO_EXT) != 0)
bp->b_xflags |= BX_ALTDATA;
Release laundered vnode pages to the head of the inactive queue. The swap pager enqueues laundered pages near the head of the inactive queue to avoid another trip through LRU before reclamation. This change adds support for this behaviour to the vnode pager and makes use of it in UFS and ext2fs. Some ioflag handling is consolidated into a common subroutine so that this support can be easily extended to other filesystems which make use of the buffer cache. No changes are needed for ZFS since its putpages routine always undirties the pages before returning, and the laundry thread requeues the pages appropriately in this case. Reviewed by: alc, kib Differential Revision: https://reviews.freebsd.org/D8589
2016-11-23 17:53:07 +00:00
if ((ioflag & (IO_VMIO | IO_DIRECT)) != 0 && LIST_EMPTY(&bp->b_dep)) {
bp->b_flags |= B_RELBUF;
if ((ioflag & IO_NOREUSE) != 0)
bp->b_flags |= B_NOREUSE;
if (release)
brelse(bp);
} else if (release)
bqrelse(bp);
}
void
vfs_bio_brelse(struct buf *bp, int ioflag)
{
b_io_dismiss(bp, ioflag, true);
}
void
vfs_bio_set_flags(struct buf *bp, int ioflag)
{
b_io_dismiss(bp, ioflag, false);
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
/*
- Always call bfreekva() w/o vm_mtx held. - Always call vfs_setdirty() with vm_mtx held. - Fix an old comment: vm_hold_unload_pages is called vm_hold_free_pages() nowadays. - Always call vm_hold_free_pages() w/o vm_mtx held.
2001-05-23 22:24:49 +00:00
* vm_hold_load_pages and vm_hold_free_pages get pages into
Only do object paging_in_progress wakeups if someone is waiting on this condition. Added some comments. Submitted by: John Dyson
1995-02-22 09:16:07 +00:00
* a buffers address space. The pages are anonymous and are
* not associated with a file object.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
Introduce a global lock for the vm subsystem (vm_mtx). vm_mtx does not recurse and is required for most low level vm operations. faults can not be taken without holding Giant. Memory subsystems can now call the base page allocators safely. Almost all atomic ops were removed as they are covered under the vm mutex. Alpha and ia64 now need to catch up to i386's trap handlers. FFS and NFS have been tested, other filesystems will need minor changes (grabbing the vm lock when twiddling page properties). Reviewed (partially) by: jake, jhb
2001-05-19 01:28:09 +00:00
static void
stylistic polishing.
2004-09-15 20:54:23 +00:00
vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
{
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
vm_offset_t pg;
vm_page_t p;
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
int index;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
BUF_CHECK_MAPPED(bp);
Rework vm_hold_{load,free}_pages to calculate an index once and use that. At the same time, be sure to page-truncate bp->b_data so that the result of the calculation isn't negative.
1996-01-06 23:23:02 +00:00
to = round_page(to);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
from = round_page(from);
Fixed two potentially serious classes of bugs: 1) The vnode pager wasn't properly tracking the file size due to "size" being page rounded in some cases and not in others. This sometimes resulted in corrupted files. First noticed by Terry Lambert. Fixed by changing the "size" pager_alloc parameter to be a 64bit byte value (as opposed to a 32bit page index) and changing the pagers and their callers to deal with this properly. 2) Fixed a bogus type cast in round_page() and trunc_page() that caused some 64bit offsets and sizes to be scrambled. Removing the cast required adding casts at a few dozen callers. There may be problems with other bogus casts in close-by macros. A quick check seemed to indicate that those were okay, however.
1998-10-13 08:24:45 +00:00
index = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT;
Rework vm_hold_{load,free}_pages to calculate an index once and use that. At the same time, be sure to page-truncate bp->b_data so that the result of the calculation isn't negative.
1996-01-06 23:23:02 +00:00
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
for (pg = from; pg < to; pg += PAGE_SIZE, index++) {
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
/*
* note: must allocate system pages since blocking here
Modify vm_hold_load_pages() to allocate pages using VM_ALLOC_NOOBJ rather than using the kernel object. This allows the elimination of page queues locking from vm_hold_free_pages().
2009-05-29 18:35:51 +00:00
* could interfere with paging I/O, no matter which
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
* process we are.
*/
Eliminate vestiges of page coloring in VM_ALLOC_NOOBJ calls to vm_page_alloc(). While I'm here, for the sake of consistency, always specify the allocation class, such as VM_ALLOC_NORMAL, as the first of the flags.
2011-10-27 16:39:17 +00:00
p = vm_page_alloc(NULL, 0, VM_ALLOC_SYSTEM | VM_ALLOC_NOOBJ |
Replace manyinstances of VM_WAIT with blocking page allocation flags similar to the kernel memory allocator. This simplifies NUMA allocation because the domain will be known at wait time and races between failure and sleeping are eliminated. This also reduces boilerplate code and simplifies callers. A wait primitive is supplied for uma zones for similar reasons. This eliminates some non-specific VM_WAIT calls in favor of more explicit sleeps that may be satisfied without new pages. Reviewed by: alc, kib, markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon
2017-11-08 02:39:37 +00:00
VM_ALLOC_WIRED | VM_ALLOC_COUNT((to - pg) >> PAGE_SHIFT) |
VM_ALLOC_WAITOK);
Convert all pmap_kenter/pmap_kremove pairs in MI code to use pmap_qenter/ pmap_qremove. pmap_kenter is not safe to use in MI code because it is not guaranteed to flush the mapping from the tlb on all cpus. If the process in question is preempted and migrates cpus between the call to pmap_kenter and pmap_kremove, the original cpu will be left with stale mappings in its tlb. This is currently not a problem for i386 because we do not use PG_G on SMP, and thus all mappings are flushed from the tlb on context switches, not just user mappings. This is not the case on all architectures, and if PG_G is to be used with SMP on i386 it will be a problem. This was committed by peter earlier as part of his fine grained tlb shootdown work for i386, which was backed out for other reasons. Reviewed by: peter
2002-03-17 00:56:41 +00:00
pmap_qenter(pg, &p, 1);
Eliminated many redundant vm_map_lookup operations for vm_mmap. Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish overhead for merged cache. Efficiency improvement for vfs_cluster. It used to do alot of redundant calls to cluster_rbuild. Correct the ordering for vrele of .text and release of credentials. Use the selective tlb update for 486/586/P6. Numerous fixes to the size of objects allocated for files. Additionally, fixes in the various pagers. Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs. Fixes in the swap pager for exhausted resources. The pageout code will not as readily thrash. Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE), thereby improving efficiency of several routines. Eliminate even more unnecessary vm_page_protect operations. Significantly speed up process forks. Make vm_object_page_clean more efficient, thereby eliminating the pause that happens every 30seconds. Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the case of filesystems mounted async. Fix a panic with busy pages when write clustering is done for non-VMIO buffers.
1996-01-19 04:00:31 +00:00
bp->b_pages[index] = p;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Re-institute a bugfix in allocation of anonymous buffer memory.
1997-09-21 04:49:30 +00:00
bp->b_npages = index;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/* Return pages associated with this buf to the vm system */
Be consistent about "static" functions: if the function is marked static in its prototype, mark it static at the definition too. Inspired by: FlexeLint warning #512
2002-09-28 17:15:38 +00:00
static void
Change the implementation of vm_hold_free_pages() so that it performs at most one call to pmap_qremove(), and thus one TLB shootdown, instead of one call and TLB shootdown per page. Simplify the interface to vm_hold_free_pages(). MFC after: 3 weeks
2010-07-11 20:11:44 +00:00
vm_hold_free_pages(struct buf *bp, int newbsize)
While in the real world, I had a bad case of being swapped out for a lot of cycles. While waiting there I added a lot of the extra ()'s I have, (I have never used LISP to any extent). So I compiled the kernel with -Wall and shut up a lot of "suggest you add ()'s", removed a bunch of unused var's and added a couple of declarations here and there. Having a lap-top is highly recommended. My kernel still runs, yell at me if you kernel breaks.
1994-09-25 19:34:02 +00:00
{
Change the implementation of vm_hold_free_pages() so that it performs at most one call to pmap_qremove(), and thus one TLB shootdown, instead of one call and TLB shootdown per page. Simplify the interface to vm_hold_free_pages(). MFC after: 3 weeks
2010-07-11 20:11:44 +00:00
vm_offset_t from;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
vm_page_t p;
Re-institute a bugfix in allocation of anonymous buffer memory.
1997-09-21 04:49:30 +00:00
int index, newnpages;
Rework vm_hold_{load,free}_pages to calculate an index once and use that. At the same time, be sure to page-truncate bp->b_data so that the result of the calculation isn't negative.
1996-01-06 23:23:02 +00:00
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
BUF_CHECK_MAPPED(bp);
Change the implementation of vm_hold_free_pages() so that it performs at most one call to pmap_qremove(), and thus one TLB shootdown, instead of one call and TLB shootdown per page. Simplify the interface to vm_hold_free_pages(). MFC after: 3 weeks
2010-07-11 20:11:44 +00:00
from = round_page((vm_offset_t)bp->b_data + newbsize);
newnpages = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT;
if (bp->b_npages > newnpages)
pmap_qremove(from, bp->b_npages - newnpages);
for (index = newnpages; index < bp->b_npages; index++) {
Changes to support 1Tb filesizes. Pages are now named by an (object,index) pair instead of (object,offset) pair.
1995-12-11 04:58:34 +00:00
p = bp->b_pages[index];
Change the implementation of vm_hold_free_pages() so that it performs at most one call to pmap_qremove(), and thus one TLB shootdown, instead of one call and TLB shootdown per page. Simplify the interface to vm_hold_free_pages(). MFC after: 3 weeks
2010-07-11 20:11:44 +00:00
bp->b_pages[index] = NULL;
p->wire_count--;
vm_page_free(p);
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
}
Make v_wire_count a per-cpu counter(9) counter. This eliminates a significant source of cache line contention from vm_page_alloc(). Use accessors and vm_page_unwire_noq() so that the mechanism can be easily changed in the future. Reviewed by: markj Discussed with: kib, glebius Tested by: pho (earlier version) Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14273
2018-02-12 22:53:00 +00:00
vm_wire_sub(bp->b_npages - newnpages);
Re-institute a bugfix in allocation of anonymous buffer memory.
1997-09-21 04:49:30 +00:00
bp->b_npages = newnpages;
The big 4.4BSD Lite to FreeBSD 2.0.0 (Development) patch. Reviewed by: Rodney W. Grimes Submitted by: John Dyson and David Greenman
1994-05-25 09:21:21 +00:00
}
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
/*
* Map an IO request into kernel virtual address space.
*
* All requests are (re)mapped into kernel VA space.
* Notice that we use b_bufsize for the size of the buffer
* to be mapped. b_bcount might be modified by the driver.
Close the remaining user address mapping races for physical I/O, CAM, and AIO. Still TODO: streamline useracc() checks. Reviewed by: alc, tegge MFC after: 7 days
2003-01-20 17:46:48 +00:00
*
* Note that even if the caller determines that the address space should
* be valid, a race or a smaller-file mapped into a larger space may
* actually cause vmapbuf() to fail, so all callers of vmapbuf() MUST
* check the return value.
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
*
* This function only works with pager buffers.
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
*/
Close the remaining user address mapping races for physical I/O, CAM, and AIO. Still TODO: streamline useracc() checks. Reviewed by: alc, tegge MFC after: 7 days
2003-01-20 17:46:48 +00:00
int
Do not remap usermode pages into KVA for physio. Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-19 14:43:57 +00:00
vmapbuf(struct buf *bp, int mapbuf)
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
{
Convert vmapbuf() from using pmap_extract() to using pmap_extract_and_hold(). Note, however, that GIANT_REQUIRED should not be removed until all platforms fully implement the "prot" parameter to pmap_extract_and_hold(). Reviewed by: tegge
2003-09-13 04:29:55 +00:00
vm_prot_t prot;
Introduce and use a new VM interface for temporarily pinning pages. This new interface replaces the combined use of vm_fault_quick() and pmap_extract_and_hold() throughout the kernel. In collaboration with: kib@
2010-12-25 21:26:56 +00:00
int pidx;
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
o Check the b_bufsize passed to vmapbuf() returning an error if it is invalid. o Remove a debugging printf() from vmapbuf(). Suggested by: tegge
2003-04-04 06:14:54 +00:00
if (bp->b_bufsize < 0)
return (-1);
stylistic polishing.
2004-09-15 20:54:23 +00:00
prot = VM_PROT_READ;
if (bp->b_iocmd == BIO_READ)
prot |= VM_PROT_WRITE; /* Less backwards than it looks */
Introduce and use a new VM interface for temporarily pinning pages. This new interface replaces the combined use of vm_fault_quick() and pmap_extract_and_hold() throughout the kernel. In collaboration with: kib@
2010-12-25 21:26:56 +00:00
if ((pidx = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map,
(vm_offset_t)bp->b_data, bp->b_bufsize, prot, bp->b_pages,
btoc(MAXPHYS))) < 0)
return (-1);
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
bp->b_npages = pidx;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
bp->b_offset = ((vm_offset_t)bp->b_data) & PAGE_MASK;
Do not remap usermode pages into KVA for physio. Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-19 14:43:57 +00:00
if (mapbuf || !unmapped_buf_allowed) {
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
pmap_qenter((vm_offset_t)bp->b_kvabase, bp->b_pages, pidx);
bp->b_data = bp->b_kvabase + bp->b_offset;
} else
Do not remap usermode pages into KVA for physio. Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-19 14:43:57 +00:00
bp->b_data = unmapped_buf;
Close the remaining user address mapping races for physical I/O, CAM, and AIO. Still TODO: streamline useracc() checks. Reviewed by: alc, tegge MFC after: 7 days
2003-01-20 17:46:48 +00:00
return(0);
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
}
/*
* Free the io map PTEs associated with this IO operation.
* We also invalidate the TLB entries and restore the original b_addr.
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
*
* This function only works with pager buffers.
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
*/
void
vunmapbuf(struct buf *bp)
{
int npages;
npages = bp->b_npages;
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (buf_mapped(bp))
Do not remap usermode pages into KVA for physio. Sponsored by: The FreeBSD Foundation Tested by: pho
2013-03-19 14:43:57 +00:00
pmap_qremove(trunc_page((vm_offset_t)bp->b_data), npages);
Implement and use a single optimized function for unholding a set of pages. Reviewed by: kib@
2010-12-17 22:41:22 +00:00
vm_page_unhold_pages(bp->b_pages, npages);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
bp->b_data = unmapped_buf;
Merge all the various copies of vmapbuf() and vunmapbuf() into a single portable copy. Note that pmap_extract() must be used instead of pmap_kextract(). This is precursor work to a reorganization of vmapbuf() to close remaining user/kernel races (which can lead to a panic).
2003-01-15 23:54:35 +00:00
}
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
void
bdone(struct buf *bp)
{
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
struct mtx *mtxp;
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
mtxp = mtx_pool_find(mtxpool_sleep, bp);
mtx_lock(mtxp);
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
bp->b_flags |= B_DONE;
wakeup(bp);
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
mtx_unlock(mtxp);
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
}
void
bwait(struct buf *bp, u_char pri, const char *wchan)
{
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
struct mtx *mtxp;
stylistic polishing.
2004-09-15 20:54:23 +00:00
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
mtxp = mtx_pool_find(mtxpool_sleep, bp);
mtx_lock(mtxp);
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
while ((bp->b_flags & B_DONE) == 0)
- Reduce contention on the global bdonelock and bpinlock by using a pool mutex to protect these sleep/wakeup/counter races. This still is preferable to bloating each bio with a mtx.
2008-03-21 10:00:05 +00:00
msleep(bp, mtxp, pri, wchan, 0);
mtx_unlock(mtxp);
- Add a lock for protecting against msleep(bp, ...) wakeup(bp) races. - Create a new function bdone() which sets B_DONE and calls wakup(bp). This is suitable for use as b_iodone for buf consumers who are not going through the buf cache. - Create a new function bwait() which waits for the buf to be done at a set priority and with a specific wmesg. - Replace several cases where the above functionality was implemented without locking with the new functions.
2003-03-13 07:31:45 +00:00
}
Add BO_SYNC() and add a default which uses the secret vnode pointer and VOP_FSYNC() for now.
2005-01-11 10:43:08 +00:00
int
Remove the struct thread unuseful argument from bufobj interface. In particular following functions KPI results modified: - bufobj_invalbuf() - bufsync() and BO_SYNC() "virtual method" of the buffer objects set. Main consumers of bufobj functions are affected by this change too and, in particular, functions which changed their KPI are: - vinvalbuf() - g_vfs_close() Due to the KPI breakage, __FreeBSD_version will be bumped in a later commit. As a side note, please consider just temporary the 'curthread' argument passing to VOP_SYNC() (in bufsync()) as it will be axed out ASAP Reviewed by: kib Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
2008-10-10 21:23:50 +00:00
bufsync(struct bufobj *bo, int waitfor)
Add BO_SYNC() and add a default which uses the secret vnode pointer and VOP_FSYNC() for now.
2005-01-11 10:43:08 +00:00
{
vfs: remove the __bo_vnode field from struct vnode The pointer can be obtained using __containerof instead. Reviewed by: kib
2016-09-30 17:11:03 +00:00
return (VOP_FSYNC(bo2vnode(bo), waitfor, curthread));
Add BO_SYNC() and add a default which uses the secret vnode pointer and VOP_FSYNC() for now.
2005-01-11 10:43:08 +00:00
}
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
void
bufstrategy(struct bufobj *bo, struct buf *bp)
{
vfs: annotate variables only used by debug builds as __unused
2018-05-19 04:59:39 +00:00
int i __unused;
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
struct vnode *vp;
vp = bp->b_vp;
Remove the last call in the system to VOP_SPECSTRATEGY(): We can no longer come through the VNODE layer to the disks since all the filesystems now go via geom_vfs to GEOM.
2004-10-29 10:52:31 +00:00
KASSERT(vp == bo->bo_private, ("Inconsistent vnode bufstrategy"));
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
("Wrong vnode in bufstrategy(bp=%p, vp=%p)", bp, vp));
Remove the last call in the system to VOP_SPECSTRATEGY(): We can no longer come through the VNODE layer to the disks since all the filesystems now go via geom_vfs to GEOM.
2004-10-29 10:52:31 +00:00
i = VOP_STRATEGY(vp, bp);
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
KASSERT(i == 0, ("VOP_STRATEGY failed bp=%p vp=%p", bp, bp->b_vp));
}
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
/*
* Initialize a struct bufobj before use. Memory is assumed zero filled.
*/
void
bufobj_init(struct bufobj *bo, void *private)
{
static volatile int bufobj_cleanq;
bo->bo_domain =
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
atomic_fetchadd_int(&bufobj_cleanq, 1) % buf_domains;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
rw_init(BO_LOCKPTR(bo), "bufobj interlock");
bo->bo_private = private;
TAILQ_INIT(&bo->bo_clean.bv_hd);
TAILQ_INIT(&bo->bo_dirty.bv_hd);
}
- Add bufobj_wrefl() to add a write ref to a bufobj that is already locked.
2005-05-30 07:01:18 +00:00
void
bufobj_wrefl(struct bufobj *bo)
{
KASSERT(bo != NULL, ("NULL bo in bufobj_wref"));
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
ASSERT_BO_WLOCKED(bo);
- Add bufobj_wrefl() to add a write ref to a bufobj that is already locked.
2005-05-30 07:01:18 +00:00
bo->bo_numoutput++;
}
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
void
bufobj_wref(struct bufobj *bo)
{
KASSERT(bo != NULL, ("NULL bo in bufobj_wref"));
BO_LOCK(bo);
bo->bo_numoutput++;
BO_UNLOCK(bo);
}
void
bufobj_wdrop(struct bufobj *bo)
{
KASSERT(bo != NULL, ("NULL bo in bufobj_wdrop"));
BO_LOCK(bo);
KASSERT(bo->bo_numoutput > 0, ("bufobj_wdrop non-positive count"));
if ((--bo->bo_numoutput == 0) && (bo->bo_flag & BO_WWAIT)) {
bo->bo_flag &= ~BO_WWAIT;
wakeup(&bo->bo_numoutput);
}
BO_UNLOCK(bo);
}
int
bufobj_wwait(struct bufobj *bo, int slpflag, int timeo)
{
int error;
KASSERT(bo != NULL, ("NULL bo in bufobj_wwait"));
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
ASSERT_BO_WLOCKED(bo);
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
error = 0;
while (bo->bo_numoutput) {
bo->bo_flag |= BO_WWAIT;
- Convert the bufobj lock to rwlock. - Use a shared bufobj lock in getblk() and inmem(). - Convert softdep's lk to rwlock to match the bufobj lock. - Move INFREECNT to b_flags and protect it with the buf lock. - Remove unnecessary locking around bremfree() and BKGRDINPROG. Sponsored by: EMC / Isilon Storage Division Discussed with: mckusick, kib, mdf
2013-05-31 00:43:41 +00:00
error = msleep(&bo->bo_numoutput, BO_LOCKPTR(bo),
Move the VI_BWAIT flag into no bo_flag element of bufobj and call it BO_WWAIT Add bufobj_wref(), bufobj_wdrop() and bufobj_wwait() to handle the write count on a bufobj. Bufobj_wdrop() replaces vwakeup(). Use these functions all relevant places except in ffs_softdep.c where the use if interlocked_sleep() makes this impossible. Rename b_vnbufs to b_bobufs now that we touch all the relevant files anyway.
2004-10-21 15:53:54 +00:00
slpflag | (PRIBIO + 1), "bo_wwait", timeo);
if (error)
break;
}
return (error);
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
/*
* Set bio_data or bio_ma for struct bio from the struct buf.
*/
void
bdata2bio(struct buf *bp, struct bio *bip)
{
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
if (!buf_mapped(bp)) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
KASSERT(unmapped_buf_allowed, ("unmapped"));
bip->bio_ma = bp->b_pages;
bip->bio_ma_n = bp->b_npages;
bip->bio_data = unmapped_buf;
bip->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK;
bip->bio_flags |= BIO_UNMAPPED;
KASSERT(round_page(bip->bio_ma_offset + bip->bio_length) /
PAGE_SIZE == bp->b_npages,
Make kassert_printf use __printflike. Fix associated errors/warnings while I'm here. Requested by: avg
2013-07-07 21:39:37 +00:00
("Buffer %p too short: %d %lld %d", bp, bip->bio_ma_offset,
(long long)bip->bio_length, bip->bio_ma_n));
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
} else {
bip->bio_data = bp->b_data;
bip->bio_ma = NULL;
}
}
[mips] enable relbuf on mips for now to work around page aliasing in mips hardware. Although the higher end MIPS hardware handles cache aliasing issues in hardware, the older cores (r4k, etc) and some compile versions of the newer cores (mips24k, mips34k, mips74k) don't have this feature. This means we end up with some very unfortunate behaviour that was made very obvious by some recent changes to the FFS pager by kib. So, flip this off until we get our MIPS pmap/cache code upgraded to handle aliased pages in software. Discussed with: kib, bsdimp, juli
2016-11-15 01:41:45 +00:00
/*
* The MIPS pmap code currently doesn't handle aliased pages.
* The VIPT caches may not handle page aliasing themselves, leading
* to data corruption.
*
* As such, this code makes a system extremely unhappy if said
* system doesn't support unaliasing the above situation in hardware.
* Some "recent" systems (eg some mips24k/mips74k cores) don't enable
* this feature at build time, so it has to be handled in software.
*
* Once the MIPS pmap/cache code grows to support this function on
* earlier chips, it should be flipped back off.
*/
#ifdef __mips__
static int buf_pager_relbuf = 1;
#else
static int buf_pager_relbuf = 0;
#endif
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
SYSCTL_INT(_vfs, OID_AUTO, buf_pager_relbuf, CTLFLAG_RWTUN,
&buf_pager_relbuf, 0,
"Make buffer pager release buffers after reading");
/*
* The buffer pager. It uses buffer reads to validate pages.
*
* In contrast to the generic local pager from vm/vnode_pager.c, this
* pager correctly and easily handles volumes where the underlying
* device block size is greater than the machine page size. The
* buffer cache transparently extends the requested page run to be
* aligned at the block boundary, and does the necessary bogus page
* replacements in the addends to avoid obliterating already valid
* pages.
*
* The only non-trivial issue is that the exclusive busy state for
* pages, which is assumed by the vm_pager_getpages() interface, is
* incompatible with the VMIO buffer cache's desire to share-busy the
* pages. This function performs a trivial downgrade of the pages'
* state before reading buffers, and a less trivial upgrade from the
* shared-busy to excl-busy state after the read.
*/
int
vfs_bio_getpages(struct vnode *vp, vm_page_t *ma, int count,
int *rbehind, int *rahead, vbg_get_lblkno_t get_lblkno,
vbg_get_blksize_t get_blksize)
{
vm_page_t m;
vm_object_t object;
struct buf *bp;
Tweaks for the buffer pager. Pass current thread credentials instead of NOCRED. Only allow unmapped buffers for filesystem which proclaimed the support. For all filesystems which currently use buffer pager (UFS, msdosfs and cd9660), the changes are effectively nop. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-11-08 10:10:55 +00:00
struct mount *mp;
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
daddr_t lbn, lbnp;
vm_ooffset_t la, lb, poff, poffe;
long bsize;
Restore vnode pager statistic for buffer pagers. Reviewed by: alc, markj Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week Differential revision: https://reviews.freebsd.org/D8585
2016-11-22 10:06:39 +00:00
int bo_bs, br_flags, error, i, pgsin, pgsin_a, pgsin_b;
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
bool redo, lpart;
object = vp->v_object;
Tweaks for the buffer pager. Pass current thread credentials instead of NOCRED. Only allow unmapped buffers for filesystem which proclaimed the support. For all filesystems which currently use buffer pager (UFS, msdosfs and cd9660), the changes are effectively nop. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-11-08 10:10:55 +00:00
mp = vp->v_mount;
vfs_bio.c: Apply cleanups motivated by Coverity analysis It is believed that the conditions Coverity indicated were actually impossible to hit. So this patch just adds a cleanup to only compute v_mount once in brelse(), and in vfs_bio_getpages() always initializes error to zero to appease the static analyzer. No functional change intended. Submitted by: Darrick Lew <darrick.freebsd AT gmail.com> Reviewed by: kib Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D14613
2018-03-14 22:11:45 +00:00
error = 0;
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
la = IDX_TO_OFF(ma[count - 1]->pindex);
if (la >= object->un_pager.vnp.vnp_size)
return (VM_PAGER_BAD);
correct read-ahead calculations in vfs_bio_getpages Previously the calculations were done as if the requested region ended at the start of the last requested page, not its end. The problem as actually quite minor as it affected only stats and page prefaulting, not the actual page data, and only with specific parameters. Reviewed by: kib (previous version) MFC after: 2 weeks
2018-01-18 12:59:04 +00:00
/*
* Change the meaning of la from where the last requested page starts
* to where it ends, because that's the end of the requested region
* and the start of the potential read-ahead region.
*/
la += PAGE_SIZE;
lpart = la > object->un_pager.vnp.vnp_size;
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
bo_bs = get_blksize(vp, get_lblkno(vp, IDX_TO_OFF(ma[0]->pindex)));
Restore vnode pager statistic for buffer pagers. Reviewed by: alc, markj Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week Differential revision: https://reviews.freebsd.org/D8585
2016-11-22 10:06:39 +00:00
/*
* Calculate read-ahead, behind and total pages.
*/
pgsin = count;
lb = IDX_TO_OFF(ma[0]->pindex);
pgsin_b = OFF_TO_IDX(lb - rounddown2(lb, bo_bs));
pgsin += pgsin_b;
if (rbehind != NULL)
*rbehind = pgsin_b;
pgsin_a = OFF_TO_IDX(roundup2(la, bo_bs) - la);
if (la + IDX_TO_OFF(pgsin_a) >= object->un_pager.vnp.vnp_size)
pgsin_a = OFF_TO_IDX(roundup2(object->un_pager.vnp.vnp_size,
PAGE_SIZE) - la);
pgsin += pgsin_a;
if (rahead != NULL)
*rahead = pgsin_a;
- Remove 'struct vmmeter' from 'struct pcpu', leaving only global vmmeter in place. To do per-cpu stats, convert all fields that previously were maintained in the vmmeters that sit in pcpus to counter(9). - Since some vmmeter stats may be touched at very early stages of boot, before we have set up UMA and we can do counter_u64_alloc(), provide an early counter mechanism: o Leave one spare uint64_t in struct pcpu, named pc_early_dummy_counter. o Point counter(9) fields of vmmeter to pcpu[0].pc_early_dummy_counter, so that at early stages of boot, before counters are allocated we already point to a counter that can be safely written to. o For sparc64 that required a whole dummy pcpu[MAXCPU] array. Further related changes: - Don't include vmmeter.h into pcpu.h. - vm.stats.vm.v_swappgsout and vm.stats.vm.v_swappgsin changed to 64-bit, to match kernel representation. - struct vmmeter hidden under _KERNEL, and only vmstat(1) is an exclusion. This is based on benno@'s 4-year old patch: https://lists.freebsd.org/pipermail/freebsd-arch/2013-July/014471.html Reviewed by: kib, gallatin, marius, lidl Differential Revision: https://reviews.freebsd.org/D10156
2017-04-17 17:34:47 +00:00
VM_CNT_INC(v_vnodein);
VM_CNT_ADD(v_vnodepgsin, pgsin);
Restore vnode pager statistic for buffer pagers. Reviewed by: alc, markj Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week Differential revision: https://reviews.freebsd.org/D8585
2016-11-22 10:06:39 +00:00
Tweaks for the buffer pager. Pass current thread credentials instead of NOCRED. Only allow unmapped buffers for filesystem which proclaimed the support. For all filesystems which currently use buffer pager (UFS, msdosfs and cd9660), the changes are effectively nop. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-11-08 10:10:55 +00:00
br_flags = (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS)
!= 0) ? GB_UNMAPPED : 0;
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
VM_OBJECT_WLOCK(object);
again:
for (i = 0; i < count; i++)
vm_page_busy_downgrade(ma[i]);
VM_OBJECT_WUNLOCK(object);
lbnp = -1;
for (i = 0; i < count; i++) {
m = ma[i];
/*
* Pages are shared busy and the object lock is not
* owned, which together allow for the pages'
* invalidation. The racy test for validity avoids
* useless creation of the buffer for the most typical
* case when invalidation is not used in redo or for
* parallel read. The shared->excl upgrade loop at
* the end of the function catches the race in a
* reliable way (protected by the object lock).
*/
if (m->valid == VM_PAGE_BITS_ALL)
continue;
poff = IDX_TO_OFF(m->pindex);
poffe = MIN(poff + PAGE_SIZE, object->un_pager.vnp.vnp_size);
for (; poff < poffe; poff += bsize) {
lbn = get_lblkno(vp, poff);
if (lbn == lbnp)
goto next_page;
lbnp = lbn;
bsize = get_blksize(vp, lbn);
Tweaks for the buffer pager. Pass current thread credentials instead of NOCRED. Only allow unmapped buffers for filesystem which proclaimed the support. For all filesystems which currently use buffer pager (UFS, msdosfs and cd9660), the changes are effectively nop. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2016-11-08 10:10:55 +00:00
error = bread_gb(vp, lbn, bsize, curthread->td_ucred,
br_flags, &bp);
Generalize UFS buffer pager to allow it serving other filesystems which also use buffer cache. Most important addition to the code is the handling of filesystems where the block size is less than the machine page size, which might require reading several buffers to validate single page. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2016-10-28 11:43:59 +00:00
if (error != 0)
goto end_pages;
if (LIST_EMPTY(&bp->b_dep)) {
/*
* Invalidation clears m->valid, but
* may leave B_CACHE flag if the
* buffer existed at the invalidation
* time. In this case, recycle the
* buffer to do real read on next
* bread() after redo.
*
* Otherwise B_RELBUF is not strictly
* necessary, enable to reduce buf
* cache pressure.
*/
if (buf_pager_relbuf ||
m->valid != VM_PAGE_BITS_ALL)
bp->b_flags |= B_RELBUF;
bp->b_flags &= ~B_NOCACHE;
brelse(bp);
} else {
bqrelse(bp);
}
}
KASSERT(1 /* racy, enable for debugging */ ||
m->valid == VM_PAGE_BITS_ALL || i == count - 1,
("buf %d %p invalid", i, m));
if (i == count - 1 && lpart) {
VM_OBJECT_WLOCK(object);
if (m->valid != 0 &&
m->valid != VM_PAGE_BITS_ALL)
vm_page_zero_invalid(m, TRUE);
VM_OBJECT_WUNLOCK(object);
}
next_page:;
}
end_pages:
VM_OBJECT_WLOCK(object);
redo = false;
for (i = 0; i < count; i++) {
vm_page_sunbusy(ma[i]);
ma[i] = vm_page_grab(object, ma[i]->pindex, VM_ALLOC_NORMAL);
/*
* Since the pages were only sbusy while neither the
* buffer nor the object lock was held by us, or
* reallocated while vm_page_grab() slept for busy
* relinguish, they could have been invalidated.
* Recheck the valid bits and re-read as needed.
*
* Note that the last page is made fully valid in the
* read loop, and partial validity for the page at
* index count - 1 could mean that the page was
* invalidated or removed, so we must restart for
* safety as well.
*/
if (ma[i]->valid != VM_PAGE_BITS_ALL)
redo = true;
}
if (redo && error == 0)
goto again;
VM_OBJECT_WUNLOCK(object);
return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK);
}
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
#include "opt_ddb.h"
#ifdef DDB
#include <ddb/ddb.h>
Document all functions, global and static variables, and sysctls. Includes some minor whitespace changes, and re-ordering to be able to document properly (e.g, grouping of variables and the SYSCTL macro calls for them, where the documentation has been added.) Reviewed by: phk (but all errors are mine)
2002-03-05 15:38:49 +00:00
/* DDB command to show buffer data */
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
DB_SHOW_COMMAND(buffer, db_show_buffer)
{
/* get args */
struct buf *bp = (struct buf *)addr;
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
#ifdef FULL_BUF_TRACKING
uint32_t i, j;
#endif
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
if (!have_addr) {
db_printf("usage: show buffer <addr>\n");
return;
}
- Add information about the buf lock to db_show_buffer. - Add a 'show lockedbufs' command that is similar to show lockedvnods. Sponsored by: Isilon Systems, Inc.
2005-03-25 00:20:37 +00:00
db_printf("buf at %p\n", bp);
Add printing of b_ioflags to DDB `show buffer' command. Sponsored by: Netflix
2019-01-25 21:24:09 +00:00
db_printf("b_flags = 0x%b, b_xflags=0x%b\n",
(u_int)bp->b_flags, PRINT_BUF_FLAGS,
(u_int)bp->b_xflags, PRINT_BUF_XFLAGS);
db_printf("b_vflags=0x%b b_ioflags0x%b\n",
(u_int)bp->b_vflags, PRINT_BUF_VFLAGS,
(u_int)bp->b_ioflags, PRINT_BIO_FLAGS);
Fixed some printf format errors (hopefully all of the remaining daddr64_t ones for GENERIC, and all others on the same line as those). Reformat the printfs if necessary to avoid new long lones or old format printf errors.
2002-03-19 04:09:21 +00:00
db_printf(
"b_error = %d, b_bufsize = %ld, b_bcount = %ld, b_resid = %ld\n"
Augment DDB "show buffer" command to print the buffer's referenced vnode pointer (b_vp). The value of b_vp can be used by "show vnode" to print the vnode and "show vnodebufs" to print all the clean and dirty buffers associated with the vnode (which should include this buffer). Sponsored by: Netflix
2019-03-11 21:49:44 +00:00
"b_bufobj = (%p), b_data = %p\n, b_blkno = %jd, b_lblkno = %jd, "
"b_vp = %p, b_dep = %p\n",
Fixed some printf format errors (hopefully all of the remaining daddr64_t ones for GENERIC, and all others on the same line as those). Reformat the printfs if necessary to avoid new long lones or old format printf errors.
2002-03-19 04:09:21 +00:00
bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid,
Add the ffs structures introspection functions for ddb. Show the b_dep value for the buffer in the show buffer command. Add a comand to dump the dirty/clean buffer list for vnode. Reviewed by: tegge Tested and used by: pho MFC after: 1 month
2008-09-16 11:19:38 +00:00
bp->b_bufobj, bp->b_data, (intmax_t)bp->b_blkno,
Augment DDB "show buffer" command to print the buffer's referenced vnode pointer (b_vp). The value of b_vp can be used by "show vnode" to print the vnode and "show vnodebufs" to print all the clean and dirty buffers associated with the vnode (which should include this buffer). Sponsored by: Netflix
2019-03-11 21:49:44 +00:00
(intmax_t)bp->b_lblkno, bp->b_vp, bp->b_dep.lh_first);
Refactor unmapped buffer address handling. - Use pointer assignment rather than a combination of pointers and flags to switch buffers between unmapped and mapped. This eliminates multiple flags and generally simplifies the logic. - Eliminate b_saveaddr since it is only used with pager bufs which have their b_data re-initialized on each allocation. - Gather up some convenience routines in the buffer cache for manipulating buf space and buf malloc space. - Add an inline, buf_mapped(), to standardize checks around unmapped buffers. In collaboration with: mlaier Reviewed by: kib Tested by: pho (many small revisions ago) Sponsored by: EMC / Isilon Storage Division
2015-07-23 19:13:41 +00:00
db_printf("b_kvabase = %p, b_kvasize = %d\n",
bp->b_kvabase, bp->b_kvasize);
Re-institute a bugfix in allocation of anonymous buffer memory.
1997-09-21 04:49:30 +00:00
if (bp->b_npages) {
int i;
db_printf("b_npages = %d, pages(OBJ, IDX, PA): ", bp->b_npages);
for (i = 0; i < bp->b_npages; i++) {
vm_page_t m;
m = bp->b_pages[i];
Let DDB's buf printer handle NULL pointers in the buf page array. A buf's b_pages and b_npages fields may be inconsistent after a panic. For instance, vfs_vmio_invalidate() sets b_npages to zero only after all pages are unwired and their page array entries are cleared. MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2016-07-14 18:49:05 +00:00
if (m != NULL)
db_printf("(%p, 0x%lx, 0x%lx)", m->object,
(u_long)m->pindex,
(u_long)VM_PAGE_TO_PHYS(m));
else
db_printf("( ??? )");
Re-institute a bugfix in allocation of anonymous buffer memory.
1997-09-21 04:49:30 +00:00
if ((i + 1) < bp->b_npages)
db_printf(",");
}
db_printf("\n");
}
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
BUF_LOCKPRINTINFO(bp);
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
#if defined(FULL_BUF_TRACKING)
db_printf("b_io_tracking: b_io_tcnt = %u\n", bp->b_io_tcnt);
i = bp->b_io_tcnt % BUF_TRACKING_SIZE;
Improve BUF_TRACKING by not displaying NULL entries. Reviewed by: cem MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D10443
2017-04-23 17:39:31 +00:00
for (j = 1; j <= BUF_TRACKING_SIZE; j++) {
if (bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)] == NULL)
continue;
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
db_printf(" %2u: %s\n", j,
bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)]);
Improve BUF_TRACKING by not displaying NULL entries. Reviewed by: cem MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D10443
2017-04-23 17:39:31 +00:00
}
Add BUF_TRACKING and FULL_BUF_TRACKING buffer debugging Upstream the BUF_TRACKING and FULL_BUF_TRACKING buffer debugging code. This can be handy in tracking down what code touched hung bios and bufs last. The full history is especially useful, but adds enough bloat that it shouldn't be enabled in release builds. Function names (or arbitrary string constants) are tracked in a fixed-size ring in bufs. Bios gain a pointer to the upper buf for tracking. SCSI CCBs gain a pointer to the upper bio for tracking. Reviewed by: markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8366
2016-10-31 23:09:52 +00:00
#elif defined(BUF_TRACKING)
db_printf("b_io_tracking: %s\n", bp->b_io_tracking);
#endif
Tweak the output of VOP_PRINT/vn_printf() some. - Align the fifo output in fifo_print() with other vn_printf() output. - Remove the leading space from lockmgr_printinfo() so its output lines up in vn_printf(). - lockmgr_printinfo() now ends with a newline, so remove an extra newline from vn_printf().
2009-02-06 20:06:48 +00:00
db_printf(" ");
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
}
- Add information about the buf lock to db_show_buffer. - Add a 'show lockedbufs' command that is similar to show lockedvnods. Sponsored by: Isilon Systems, Inc.
2005-03-25 00:20:37 +00:00
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
DB_SHOW_COMMAND(bufqueues, bufqueues)
{
struct bufdomain *bd;
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
struct buf *bp;
long total;
int i, j, cnt;
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
db_printf("bqempty: %d\n", bqempty.bq_len);
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
for (i = 0; i < buf_domains; i++) {
bd = &bdomain[i];
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
db_printf("Buf domain %d\n", i);
db_printf("\tfreebufs\t%d\n", bd->bd_freebuffers);
db_printf("\tlofreebufs\t%d\n", bd->bd_lofreebuffers);
db_printf("\thifreebufs\t%d\n", bd->bd_hifreebuffers);
db_printf("\n");
db_printf("\tbufspace\t%ld\n", bd->bd_bufspace);
db_printf("\tmaxbufspace\t%ld\n", bd->bd_maxbufspace);
db_printf("\thibufspace\t%ld\n", bd->bd_hibufspace);
db_printf("\tlobufspace\t%ld\n", bd->bd_lobufspace);
db_printf("\tbufspacethresh\t%ld\n", bd->bd_bufspacethresh);
db_printf("\n");
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
db_printf("\tnumdirtybuffers\t%d\n", bd->bd_numdirtybuffers);
db_printf("\tlodirtybuffers\t%d\n", bd->bd_lodirtybuffers);
db_printf("\thidirtybuffers\t%d\n", bd->bd_hidirtybuffers);
db_printf("\tdirtybufthresh\t%d\n", bd->bd_dirtybufthresh);
db_printf("\n");
total = 0;
TAILQ_FOREACH(bp, &bd->bd_cleanq->bq_queue, b_freelist)
total += bp->b_bufsize;
db_printf("\tcleanq count\t%d (%ld)\n",
bd->bd_cleanq->bq_len, total);
total = 0;
TAILQ_FOREACH(bp, &bd->bd_dirtyq.bq_queue, b_freelist)
total += bp->b_bufsize;
db_printf("\tdirtyq count\t%d (%ld)\n",
bd->bd_dirtyq.bq_len, total);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
db_printf("\twakeup\t\t%d\n", bd->bd_wanted);
db_printf("\tlim\t\t%d\n", bd->bd_lim);
db_printf("\tCPU ");
Fix issues with sparse cpu allocation. Consistently use mp_maxid + 1. Reported by: pho Reviewed by: markj Sponsored by: Netflix, Dell/EMC Isilon
2018-02-25 00:35:21 +00:00
for (j = 0; j <= mp_maxid; j++)
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
db_printf("%d, ", bd->bd_subq[j].bq_len);
db_printf("\n");
Move the dirty queues inside the per-domain structure. This resolves a bug where we had not hit global dirty limits but a single queue was starved for space by dirty buffers. A single buf_daemon is maintained for now. Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ keeping many bufs locked until a cg block is written. Document this with a comment. Fix sysctls to work with per-domain variables. Add more ddb debugging. Reported by: pho Reviewed by: kib Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14705
2018-03-17 18:14:49 +00:00
cnt = 0;
total = 0;
for (j = 0; j < nbuf; j++)
if (buf[j].b_domain == i && BUF_ISLOCKED(&buf[j])) {
cnt++;
total += buf[j].b_bufsize;
}
db_printf("\tLocked buffers: %d space %ld\n", cnt, total);
cnt = 0;
total = 0;
for (j = 0; j < nbuf; j++)
if (buf[j].b_domain == i) {
cnt++;
total += buf[j].b_bufsize;
}
db_printf("\tTotal buffers: %d space %ld\n", cnt, total);
Further parallelize the buffer cache. Provide multiple clean queues partitioned into 'domains'. Each domain manages its own bufspace and has its own bufspace daemon. Each domain has a set of subqueues indexed by the current cpuid to reduce lock contention on the cleanq. Refine the sleep/wakeup around the bufspace daemon to use atomics as much as possible. Add a B_REUSE flag that is used to requeue bufs during the scan to approximate LRU rather than locking the queue on every use of a frequently accessed buf. Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts. Reviewed by: markj Tested by: pho Sponsored by: Netflix, Dell/EMC Isilon Differential Revision: https://reviews.freebsd.org/D14274
2018-02-20 00:06:07 +00:00
}
}
- Add information about the buf lock to db_show_buffer. - Add a 'show lockedbufs' command that is similar to show lockedvnods. Sponsored by: Isilon Systems, Inc.
2005-03-25 00:20:37 +00:00
DB_SHOW_COMMAND(lockedbufs, lockedbufs)
{
struct buf *bp;
int i;
for (i = 0; i < nbuf; i++) {
bp = &buf[i];
- Introduce the function lockmgr_recursed() which returns true if the lockmgr lkp, when held in exclusive mode, is recursed - Introduce the function BUF_RECURSED() which does the same for bufobj locks based on the top of lockmgr_recursed() - Introduce the function BUF_ISLOCKED() which works like the counterpart VOP_ISLOCKED(9), showing the state of lockmgr linked with the bufobj BUF_RECURSED() and BUF_ISLOCKED() entirely replace the usage of bogus BUF_REFCNT() in a more explicative and SMP-compliant way. This allows us to axe out BUF_REFCNT() and leaving the function lockcount() totally unused in our stock kernel. Further commits will axe lockcount() as well as part of lockmgr() cleanup. KPI results, obviously, broken so further commits will update manpages and freebsd version. Tested by: kris (on UFS and NFS)
2008-01-19 17:36:23 +00:00
if (BUF_ISLOCKED(bp)) {
- Add information about the buf lock to db_show_buffer. - Add a 'show lockedbufs' command that is similar to show lockedvnods. Sponsored by: Isilon Systems, Inc.
2005-03-25 00:20:37 +00:00
db_show_buffer((uintptr_t)bp, 1, 0, NULL);
db_printf("\n");
Make it possible to terminate "show lockedbufs" by pressing "q". MFC after: 2 weeks
2017-04-23 22:20:25 +00:00
if (db_pager_quit)
break;
- Add information about the buf lock to db_show_buffer. - Add a 'show lockedbufs' command that is similar to show lockedvnods. Sponsored by: Isilon Systems, Inc.
2005-03-25 00:20:37 +00:00
}
}
}
Add the ffs structures introspection functions for ddb. Show the b_dep value for the buffer in the show buffer command. Add a comand to dump the dirty/clean buffer list for vnode. Reviewed by: tegge Tested and used by: pho MFC after: 1 month
2008-09-16 11:19:38 +00:00
DB_SHOW_COMMAND(vnodebufs, db_show_vnodebufs)
{
struct vnode *vp;
struct buf *bp;
if (!have_addr) {
db_printf("usage: show vnodebufs <addr>\n");
return;
}
vp = (struct vnode *)addr;
db_printf("Clean buffers:\n");
TAILQ_FOREACH(bp, &vp->v_bufobj.bo_clean.bv_hd, b_bobufs) {
db_show_buffer((uintptr_t)bp, 1, 0, NULL);
db_printf("\n");
}
db_printf("Dirty buffers:\n");
TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) {
db_show_buffer((uintptr_t)bp, 1, 0, NULL);
db_printf("\n");
}
}
Add INVARIANTS checking that numfreebufs values are sane. Also add a per-buf flag to catch if a buf is double-counted in the free count. This code was useful to debug an instance where a local patch at Isilon was incorrectly managing numfreebufs for a new buf state. Reviewed by: jeff Approved by: zml (mentor)
2010-06-11 17:03:26 +00:00
DB_COMMAND(countfreebufs, db_coundfreebufs)
{
struct buf *bp;
int i, used = 0, nfree = 0;
if (have_addr) {
db_printf("usage: countfreebufs\n");
return;
}
for (i = 0; i < nbuf; i++) {
bp = &buf[i];
Parallelize the buffer cache and rewrite getnewbuf(). This results in a 8x performance improvement in a micro benchmark on a 4 socket machine. - Get buffer headers from a per-cpu uma cache that sits in from of the free queue. - Use a per-cpu quantum cache in vmem to eliminate contention for kva. - Use multiple clean queues according to buffer cache size to eliminate clean queue lock contention. - Introduce a bufspace daemon that attempts to prevent getnewbuf() callers from blocking or doing direct recycling. - Close some bufspace allocation races that could lead to endless recycling. - Further the transition to a more modern style of small functions grouped by prefix in order to improve growing complexity. Sponsored by: EMC / Isilon Reviewed by: kib Tested by: pho
2015-10-14 02:10:07 +00:00
if (bp->b_qindex == QUEUE_EMPTY)
Add INVARIANTS checking that numfreebufs values are sane. Also add a per-buf flag to catch if a buf is double-counted in the free count. This code was useful to debug an instance where a local patch at Isilon was incorrectly managing numfreebufs for a new buf state. Reviewed by: jeff Approved by: zml (mentor)
2010-06-11 17:03:26 +00:00
nfree++;
else
used++;
}
db_printf("Counted %d free, %d used (%d tot)\n", nfree, used,
nfree + used);
db_printf("numfreebuffers is %d\n", numfreebuffers);
}
Add a DDB command `show buffer', to display a struct buf. It's impossible to display everything, so i've chosen a small subset. Add more to this as you think seems useful.
1997-05-10 09:09:42 +00:00
#endif /* DDB */
Reference in New Issue Copy Permalink