Restructure the zfsdev_state_list to allow for lock-free reading by
converting to a simple singly-linked list from which items are never
deleted and over which only forward iterations are performed. It depends
on, among other things, the atomicity of accessing the zs_minor integer
and zs_next pointer.
This fixes a lock inversion in which the zfsdev_state_lock is used by
both the sync task (txg_sync) and indirectly by any user program which
uses /dev/zfs; the zfsdev_release method uses the same lock and then
blocks on the sync task.
The most typical failure scenerio occurs when the sync task is cleaning
up a user hold while various concurrent "zfs" commands are in progress.
Neither Illumos nor Solaris are affected by this issue because they use
DDI interface which provides lock-free reading of device state via the
ddi_get_soft_state() function.
Signed-off-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2301
Originally, vdev_file used system_taskq. This would cause a deadlock,
especially on system with few CPUs. The reason is that the prefetcher
threads, which are on system_taskq, will sometimes be blocked waiting
for I/O to finish. If the prefetcher threads consume all the tasks in
system_taskq, the I/O cannot be served and thus results in a deadlock.
We fix this by creating a dedicated vdev_file_taskq for vdev_file I/O.
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2270
The dva_get_dsize_sync() function incorrectly assumes that the call
to vdev_lookup_top() cannot fail. However, the NULL dereference at
clearly shows that under certain circumstances it is possible. Note
that offset 0x570 (1376) maps as expected to vd->vdev_deflate_ratio.
BUG: unable to handle kernel NULL pointer dereference at 00000570
crash> struct -o vdev
struct vdev {
[0] uint64_t vdev_id;
... ...
[1376] uint64_t vdev_deflate_ratio;
Given that this can happen this patch add the required error handling.
In the case where vdev_lookup_top() fails assume that no deflation
will occur for the DVA and use the asize.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
Closes#1707Closes#1987Closes#1891
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
When fetching property values of snapshots, a check against the head
dataset type must be performed. Previously, this additional check was
performed only when fetching "version", "normalize", "utf8only" or "case".
This caused the ZPL properties "acltype", "exec", "devices", "nbmand",
"setuid" and "xattr" to be erroneously displayed with meaningless values
for snapshots of volumes. It also did not allow for the display of
"volsize" of a snapshot of a volume.
This patch adds the headcheck flag paramater to zfs_prop_valid_for_type()
and zprop_valid_for_type() to indicate the check is being done
against a head dataset's type in order that properties valid only for
snapshots are handled correctly. This allows the the head check in
get_numeric_property() to be performed when fetching a property for
a snapshot.
Signed-off-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2265
A minor style issue was accidentally introduced by aa7d06a.
This change resolves that style problem.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Today the metaslab_debug logic performs two tasks:
- load all metaslabs on import/open
- don't unload metaslabs at the end of spa_sync
This change provides knobs for each of these independently.
References:
https://illumos.org/issues/4101https://github.com/illumos/illumos-gate/commit/0713e23
Notes:
1) This is a small piece of the metaslab improvement patch from
Illumos. It was worth bringing over before the rest, since it's
low risk and it can be useful on fragmented pools (e.g. Lustre
MDTs). metaslab_debug_unload would give the performance benefit
of the old metaslab_debug option without causing unwanted delay
during pool import.
Ported-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2227
When the system attributes (SAs) for an object exceed what can
can be stored in the bonus area of a dnode a spill block is
allocated. These spill blocks are currently considered data
blocks. However, they should be accounted for as meta data
because they are effectively an extension of the dnode.
While this may seem like a minor accounting issue it has broader
implications. The key thing to be aware of is that each spill
block will hold a reference on its parent dnode. The dnode in
turn holds a reference on its dbuf in the dnode object. This
means that a single 512 byte data buffer for a spill block can
pin over 16k of meta data. This is analogous to the small file
situation described in 2b13331 where a relatively small number
of data buffer can cause the ARC to exceed the meta limit.
However, unlike the small file case a spill block can legitimately
be considered meta data. By changing the spill block to meta data
they will now be dropped from the cache when the meta limit is
reached. This then allows the dnodes and dbufs which the spill
block was pinning to be released.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Closes#2294
As described in the comment above dmu_tx_assign() this function must
only fail if the pool is out of space. If for some other reason the
TX cannot be assigned (such as memory pressure) ERESTART must be
returned. Alternately, EAGAIN could be returned to inject a delay
but that isn't required because the caller will block on the condition
variable waiting for the next TXG.
/*
* Assign tx to a transaction group. txg_how can be one of:
*
* (1) TXG_WAIT. If the current open txg is full, waits until there's
* a new one. This should be used when you're not holding locks.
* It will only fail if we're truly out of space (or over quota).
* ...
*/
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ned Bass <bass6@llnl.gov>
Closes#2287
We add support for lsattr and chattr to resolve a regression caused
by 88c283952f that broke Python's
xattr.list(). That changet broke Gentoo Portage's FEATURES=xattr,
which depended on Python's xattr.list().
Only attributes common to both Solaris and Linux are supported. These
are 'a', 'd' and 'i' in Linux's lsattr and chattr commands. File
attributes exclusive to Solaris are present in the ZFS code, but cannot
be accessed or modified through this method. That was the case prior to
this patch. The resolution of issue zfsonlinux/zfs#229 should implement
some method to permit access and modification of Solaris-specific
attributes.
References:
https://bugs.gentoo.org/show_bug.cgi?id=483516
Original-patch-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1691
Some network related block device uses tcp_sendpage, which doesn't
behave well when using 0-count page. Add assertion to catch them.
This has a runtime dependency on:
zfsonlinux/spl@ae16ed9 Fix crash when using ZFS on Ceph rbd
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2277
Endianness detection in LZ4 is broken in user-space builds. This
bug corrupts compressed data and manifests itself in several ztest
failures. When LZ4 was originally ported to Illumos ZFS, the proper
checks for Linux were stripped out. The Linux port then inherited
the remaining detection code that works on Illumos but not on Linux.
The current LZ4 endianness check misuses the condition
defined(__BIG_ENDIAN) to indicate a big-endian system. On Linux
__BIG_ENDIAN is defined uncondtionally in the user-space header
/usr/include/endian.h, regardless of the endianness of the system.
The kernel does not use this header, so only user-space builds are
affected.
While we could fix this by restoring the upstream LZ4 endianness
detection code, reliable checks already exist in
libspl/include/sys/isa_defs.h. This change uses the libspl results
to replace the word-size and endianness checks in LZ4, simplifying
the code and reducing duplication.
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: DHE <git@dehacked.net>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Fixes#1963Fixes#1964Fixes#1965
Due to an asymmetry in the kmem accounting a memory leak was being
reported when it was only an accounting issue. All memory allocated
with kmem_alloc() must be released with kmem_free() or it will not
be properly accounted for.
In this case the code used strfree() to release the memory allocated
by kmem_alloc(). Presumably this was done because the size of the
memory region wasn't available when the memory needed to be freed.
To resolve this issue the code has been updated to use strdup() instead
of kmem_alloc() to allocate the memory. Like strfree(), strdup() is
not integrated with the memory accounting. This means we can use
strfree() to release it like Illumos.
SPL: kmem leaked 10/4368729 bytes
address size data func:line
ffff880067e9aa40 10 ZZZZZZZZZZ zfsdev_ioctl:5655
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Closes#2262
Also, make sure we use clock_t for ddi_get_lbolt to prevent type conversion
from screwing things.
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2142
rq_for_each_segment changed from taking bio_vec * to taking bio_vec.
We provide rq_for_each_segment4 which takes both.
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2124
After the dmu_req_copy change, bi_io_vecs are not touched, so this is no
longer needed.
This reverts commit e26ade5101.
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2124
Originally, dmu_req_copy modifies bv_len and bv_offset in bio_vec so that it
can continue in subsequent passes. However, after the immutable biovec changes
in Linux 3.14, this is not allowed. So instead, we just tell dmu_req_copy how
many bytes are already copied and it will skip to the right spot accordingly.
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2124
bi_sector, bi_size and bi_idx are moved from bio to bio->bi_iter.
This patch creates BIO_BI_*(bio) macros to hide the differences.
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2124
posix_acl_{create,chmod} is changed to __posix_acl_{create_chmod}
Signed-off-by: Chunwei Chen <tuxoko@gmail.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2124
zfsonlinux/zfs#180 occurred because of a race between inode eviction and
zfs_zget(). zfsonlinux/zfs@36df284 tried to address it by making a call
to the VFS to learn whether an inode is being evicted. If it was being
evicted the operation was retried after dropping and reacquiring the
relevant resources. Unfortunately, this introduced another deadlock.
INFO: task kworker/u24:6:891 blocked for more than 120 seconds.
Tainted: P O 3.13.6 #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
kworker/u24:6 D ffff88107fcd2e80 0 891 2 0x00000000
Workqueue: writeback bdi_writeback_workfn (flush-zfs-5)
ffff8810370ff950 0000000000000002 ffff88103853d940 0000000000012e80
ffff8810370fffd8 0000000000012e80 ffff88103853d940 ffff880f5c8be098
ffff88107ffb6950 ffff8810370ff980 ffff88103a9a5b78 0000000000000000
Call Trace:
[<ffffffff813dd1d4>] schedule+0x24/0x70
[<ffffffff8115fc09>] __wait_on_freeing_inode+0x99/0xc0
[<ffffffff8115fdd8>] find_inode_fast+0x78/0xb0
[<ffffffff811608c5>] ilookup+0x65/0xd0
[<ffffffffa035c5ab>] zfs_zget+0xdb/0x260 [zfs]
[<ffffffffa03589d6>] zfs_get_data+0x46/0x340 [zfs]
[<ffffffffa035fee1>] zil_add_block+0xa31/0xc00 [zfs]
[<ffffffffa0360642>] zil_commit+0x12/0x20 [zfs]
[<ffffffffa036a6e4>] zpl_putpage+0x174/0x840 [zfs]
[<ffffffff811071ec>] do_writepages+0x1c/0x40
[<ffffffff8116df2b>] __writeback_single_inode+0x3b/0x2b0
[<ffffffff8116ecf7>] writeback_sb_inodes+0x247/0x420
[<ffffffff8116f5f3>] wb_writeback+0xe3/0x320
[<ffffffff81170b8e>] bdi_writeback_workfn+0xfe/0x490
[<ffffffff8106072c>] process_one_work+0x16c/0x490
[<ffffffff810613f3>] worker_thread+0x113/0x390
[<ffffffff81066edf>] kthread+0xdf/0x100
This patch implements the original fix in a slightly different manner in
order to avoid both deadlocks. Instead of relying on a call to ilookup()
which can block in __wait_on_freeing_inode() the return value from igrab()
is used. This gives us the information that ilookup() provided without
the risk of a deadlock.
Alternately, this race could be closed by registering an sops->drop_inode()
callback. The callback would need to detect the active SA hold thereby
informing the VFS that this inode should not be evicted.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #180
When a vdev starts getting I/O or checksum errors it is now
possible to automatically rebuild to a hot spare device.
To cleanly support this functionality in a shell script some
additional information was added to all zevent ereports which
include a vdev. This covers both io and checksum zevents but
may be used but other scripts.
In the Illumos FMA solution the same information is required
but it is retrieved through the libzfs library interface.
Specifically the following members were added:
vdev_spare_paths - List of vdev paths for all hot spares.
vdev_spare_guids - List of vdev guids for all hot spares.
vdev_read_errors - Read errors for the problematic vdev
vdev_write_errors - Write errors for the problematic vdev
vdev_cksum_errors - Checksum errors for the problematic vdev.
By default the required hot spare scripts are installed but this
functionality is disabled. To enable hot sparing uncomment the
ZED_SPARE_ON_IO_ERRORS and ZED_SPARE_ON_CHECKSUM_ERRORS in the
/etc/zfs/zed.d/zed.rc configuration file.
These scripts do no add support for the autoexpand property. At
a minimum this requires adding a new udev rule to detect when
a new device is added to the system. It also requires that the
autoexpand policy be ported from Illumos, see:
https://github.com/illumos/illumos-gate/blob/master/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c
Support for detecting the correct name of a vdev when it's not
a whole disk was added by Turbo Fredriksson.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chris Dunlap <cdunlap@llnl.gov>
Signed-off-by: Turbo Fredriksson <turbo@bayour.com>
Issue #2
Due to the very poorly chosen argument name 'cleanup_fd' it was
completely unclear that this file descriptor is used to track the
current cursor location. When the file descriptor is created by
opening ZFS_DEV a private cursor is created in the kernel for the
returned file descriptor. Subsequent calls to zpool_events_next()
and zpool_events_seek() then require the file descriptor as an
argument to reposition the cursor. When the file descriptor is
closed the kernel state tracking the cursor is destroyed.
This patch contains no functional change, it just changes a
few variable names and clarifies the documentation.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chris Dunlap <cdunlap@llnl.gov>
Issue #2
The ZFS_IOC_EVENTS_SEEK ioctl was added to allow user space callers
to seek around the zevent file descriptor by EID. When a specific
EID is passed and it exists the cursor will be positioned there.
If the EID is no longer cached by the kernel ENOENT is returned.
The caller may also pass ZEVENT_SEEK_START or ZEVENT_SEEK_END to seek
to those respective locations.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chris Dunlap <cdunlap@llnl.gov>
Issue #2
Tagging each zevent with a unique monotonically increasing EID
(Event IDentifier) provides the required infrastructure for a user
space daemon to reliably process zevents. By writing the EID to
persistent storage the daemon can safely resume where it left off
in the event stream when it's restarted.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chris Dunlap <cdunlap@llnl.gov>
Issue #2
Originally, users had to handle spa namespace collisions by either
exporting the already imported pool or by specifying a new name for the
pool with a conflicting name. In the case of root pools from virtual
guests, neither approach to collision resolution is reasonable. This is
addressed by extending the new name syntax with a -t option to specify
that the new name is temporary. When specified, this sets an internal
flag that is passed into the kernel to tell it that all label updates
should refer to the name used in the original label. Consequently, the
original pool name will be retained on export.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2189
Remove the redundant call to zfs_unmount_snap() which was being
done after char array was freed,
This fixes an upstream regression that was introduced in commit
zfsonlinux/zfs@d09f25dc66, which
ported the Illumos 3744 changes.
Signed-off-by: Andrey Vesnovaty <andrey.vesnovaty@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Closes#2156
4088 use after free in arc_release()
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
https://www.illumos.org/issues/4088illumos/illumos-gate@ccc22e1304
From the illumos issue:
A race-induced use after free occurs in arc_release() where the
ARC header is used outside the critical section protected by the
hash_lock.
Ported by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Closes#2162
The spa_label_features nvlist is used in the sync context during pool
version upgrade.
Signed-off-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2168
These are needed by consumers (i.e. Lustre) who wish to perform a
callback in the syncing context. Both a blocking and non-blocking
version are available to callers.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Some callers of dmu_tx_assign() use the TXG_NOWAIT flag and call
dmu_tx_wait() themselves before retrying if the assignment fails.
The wait times for such callers are not accounted for in the
dmu_tx_assign kstat histogram, because the histogram only records
time spent in dmu_tx_assign(). This change moves the histogram
update to dmu_tx_wait() to properly account for all time spent there.
One downside of this approach is that it is possible to call
dmu_tx_wait() multiple times before successfully assigning a
transaction, in which case the cumulative wait time would not be
recorded. However, this case should not often arise in practice,
because most callers currently use one of these forms:
dmu_tx_assign(tx, TXG_WAIT);
dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
The first form should make just one call to dmu_tx_delay() inside of
dmu_tx_assign(). The second form retries with TXG_WAITED if the first
assignment fails and incurs a delay, in which case no further waiting
is performed. Therefore transaction delays normally occur in one
call to dmu_tx_wait() so the histogram should be fairly accurate.
Another possible downside of this approach is that the histogram will
no longer record overhead outside of dmu_tx_wait() such as in
dmu_tx_try_assign(). While I'm not aware of any reason for concern on
this point, it is conceivable that lock contention, long list
traversal, etc. could cause assignment delays that would not be
reflected in the histogram. Therefore the histogram should strictly
be used for visibility in to the normal delay mechanisms and not as a
profiling tool for code performance.
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1915
The nreserved column in the txgs kstat file always contains 0
following the write throttle restructuring of commit
e8b96c6007.
Prior to that commit, the nreserved column showed the number of bytes
temporarily reserved in the pool by a transaction group at sync time.
The new write throttle did away with temporary reservations and uses
the amount of dirty data instead. To approximate the old output of
the txgs kstat, the number of dirty bytes per-txg was passed in as
the nreserved value to spa_txg_history_set_io(). This approach did
not work as intended, because the per-txg dirty value is decremented
as data is written out to disk, so it is zero by the time we call
spa_txg_history_set_io(). To fix this, save the number of dirty
bytes before calling spa_sync(), and pass this value in to
spa_txg_history_set_io().
Also, since the name "nreserved" is now a misnomer, the column
heading is now labeled "ndirty".
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1696
A few counters in the dmu_tx kstats are obsolete or no longer
bumped properly.
- The sync task restructuring commit
13fe019870 removed the code
that bumpted dmu_tx_quota. The counter is now bumped in two
cases, instead of just the one case as before (after the result
of dsl_dataset_check_quota call). The second case is where
we check the requested reservation against the actual pool size,
as this is an implicit quota of sorts.
- The write throttle restructuring commit
e8b96c6007 makes dmu_tx_how and
dmu_tx_inflight obsolete, so they are removed.
Signed-off-by: Kohsuke Kawaguchi <kk@kohsuke.org>
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1914
Userland tools such as blkid, grub2-probe and zdb will go through the
buffer cache. However, ZFS uses on submit_bio() to bypass the buffer
cache when performing IO operations on vdevs for efficiency purposes.
This permits the on-disk state and buffer cache to fall out of
synchronization. That causes seemingly random failures when tools
reading stale metadata from the buffer cache try to access references to
data that is no longer there.
A particularly bad failure this causes involves grub2-probe, which is
used by grub2-mkconfig. Ordinarily, a rootfs might be called
rpool/ROOT/gentoo. However, when a failure occurs in grub2-probe,
grub2-mkconfig will generate a configuration file containing
/ROOT/gentoo, which omits the pool name and causes a boot failure.
This is avoidable by calling invalidate_bdev() on each flush, which is a
simple way to ensure that all non-dirty pages are wiped. Since userland
tools rarely access vdevs directly, this should be a fancy noop >99.999%
of the time and have little impact on IO. We could have tried a finer
grained approach for the rare instances in which the vdevs are accessed
frequently by userland. However, that would require consideration of
corner cases and it is not worth the effort.
Memory-wise, it would have been better to use a Linux kernel API hook to
disable the buffer cache on such devices, but it provides us no way of
doing that, so we opt for this approach instead. We should revisit that
idea in the future when higher priority issues have been tackled.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2150
4574 get_clones_stat does not call zap_count in non-debug kernel
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Marcel Telka <marcel@telka.sk>
Approved by: Gordon Ross <gwr@nexenta.com>
References:
https://www.illumos.org/issues/4574illumos/illumos-gate@03d1795fa6
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2147
The length (number of integers) argument passed to zap_lookup was wrong;
likely as a result of performing stack-reduction on the function.
Signed-off-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2141
Remove recursion from dsl_dir_willuse_space() to reduce stack usage.
Issues with stack overflow were observed in zfs recv of zvols,
likelihood of an overflow is proportional to the depth of the dataset
as dsl_dir_willuse_space() recurses to parent datasets.
Signed-off-by: Andrew Barnes <barnes333@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2069
Unfortunately, this change is an cheap attempt to work around a
pathological workload for the ARC. A "real" solution still needs to be
fleshed out, so this patch is intended to alleviate the situation in the
meantime. Let me try and describe the problem..
Data buffers residing in the dbuf hash table (dbuf cache) will keep a
hold on their respective dnode, this dnode will in turn keep a hold on
its backing dbuf (the physical block of the dnode object backing it).
Since the dnode has a hold on its backing dbuf, the arc buffer for this
dbuf is unevictable. What this essentially boils down to, "data" buffers
have the potential to pin "metadata" in the arc (as a result of these
dnode object buffers being unevictable).
This scenario becomes a real problem when the workload consists of many
small files (e.g. creating millions of 4K files). With this workload,
the arc's "arc_meta_used" space get filled up with buffers for any
resident directories as well as buffers for the objset's dnode object.
Once the "arc_meta_limit" is reached, the directory buffers will be
evicted and only the unevictable dnode object buffers will reside. If
the workload is simply creating new small files, these dnode object
buffers will never even be needed again, whereas the directory buffers
will be used constantly until the creates move to a new directory.
If "arc_c" and "arc_meta_limit" are sized appropriately, this
situation wont occur. This is because as the data buffers accumulate,
"arc_size" will eventually approach "arc_c" (before "arc_meta_used"
reaches "arc_meta_limit"); at that point the data buffers will be
evicted, which releases the hold on the dnode, which releases the hold
on the dnode object's dbuf, which allows that buffer to be evicted from
the arc in preference to more "useful" metadata.
So, to side step the issue, we simply need to ensure "arc_size" reaches
"arc_c" before "arc_meta_used" reaches "arc_meta_limit". In order to
pick a proper limit, we have to do some math.
To make things a little easier to follow, it is assumed that there will
only be a single data buffer per file (which is probably always the case
for "small" files anyways).
Based on the current internals of the arc, if N files residing in the
dbuf cache all pin a single dnode buffer (i.e. their dnodes all share
the same physical dnode object block), then the following amount of
"arc_meta_used" space will be consumed:
- 16K for the dnode object's block - [ 16384 bytes]
- N * sizeof(dnode_t) -------------- [ N * 928 bytes]
- (N + 1) * sizeof(arc_buf_t) ------ [(N + 1) * 72 bytes]
- (N + 1) * sizeof(arc_buf_hdr_t) -- [(N + 1) * 264 bytes]
- (N + 1) * sizeof(dmu_buf_impl_t) - [(N + 1) * 280 bytes]
To simplify, these N files will pin the following amount of
"arc_meta_used" space as unevictable:
Pinned "arc_meta_used" bytes = 16384 + N * 928 + (N + 1) * (72 + 264 + 280)
Pinned "arc_meta_used" bytes = 17000 + N * 1544
This pinned space is regardless of the size of the files, and is only
dependent on the number of pinned dnodes sharing a physical block
(i.e. N). For example, 32 512b files sharing a single dnode object
block would consume the same "arc_meta_used" space as 32 4K files
sharing a single dnode object block.
Now, given a files size of S, we can determine the total amount of
space that will be consumed in the arc:
Total = 17000 + N * 1544 + S * N
^^^^^^^^^^^^^^^^ ^^^^^
metadata data
So, given these formulas, we can generate a table which states the ratio
of pinned metadata to total arc (meta + data) using different values of
N (number of pinned dnodes per pinned physical dnode block) and S (size
of the file).
File Sizes (S)
| 512 | 1024 | 2048 | 4096 | 8192 | 16384 |
---+----------+----------+----------+----------+----------+----------+
1 | 0.973132 | 0.947670 | 0.900544 | 0.819081 | 0.693597 | 0.530921 |
2 | 0.951497 | 0.907481 | 0.830632 | 0.710325 | 0.550779 | 0.380051 |
N 4 | 0.918807 | 0.849809 | 0.738842 | 0.585844 | 0.414271 | 0.261250 |
8 | 0.877541 | 0.781803 | 0.641770 | 0.472505 | 0.309333 | 0.182965 |
16 | 0.835819 | 0.717945 | 0.559996 | 0.388885 | 0.241376 | 0.137253 |
32 | 0.802106 | 0.669597 | 0.503304 | 0.336277 | 0.202123 | 0.112423 |
As you can see, if we wanted to support the absolute worst case of 1
dnode per physical dnode block and 512b files, we would have to set the
"arc_meta_limit" to something greater than 97.3132% of "arc_c_max". At
that point, it essentially defeats the purpose of having an
"arc_meta_limit" at all.
This patch changes the default value of "arc_meta_limit" to be 75% of
"arc_c_max", which should be good enough for "most" workloads (I think).
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
Previously, the "data_size" field in the arcstats kstat contained the
amount of cached "metadata" and "data" in the ARC. The problem is this
then made it difficult to extract out just the "metadata" size, or just
the "data" size.
To make it easier to distinguish the two values, "data_size" has been
modified to count only buffers of type ARC_BUFC_DATA, and "meta_size"
was added to count only buffers of type ARC_BUFC_METADATA. If one wants
the old "data_size" value, simply sum the new "data_size" and
"meta_size" values.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
When the arc is at it's size limit and a new buffer is added, data will
be evicted (or recycled) from the arc to make room for this new buffer.
As far as I can tell, this is to try and keep the arc from over stepping
it's bounds (i.e. keep it below the size limitation placed on it).
This makes sense conceptually, but there appears to be a subtle flaw in
its current implementation, resulting in metadata buffers being
throttled. When it evicts from the arc's lists, it also passes in a
"type" so as to remove a buffer of the same type that it is adding. The
problem with this is that once the size limit is hit, the ratio of
"metadata" to "data" contained in the arc essentially becomes fixed.
For example, consider the following scenario:
* the size of the arc is capped at 10G
* the meta_limit is capped at 4G
* 9G of the arc contains "data"
* 1G of the arc contains "metadata"
Now, every time a new "metadata" buffer is created and added to the arc,
an older "metadata" buffer(s) will be removed from the arc; preserving
the 9G "data" to 1G "metadata" ratio that was in-place when the size
limit was reached. This occurs even though the amount of "metadata" is
far below the "metadata" limit. This can result in the arc behaving
pathologically for certain workloads.
To fix this, the arc_get_data_buf function was modified to evict "data"
from the arc even when adding a "metadata" buffer; unless it's at the
"metadata" limit. In addition, arc_evict now more closely resembles
arc_evict_ghost; such that when evicting "data" from the arc, it may
make a second pass over the arc lists and evict "metadata" if it cannot
meet the eviction size the first time around.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
Using "arc_meta_used" to determine if the arc's mru list is over it's
target value of "arc_p" doesn't seem correct. The size of the mru list
and the value of "arc_meta_used", although related, are completely
independent. Buffers contained in "arc_meta_used" may not even be
contained in the arc's mru list. As such, this patch removes
"arc_meta_used" from the calculation in arc_adjust.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
To maintain a strict limit on the metadata contained in the arc, while
preventing the arc buffer headers from completely consuming the
"arc_meta_used" space, we need to evict metadata buffers from the arc's
ghost lists along with the regular lists.
This change modifies arc_adjust_meta such that it more closely models
the adjustments made in arc_adjust. "arc_meta_used" is used similarly to
"arc_size", and "arc_meta_limit" is used similarly to "arc_c".
Testing metadata intensive workloads (e.g. creating, copying, and
removing millions of small files and/or directories) has shown this
change to make a dramatic improvement to the hit rate maintained in the
arc. While I think there is still room for improvement, this is a big
step in the right direction.
In addition, zpl_free_cached_objects was made into a no-op as I'm not
yet sure how to properly implement that function.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
This reverts commit c11a12bc3b.
Out of memory events were fixed by reverting this patch.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
It's unclear why adjustments to arc_p need to be dampened as they are in
arc_adjust. With that said, it's removal significantly improves the arc's
ability to "warm up" to a given workload. Thus, I'm disabling by default
until its usefulness is better understood.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
Setting a limit on the minimum value of "arc_p" has been shown to have
detrimental effects on the arc hit rate for certain "metadata" intensive
workloads. Specifically, this has been exhibited with a workload that
constantly dirties new "metadata" but also frequently touches a "small"
amount of mfu data (e.g. mkdir's).
What is seen is that the new anon data throttles the mfu list to a
negligible size (because arc_p > anon + mru in arc_get_data_buf), even
though the mfu ghost list receives a constant stream of hits. To remedy
this, arc_p is now allowed to drop to zero if the algorithm deems it
necessary.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
For specific workloads consisting mainly of mfu data and new anon data
buffers, the aggressive growth of arc_p found in the arc_get_data_buf()
function can have detrimental effects on the mfu list size and ghost
list hit rate.
Running a workload consisting of two processes:
* Process 1 is creating many small files
* Process 2 is tar'ing a directory consisting of many small files
I've seen arc_p and the mru grow to their maximum size, while the mru
ghost list receives 100K times fewer hits than the mfu ghost list.
Ideally, as the mfu ghost list receives hits, arc_p should be driven
down and the size of the mfu should increase. Given the specific
workload I was testing with, the mfu list size should grow to a point
where almost no mfu ghost list hits would occur. Unfortunately, this
does not happen because the newly dirtied anon buffers constancy drive
arc_p to its maximum value and keep it there (effectively prioritizing
the mru list and starving the mfu list down to a negligible size).
The logic to increment arc_p from within the arc_get_data_buf() function
was introduced many years ago in this upstream commit:
commit 641fbdae3a027d12b3c3dcd18927ccafae6d58bc
Author: maybee <none@none>
Date: Wed Dec 20 15:46:12 2006 -0800
6505658 target MRU size (arc.p) needs to be adjusted more aggressively
and since I don't fully understand the motivation for the change, I am
reluctant to completely remove it.
As a way to test out how it's removal might affect performance, I've
disabled that code by default, but left it tunable via a module option.
Thus, if its removal is found to be grossly detrimental for certain
workloads, it can be re-enabled on the fly, without a code change.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
In an attempt to prevent arc_c from collapsing "too fast", the
arc_shrink() function was updated to take a "bytes" parameter by this
change:
commit 302f753f16
Author: Brian Behlendorf <behlendorf1@llnl.gov>
Date: Tue Mar 13 14:29:16 2012 -0700
Integrate ARC more tightly with Linux
Unfortunately, that change failed to make a similar change to the way
that arc_p was updated. So, there still exists the possibility for arc_p
to collapse to near 0 when the kernel start calling the arc's shrinkers.
This change attempts to fix this, by decrementing arc_p by the "bytes"
parameter in the same way that arc_c is updated.
In addition, a minimum value of arc_p is attempted to be maintained,
similar to the way a minimum arc_p value is maintained in arc_adapt().
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2110
Decrease the mimimum ARC size from 1/32 of total system memory
(or 64MB) to a much smaller 4MB.
1) Large systems with over a 1TB of memory are being deployed
and reserving 1/32 of this memory (32GB) as the mimimum
requirement is overkill.
2) Tiny systems like the raspberry pi may only have 256MB of
memory in which case 64MB is far too large.
The ARC should be reclaimable if the VFS determines it needs
the memory for some other purpose. If you want to ensure the
ARC is never completely reclaimed due to memory pressure you
may still set a larger value with zfs_arc_min.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Issue #2110
