The spa_deadman() and spa_sync() functions can both be run in the
spa_sync context and therefore should use TQ_PUSHPAGE instead of
TQ_SLEEP.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1734Closes#1749
Locking mutex &vq->vq_lock in vdev_mirror_pending is unneeded:
* no data is modified
* only vq_pending_tree is read
* in case garbage is returned (eg. vq_pending_tree being updated
while the read is made) the worst case would be that a single
read could be queued on a mirror side which more busy than thought
The benefit of this change is streamlining of the code path since
it is taken for *every* mirror member on *every* read.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1739
dataset_remove_clones_key does recursion, so if the recursion goes
deep it can overrun the linux kernel stack size of 8KB. I have seen
this happen in the actual deployment, and subsequently confirmed it by
running a test workload on a custom-built kernel that uses 32KB stack.
See the following stack trace as an example of the case where it would
have run over the 8KB stack kernel:
Depth Size Location (42 entries)
----- ---- --------
0) 11192 72 __kmalloc+0x2e/0x240
1) 11120 144 kmem_alloc_debug+0x20e/0x500
2) 10976 72 dbuf_hold_impl+0x4a/0xa0
3) 10904 120 dbuf_prefetch+0xd3/0x280
4) 10784 80 dmu_zfetch_dofetch.isra.5+0x10f/0x180
5) 10704 240 dmu_zfetch+0x5f7/0x10e0
6) 10464 168 dbuf_read+0x71e/0x8f0
7) 10296 104 dnode_hold_impl+0x1ee/0x620
8) 10192 16 dnode_hold+0x19/0x20
9) 10176 88 dmu_buf_hold+0x42/0x1b0
10) 10088 144 zap_lockdir+0x48/0x730
11) 9944 128 zap_cursor_retrieve+0x1c4/0x2f0
12) 9816 392 dsl_dataset_remove_clones_key.isra.14+0xab/0x190
13) 9424 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
14) 9032 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
15) 8640 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
16) 8248 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
17) 7856 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
18) 7464 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
19) 7072 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
20) 6680 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
21) 6288 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
22) 5896 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
23) 5504 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
24) 5112 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
25) 4720 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
26) 4328 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
27) 3936 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
28) 3544 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
29) 3152 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
30) 2760 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
31) 2368 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
32) 1976 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
33) 1584 392 dsl_dataset_remove_clones_key.isra.14+0x10c/0x190
34) 1192 232 dsl_dataset_destroy_sync+0x311/0xf60
35) 960 72 dsl_sync_task_group_sync+0x12f/0x230
36) 888 168 dsl_pool_sync+0x48b/0x5c0
37) 720 184 spa_sync+0x417/0xb00
38) 536 184 txg_sync_thread+0x325/0x5b0
39) 352 48 thread_generic_wrapper+0x7a/0x90
40) 304 128 kthread+0xc0/0xd0
41) 176 176 ret_from_fork+0x7c/0xb0
This change reduces the stack usage in dsl_dataset_remove_clones_key
by allocating structures in heap, not in stack. This is not a fundamental
fix, as one can create an arbitrary large data set that runs over any
fixed size stack, but this will make the problem far less likely.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Kohsuke Kawaguchi <kk@kohsuke.org>
Closes#1726
The zpl_mknod() function was incorrectly negating its return value.
This doesn't cause any problems in the success case, but it does
prevent us from returning the correct error code for a failure.
The implementation of this function is now consistent with all
the other zpl_* functions.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1717
When compiling on an ARM device using gcc 4.7.3 several variables
in the zfs_obj_to_path_impl() function were flagged as uninitialized.
To resolve the warnings explicitly initialize them to zero.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1716
After the restructuring in 13fe019 The 'zfs rename' command will
result in a KM_SLEEP being called in the sync context. This may
deadlock due to reclaim so it was changed to KM_PUSHPAGE.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1711
2882 implement libzfs_core
2883 changing "canmount" property to "on" should not always remount dataset
2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Chris Siden <christopher.siden@delphix.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Reviewed by: Bill Pijewski <wdp@joyent.com>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Eric Schrock <Eric.Schrock@delphix.com>
References:
https://www.illumos.org/issues/2882https://www.illumos.org/issues/2883https://www.illumos.org/issues/2900illumos/illumos-gate@4445fffbbb
Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1293
Porting notes:
WARNING: This patch changes the user/kernel ABI. That means that
the zfs/zpool utilities built from master are NOT compatible with
the 0.6.2 kernel modules. Ensure you load the matching kernel
modules from master after updating the utilities. Otherwise the
zfs/zpool commands will be unable to interact with your pool and
you will see errors similar to the following:
$ zpool list
failed to read pool configuration: bad address
no pools available
$ zfs list
no datasets available
Add zvol minor device creation to the new zfs_snapshot_nvl function.
Remove the logging of the "release" operation in
dsl_dataset_user_release_sync(). The logging caused a null dereference
because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the
logging functions try to get the ds name via the dsl_dataset_name()
function. I've got no idea why this particular code would have worked
in Illumos. This code has subsequently been completely reworked in
Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring).
Squash some "may be used uninitialized" warning/erorrs.
Fix some printf format warnings for %lld and %llu.
Apply a few spa_writeable() changes that were made to Illumos in
illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and
3115 fixes.
Add a missing call to fnvlist_free(nvl) in log_internal() that was added
in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time
(zfsonlinux/zfs@9e11c73) because it depended on future work.
There is currently a subtle bug in the SA implementation which
can crop up which prevents us from safely using multiple variable
length SAs in one object.
Fortunately, the only existing use case for this are symlinks with
SA based xattrs. Therefore, until the root cause in the SA code
can be identified and fixed we prevent adding SA xattrs to symlinks.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1468
This reverts commit fadd0c4da1 which
introduced a regression in honoring the meta limit.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Close#1660
Commit torvalds/linux@2233f31aad
replaced ->readdir() with ->iterate() in struct file_operations.
All filesystems must now use the new ->iterate method.
To handle this the code was reworked to use the new ->iterate
interface. Care was taken to keep the majority of changes
confined to the ZPL layer which is already Linux specific.
However, minor changes were required to the common zfs_readdir()
function.
Compatibility with older kernels was accomplished by adding
versions of the trivial dir_emit* helper functions. Also the
various *_readdir() functions were reworked in to wrappers
which create a dir_context structure to pass to the new
*_iterate() functions.
Unfortunately, the new dir_emit* functions prevent us from
passing a private pointer to the filldir function. The xattr
directory code leveraged this ability through zfs_readdir()
to generate the list of xattr names. Since we can no longer
use zfs_readdir() a simplified zpl_xattr_readdir() function
was added to perform the same task.
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1653
Issue #1591
Because we need to be more frugal about our stack usage under
Linux. The __zio_execute() function was modified to re-dispatch
zios to a ZIO_TASKQ_ISSUE thread when we're in a context which
is known to be stack heavy. Those two contexts are the sync
thread and what ever thread is performing spa initialization.
Unfortunately, this change introduced an unlikely bug which can
result in a zio being re-dispatched indefinitely and never being
executed. If during spa initialization we handle a zio with
ZIO_PRIORITY_NOW it will be moved to the high priority queue.
When __zio_execute() is called again for the zio it will mis-
interpret the context and re-dispatch it again. The system
will get stuck spinning re-dispatching the zio and making no
forward progress.
To fix this rare issue __zio_execute() has been updated not
to re-dispatch zios on either the ZIO_TASKQ_ISSUE or
ZIO_TASKQ_ISSUE_HIGH task queues.
In practice this issue was rarely reported and can usually
be fixed by rebooting the system and importing the pool again.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1455
3618 ::zio dcmd does not show timestamp data
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: George Wilson <gwilson@zfsmail.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
http://www.illumos.org/issues/3618illumos/illumos-gate@c55e05cb35
Notes on porting to ZFS on Linux:
The original changeset mostly deals with mdb ::zio dcmd.
However, in order to provide the requested functionality
it modifies vdev and zio structures to keep the timing data
in nanoseconds instead of ticks. It is these changes that
are ported over in the commit in hand.
One visible change of this commit is that the default value
of 'zfs_vdev_time_shift' tunable is changed:
zfs_vdev_time_shift = 6
to
zfs_vdev_time_shift = 29
The original value of 6 was inherited from OpenSolaris and
was subotimal - since it shifted the raw tick value - it
didn't compensate for different tick frequencies on Linux and
OpenSolaris. The former has HZ=1000, while the latter HZ=100.
(Which itself led to other interesting performance anomalies
under non-trivial load. The deadline scheduler delays the IO
according to its priority - the lower priority the further
the deadline is set. The delay is measured in units of
"shifted ticks". Since the HZ value was 10 times higher,
the delay units were 10 times shorter. Thus really low
priority IO like resilver (delay is 10 units) and scrub
(delay is 20 units) were scheduled much sooner than intended.
The overall effect is that resilver and scrub IO consumed
more bandwidth at the expense of the other IO.)
Now that the bookkeeping is done is nanoseconds the shift
behaves correctly for any tick frequency (HZ).
Ported-by: Cyril Plisko <cyril.plisko@mountall.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1643
When CONFIG_UIDGID_STRICT_TYPE_CHECKS is enabled uid_t/git_t are
replaced by kuid_t/kgid_t, which are structures instead of integral
types. This causes any code that uses an integral type to fail to build.
The User Namespace functionality introduced in Linux 3.8 requires
CONFIG_UIDGID_STRICT_TYPE_CHECKS, so we could not build against any
kernel that supported it.
We resolve this by converting between the new kuid_t/kgid_t structures
and the original uid_t/gid_t types.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1589
When the meta limit is exceeded the ARC evicts some meta data
buffers from the mfu+mru lists. Unfortunately, for meta data
heavy workloads it's possible for these buffers to accumulate
on the ghost lists if arc_c doesn't exceed arc_size.
To handle this case arc_adjust_meta() has been entended to
explicitly evict meta data buffers from the ghost lists in
proportion to what was evicted from the mfu+mru lists.
If this is insufficient we request that the VFS release
some inodes and dentries. This will result in the release
of some dnodes which are counted as 'other' metadata.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The default behavior of arc_evict_ghost() is to start by evicting
data buffers. Then only if the requested number of bytes to evict
cannot be satisfied by data buffers move on to meta data buffers.
This is ideal for honoring arc_c since it's preferable to keep the
meta data cached. However, if we're trying to free memory from the
arc to honor the meta limit it's a problem because we will need to
discard all the data to get to the meta data.
To avoid this issue the arc_evict_ghost() is now passed a fourth
argumented describing which buffer type to start with. The
arc_evict() function already behaves exactly like this for a
same reason so this is consistent with the existing code.
All existing callers have been updated to pass ARC_BUFC_DATA so
this patch introduces no functional change. New callers may
pass ARC_BUFC_METADATA to skip immediately to evicting meta
data leaving the normal data untouched.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
3137 L2ARC compression
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
illumos/illumos-gate@aad02571bchttps://www.illumos.org/issues/3137http://wiki.illumos.org/display/illumos/L2ARC+Compression
Notes for Linux port:
A l2arc_nocompress module option was added to prevent the
compression of l2arc buffers regardless of how a dataset's
compression property is set. This allows the legacy behavior
to be preserved.
Ported by: James H <james@kagisoft.co.uk>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1379
This is analogous to SPL commit zfsonlinux/spl@b9b3715. While
we don't have clear evidence of systems getting caught here
indefinately like in the SPL this ensures that it will never
happen.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1579
zfs_readdir() is used by getdents(), which provides a list of all files
in directory, their types and an offset that be used by llseek() to seek
to the next directory entry.
On Solaris, the first two directory entries "." and ".." respectively
have offsets 1 and 2 on ZFS while the other files have rather large
numbers. Currently, ZFSOnLinux is giving "." offset 0 and all other
entries large numbers. The first entry's next entry offset points to
itself, which causes software that uses llseek() in conjunction with
getdents() for filesystem navigation to enter an infinite loop. The
offsets used for each directory entry are filesystem specific on all
platforms, so we can fix this by adopting the Solaris behavior.
Also, we currently report each directory entry as having type 0 (???).
This is not wrong, but we can do better. getdents() on Solaris does not
appear to provide this information, but it does on Linux and Mac OS X
do. ZFS provides easy access to type information in zfs_readdir(), so
this patch provides this as well.
Reported-by: Andrey <andrey@kudinov.su>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1624
3639 zpool.cache should skip over readonly pools
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Basil Crow <basil.crow@delphix.com>
Approved by: Gordon Ross <gwr@nexenta.com>
References:
illumos/illumos-gate@fb02ae0252https://www.illumos.org/issues/3639
Normally we don't list pools that are imported read-only in the cache
file, however you can accidentally get one into the cache file by
importing and exporting a read-write pool while a read-only pool is
imported:
$ zpool import -o readonly test1
$ zpool import test2
$ zpool export test2
$ zdb -C
This is a problem because if the machine reboots we import all pools in
the cache file as read-write.
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
When the property atime=on is set operations which only access
and inode do cause an atime update. However, it turns out that
dirty inodes with updated atimes are only written to disk when
the inodes get evicted from the cache. Somewhat surprisingly
the source suggests that this isn't a ZoL specific issue.
This behavior may in part explain why zfs's reclaim logic has
been observed to be slow. When reclaiming inodes its likely
that they have a dirty atime which will force a write to disk.
Obviously we don't want to force a write to disk for every
atime update, these needs to be batched. The right way to
do this is to fully implement the .dirty_inode and .write_inode
callbacks. However, to do that right requires proper unification
of some fields in the znode/inode. Then we could just mark the
inode dirty and leave it to the VFS to call .write_inode
periodically.
Until that work gets done we have to settle for some middle
ground. The simplest and safest thing we can do for now is
to write the dirty inode on last close. This should prevent
the majority of inodes in the cache from having dirty atimes
and not drastically increase the number of writes.
Some rudimentally testing to show how long it takes to drop
500,000 inodes from the cache shows promising results. This
is as expected because we're no longer do lots of IO as part
of the eviction, it was done earlier during the close.
w/out patch: ~30s to drop 500,000 inodes with drop_caches.
with patch: ~3s to drop 500,000 inodes with drop_caches.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The dmu_prefetch, dmu_free_long_range, dmu_free_object,
dmu_prealloc, dmu_write_policy, and dmu_sync symbols have
been exported so they may be used by other modules.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
dmu_tx_hold_object_impl can return NULL on error. Check for this
condition prior to dereferencing pointer. This can only occur if
the passed object was invalid or unallocated.
Signed-off-by: Nathaniel Clark <Nathaniel.Clark@misrule.us>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1610
The code involving b_thawed appears to be dead, so lets discard it.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1614
These functions are used in neither Illumos nor ZFSOnLinux. They appear
to have been replaced by arc_buf_alloc()/arc_buf_free(), so lets remove
them.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1614
We declare zio_alloc_arena using extern, but it does not appear to exist
anywhere in the code. This permits undefined behavior, so lets remove
it.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1614
The l2arc module options can be made safely writable. This allows
the options to be changed without unloading/loading the modules.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
These days modern SSDs can efficiently service concurrent reads
and writes. When this flag was added that wasn't really the
case for a variety of SSD controllers. But now we can set the
default value to take advantage of this parallelism and only
disable this as needed for specific troublesome hardware.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Based on the comments in arc.c we know that buffers can exist both
in arc and l2arc, under this circumstance both arc_buf_hdr_t and
l2arc_buf_hdr_t will be allocated. However the current logic only
cares for memory that l2arc_buf_hdr takes up when the buffer's
state transfers from or to arc_l2c_only. This will cause obvious
deviations for illumos's zfs version since the sizeof(l2arc_buf_hdr)
is larger than ZOL's. We can implement the calcuation in the
following simple way:
1. When allocate a l2arc_buf_hdr_t we add its memory consumption
instantly and subtract it when we free or evict the l2arc buf.
2. According to l2arc_hdr_stat_add and l2arc_hdr_stat_remove, if
the buffer only stays in l2arc we should also add the memory
its arc_buf_hdr_t consumes, so we only need to add HDR_SIZE to
arcstat_l2_hdr_size since we already concerned with L2HDR_SIZE
in step 1 and the same for transfering arc bufs from l2arc only
state.
The testbox has 2 4-core Intel Xeon CPUs(2.13GHz), with 16GB memory
and tests were set upped in the following way:
1. Fdisked a SATA disk into two partitions, one partition for zpool
storage and the other one was used as the cache device.
2. Generated some files occupying 14GB altogether in the zpool
prepared in step 1 using iozone.
3. Read them all using md5sum and watched the l2arc related statistics
in /proc/spl/kstat/zfs/arcstats. After the reading ended the
l2_hdr_size and l2_size were shown like this:
l2_size 4 4403780608
l2_hdr_size 4 0
which was weird.
4. After applying this patch and reran step 1-3, the results were
as following:
l2_size 4 4306443264
l2_hdr_size 4 535600
these numbers made sense, on 64-bit systems the
sizeof(l2arc_buf_hdr_t) is 16 bytes. Assue all blocks cached by
l2arc are 128KB, so 535600/16*128*1024=4387635200, since not all
blocks are equal-sized, the theoretical result will be a little
bigger, as we can see.
Since I'm familiar with systemtap instrumentation tool I used it to
examine what had happened. The script looked like this:
probe module("zfs").function("arc_chage_state")
{
if ($new_state == $arc_l2_only)
printf("change arc buf to arc_l2_only\n")
}
It will print out some information each time we call funciton
arc_chage_state if the argument new_state is arc_l2_only. I
gathered the trace logs and found that none of the arc bufs ran
into arc state arc_l2_only when the tests was running, this was
the reason why l2_hdr_size in step 3 was 0. The arc bufs fell into
arc_l2_only when the pool or the filesystem was offlined.
Signed-off-by: Ying Zhu <casualfisher@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The iterate_supers_type() function which was introduced in the
3.0 kernel was supposed to provide a safe way to call an arbitrary
function on all super blocks of a specific type. Unfortunately,
because a list_head was used a bug was introduced which made it
possible for iterate_supers_type() to get stuck spinning on a
super block which was just deactivated.
This can occur because when the list head is removed from the
fs_supers list it is reinitialized to point to itself. If the
iterate_supers_type() function happened to be processing the
removed list_head it will get stuck spinning on that list_head.
The bug was fixed in the 3.3 kernel by converting the list_head
to an hlist_node. However, to resolve the issue for existing
3.0 - 3.2 kernels we detect when a list_head is used. Then to
prevent the spinning from occurring the .next pointer is set to
the fs_supers list_head which ensures the iterate_supers_type()
function will always terminate.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1045Closes#861Closes#790
During mount a filesystem dataset would have the MS_RDONLY bit
incorrectly cleared even if the entire pool was read-only.
There is existing to code to handle this case but it was being run
before the property callbacks were registered. To resolve the
issue we move this read-only code after the callback registration.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1338
It is possible for an automounted snapshot which is expiring to
deadlock with a manual unmount of the snapshot. This can occur
because taskq_cancel_id() will block if the task is currently
executing until it completes. But it will never complete because
zfsctl_unmount_snapshot() is holding the zsb->z_ctldir_lock which
zfsctl_expire_snapshot() must acquire.
---------------------- z_unmount/0:2153 ---------------------
mutex_lock <blocking on zsb->z_ctldir_lock>
zfsctl_unmount_snapshot
zfsctl_expire_snapshot
taskq_thread
------------------------- zfs:10690 -------------------------
taskq_wait_id <waiting for z_unmount to exit>
taskq_cancel_id
__zfsctl_unmount_snapshot
zfsctl_unmount_snapshot <takes zsb->z_ctldir_lock>
zfs_unmount_snap
zfs_ioc_destroy_snaps_nvl
zfsdev_ioctl
do_vfs_ioctl
We resolve the deadlock by dropping the zsb->z_ctldir_lock before
calling __zfsctl_unmount_snapshot(). The lock is only there to
prevent concurrent modification to the zsb->z_ctldir_snaps AVL
tree. Moreover, we're careful to remove the zfs_snapentry_t from
the AVL tree before dropping the lock which ensures no other tasks
can find it. On failure it's added back to the tree.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chris Dunlap <cdunlap@llnl.gov>
Closes#1527
The read bandwidth of an N-way mirror can by increased by 50%,
and the IOPs by 10%, by more carefully selecting the preferred
leaf vdev.
The existing algorthm selects a perferred leaf vdev based on
offset of the zio request modulo the number of members in the
mirror. It assumes the drives are of equal performance and
that spreading the requests randomly over both drives will be
sufficient to saturate them. In practice this results in the
leaf vdevs being under utilized.
Utilization can be improved by preferentially selecting the leaf
vdev with the least pending IO. This prevents leaf vdevs from
being starved and compensates for performance differences between
disks in the mirror. Faster vdevs will be sent more work and
the mirror performance will not be limitted by the slowest drive.
In the common case where all the pending queues are full and there
is no single least busy leaf vdev a batching stratagy is employed.
Of the N least busy vdevs one is selected with equal probability
to be the preferred vdev for T microseconds. Compared to randomly
selecting a vdev to break the tie batching the requests greatly
improves the odds of merging the requests in the Linux elevator.
The testing results show a significant performance improvement
for all four workloads tested. The workloads were generated
using the fio benchmark and are as follows.
1) 1MB sequential reads from 16 threads to 16 files (MB/s).
2) 4KB sequential reads from 16 threads to 16 files (MB/s).
3) 1MB random reads from 16 threads to 16 files (IOP/s).
4) 4KB random reads from 16 threads to 16 files (IOP/s).
| Pristine | With 1461 |
| Sequential Random | Sequential Random |
| 1MB 4KB 1MB 4KB | 1MB 4KB 1MB 4KB |
| MB/s MB/s IO/s IO/s | MB/s MB/s IO/s IO/s |
---------------+-----------------------+------------------------+
2 Striped | 226 243 11 304 | 222 255 11 299 |
2 2-Way Mirror | 302 324 16 534 | 433 448 23 571 |
2 3-Way Mirror | 429 458 24 714 | 648 648 41 808 |
2 4-Way Mirror | 562 601 36 849 | 816 828 82 926 |
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1461
This change adds a new kstat to gain some visibility into the amount of
time spent in each call to dmu_tx_assign. A histogram is exported via
a new dmu_tx_assign_histogram-$POOLNAME file. The information contained
in this histogram is the frequency dmu_tx_assign took to complete given
an interval range. For example, given the below histogram file:
$ cat /proc/spl/kstat/zfs/dmu_tx_assign_histogram-tank
12 1 0x01 32 1536 19792068076691 20516481514522
name type data
1 us 4 859
2 us 4 252
4 us 4 171
8 us 4 2
16 us 4 0
32 us 4 2
64 us 4 0
128 us 4 0
256 us 4 0
512 us 4 0
1024 us 4 0
2048 us 4 0
4096 us 4 0
8192 us 4 0
16384 us 4 0
32768 us 4 1
65536 us 4 1
131072 us 4 1
262144 us 4 4
524288 us 4 0
1048576 us 4 0
2097152 us 4 0
4194304 us 4 0
8388608 us 4 0
16777216 us 4 0
33554432 us 4 0
67108864 us 4 0
134217728 us 4 0
268435456 us 4 0
536870912 us 4 0
1073741824 us 4 0
2147483648 us 4 0
one can see most calls to dmu_tx_assign completed in 32us or less, but a
few outliers did not. Specifically, 4 of the calls took between 262144us
and 131072us. This information is difficult, if not impossible, to gather
without this change.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1584
In the event that a pool gets suspended log this information to
the console. This is critical information and we want to make
sure it gets logged.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1555
To avoid a potential deadlock when using a zvol as a swap
device prevent vdev_disk_io_flush() from performing IO during
the bio_alloc().
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1508
When we remove references of arc bufs in the arc_anon state we
needn't take its header's hash_lock, so postpone it to where we
really need it to avoid unnecessary invocations of function buf_hash.
Signed-off-by: Ying Zhu <casualfisher@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1557
There are times when it is desirable for zfs to not automatically
populate the spa namespace at module load time using the pools
in the /etc/zfs/zpool.cache file. The zfs_autoimport_disable
module option has been added to control this behavior.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #330
Linux kernel commit torvalds/linux@db2a144 changed the return type
of block_device_operations->release() to void. Detect the expected
prototype and defined our callout accordingly.
Signed-off-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1494
One of the side effects of calling zvol_create_minors() in
zvol_init() is that all pools listed in the cache file will
be opened. Depending on the state and contents of your pool
this operation can take a considerable length of time.
Doing this at load time is undesirable because the kernel
is holding a global module lock. This prevents other modules
from loading and can serialize an otherwise parallel boot
process. Doing this after module inititialization also
reduces the chances of accidentally introducing a race
during module init.
To ensure that /dev/zvol/<pool>/<dataset> devices are
still automatically created after the module load completes
a udev rules has been added. When udev notices that the
/dev/zfs device has been create the 'zpool list' command
will be run. This then will cause all the pools listed
in the zpool.cache file to be opened.
Because this process in now driven asynchronously by udev
there is the risk of problems in downstream distributions.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #756
Issue #1020
Issue #1234
The following error will occur on some (possibly all) kernels
because blk_init_queue() will try to take the spinlock before
we initialize it.
BUG: spinlock bad magic on CPU#0, zpool/4054
lock: 0xffff88021a73de60, .magic: 00000000,
.owner: <none>/-1, .owner_cpu: 0
Pid: 4054, comm: zpool Not tainted 3.9.3 #11
Call Trace:
[<ffffffff81478ef8>] spin_dump+0x8c/0x91
[<ffffffff81478f1e>] spin_bug+0x21/0x26
[<ffffffff812da097>] do_raw_spin_lock+0x127/0x130
[<ffffffff8147d851>] _raw_spin_lock_irq+0x21/0x30
[<ffffffff812c2c1e>] cfq_init_queue+0x1fe/0x350
[<ffffffff812aacb8>] elevator_init+0x78/0x140
[<ffffffff812b2677>] blk_init_allocated_queue+0x87/0xb0
[<ffffffff812b26d5>] blk_init_queue_node+0x35/0x70
[<ffffffff812b271e>] blk_init_queue+0xe/0x10
[<ffffffff8125211b>] __zvol_create_minor+0x24b/0x620
[<ffffffff81253264>] zvol_create_minors_cb+0x24/0x30
[<ffffffff811bd9ca>] dmu_objset_find_spa+0xea/0x510
[<ffffffff811bda71>] dmu_objset_find_spa+0x191/0x510
[<ffffffff81253ea2>] zvol_create_minors+0x92/0x180
[<ffffffff811f8d80>] spa_open_common+0x250/0x380
[<ffffffff811f8ece>] spa_open+0xe/0x10
[<ffffffff8122817e>] pool_status_check.part.22+0x1e/0x80
[<ffffffff81228a55>] zfsdev_ioctl+0x155/0x190
[<ffffffff8116a695>] do_vfs_ioctl+0x325/0x5a0
[<ffffffff8116a950>] sys_ioctl+0x40/0x80
[<ffffffff814812c9>] ? do_page_fault+0x9/0x10
[<ffffffff81483929>] system_call_fastpath+0x16/0x1b
zd0: unknown partition table
We fix this by calling spin_lock_init before blk_init_queue.
The manner in which zvol_init() initializes structures is
suspectible to a race between initialization and a probe on
a zvol. We reorganize zvol_init() to prevent that.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
There is an extremely odd bug that causes zvols to fail to appear on
some systems, but not others. Recently, I was able to consistently
reproduce this issue over a period of 1 month. The issue disappeared
after I applied this change from FreeBSD.
This is from FreeBSD's pool version 28 import, which occurred in
revision 219089.
Ported-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #441
Issue #599
3122 zfs destroy filesystem should prefetch blocks
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>
References:
illumos/illumos-gate@b4709335aahttps://www.illumos.org/issues/3122
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1565
Commit 55d85d5a8c (backport of
the upstream changes) replaced three hardcoded constants:
#define SYNC_PASS_DEFERRED_FREE 2 /* defer frees after this pass */
#define SYNC_PASS_DONT_COMPRESS 4 /* don't compress after this pass */
#define SYNC_PASS_REWRITE 1 /* rewrite new bps after this pass */
with a tunable parameters:
int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */
int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */
int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */
This commit makes these tunables available as module parameters
in Linux. They should only be used for performance analysis
because changing them can result in subtle and pathological
performance problems.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1562
The approach taken was the rework zfs_holey() as little as
possible and then just wrap the code as needed to ensure
correct locking and error handling.
Tested with xfstests 285 and 286. All tests pass except for
7-9 of 285 which try to reserve blocks first via fallocate(2)
and fail because fallocate(2) is not yet supported.
Note that the filp->f_lock spinlock did not exist prior to
Linux 2.6.30, but we avoid the need for autotools check by
virtue of the fact that SEEK_DATA/SEEK_HOLE support was not
added until Linux 3.1.
An autoconf check was added for lseek_execute() which is
currently a private function but the expectation is that it
will be exported perhaps as early as Linux 3.11.
Reviewed-by: Richard Laager <rlaager@wiktel.com>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1384
This patch restores the zfs_holey() function from OpenSolaris.
This was removed by commit 3558fd7 because it wasn't clear we
had a use for it in ZoL. However, this functionality is a
prerequisite for adding SEEK_DATA/SEEK_HOLE support to the ZPL.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Issue #1384
By definitition these allocations will never fail. For
consistency with the rest of the code remove this dead error
handling code.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1558
Fix a pair of conditions in which a concurrent umount can cause
NULL pointer dereferences:
* zfs_sb_teardown - prevent a NULL dereference by not calling
dmu_objset_pool with a null z_os.
* zfs_resume_fs - don't try to unmount with a null z_os. This
change makes the ZoL code more consistent
with both Illumos and FreeBSD.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1543
Previous commit 7ef5e54e2e caused
module probe failure on 32-bit systems, dmesg showed
Unknown symbol __moddi3
This was caused by the modulo operation 'gethrtime() % tqs->stqs_count'
in the committed code. Instead of implementing __moddi3 for all 32-bit
systems, Behlendorf advised we can just cast the return value of
gethrtime() into a uint64_t, since gethrtime does not return negative
value on all circumstances we need not care about the potential overflow.
Signed-off-by: Ying Zhu <casualfisher@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1551
Until these hooks are fully implemented return the expected
-EOPNOTSUPP error to indicate they are not functional. This
allows test suites such as xfstests to cleanly skip testing
this functionality until it's implemented.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #229
The non-blocking allocation handlers in nvlist_alloc() would be
mistakenly assigned if any flags other than KM_SLEEP were passed.
This meant that nvlists allocated with KM_PUSHPUSH or other KM_*
debug flags were effectively always using atomic allocations.
While these failures were unlikely it could lead to assertions
because KM_PUSHPAGE allocations in particular are guaranteed to
succeed or block. They must never fail.
Since the existing API does not allow us to pass allocation
flags to the private allocators the cleanest thing to do is to
add a KM_PUSHPAGE allocator.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closeszfsonlinux/spl#249
3805 arc shouldn't cache freed blocks
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Richard Elling <richard.elling@dey-sys.com>
Reviewed by: Will Andrews <will@firepipe.net>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
illumos/illumos-gate@6e6d5868f5https://www.illumos.org/issues/3805
ZFS should proactively evict freed blocks from the cache.
On dcenter, we saw that we were caching ~256GB of metadata, while the
pool only had <4GB of metadata on disk. We were wasting about half the
system's RAM (252GB) on blocks that have been freed.
Even though these freed blocks will never be used again, and thus will
eventually be evicted, this causes us to use memory inefficiently for 2
reasons:
1. A block that is freed has no chance of being accessed again, but will
be kept in memory preferentially to a block that was accessed before it
(and is thus older) but has not been freed and thus has at least some
chance of being accessed again.
2. We partition the ARC into several buckets:
user data that has been accessed only once (MRU)
metadata that has been accessed only once (MRU)
user data that has been accessed more than once (MFU)
metadata that has been accessed more than once (MFU)
The user data vs metadata split is somewhat arbitrary, and the primary
control on how much memory is used to cache data vs metadata is to
simply try to keep the proportion the same as it has been in the past
(each bucket "evicts against" itself). The secondary control is to
evict data before evicting metadata.
Because of this bucketing, we may end up with one bucket mostly
containing freed blocks that are very old, while another bucket has more
recently accessed, still-allocated blocks. Data in the useful bucket
(with still-allocated blocks) may be evicted in preference to data in
the useless bucket (with old, freed blocks).
On dcenter, we saw that the MFU metadata bucket was 230MB, while the MFU
data bucket was 27GB and the MRU metadata bucket was 256GB. However,
the vast majority of data in the MRU metadata bucket (256GB) was freed
blocks, and thus useless. Meanwhile, the MFU metadata bucket (230MB)
was constantly evicting useful blocks that will be soon needed.
The problem of cache segmentation is a larger problem that needs more
investigation. However, if we stop caching freed blocks, it should
reduce the impact of this more fundamental issue.
Ported-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1503
3552 condensing one space map burns 3 seconds of CPU in spa_sync() thread
3564 spa_sync() spends 5-10% of its time in metaslab_sync() (when not condensing)
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
References:
illumos/illumos-gate@16a4a80742https://www.illumos.org/issues/3552https://www.illumos.org/issues/3564
Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1513
3006 VERIFY[S,U,P] and ASSERT[S,U,P] frequently check if first
argument is zero
Reviewed by Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by George Wilson <george.wilson@delphix.com>
Approved by Eric Schrock <eric.schrock@delphix.com>
References:
illumos/illumos-gate@fb09f5aad4https://illumos.org/issues/3006
Requires:
zfsonlinux/spl@1c6d149feb
Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1509
When SA xattrs are enabled only fallback to checking the directory
xattrs when the name is not found as a SA xattr. Otherwise, the SA
error which should be returned to the caller is overwritten by the
directory xattr errors. Positive return values indicating success
will also be immediately returned.
In the case of #1437 the ERANGE error was being correctly returned
by zpl_xattr_get_sa() only to be overridden with ENOENT which was
returned by the subsequent unnessisary call to zpl_xattr_get_dir().
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1437
As a part of scrub/resilver tuning zfs_scrub_limit fell out of use,
but the definition of the variable remained in place.
Moreover various guides still (misleadingly) mention it as a way
to influence resilver/scrub behavior.
This commit removes its finally.
Signed-off-by: Cyril Plisko <cyril.plisko@mountall.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1444
The assertions in ddt_phys_decref and ddt_sync_entry cast ddp->ddp_refcnt
from uint64_t to int64_t, with a reference count bigger than 2^63, e.g. the
reference count of zero blocks commonly available in spare files, we may
mistakenly hit these assertations, so drop the type conversions here.
Signed-off-by: Ying Zhu <casualfisher@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1436
The vn_rdwr() function performs I/O by calling the vfs_write() or
vfs_read() functions. These functions reside just below the system
call layer and the expectation is they have almost the entire 8k of
stack space to work with. In fact, certain layered configurations
such as ext+lvm+md+multipath require the majority of this stack to
avoid stack overflows.
To avoid this posibility the vn_rdwr() call in dump_bytes() has been
moved to the ZIO_TYPE_FREE, taskq. This ensures that all I/O will be
performed with the majority of the stack space available. This ends
up being very similiar to as if the I/O were issued via sys_write()
or sys_read().
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1399Closes#1423
3581 spa_zio_taskq[ZIO_TYPE_FREE][ZIO_TASKQ_ISSUE]->tq_lock is piping hot
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Gordon Ross <gordon.ross@nexenta.com>
Approved by: Richard Lowe <richlowe@richlowe.net>
References:
illumos/illumos-gate@ec94d32https://illumos.org/issues/3581
Notes for Linux port:
Earlier commit 08d08eb reduced contention on this taskq lock by simply
reducing the number of z_fr_iss threads from 100 to one-per-CPU. We
also optimized the taskq implementation in zfsonlinux/spl@3c6ed54.
These changes significantly improved unlink performance to acceptable
levels.
This patch further reduces time spent spinning on this lock by
randomly dispatching the work items over multiple independent task
queues. The Illumos ZFS developers stated that this lock contention
only arose after "3329 spa_sync() spends 10-20% of its time in
spa_free_sync_cb()" was landed. It's not clear if 3329 affects the
Linux port or not. I didn't see spa_free_sync_cb() show up in
oprofile sessions while unlinking large files, but I may just not
have used the right test case.
I tested unlinking a 1 TB of data with and without the patch and
didn't observe a meaningful difference in elapsed time. However,
oprofile showed that the percent time spent in taskq_thread() was
reduced from about 16% to about 5%. Aside from a possible slight
performance benefit this may be worth landing if only for the sake of
maintaining consistency with upstream.
Ported-by: Ned Bass <bass6@llnl.gov>
Closes#1327
3329 spa_sync() spends 10-20% of its time in spa_free_sync_cb()
3330 space_seg_t should have its own kmem_cache
3331 deferred frees should happen after sync_pass 1
3335 make SYNC_PASS_* constants tunable
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan McDonald <danmcd@nexenta.com>
Approved by: Eric Schrock <eric.schrock@delphix.com>
References:
illumos/illumos-gate@01f55e48fbhttps://www.illumos.org/issues/3329https://www.illumos.org/issues/3330https://www.illumos.org/issues/3331https://www.illumos.org/issues/3335
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
3306 zdb should be able to issue reads in parallel
3321 'zpool reopen' command should be documented in the man
page and help
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>
References:
illumos/illumos-gate@31d7e8fa33https://www.illumos.org/issues/3306https://www.illumos.org/issues/3321
The vdev_file.c implementation in this patch diverges significantly
from the upstream version. For consistenty with the vdev_disk.c
code the upstream version leverages the Illumos bio interfaces.
This makes sense for Illumos but not for ZoL for two reasons.
1) The vdev_disk.c code in ZoL has been rewritten to use the
Linux block device interfaces which differ significantly
from those in Illumos. Therefore, updating the vdev_file.c
to use the Illumos interfaces doesn't get you consistency
with vdev_disk.c.
2) Using the upstream patch as is would requiring implementing
compatibility code for those Solaris block device interfaces
in user and kernel space. That additional complexity could
lead to confusion and doesn't buy us anything.
For these reasons I've opted to simply move the existing vn_rdwr()
as is in to the taskq function. This has the advantage of being
low risk and easy to understand. Moving the vn_rdwr() function
in to its own taskq thread also neatly avoids the possibility of
a stack overflow.
Finally, because of the additional work which is being handled by
the free taskq the number of threads has been increased. The
thread count under Illumos defaults to 100 but was decreased to 2
in commit 08d08e due to contention. We increase it to 8 until
the contention can be address by porting Illumos #3581.
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1354
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>
NOTES: This patch has been reworked from the original in the
following ways to accomidate Linux ZFS implementation
*) Usage of the cyclic interface was replaced by the delayed taskq
interface. This avoids the need to implement new compatibility
code and allows us to rely on the existing taskq implementation.
*) An extern for zfs_txg_synctime_ms was added to sys/dsl_pool.h
because declaring externs in source files as was done in the
original patch is just plain wrong.
*) Instead of panicing the system when the deadman triggers a
zevent describing the blocked vdev and the first pending I/O
is posted. If the panic behavior is desired Linux provides
other generic methods to panic the system when threads are
observed to hang.
*) For reference, to delay zios by 30 seconds for testing you can
use zinject as follows: 'zinject -d <vdev> -D30 <pool>'
References:
illumos/illumos-gate@283b84606bhttps://www.illumos.org/issues/3246
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1396
A deadlock was accidentally introduced by commit e95853a which
can occur when the system is under memory pressure. What happens
is that while the txg_quiesce thread is holding the tx->tx_cpu
locks it enters memory reclaim. In the context of this memory
reclaim it then issues synchronous I/O to a ZVOL swap device.
Because the txg_quiesce thread is holding the tx->tx_cpu locks
a new txg cannot be opened to handle the I/O. Deadlock.
The fix is straight forward. Move the memory allocation outside
the critical region where the tx->tx_cpu locks are held. And for
good measure change the offending allocation to KM_PUSHPAGE to
ensure it never attempts to issue I/O during reclaim.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1274
According to the getxattr(2) man page the ERANGE errno should be
returned when the size of the value buffer is to small to hold the
result. Prior to this patch the implementation would just truncate
the value to size bytes.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1408
The zpl_readdir() function shouldn't be registered as part of
the zpl_file_operations table, it must only be part of the
zpl_dir_file_operations table. By removing this callback
the VFS will now correctly return ENOTDIR when calling
getdents() on a file.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1404
Previous patches have allowed you to set an increased ashift to
avoid doing 512b IO with 4k sector devices. However, it was not
possible to set the ashift lower than the reported physical sector
size even when a smaller logical size was supported. In practice,
there are several cases where settong a lower ashift is useful:
* Most modern drives now correctly report their physical sector
size as 4k. This causes zfs to correctly default to using a 4k
sector size (ashift=12). However, for some usage models this
new default ashift value causes an unacceptable increase in
space usage. Filesystems with many small files may see the
total available space reduced to 30-40% which is unacceptable.
* When replacing a drive in an existing pool which was created
with ashift=9 a modern 4k sector drive cannot be used. The
'zpool replace' command will issue an error that the new drive
has an 'incompatible sector alignment'. However, by allowing
the ashift to be manual specified as smaller, non-optimal,
value the device may still be safely used.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1381Closes#1328
Issue #967
Issue #548
3422 zpool create/syseventd race yield non-importable pool
3425 first write to a new zvol can fail with EFBIG
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>
References:
illumos/illumos-gate@bda8819455https://www.illumos.org/issues/3422https://www.illumos.org/issues/3425
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1390
The assumption in zio_ddt_free() is that ddt_phys_select() must
always find a match. However, if that fails due to a damaged
DDT or some other reason the code will NULL dereference in
ddt_phys_decref().
While this should never happen it has been observed on various
platforms. The result is that unless your willing to patch the
ZFS code the pool is inaccessible. Therefore, we're choosing
to more gracefully handle this case rather than leave it fatal.
http://mail.opensolaris.org/pipermail/zfs-discuss/2012-February/050972.html
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1308
Enabling metaslab debugging will prevent space maps from being
automatically unloaded. This can significantly increase the
memory footprint but being able to dynamically control this is
helpful for debugging and certain performance testing.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The mainline kernel started defining GCC_VERSION with commit
torvalds/linux@3f3f8d2f48.
Unfortunately, LZ4 also defines this macro, but the two
defintions are incompatible. We undefine GCC_VERSION in lz4.c
to handle this.
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1339
A few files still refer to @behlendorf's private fork on
github. Use the primary web site URL instead. Two typos
are also corrected.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Provide a mechanism to control the directory name the modules
are installed in. The kernel privdes INSTALL_MOD_DIR for
this but it was hardcoded to be 'addon/zfs'.
Add a KMODDIR variable which can be passed to 'make install'
to override the default directory name. While we're here
change the default from 'addon/zfs' to 'extra' which is the
kernel.org default.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The new snapdev dataset property may be set to control the
visibility of zvol snapshot devices. By default this value
is set to 'hidden' which will prevent zvol snapshots from
appearing under /dev/zvol/ and /dev/<dataset>/. When set to
'visible' all zvol snapshots for the dataset will be visible.
This functionality was largely added because when automatic
snapshoting is enabled large numbers of read-only zvol snapshots
will be created. When creating these devices the kernel will
attempt to read their partition tables, and blkid will attempt
to identify any filesystems on those partitions. This leads
to a variety of issues:
1) The zvol partition tables will be read in the context of
the `modprobe zfs` for automatically imported pools. This
is undesirable and should be done asynchronously, but for
now reducing the number of visible devices helps.
2) Udev expects to be able to complete its work for a new
block devices fairly quickly. When many zvol devices are
added at the same time this is no longer be true. It can
lead to udev timeouts and missing /dev/zvol links.
3) Simply having lots of devices in /dev/ can be aukward from
a management standpoint. Hidding the devices your unlikely
to ever use helps with this. Any snapshot device which is
needed can be made visible by changing the snapdev property.
NOTE: This patch changes the default behavior for zvols which
was effectively 'snapdev=visible'.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1235Closes#945
Issue #956
Issue #756
The changes to zvol.c were never merged from the last onnv_147
bulk update. This was because zvol.c was largely rewritten
for Linux making it fairly easy to miss these sorts of changes.
This causes a regression when importing a zpool with zvols
read-only. This does not impact pool which only contain
filesystem datasets.
References:
illumos/illumos-gate@f9af39b
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1332Closes#1333
The PaX team modified the kernel's modpost to report writeable function
pointers as section mismatches because they are potential exploit
targets. We could ignore the warnings, but their presence can obscure
actual issues. Proper const correctness can also catch programming
mistakes.
Building the kernel modules against a PaX/GrSecurity patched Linux 3.4.2
kernel reports 133 section mismatches prior to this patch. This patch
eliminates 130 of them. The quantity of writeable function pointers
eliminated by constifying each structure is as follows:
vdev_opts_t 52
zil_replay_func_t 24
zio_compress_info_t 24
zio_checksum_info_t 9
space_map_ops_t 7
arc_byteswap_func_t 5
The remaining 3 writeable function pointers cannot be addressed by this
patch. 2 of them are in zpl_fs_type. The kernel's sget function requires
that this be non-const. The final writeable function pointer is created
by SPL_SHRINKER_DECLARE. The kernel's set_shrinker() and
remove_shrinker() functions also require that this be non-const.
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1300
The issue with hot spares in ZoL is because it opens all leaf
vdevs exclusively (O_EXCL). On Linux, exclusive opens cause
subsequent exclusive opens to fail with EBUSY.
This could be resolved by not opening any of the devices
exclusively, which is what Illumos does, but the additional
protection offered by exclusive opens is desirable. It cleanly
prevents you from accidentally adding an in-use non-ZFS device
to your pool.
To fix this we very slightly relaxed the usage of O_EXCL in
the following ways.
1) Functions which open the device but only read had the
O_EXCL flag removed and were updated to use O_RDONLY.
2) A common holder was added to the vdev disk code. This
allow the ZFS code to internally open the device multiple
times but non-ZFS callers may not.
3) An exception was added to make_disks() for hot spare when
creating partition tables. For hot spare devices which
are already opened exclusively we skip creating the partition
table because this must already have been done when the disk
was originally added as a hot spare.
Additional minor changes include fixing check_in_use() to use
a partition instead of a slice suffix. And is_spare() was moved
above make_disks() to avoid adding a forward reference.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#250
As described by the comment and enforced the by assertion the
v->vdev_wholedisk will never be -1. The wholedisk handling
is performed by the user space utilities. To prevent confusion
this dead code is being removed.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
When vdev_disk.c was implemented for Linux we failed to handle the
reopen case. According to the vdev_reopen() comment leaf vdevs should
not be closed or opened when v->vdev_reopening is set. Under Linux
we would always close and open the device.
This issue was only noticed when a 'zpool scrub' command was run while
the leaf vdev device names in /dev/disk/by-vdev were missing. The
scrub command calls vdev_reopen() which caused the vdevs to be closed
but they couldn't be reopened due to the missing links. The result
was that all the vdevs were marked unavailable and the pool was
halted due to failmode=wait.
This patch adds the missing functionality in a similiar fashion to
to the Illumos code.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
To determine whether the kernel is capable of handling empty barrier
BIOs, we check for the presence of the bio_empty_barrier() macro,
which was introduced in 2.6.24. If this macro is defined, then we can
flush disk vdevs; if it isn't, then flushing is disabled.
Unfortunately, the bio_empty_barrier() macro was removed in 2.6.37,
even though the kernel is still capable of handling empty barrier BIOs.
As a result, flushing is effectively disabled on kernels >= 2.6.37,
meaning that starting from this kernel version, zfs doesn't use
barriers to guarantee on-disk data consistency. This is quite bad and
can lead to potential data corruption on power failures.
This patch fixes the issue by removing the configure check for
bio_empty_barrier(), as we don't support kernels <= 2.6.24 anymore.
Thanks to Richard Kojedzinszky for catching this nasty bug.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1318
The zfs_arc_memory_throttle_disable module option was introduced
by commit 0c5493d470 to resolve a
memory miscalculation which could result in the txg_sync thread
spinning.
When this was first introduced the default behavior was left
unchanged until enough real world usage confirmed there were no
unexpected issues. We've now reached that point. Linux's
direct reclaim is working as expected so we're enabling this
behavior by default.
This helps pave the way to retire the spl_kmem_availrmem()
functionality in the SPL layer. This was the only caller.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #938
A couple of assertions in spa.c were designed to prevent the use of
invalid pool versions. They were written under the assumption
that all valid pools are less than SPA_VERSION. Since feature flags
jumped from 28 to 5000, any numbers in the range 28 to 5000
non-inclusive will fail to trigger them. We switch to the new
SPA_VERSION_IS_SUPPORTED macro to correct this.
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1282
It turns out that the Linux VFS doesn't strictly handle all cases
where a component path name exceeds MAXNAMELEN. It does however
appear to correctly handle MAXPATHLEN for us.
The right way to handle this appears to be to add an explicit
check to the zpl_lookup() function. Several in-tree filesystems
handle this case the same way.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1279
Two more locations where KM_SLEEP was used in a call which must
use KM_PUSHPAGE were found while using the zpool upgrade command.
See commit b8d06fc for additional details.
Also make a small correction to the comment block above
dsl_dir_open_spa().
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1268
Explicitly case this value to an unsigned long long for 32-bit
systems to inform the compiler that a long type should not be
used. Otherwise we get the following compiler error:
dmu_send.c:376: error: integer constant is too large for
‘long’ type
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The way in which virtual box ab(uses) memory can throw off the
free memory calculation in arc_memory_throttle(). The result is
the txg_sync thread will effectively spin waiting for memory to
be released even though there's lots of memory on the system.
To handle this case I'm adding a zfs_arc_memory_throttle_disable
module option largely for virtual box users. Setting this option
disables free memory checks which allows the txg_sync thread to
make progress.
By default this option is disabled to preserve the current
behavior. However, because Linux supports direct memory reclaim
it's doubtful throttling due to perceived memory pressure is ever
a good idea. We should enable this option by default once we've
done enough real world testing to convince ourselve there aren't
any unexpected side effects.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#938
Commit 1eb5bfa introduced a new zfs_disable_dup_eviction tunable.
It should have been made available as a module option in the
original patch but was overlooked.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
When a system attribute layout is created an inconsistency may occur
between the system attribute header (sa_hdr_phys_t) size and the
variable-sized attribute count stored in the layout. The inconsistency
results in the following failed assertion when SA_HDR_SIZE_MATCH_LAYOUT
returns false:
SPLError: 11315:0:(sa.c:1541:sa_find_idx_tab())
ASSERTION((IS_SA_BONUSTYPE(bonustype) && SA_HDR_SIZE_MATCH_LAYOUT(hdr,
tb)) || !IS_SA_BONUSTYPE(bonustype) || (IS_SA_BONUSTYPE(bonustype) &&
hdr->sa_layout_info == 0)) failed
The bug originates in this snippet from sa_find_sizes().
if (is_var_sz && var_size > 1) {
if (P2ROUNDUP(hdrsize + sizeof (uint16_t),
*total < full_space) {
hdrsize += sizeof (uint16_t);
This assumes that the current variable-sized attribute will be stored in
the current buffer and accounts for the space needed to store its size
in the sa_hdr_phys_t. However if the next attribute spills over we need
to store a blkptr_t at the end of the bonus buffer to point to the spill
block. If the current attribute is in the way of the blkptr_t then it
too will be relocated into the spill block. But since we've already
accounted for it in the header size we get the inconsistency described
above.
To avoid this, record the index of the last variable-sized attribute
that prompted a hdrsize increase, and reverse the increase if we later
determine that that attribute will be relocated to the spill block.
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1250
A rounding discrepancy exists between how sa_build_layouts() and
sa_find_sizes() calculate when the spill block needs to be kicked in.
This results in a narrow size range where sa_build_layouts() believes
there must be a spill block allocated but due to the discrepancy there
isn't. A panic then occurs when the hdl->sa_spill NULL pointer is
dereferenced.
The following reproducer for this bug was isolated:
truncate -s 128m /tmp/tank
zpool create tank /tmp/tank
zfs create -o xattr=sa tank/fish
ln -s `perl -e 'print "z" x 41'` /tank/fish/z
setfattr -hn trusted.foo -v`perl -e 'print "z"x45'` /tank/fish/z
This test results in roughly the following system attribute (SA)
layout:
176 bytes - "standard" SA's
41 bytes - name of symbolic link target
100 bytes - XDR encoded nvlist for xattr
---
317 bytes - total
Because 317 is less than DN_MAX_BONUSLEN (320), sa_find_sizes()
decides no spill block is needed. But sa_build_layouts() rounds 41 up
to 48 when computing the space requirements so it tries to switch to
the spill block.
Note that we were only able to reproduce this bug using a combination
of symbolic links and the Linux-specific xattr=sa dataset property.
So while this issue is not technically Linux-specific, it may be
difficult or impossible to hit the narrow size range needed to
reproduce it on other platforms.
To fix the discrepancy, round the running total in sa_find_sizes() up
to an 8-byte boundary before accounting for each SA, since this is how
they will be stored in the bonus and (possibly) spill buffers.
To make the intent of the code more clear, explicitly assert key
assumptions about expected alignment of data and whether spill-over
will occur.
Signed-off-by: Matthew Ahrens <mahrens@delphix.com
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1240
3035 LZ4 compression support in ZFS and GRUB
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Christopher Siden <csiden@delphix.com>
References:
illumos/illumos-gate@a6f561b4aehttps://www.illumos.org/issues/3035http://wiki.illumos.org/display/illumos/LZ4+Compression+In+ZFS
This patch has been slightly modified from the upstream Illumos
version to be compatible with Linux. Due to the very limited
stack space in the kernel a lz4 workspace kmem cache is used.
Since we are using gcc we are also able to take advantage of the
gcc optimized __builtin_ctz functions.
Support for GRUB has been dropped from this patch. That code
is available but those changes will need to made to the upstream
GRUB package.
Lastly, several hunks of dead code were dropped for clarity. They
include the functions real_LZ4_uncompress(), LZ4_compressBound()
and the Visual Studio specific hunks wrapped in _MSC_VER.
Ported-by: Eric Dillmann <eric@jave.fr>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1217
Explicitly set acl details to zero to silence gcc (zfs_acl_node_read
can't be sure zfs_acl_znode_info will set acl_count and aclsize).
Normally suppressing these warnings by setting this to zero at
declaration time is a bad idea but in this instance it's hard to
avoid and should be fairly safe.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1244
Retire the dmu_snapshot_id() function which was introduced in the
initial .zfs control directory implementation. There is already
an existing dsl_dataset_snap_lookup() which does exactly what we
need, and the dmu_snapshot_id() function as implemented is racy.
https://github.com/zfsonlinux/zfs/issues/1215#issuecomment-12579879
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1238
Added d_clear_d_op() helper function which clears some flags and the
registered dentry->d_op table. This is required because d_set_d_op()
issues a warning when the dentry operations table is already set.
For the .zfs control directory to work properly we must be able to
override the default operations table and register custom .d_automount
and .d_revalidate callbacks.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ned Bass <bass6@llnl.gov>
Closes#1230
Callers of zap_deref_leaf() must be careful to drop leaf->l_rwlock
since that function returns with the lock held on success. All other
callers drop the lock correctly but it seems fzap_cursor_move_to_key()
does not. This may block writers or cause VERIFY failures when the
lock is freed.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1215Closeszfsonlinux/spl#143Closeszfsonlinux/spl#97
In zpl_revalidate() it's possible for the nameidata to be NULL
for kernels which still accept the parameter. In particular,
lookup_one_len() calls d_revalidate() with a NULL nameidata.
Resolve the issue by checking for a NULL nameidata in which case
just set the flags to 0.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1226
As of Linux 2.6.37 the right way to register custom dentry
operations is to use the super block's ->s_d_op field.
For older kernels they should be registered as part of the
lookup operation.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1223
Commit 65d56083b4 fixes the lock
inversion between spa_namespace_lock and bdev->bd_mutex but only
for the first user of spa_namespace_lock: dmu_objset_own().
Later spa_namespace_lock gets acquired by dsl_prop_get_integer()
though dsl_prop_get()->dsl_dataset_hold()->dsl_dir_open_spa()->
spa_open()->spa_open_common() without this "protection". By
moving the mutex release after this second use, even this
acquisition of the lock is "protected" by the ERESTARTSYS trick.
Signed-off-by: Massimo Maggi <me@massimo-maggi.eu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1220
This reverts commit 53c7411919
effectively reinstating the asynchronous xattr cleanup code.
These Linux changes were reverted because after testing
and careful contemplation I was convinced that due to the
89260a1c8851ce05ea04b23606ba438b271d890 commit they were no
longer required.
Unfortunately, the deadlock described in #1176 was a case
which wasn't considered. At mount zfs_unlinked_drain() can
occur which will unlink a list of znodes in effectively a
random order which isn't safe. The only reason it was safe
to originally revert this change was the we could guarantee
that the VFS would always prune the xattr leaves before the
parents.
Therefore, until we can cleanly resolve this deadlock for
all cases we need to keep this change in spite of the xattr
unlink performance penalty associated with it.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1176
Issue #457
Rolling back a mounted filesystem with open file handles and
cached dentries+inodes never worked properly in ZoL. The
major issue was that Linux provides no easy mechanism for
modules to invalidate the inode cache for a file system.
Because of this it was possible that an inode from the previous
filesystem would not get properly dropped from the cache during
rolling back. Then a new inode with the same inode number would
be create and collide with the existing cached inode. Ideally
this would trigger an VERIFY() but in practice the error wasn't
handled and it would just NULL reference.
Luckily, this issue can be resolved by sprucing up the existing
Solaris zfs_rezget() functionality for the Linux VFS.
The way it works now is that when a file system is rolled back
all the cached inodes will be traversed and refetched from disk.
If a version of the cached inode exists on disk the in-core
copy will be updated accordingly. If there is no match for that
object on disk it will be unhashed from the inode cache and
marked as stale.
This will effectively make the inode unfindable for lookups
allowing the inode number to be immediately recycled. The inode
will then only be accessible from the cached dentries. Subsequent
dentry lookups which reference a stale inode will result in the
dentry being invalidated. Once invalidated the dentry will drop
its reference on the inode allowing it to be safely pruned from
the cache.
Special care is taken for negative dentries since they do not
reference any inode. These dentires will be invalidate based
on when they were added to the dentry cache. Entries added
before the last rollback will be invalidate to prevent them
from masking real files in the dataset.
Two nice side effects of this fix are:
* Removes the dependency on spl_invalidate_inodes(), it can now
be safely removed from the SPL when we choose to do so.
* zfs_znode_alloc() no longer requires a dentry to be passed.
This effectively reverts this portition of the code to its
upstream counterpart. The dentry is not instantiated more
correctly in the Linux ZPL layer.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ned Bass <bass6@llnl.gov>
Closes#795
Lookups in the snapshot control directory for an existing snapshot
fail with ENOENT if an earlier lookup failed before the snapshot was
created. This is because the earlier lookup causes a negative dentry
to be cached which is never invalidated.
The bug can be reproduced as follows (the second ls should succeed):
$ ls /tank/.zfs/snapshot/s
ls: cannot access /tank/.zfs/snapshot/s: No such file or directory
$ zfs snap tank@s
$ ls /tank/.zfs/snapshot/s
ls: cannot access /tank/.zfs/snapshot/s: No such file or directory
To remedy this, always invalidate cached dentries in the snapshot
control directory. Since these entries never exist on disk there is
no significant performance penalty for the extra lookups.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1192
A misplaced single quote caused the umount command to fail with a
syntax error when unmounting snapshots under the .zfs/snapshot
control directory.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1210
3189 kernel panic in ZFS test suite during hotspare_onoffline_004_neg
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Arne Jansen <sensille@gmx.net>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
illumos/illumos-gate@8f0b538d1d
changeset: 13818:e9ad0a945d45
https://www.illumos.org/issues/3189
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2618 arc.c mistypes in the comments
Reviewed by: Jason King <jason.brian.king@gmail.com>
Reviewed by: Josef Sipek <jeffpc@josefsipek.net>
Approved by: Richard Lowe <richlowe@richlowe.net>
References:
illumos/illumos-gate@fc98fea58e
illumos changeset: 13721:5b51a16a186f
https://www.illumos.org/issues/2618
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
As of Linux 3.4 the UMH_WAIT_* constants were renumbered. In
particular, the meaning of "1" changed from UMH_WAIT_PROC (wait for
process to complete), to UMH_WAIT_EXEC (wait for the exec, but not the
process). A number of call sites used the number 1 instead of the
constant name, so the behavior was not as expected on kernels with this
change.
One visible consequence of this change was that processes accessing
automounted snapshots received an ELOOP error because they failed to
wait for zfs.mount to complete.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#816
This reverts commit 7afcf5b1da which
accidentally introduced a regression with the .zfs snapshot directory.
While the updated code still does correctly mount the requested
snapshot. It updates the vfsmount such that it references the
original dataset vfsmount. The result is that the snapshot itself
isn't visible.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #816
Related to 91579709fc we need to
be very careful about not overrunning the stack in kernel space.
However, in user space we're already allowing slightly larger
stacks so this stack usage optimization is not required there.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
To save valuable stack all zio's were made asynchronous when in the
tgx_sync_thread context or during pool initialization. See commit
2fac4c2 for the original patch and motivation.
Unfortuantely, the changes to dsl_pool_sync_context() made by the
feature flags broke this logic causing in __zio_execute() to dispatch
itself infinitely when called during pool initialization. This
commit refines the existing logic to specificly target only the two
cases we care about.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
3349 zpool upgrade -V bumps the on disk version number, but leaves
the in core version
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
illumos/illumos-gate@25345e4666https://www.illumos.org/issues/3349
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
2762 zpool command should have better support for feature flags
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Eric Schrock <Eric.Schrock@delphix.com>
References:
illumos/illumos-gate@57221772c3https://www.illumos.org/issues/2762
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
3090 vdev_reopen() during reguid causes vdev to be treated as corrupt
3102 vdev_uberblock_load() and vdev_validate() may read the wrong label
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Approved by: Eric Schrock <Eric.Schrock@delphix.com>
References:
illumos/illumos-gate@dfbb943217
illumos changeset: 13777:b1e53580146d
https://www.illumos.org/issues/3090https://www.illumos.org/issues/3102
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#939
2619 asynchronous destruction of ZFS file systems
2747 SPA versioning with zfs feature flags
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <gwilson@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Eric Schrock <Eric.Schrock@delphix.com>
References:
illumos/illumos-gate@53089ab7c8illumos/illumos-gate@ad135b5d64
illumos changeset: 13700:2889e2596bd6
https://www.illumos.org/issues/2619https://www.illumos.org/issues/2747
NOTE: The grub specific changes were not ported. This change
must be made to the Linux grub packages.
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
In a debug build, certain GCC versions flag an array bounds warning in
the below code from dnode_sync.c
} else {
int i;
ASSERT(dn->dn_next_nblkptr[txgoff] < dnp->dn_nblkptr);
/* the blkptrs we are losing better be unallocated */
for (i = dn->dn_next_nblkptr[txgoff];
i < dnp->dn_nblkptr; i++)
ASSERT(BP_IS_HOLE(&dnp->dn_blkptr[i]));
This usage is in fact safe, since the ASSERT ensures the index does
not exceed to maximum possible number of block pointers. However gcc
can't determine that the assignment 'i = dn->dn_next_nblkptr[txgoff];'
falls within the array bounds so it issues a warning. To avoid this,
initialize i to zero to make gcc happy but skip the elements before
dn->dn_next_nblkptr[txgoff] in the loop body. Since a dnode contains
at most 3 block pointers this overhead should be negligible.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#950
This reverts commit 9dcb971983
which was originally introduced to debug occasional slow I/Os.
These I/Os would complete eventually but were observed to take
several 100 seconds.
The root cause of this issue was the CFQ scheduler which can,
under certain conditions, excessively delay an I/O from being
issued to the device. This issue was mitigated somewhat by
commit 84daaddedb which ensures
the I/O elevator gets changed even for DM style devices.
This change isn't in any way harmful but it does conflict with
a required change to properly account from I/O wait time.
Because Linux does not export the io_schedule_timeout() function
we must instead rely on io_schedule() via cv_wait_io().
The additional debugging information which was added to the
delay event has been intentionally left in place.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
In all but one case the spa_namespace_lock is taken before the
bdev->bd_mutex lock. But Linux __blkdev_get() function calls
fops->open() with the bdev->bd_mutex lock held and we must
somehow still safely acquire the spa_namespace_lock.
To avoid a potential lock inversion deadlock we preemptively
try to take the spa_namespace_lock(). Normally it will not
be contended and this is safe because spa_open_common() handles
the case where the caller already holds the spa_namespace_lock.
When it is contended we risk a lock inversion if we were to
block waiting for the lock. Luckily, the __blkdev_get()
function allows us to return -ERESTARTSYS which will result in
bdev->bd_mutex being dropped, reacquired, and fops->open() being
called again. This process can be repeated safely until both
locks are acquired.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jorgen Lundman <lundman@lundman.net>
Closes#612
This reverts commit 31f2b5abdf back
to the original code until the fsync(2) performance regression
can be addressed.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
It's my understanding that the zfs_fsyncer_key TSD was added as
a performance omtimization to reduce contention on the zl_lock
from zil_commit(). This issue manifested itself as very long
(100+ms) fsync() system call times for fsync() heavy workloads.
However, under Linux I'm not seeing the same contention that
was originally described. Therefore, I'm removing this code
in order to ween ourselves off any dependence on TSD. If the
original performance issue reappears on Linux we can revisit
fixing it without resorting to TSD.
This just leaves one small ZFS TSD consumer. If it can be
cleanly removed from the code we'll be able to shed the SPL
TSD implementation entirely.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closeszfsonlinux/spl#174
The current state of udev and devicer-mapper devices makes it difficult
to construct a mapping of DM partitions and their underlying DM device.
For example, with a /dev directory with the following contents:
$ ls -d /dev/dm-*
/dev/dm-0
/dev/dm-1
/dev/dm-2
/dev/dm-3
it is not immediately apparent if these are completely separate devices,
or partitions and real devices intermixed. In contrast, SCSI devices
would appear as so:
$ ls -d /dev/sd*
/dev/sda
/dev/sda1
/dev/sdb
/dev/sdb1
Here, one can immediately determine that there are two devices (sda and
sdb), each containing a single partition. The lack of a predictable and
consistent mapping from DM devices to DM device partitions makes it
difficult for user space to process these devices the same way it does
SCSI devices.
As a result, the ZFS utilities do not partition DM devices, and instead
set the "vdev_wholedisk" label to 0 and treat them as partitions. This
has the side effect that, even if ZFS has sole ownership of the device,
the IO scheduler will not be modified because it is treated as a
partition.
This change adds an exception for DM devices in vdev_elevator_switch,
allowing the elevator to be modified even though the "vdev_wholedisk"
property is not set.
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1149
During the original ZoL port the vdev_uses_zvols() function was
disabled until it could be properly implemented. This prevented
a zpool from use a zvol for its slog device.
This patch implements that missing functionality by adding a
zvol_is_zvol() function to zvol.c. Given the full path to a
device it will lookup the device and verify its major number
against the registered zvol major number for the system. If
they match we know the device is a zvol.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1131
Revert the portion of commit d3aa3ea which always resulted in the
SAs being update when an mmap()'ed file was closed. That change
accidentally resulted in unexpected ctime updates which upset tools
like git. That was always a horrible hack and I'm happy it will
never make it in to a tagged release.
The right fix is something I initially resisted doing because I
was worried about the additional overhead. However, in hindsight
the overhead isn't as bad as I feared.
This patch implemented the sops->dirty_inode() callback which is
unsurprisingly called when an inode is dirtied. We leverage this
callback to keep the znode SAs strictly in sync with the inode.
However, for now we're going to go slowly to avoid introducing
any new unexpected issues by only updating the atime, mtime, and
ctime. This will cover the callpath of most concern to us.
->filemap_page_mkwrite->file_update_time->update_time->
mark_inode_dirty_sync->__mark_inode_dirty->dirty_inode
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#764Closes#1140
Ensure that the path member pointers are associated with the
newly-mounted snapshot when zpl_snapdir_automount() returns. Otherwise
the follow_automount() function may be called repeatedly, leading to an
incorrect ELOOP error return. This problem was observed as a 'Too many
levels of symbolic links' error from user-space commands accessing an
unmounted snapshot in the .zfs/snapshot directory.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#816
Linux kernel commit d8e794d accidentally broke the delayed work
APIs for non-GPL callers. While the APIs to schedule a delayed
work item are still available to all callers, it is no longer
possible to initialize the delayed work item.
I'm cautiously optimistic we could get the delayed_work_timer_fn
exported for all callers in the upstream kernel. But frankly
the compatibility code to use this kernel interface has always
been problematic.
Therefore, this patch abandons direct use the of the Linux
kernel interface in favor of the new delayed taskq interface.
It provides roughly the same functionality as delayed work queues
but it's a stable interface under our control.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1053
When writes to zvols invoke ZIL, zfs_range_new_proxy() is called,
which allocates memory using KM_SLEEP, triggering a warning.
Switch to KM_PUSHPAGE to silence that warning. See commit
b8d06fca08 for additional details.
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1138
This reverts commit b00131d43c which
is no longer needed due to e89260a1c8.
This change forces all xattr znodes to hold a reference on their
parent which ensures prune_icache() will never attempt to evict
both the parent and child concurrently. This effectively prevents
the deadlock condition from ever occuring.
Therefore we can safely revert back to the upstream synchronous
cleanup code. This is nice because it keeps our code base closer
to upstream and resolves the performance issues introduced by the
original deadlock fix.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#457
When updating a file via mmap()'ed I/O preserve the mtime/ctime
which were updated when the page was made writable by the generic
callback filemap_page_mkwrite().
But more importantly than preserving the exact time add the missing
call to sa_bulk_update(). This ensures that the znode modifications
are written to disk as part of the transaction. Without this the
inode may mistaken rollback to the previous on-disk znode state.
Additionally, for mmap()'ed znodes explicitly set the atime, mtime,
and ctime on close using the up to date values in the inode. This
is critical because writepage() may occur after close and on close
we need to ensure the values are correct.
Original-patch-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#764
Unlike normal file or directory znodes, an xattr znode is
guaranteed to only have a single parent. Therefore, we can
take a refernce on that parent if it is provided at create
time and cache it. Additionally, we take care to cache it
on any subsequent zfs_zaccess() where the parent is provided
as an optimization.
This allows us to avoid needing to do a zfs_zget() when
setting up the SELinux security xattr in the create path.
This is critical because a hash lookup on the directory
will deadlock since it is locked.
The zpl_xattr_security_init() call has also been moved up
to the zpl layer to ensure TXs to create the required
xattrs are performed after the create TX. Otherwise we
run the risk of deadlocking on the open create TX.
Ideally the security xattr should be fully constructed
before the new inode is unlocked. However, doing so would
require far more extensive changes to ZFS.
This change may also have the benefitial side effect of
ensuring xattr directory znodes are evicted from the cache
before normal file or directory znodes due to the extra
reference.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#671
Add the missing error handling to load_nvlist(). There's no good
reason this needs to be fatal. All callers of load_nvlist() do
correctly handle an error condition and it is preferable that an
error be returned. This will allow 'zpool import -FX' to safely
attempt to rollback through previous txgs looking for a good one.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1120
Due to the slightly increased size of the ZFS super block
caused by 30315d2 there are now allocation warnings. The
allocation size is still small (just over 8k) and super
blocks are rarely allocated so we suppress the warning.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1101
If zvol_alloc() fails zv will be set to NULL and dereferenced
in out_dmu_objset_disown. To avoid this entirely the zv->objset
line is moved up in to the success block.
Original-patch-by: Jorgen Lundman <lundman@lundman.net>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1109
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Richard Elling <richard.elling@richardelling.com>
Reviewed by: Gordon Ross <gwr@nexenta.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Approved by: Richard Lowe <richlowe@richlowe.net>
Refererces to Illumos issue:
https://www.illumos.org/issues/2671
This patch has been slightly modified from the upstream Illumos
version. In the upstream implementation a warning message is
logged to the console. To prevent pointless console noise this
notification is now posted as a "ereport.fs.zfs.vdev.bad_ashift"
event.
The event indicates a non-optimial (but entirely safe) ashift
value was used to create the pool. Depending on your workload
this may impact pool performance. Unfortunately, the only way
to correct the issue is to recreate the pool with a new ashift.
NOTE: The unrelated fix to the comment in zpool_main.c appears
in the upstream commit and was preserved for consistnecy.
Ported-by: Cyril Plisko <cyril.plisko@mountall.com>
Reworked-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#955
Gunnar Beutner did all the hard work on this one by correctly
identifying that this issue is a race between dmu_sync() and
dbuf_dirty().
Now in all cases the caller is responsible for preventing this
race by making sure the zfs_range_lock() is held when dirtying
a buffer which may be referenced in a log record. The mmap
case which relies on zfs_putpage() was not taking the range
lock. This code was accidentally dropped when the function
was rewritten for the Linux VFS.
This patch adds the required range locking to zfs_putpage().
It also adds the missing ZFS_ENTER()/ZFS_EXIT() macros which
aren't strictly required due to the VFS holding a reference.
However, this makes the code more consistent with the upsteam
code and there's no harm in being extra careful here.
Original-patch-by: Gunnar Beutner <gunnar@beutner.name>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#541
When using a zvol to back a btrfs filesystem the btrfs mount
would hang. This was due to the bio completion callback used
in btrfs assuming that lower level drivers would never modify
the bio->bi_io_vecs after they were submitted via bio_submit().
If they are modified btrfs will miscalculate which pages need
to be unlocked resulting in a hang.
It's worth mentioning that other file systems such as ext[234]
and xfs work fine because they do not make the same assumption
in the bio completion callback.
The most straight forward way to fix the issue is to present
the semantics expected by btrfs. This is done by cloning the
bios attached to each request and then using the clones bvecs
to perform the required accounting. The clones are freed after
each read/write and the original unmodified bios are linked back
in to the request.
Signed-off-by: Chris Wedgwood <cw@f00f.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#469
There have been reports of ZFS deadlocking due to what appears to
be a lost IO. This patch addes some debugging to determine the
exact state of the IO which neither 1) completed, 2) failed, or
3) timed out after zio_delay_max (30) seconds.
This information will be logged using the ZFS FMA infrastructure
as a 'delay' event and posted to the internal zevent log. By
default the last 64 events will be kept in the log but the limit
is configurable via the zfs_zevent_len_max module option.
To dump the contents of the log use the 'zpool events -v' command
and look for the resource.fs.zfs.delay event. It will include
various information about the pool, vdev, and zio which may shed
some light on the issue.
In the context of this change the 120 second kernel blocked thread
watchdog has been disabled for synchronous IOs.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #930
Create a kstat file which contains useful statistics about the
last N txgs processed. This can be helpful when analyzing pool
performance. The new KSTAT_TYPE_TXG type was added for this
purpose and it tracks the following statistics per-txg.
txg - Unique txg number
state - State (O)pen/(Q)uiescing/(S)yncing/(C)ommitted
birth; - Creation time
nread - Bytes read
nwritten; - Bytes written
reads - IOPs read
writes - IOPs write
open_time; - Length in nanoseconds the txg was open
quiesce_time - Length in nanoseconds the txg was quiescing
sync_time; - Length in nanoseconds the txg was syncing
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The interface for the ddt_zap_count() function assumes it can
never fail. However, internally ddt_zap_count() is implemented
with zap_count() which can potentially fail. Now because there
was no way to return the error to the caller a VERIFY was used
to ensure this case never happens.
Unfortunately, it has been observed that pools can be damaged in
such a way that zap_count() fails. The result is that the pool can
not be imported without hitting the VERIFY and crashing the system.
This patch reworks ddt_object_count() so the error can be safely
caught and returned to the caller. This allows a pool which has
be damaged in this way to be safely rewound for import.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#910
This reverts commit a5c20e2a0a which
accidentally introduced a regression for real 4k sector devices.
See issue #1065 for details.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1065
The following warning was originally added to provide visibility
in to how often a dio gets heavily fragmented in to over 16 bios.
This can happen due to constraints imposed by the block device
and may have a negitive impact on performance but is otherwise
harmless. To prevent needless confusion and worry the message
has been removed.
kernel: WARNING: Resized bio's/dio to 32
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
When automounting a snapshot in the .zfs/snapshot directory
make sure to quote both the dataset name and the mount point.
This ensures that if either component contains spaces, which
are allowed, they get handled correctly.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1027
In the current code, logbias=throughput implies the following:
1) All synchronous writes are logged in indirect mode.
2) The slog is not used.
(1) makes sense because it avoids writing the data twice, which is
obviously a good thing when the user wants maximum pool throughput.
(2), however, is a surprising decision. Considering all writes are
indirect, the log record doesn't contain the actual data, only pointers
to DMU blocks. As a result, log records written in logbias=throughput
mode are quite small, and as such, it doesn't make any sense to write
them to the main pool since slogs are usually optimized for small
synchronous writes.
In fact, the current behavior is actually harmful for performance,
because log blocks and data blocks from dmu_sync() seldom have the same
allocation size and as a result are usually allocated from different
metaslabs. This means that if a spindle has to write both log blocks and
DMU blocks (which is likely to happen under heavy load), it will have to
seek between the two. Allocating the log blocks from the slog pool
instead of the main pool avoids these unnecessary seeks.
This commit makes ZFS use the slog on datasets with logbias=throughput.
Real-life performance testing shows a 50% synchronous write performance
increase with some large commit sizes, and no negative effect in other
cases.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1013
Currently, ZIL blocks are spread over vdevs using hint block pointers
managed by the ZIL commit code and passed to metaslab_alloc(). Spreading
log blocks accross vdevs is important for performance: indeed, using
mutliple disks in parallel decreases the ZIL commit latency, which is
the main performance metric for synchronous writes. However, the current
implementation suffers from the following issues:
1) It would be best if the ZIL module was not aware of such low-level
details. They should be handled by the ZIO and metaslab modules;
2) Because the hint block pointer is managed per log, simultaneous
commits from multiple logs might use the same vdevs at the same time,
which is inefficient;
3) Because dmu_write() does not honor the block pointer hint, indirect
writes are not spread.
The naive solution of rotating the metaslab rotor each time a block is
allocated for the ZIL or dmu_sync() doesn't work in practice because the
first ZIL block to be written is actually allocated during the previous
commit. Consequently, when metaslab_alloc() decides the vdev for this
block, it will do so while a bunch of other allocations are happening at
the same time (from dmu_sync() and other ZILs). This means the vdev for
this block is chosen more or less at random. When the next commit
happens, there is a high chance (especially when the number of blocks
per commit is slightly less than the number of the disks) that one disk
will have to write two blocks (with a potential seek) while other disks
are sitting idle, which defeats spreading and increases the commit
latency.
This commit introduces a new concept in the metaslab allocator:
fastwrites. Basically, each top-level vdev maintains a counter
indicating the number of synchronous writes (from dmu_sync() and the
ZIL) which have been allocated but not yet completed. When the metaslab
is called with the FASTWRITE flag, it will choose the vdev with the
least amount of pending synchronous writes. If there are multiple vdevs
with the same value, the first matching vdev (starting from the rotor)
is used. Once metaslab_alloc() has decided which vdev the block is
allocated to, it updates the fastwrite counter for this vdev.
The rationale goes like this: when an allocation is done with
FASTWRITE, it "reserves" the vdev until the data is written. Until then,
all future allocations will naturally avoid this vdev, even after a full
rotation of the rotor. As a result, pending synchronous writes at a
given point in time will be nicely spread over all vdevs. This contrasts
with the previous algorithm, which is based on the implicit assumption
that blocks are written instantaneously after they're allocated.
metaslab_fastwrite_mark() and metaslab_fastwrite_unmark() are used to
manually increase or decrease fastwrite counters, respectively. They
should be used with caution, as there is no per-BP tracking of fastwrite
information, so leaks and "double-unmarks" are possible. There is,
however, an assert in the vdev teardown code which will fire if the
fastwrite counters are not zero when the pool is exported or the vdev
removed. Note that as stated above, marking is also done implictly by
metaslab_alloc().
ZIO also got a new FASTWRITE flag; when it is used, ZIO will pass it to
the metaslab when allocating (assuming ZIO does the allocation, which is
only true in the case of dmu_sync). This flag will also trigger an
unmark when zio_done() fires.
A side-effect of the new algorithm is that when a ZIL stops being used,
its last block can stay in the pending state (allocated but not yet
written) for a long time, polluting the fastwrite counters. To avoid
that, I've implemented a somewhat crude but working solution which
unmarks these pending blocks in zil_sync(), thus guaranteeing that
linguering fastwrites will get pruned at each sync event.
The best performance improvements are observed with pools using a large
number of top-level vdevs and heavy synchronous write workflows
(especially indirect writes and concurrent writes from multiple ZILs).
Real-life testing shows a 200% to 300% performance increase with
indirect writes and various commit sizes.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1013
Brian Behlendorf
GregorKopka
Kohsuke Kawaguchi
Tim Chase
Matthew Ahrens
Richard Yao
Saso Kiselkov
George Wilson
Nathaniel Clark
Ying Zhu
Prakash Surya
Shen Yan
Chris Dunlop
Pawel Jakub Dawidek
Cyril Plisko
Li Dongyang
Madhav Suresh
Adam Leventhal
Caleb James DeLisle
Martin Matuska
Jan Engelhardt
Richard Yao
Ned Bass
Eric Dillmann
George Wilson
Etienne Dechamps
Chris Wedgwood
Massimo Maggi
Christopher Siden
Arne Jansen
Bart Coddens
Matt Johnston
Jorgen Lundman
George Wilson
Etienne Dechamps