Cleanup the kernel SIMD code by removing kernel dependencies.
- Replace XSTATE_XSAVE with our own XSAVE implementation for all
kernels not exporting kernel_fpu{begin,end}(), see #13059
- Replace union fpregs_state by a uint8_t * buffer and get the size
of the buffer from the hardware via the CPUID instruction
- Replace kernels xgetbv() by our own implementation which was
already there for userspace.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Attila Fülöp <attila@fueloep.org>
Closes#13102
This reduces scheduler overhead by letting the reader consume bigger
chunks (64k => 128k at full throttle)
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#13133
Instead of writing to "devnull" and rming it later, just
> /dev/null to not have to cleanup later.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Co-authored-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#13133
By introducing lzc_send_wrapper() and routing all ZFS_IOC_SEND*
users through it, we fix a Linux 5.10-introduced bug (see comment)
This is all /transparent/ to the users API, ABI, and usage-wise,
and disabled on FreeBSD and if the output is already a pipe,
and transparently nestable (i.e. zfs_send_one() is wrapped,
but so is lzc_send_redacted() it calls to ‒ this wouldn't be strictly
necessary if ZFS_IOC_SEND_PROGRESS wasn't strictly denominational w.r.t.
the descriptor the send is happening on)
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Co-authored-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#11445Closes#13133
We do still check and only optionally set because all F_SETPIPE_SZ
directives reallocate the pipe buffer
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#13133
Notably, this also means that the pipe is expanded before each
dataset is received, so updates to /p/s/f/pipe-max-size are reflected
for each new dataset
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#13133
With the zfs_destroy ZTS test case the setup script needed to call
default_setup_noexit so compression could be turned off. Also, added
log_must to setting compression off in the reservation setup script for
turning off compression.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Brian Atkinson <batkinson@lanl.gov>
Closes#13173
The regular default_raidz_setup function in the ZFS test suite called
log_pass after creating the zpool. However, with compression now being
on by default 56fa4aa, there is no way to turn compression off in the
setup.ksh scripts when creating a raidz VDEV. The addition of the
function default_raidz_setup_noexit allows for a raidz VDEV to be
created, additional zfs property settings to be applied and for the
setup.ksh script itself to call log_pass.
With the addition of default_raidz_setup_noexit some stray log_pass
calls were removed from any setup.ksh scripts that call
default_raidz_setup.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Brian Atkinson <batkinson@lanl.gov>
Closes#13173
When unlinking multiple files from a pool at 100% capacity, it was
possible for ENOSPC to be returned after the first unlink. e.g.
rm -f /mnt/fs/test1.0.0 /mnt/fs/test1.1.0 /mnt/fs/test1.2.0
rm: cannot remove '/mnt/fs/test1.1.0': No space left on device
rm: cannot remove '/mnt/fs/test1.2.0': No space left on device
After waiting for the pending deferred frees from the first unlink to
be processed the remaining files can then be unlinked. This is caused
by the quota limit in dsl_dir_tempreserve_impl() being temporarily
decreased to the allocatable pool capacity less any deferred free
space.
This is resolved using the existing mechanism of returning ERESTART
when over quota as long as we know enough space will shortly be
available after processing the pending deferred frees.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#13172
ZFS allows to update and retrieve additional file level attributes for
FreeBSD. This commit allows additional file level attributes to be
updated and retrieved for Linux. These include the flags stored in the
upper half of z_pflags only.
Two new IOCTLs have been added for this purpose. ZFS_IOC_GETDOSFLAGS
can be used to retrieve the attributes, while ZFS_IOC_SETDOSFLAGS can
be used to update the attributes.
Attributes that are allowed to be updated include ZFS_IMMUTABLE,
ZFS_APPENDONLY, ZFS_NOUNLINK, ZFS_ARCHIVE, ZFS_NODUMP, ZFS_SYSTEM,
ZFS_HIDDEN, ZFS_READONLY, ZFS_REPARSE, ZFS_OFFLINE and ZFS_SPARSE.
Flags can be or'd together while calling ZFS_IOC_SETDOSFLAGS.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Umer Saleem <usaleem@ixsystems.com>
Closes#13118
aarch64 is a different architecture than arm. Some
compilers might choke when both __arm__ and __aarch64__
are defined.
This change separates the checks for arm and for
aarch64 in the isa_defs.h header files.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Windel Bouwman <windel@windel.nl>
Closes#10335Closes#13151
A function that returns with no value is a different thing from a
function that doesn't return at all. Those are two orthogonal
concepts, commonly confused.
pthread_create(3) expects a pointer to a start routine that has a
very precise prototype:
void *(*start_routine)(void *);
However, other thread functions, such as kernel ones, expect:
void (*start_routine)(void *);
Providing a different one is incorrect, and has only been working
because the ABIs happen to produce a compatible function.
We should use '_Noreturn void', since it's the natural type, and
then provide a '_Noreturn void *' wrapper for pthread functions.
For consistency, replace most cases of __NORETURN or
__attribute__((noreturn)) by _Noreturn. _Noreturn is understood
by -std=gnu89, so it should be safe to use everywhere.
Ref: https://github.com/openzfs/zfs/pull/13110#discussion_r808450136
Ref: https://software.codidact.com/posts/285972
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Co-authored-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
Closes#13120
Before this, /all/ numerical properties 1 (ZFS_PROP_CREATION,
ZPOOL_PROP_SIZE, VDEV_PROP_CAPACITY) would be non-fixed and
/all/ numerical properties 5 (ZFS_PROP_COMPRESSRATIO,
ZPOOL_PROP_HEALTH, VDEV_PROP_PSIZE) would be 8-wide
Realistically, this doesn't appear to be much of a problem
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#13125
07a3312f17, which introduced this in
October of 2014, didn't have zpool export -a available; we do
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Closes#13093
A simple change, but so many tests break with it,
and those are the majority of this.
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes#13078
Related to commit 90b77a036. Retry the `zpool export` if the pool
is "busy" indicating there is a process accessing the mount point.
This can happen after an import, allowing it to be retried will
avoid spurious test failures.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#13169
It's much nicer to import from upstream this way, and compiles
faster too.
Everything in lib/ is unmodified 1.4.5.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rich Ercolani <rincebrain@gmail.com>
Closes#12978
While "diff -r" is the most straightforward way of comparing directory
trees for differences, it has two major issues:
* File metadata is not compared, which means that subtle bugs may be
missed even if a test is written that exercises the buggy behaviour.
* diff(1) doesn't know how to compare special files -- it assumes they
are always different, which means that a test using diff(1) on
special files will always fail (resulting in such tests not being
added).
rsync can be used in a very similar manner to diff (with the -ni flags),
but has the additional benefit of being able to detect and resolve many
more differences between directory trees. In addition, rsync has a
standard set of features and flags while diffs feature set depends on
whether you're using GNU or BSD binutils.
Note that for several of the test cases we expect that file timestamps
will not match. For example, the ctime for a file creation or modify
event is stored in the intent log but not the mtime. Thus when replaying
the log the correct ctime is set but the current mtime is used. This is
the expected behavior, so to prevent these tests from failing, there's a
replay_directory_diff function which ignores those kinds of changes.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Closes#12588
As previously noted in #12272 the receive-o-x_props_override.ksh test
reliably fails on FreeBSD. Since we don't expect this test to pass
move the exception from the "maybe" to "known" section. This way we
don't retry the FAILED test when it is not expected to pass.
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#13167
On FreeBSD pools are not allowed to be created using vdevs which are
backed by ZFS volumes. This configuration is not recommended for any
supported platform, nevertheless the largest_pool_001_pos.ksh test
case makes use of it as a convenience. This causes the test case to
fail reliably on FreeBSD. The layout is still tolerated on Linux
so only perform this test on Linux.
Reviewed-by: Igor Kozhukhov <igor@dilos.org>
Reviewed by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#13166
PR 1711 (https://github.com/dracutdevs/dracut/pull/1711) adds a zfs_devs
function to dracut to detect the physical devices backing zfs pools. If
this function exists in the version of dracut this module is being
called from, then it does not need to run.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Signed-off-by: Savyasachee Jha <hi@savyasacheejha.com>
Closes#13121
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Amanakis <gamanakis@gmail.com>
Signed-off-by: Paul Dagnelie <pcd@delphix.com>
Closes#13154
The dsl_destroy_snapshot() call in ztest_objset_destroy_cb() may
encounter a runtime error when the pool is out of space. This is
similar to the error handling for the dsl_destroy_head() case,
but since dsl_destroy_snapshot() is implemented as a channel
program ECHRNG is returned instead of ENOSPC. ECHRNG may also
be returned instead of EBUSY if there is a hold on the snapshot.
Reviewed by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#13155
We recently had a case where our operators replaced a bad
multipathed disk, only to see it fail to autoreplace. The
zed logs showed that the multipath replacement disk did not pass
the 'is_dm' test in zfs_process_add() even though it should have.
is_dm is set if there exists a sysfs entry for to the
underlying /dev/sd* paths for the multipath disk. It's
possible this path didn't exist due to a race condition where
the sysfs paths weren't created at the time the udev event came
in to zed, but this was never verified.
This patch updates the check to look for udev properties that
indicate if the new autoreplace disk is an empty multipath disk,
rather than looking for the underlying sysfs entries. It also
adds in additional logging, and fixes a bug where zed allowed
you to use an already zfs-formatted disk from another pool
as a multipath auto-replacement disk.
Furthermore, while testing this patch, I also ran across a case
where a force-faulted disk did not have a ZPOOL_CONFIG_PHYS_PATH
entry in its config. This prevented it from being autoreplaced.
I added additional logic to derive the PHYS_PATH from the PATH if
the PATH was a /dev/disk/by-vdev/ path. For example, if PATH
was /dev/disk/by-vdev/L28, then PHYS_PATH would be L28. This is
safe since by-vdev paths represent physical locations and do not
change between boots.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#13023
Fix `zfs-dkms` installation on Debian-derived distributions by
aligning the directory detection logic to #13096.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: szubersk <szuberskidamian@gmail.com>
Closes#11449Closes#13141
- Kmemleak `clear` is invoked right before every test case run.
- Kmemleak `scan` is requested right after each test case is finished.
- Kmemleak instrumentation is not used for
setup/cleanup/pretest/posttest/failsafe stages to shorten the test
case execution time.
- Kmemleak periodic scan is disabled (`scan=0`) before the test suite
run to avoid interfering with the on-demand scan results.
- There are unavoidable potential false positives coming from kernel
areas other than OpenZFS module.
- The ZTS with kmemleak enabled duration is increased by ~50%.
Example run
```
Running Time: 07:12:13
Percent passed: 98.3%
unreferenced object 0xffff9da82aea5410 (size 80):
comm "kworker/u32:10", pid 942206, jiffies 4296749716 (age 2615.516s)
hex dump (first 32 bytes):
00 30 30 00 00 00 00 00 ff 8f 30 00 00 00 00 00 .00.......0.....
51 e6 77 05 a8 9d ff ff 00 00 00 00 00 00 00 00 Q.w.............
backtrace:
[<000000005cf1fea2>] alloc_extent_state+0x1d/0xb0 [btrfs]
[<0000000083f78ae5>] set_extent_bit+0x2ff/0x670 [btrfs]
[<00000000de29249e>] lock_extent_bits+0x6b/0xa0 [btrfs]
[<00000000b241f424>] lock_and_cleanup_extent_if_need+0xaf/0x1c0
[btrfs]
[<0000000093ca72b5>] btrfs_buffered_write+0x297/0x7d0 [btrfs]
[<000000002c2938c8>] btrfs_file_write_iter+0x127/0x390 [btrfs]
[<00000000b888f720>] do_iter_readv_writev+0x152/0x1b0
[<00000000320f0bcc>] do_iter_write+0x7c/0x1c0
[<000000000b5a8fe0>] lo_write_bvec+0x62/0x150 [loop]
[<000000009aa03c73>] loop_process_work+0x250/0xbd0 [loop]
[<00000000c7487d8a>] process_one_work+0x1f1/0x390
[<000000000b236831>] worker_thread+0x53/0x3e0
[<0000000023cb3e57>] kthread+0x127/0x150
[<000000002d48676a>] ret_from_fork+0x22/0x30
```
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: szubersk <szuberskidamian@gmail.com>
Closes#13084
Linux 5.11 changed kernel_fpu_begin() to an inlined function and
moved the functionality to kernel_fpu_begin_mask(). This breaks the
existing detection mechanism since it checks if kernel_fpu_begin is
an exported kernel symbol, which isn't the case for an inlined
function.
To avoid assumptions about internal implementation, replace
ZFS_LINUX_TEST_RESULT_SYMBOL in favor of ZFS_LINUX_TEST_RESULT
which already makes sure kernel_fpu_{begin,end}() is usable by us.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Attila Fülöp <attila@fueloep.org>
Closes#13147
As such, there are no specific synchronous semantics defined for
the xattrs. But for xattr=on, it does log to ZIL and zil_commit() is
done, if sync=always is set on dataset. This provides sync semantics
for xattr=on with sync=always set on dataset.
For the xattr=sa implementation, it doesn't log to ZIL, so, even with
sync=always, xattrs are not guaranteed to be synced before xattr call
returns to caller. So, xattr can be lost if system crash happens, before
txg carrying xattr transaction is synced.
This change adds xattr=sa logging to ZIL on xattr create/remove/update
and xattrs are synced to ZIL (zil_commit() done) for sync=always.
This makes xattr=sa behavior similar to xattr=on.
Implementation notes:
The actual logging is fairly straight-forward and does not warrant
additional explanation.
However, it has been 14 years since we last added new TX types
to the ZIL [1], hence this is the first time we do it after the
introduction of zpool features. Therefore, here is an overview of the
feature activation and deactivation workflow:
1. The feature must be enabled. Otherwise, we don't log the new
record type. This ensures compatibility with older software.
2. The feature is activated per-dataset, since the ZIL is per-dataset.
3. If the feature is enabled and dataset is not for zvol, any append to
the ZIL chain will activate the feature for the dataset. Likewise
for starting a new ZIL chain.
4. A dataset that doesn't have a ZIL chain has the feature deactivated.
We ensure (3) by activating on the first zil_commit() after the feature
was enabled. Since activating the features requires waiting for txg
sync, the first zil_commit() after enabling the feature will be slower
than usual. The downside is that this is really a conservative
approximation: even if we never append a 'TX_SETSAXATTR' to the ZIL
chain, we pay the penalty for feature activation. The upside is that the
user is in control of when we pay the penalty, i.e., upon enabling the
feature.
We ensure (4) by hooking into zil_sync(), where ZIL destroy actually
happens.
One more piece on feature activation, since it's spread across
multiple functions:
zil_commit()
zil_process_commit_list()
if lwb == NULL // first zil_commit since zil_open
zil_create()
if no log block pointer in ZIL header:
if feature enabled and not active:
// CASE 1
enable, COALESCE txg wait with dmu_tx that allocated the
log block
else // log block was allocated earlier than this zil_open
if feature enabled and not active:
// CASE 2
enable, EXPLICIT txg wait
else // already have an in-DRAM LWB
if feature enabled and not active:
// this happens when we enable the feature after zil_create
// CASE 3
enable, EXPLICIT txg wait
[1] da6c28aaf6
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Christian Schwarz <christian.schwarz@nutanix.com>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jitendra Patidar <jitendra.patidar@nutanix.com>
Closes#8768Closes#9078
Systemd units do not read @initconfdir@ but refer to variables defined
there, also a minor fixup in zfs-scrub service file.
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Damian Szuberski <szuberskidamian@gmail.com>
Signed-off-by: Krzysztof Piecuch <piecuch@kpiecuch.pl>
Closes#12946
