Age | Commit message (Collapse) | Author |
|
When running btrfs/287 with features that create extra trees or don't
the need to create some trees, such as when using the free space tree
(default for several btrfs-progs releases now) versus when not using
it (by passing -R ^free-space-tree in MKFS_OPTIONS), the test can fail
because the IDs for the two snapshots it creates changes, and the golden
output is requiring the numeric IDs of the snapshots.
For example, when disabling the free space tree, the test fails like this:
$ MKFS_OPTIONS="-R ^free-space-tree" ./check btrfs/287
FSTYP -- btrfs
PLATFORM -- Linux/x86_64 debian0 6.6.0-rc2-btrfs-next-138+ #1 SMP PREEMPT_DYNAMIC Thu Sep 21 17:58:48 WEST 2023
MKFS_OPTIONS -- -R ^free-space-tree /dev/sdc
MOUNT_OPTIONS -- /dev/sdc /home/fdmanana/btrfs-tests/scratch_1
btrfs/287 1s ... - output mismatch (see /home/fdmanana/git/hub/xfstests/results//btrfs/287.out.bad)
--- tests/btrfs/287.out 2023-09-22 12:39:43.060761389 +0100
+++ /home/fdmanana/git/hub/xfstests/results//btrfs/287.out.bad 2023-09-22 12:40:54.238849251 +0100
@@ -44,52 +44,52 @@
Create a readonly snapshot of 'SCRATCH_MNT' in 'SCRATCH_MNT/snap1'
Create a readonly snapshot of 'SCRATCH_MNT' in 'SCRATCH_MNT/snap2'
resolve first extent:
-inode 257 offset 16777216 root 257
-inode 257 offset 8388608 root 257
-inode 257 offset 16777216 root 256
-inode 257 offset 8388608 root 256
...
(Run 'diff -u /home/fdmanana/git/hub/xfstests/tests/btrfs/287.out /home/fdmanana/git/hub/xfstests/results//btrfs/287.out.bad' to see the entire diff)
HINT: You _MAY_ be missing kernel fix:
0cad8f14d70c btrfs: fix backref walking not returning all inode refs
Ran: btrfs/287
Failures: btrfs/287
Failed 1 of 1 tests
So add a filter to logical reserve calls to replace snapshot root IDs with
a logical name (snap1 and snap2).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Disseldorp <ddiss@suse.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Avoid using the shortcut "sub" for the "subvolume" command, as this is the
standard practice because such shortcuts are not guaranteed to exist in
every btrfs-progs release (they may come and go). Also make the variables
local.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Disseldorp <ddiss@suse.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
It prints "File extent layout before defrag" for the both outputs, but the
latter one should be "after defrag".
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Test that if names are added to a directory after an opendir(3) call and
before a rewinddir(3) call, future readdir(3) calls will return the names.
This is mandated by POSIX:
https://pubs.opengroup.org/onlinepubs/007904875/functions/rewinddir.html
This exercises a regression in btrfs which is fixed by a kernel patch that
has the following subject:
""btrfs: refresh dir last index during a rewinddir(3) call""
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: David Disseldorp <ddiss@suse.de>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Running btrfs/239 on a zoned device often fails with the following error.
btrfs/239 5s ... - output mismatch (see /host/btrfs/239.out.bad)
--- tests/btrfs/239.out 2023-09-21 16:56:37.735204924 +0900
+++ /host/btrfs/239.out.bad 2023-09-21 18:22:45.401433408 +0900
@@ -1,4 +1,6 @@
QA output created by 239
+/testdir/dira still exists
+/dira does not exists
File SCRATCH_MNT/testdir/file1 data:
0000000 ab ab ab ab ab ab ab ab ab ab ab ab ab ab ab ab
*
...
This happens because "testdir" and "dira" are not logged on the first fsync
(fsync $SCRATCH_MNT/testdir), but are written as a full commit. That
prevents updating the log on "mv" time, leaving them pre-mv state.
The full commit is induced by the creation of a new block group. On the
zoned mode, we use a dedicated block group for tree-log. That block group
is created on-demand or assigned to a metadata block group if there is
none. On the first mount of a file system, we need to create one because
there is only one metadata block group available for the regular
metadata. That creation of a new block group forces tree-log to be a full
commit on that transaction, which prevents logging "testdir" and "dira".
Fix the issue by calling fsync before the first "sync", which creates the
dedicated block group and let the files be properly logged.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
A few tests were still using the older style of mentioning the fix in the
comment section. This patch migrates them to using
_fixed_by_kernel_commit.
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Test for a regression in copy up of symlink that has the S_NOATIME
inode flag.
This is a regression from v5.15 reported by Ruiwen Zhao:
https://lore.kernel.org/linux-unionfs/CAKd=y5Hpg7J2gxrFT02F94o=FM9QvGp=kcH1Grctx8HzFYvpiA@mail.gmail.com/
In the bug report, the symlink has the S_NOATIME inode flag because it is
on an NFS/FUSE filesystem that sets S_NOATIME for all inodes.
The reproducer uses another technique to create a symlink with
S_NOATIME inode flag by using chattr +A inheritance on filesystems
that inherit chattr flags to symlinks.
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The tests btrfs/288, btrfs/289 and btrfs/300 are using the "xxxx..." stub
for commit ids, as when they were submitted/merged the corresponding
btrfs patches were not yet in Linus' tree. So replace the stubs with the
commit ids.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The fix commit is written in the comment without a commit hash. Use
_fixed_by_kernel_commit command to describe it.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Running btrfs/076 on a zoned null_blk device will fail with the following error.
- output mismatch (see /host/results/btrfs/076.out.bad)
--- tests/btrfs/076.out 2021-02-05 01:44:20.000000000 +0000
+++ /host/results/btrfs/076.out.bad 2023-09-15 01:49:36.000000000 +0000
@@ -1,3 +1,3 @@
QA output created by 076
-80
-80
+83
+83
...
This is because the default value of zone_append_max_bytes is 127.5 KB
which is smaller than BTRFS_MAX_UNCOMPRESSED (128K). So, the extent size is
limited to 126976 (= ROUND_DOWN(127.5K, 4096)), which makes the number of
extents larger, and fails the test.
Instead of hard-coding the number of extents, we can calculate it using the
max extent size of an extent. It is limited by either
BTRFS_MAX_UNCOMPRESSED or zone_append_max_bytes.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
I wanted to verify tests using the command "btrfs scrub start" and
found that there are many more test cases using "btrfs scrub start"
than what is listed in the group.list file. So, get them to the scrub
group.
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Currently, _get_max_file_size finds max file size on $TEST_DIR.
The tests/generic/692 uses this function to detect file size and
then tries to create a file on $SCRATCH_MNT.
This works fine when test and scratch filesystems have the same
block size. However, it will fail if they differ.
Make _get_max_file_size accept mount point on which to detect max
file size.
Signed-off-by: Andrey Albershteyn <aalbersh@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Some test cases lack executable permission ('x'). Before running each
test case, `./check` checks and grants them 'x' permission. However,
this always leads to a dirty git repo. And the absence of 'x' permission
in test cases is often overlooked during reviews.
Since maintainers use mvtest to assign new case, add this change for
convenience of maintainers.
Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
As we are obliterating the need for the device scan for the single device,
which will return success if the basic superblock verification passes,
even for the duplicate device of the mounted filesystem, drop the check
for the return code in this testcase and continue to verify if the device
path of the mounted filesystem remains unaltered after the scan.
Also, if the test fails, it leaves the local non-standard mount point
remained mounted, leading to further test cases failing. Call unmount
in _cleanup().
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Boris Burkov <boris@bur.io>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Test overlayfs fsid behavior with new mount options uuid=null/on
that were introduced in kernel v6.6:
- Test inherited upper fs fsid with mount option uuid=off/null
- Test uuid=null behavior for existing overlayfs by default
- Test persistent unique fsid with mount option uuid=on
- Test uuid=on behavior for new overlayfs by default
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Make sure that log recovery will not succeed if there are unknown
rocompat features in the superblock and the log is dirty.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Make sure we can actually read files off the ro mounted filesystem that
has an unknown rocompat feature set.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
We don't need scrub status; it is okay to ignore the warnings due to
the readonly /var/lib/btrfs if any. Redirect stderr to seqres.full.
We check the scrub return status.
+WARNING: failed to open the progress status socket at /var/lib/btrfs/scrub.progress.42fad803-d505-48f4-a04d-612dbf8bd724: Read-only file system. Progress cannot be queried
+WARNING: failed to write the progress status file: Read-only file system. Status recording disabled
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
There are two repair functions: _repair_scratch_fs() and
_repair_test_fs(). As the names suggest, these functions are designed to
repair the filesystems SCRATCH_DEV and TEST_DEV, respectively. However,
these functions never called proper comamnd for the filesystem type btrfs.
This patch fixes it. Thx.
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The write invalidation code in iomap can only be triggered for writes
that span multiple folios. If the kernel reports a huge page size,
scale up the write size.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
This function does not follow the naming convention that common helpers
must start with an underscore. Fix this.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
This helper has two parts -- querying the value, and _notrun'ing the
test if huge pages aren't turned on. Break these into the usual
_require_hugepages and _get_hugepagesize predicates so that we can adapt
xfs/559 to large folios being used for writes.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
I run fstests in a readonly container, and accidentally uninstalled the
btrfsprogs package. When I did, this test started faililng:
--- btrfs/282.out
+++ btrfs/282.out.bad
@@ -1,3 +1,7 @@
QA output created by 282
wrote 2147483648/2147483648 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+WARNING: cannot create scrub data file, mkdir /var/lib/btrfs failed: Read-only file system. Status recording disabled
+WARNING: failed to open the progress status socket at /var/lib/btrfs/scrub.progress.3e1cf8c6-8f8f-4b51-982c-d6783b8b8825: No such file or directory. Progress cannot be queried
+WARNING: cannot create scrub data file, mkdir /var/lib/btrfs failed: Read-only file system. Status recording disabled
+WARNING: failed to open the progress status socket at /var/lib/btrfs/scrub.progress.3e1cf8c6-8f8f-4b51-982c-d6783b8b8825: No such file or directory. Progress cannot be queried
Skip the test if /var/lib/btrfs isn't writable, or if /var/lib isn't
writable, which means we cannot create /var/lib/btrfs.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
This test is unreliable on NFS. It fails consistently when run vs. a
server exporting btrfs, but passes when the server exports xfs. Since we
don't have any sort of attribute that we can require to test this, just
skip this one on NFS.
Also, subsume the check for btrfs into the _supported_fs check, and add
a comment for it.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
NFS doesn't keep track of whether a file is reflinked or not, so it
doesn't prevent this behavior. It shouldn't be a problem for NFS anyway,
so just skip this test there.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
When creating a new dentry (of any type), NFS will optimize away any
on-the-wire lookups prior to the create since that means an extra
round trip to the server. Because of that, it consistently fails this
test.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
There are several generic tests that require "setcap", but don't check
whether the underlying fs supports security attrs. Add the appropriate
checks.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
This test requires FIEMAP support.
Suggested-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
_require_acl tests whether you're able to fetch the ACL from a file
using chacl, and then tests for an -EOPNOTSUPP error return.
Unfortunately, filesystems that don't support them (like NFSv4) just
return -ENODATA when someone calls getxattr for the POSIX ACL, so the
test doesn't work.
Fix the test to have chacl set an ACL on the file instead, which should
reliably fail on filesystems that don't support them.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Now that I've finally gotten liburing installed on my test machine, I
can actually test io_uring. Adapt these two tests to support
SOAK_DURATION so I can add it to that too.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Ritesh Harjani reported that mount and unmount can race with the xfs cpu
hotplug notifier hooks and crash the kernel, which isfixed by:
https://lore.kernel.org/linux-xfs/ZO6J4W9msOixUk05@dread.disaster.area/T/#t
Extend this test to include that.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Make this test controllable via SOAK_DURATION, for anyone who wants to
perform a long soak test of filesystem vs. cpu hotplug.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Since commit 3687fcb0752a ("btrfs: zoned: make auto-reclaim less
aggressive"), the reclaim process won't run unless the 75% (by default) of
the filesystem volume is allocated as block groups. As a result, btrfs/237
won't success when it is run with a large volume.
To run the reclaim process, we need to either fill the FS to the desired
level, or make a small FS so that the test write can go over the level.
Since the current test code expects the FS has only one data block group,
filling the FS is both cumbersome and need effort to rewrite the test code.
So, we take the latter method. We create a small (16 * zone size) FS. The
size is chosen to hold a minimal FS with DUP metadata setup.
However, creating a small FS is not enough. With SINGLE metadata setup, we
allocate 3 zones (one for each DATA, METADATA and SYSTEM), which is less
than 75% of 16 zones. We can tweak the threshold to 51% on regular btrfs
kernel config (!CONFIG_BTRFS_DEBUG), but that is still not enough to start
the reclaim process. So, this test requires CONFIG_BTRFS_DEBUG to set the
threshold to 1%.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
xfs_io pwrite issues a series of block size writes, but there is no
guarantee that the resulting extent(s) will be singular or contiguous.
This behavior is acceptable, but the test is flawed in that it expects
a single extent for a pwrite.
Modify test to limit pwrite and reflink to a single block.
Signed-off-by: Bill O'Donnell <bodonnel@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Dave Chinner reported that xfs/273 fails if the AG size happens to be an
exact power of two. I traced this to an agbno integer overflow when the
current GETFSMAP call is a continuation of a previous GETFSMAP call, and
the last record returned was non-shareable space at the end of an AG.
This is the regression test for that bug.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
When the AIO program failed, it is better to bail out the test to keep the
failed state intact.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
When short read or corruption happened, it is difficult to locate which IO
event failed. Print the address to make it identifiable.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The async write IOs can return some errors, which may lead to a short read or
corruption in io_verify() stage. Catch an error early to identify the root
cause easily.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
I remember this case fails on last year becuase of
kernel commit cae2de69 ("iomap: Add async buffered write support")
kernel commit 1aa91d9 ("xfs: Add async buffered write support").
as below:
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took seconds
So For async buffered write requests, the request will return -EAGAIN
if the ilock cannot be obtained immediately.
Here also a discussion[1] that seems generic/471 has been broken.
Now, I met this problem in my linux distribution, then I found the above
discussion. IMO, remove this case is ok and then we can avoid to meet this
false report again.
[Additional information from Dave Chinner]
We changed how timestamps are updated so that they are aware of
IOCB_NOWAIT. If the IOCB_NOWIAT DIO write now needs to update the
inode timestamps, it will return -EAGAIN instead of doing
potentially blocking operations that require IO to complete (i.e.
taking a transaction reservation).
Hence the first time we go to do a DIO read an inode, it's going to
do an atime update, which now occurrs from an IOCB_NOWAIT context
and we return -EAGAIN....
Yes, we added non-blocking timestamp updates as part of the async
buffered write support, but this was a general XFS IO path change of
behaviour to address a potential blocking point in *all* IOCB_NOWAIT
reads and writes, buffered or direct.
The test is not validating that RWF_NOWAIT is behaving correctly - it
just was a simple operation that kinda exercised RWF_NOWAIT semantics
when we had no other way to test this code. It has outlived it's
original purpose, so it should be removed...
[1]https://lore.kernel.org/linux-xfs/b2865bd6-2346-8f4d-168b-17f06bbedbed@kernel.dk/
Signed-off-by: Yang Xu <xuyang2018.jy@fujitsu.com>
Reviewed-by: Bill O'Donnell <bodonnel@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
[BUG]
There is a very low chance to hit data csum mismatch (caught by scrub)
during test case btrfs/06[234567].
After some extra digging, it turns out that plain fsstress itself is
enough to cause the problem:
```
workload()
{
mkfs.btrfs -f -m single -d single --csum sha256 $dev1 > /dev/null
mount $dev1 $mnt
#$fsstress -p 10 -n 1000 -w -d $mnt
umount $mnt
btrfs check --check-data-csum $dev1 || fail
}
runtime=1024
for (( i = 0; i < $runtime; i++ )); do
echo "=== $i / $runtime ==="
workload
done
```
Inside a VM which has only 6 cores, above script can trigger with 1/20
possibility.
[CAUSE]
Locally I got a much smaller workload to reproduce:
$fsstress -p 7 -n 50 -s 1691396493 -w -d $mnt -v > /tmp/fsstress
With extra kernel trace_prinkt() on the buffered/direct writes.
It turns out that the following direct write is always the cause:
btrfs_do_write_iter: r/i=5/283 buffered fileoff=708608(709121) len=12288(7712)
btrfs_do_write_iter: r/i=5/283 direct fileoff=8192(8192) len=73728(73728) <<<<<
btrfs_do_write_iter: r/i=5/283 direct fileoff=589824(589824) len=16384(16384)
With the involved byte number, it's easy to pin down the fsstress
opeartion:
0/31: writev d0/f3[285 2 0 0 296 1457078] [709121,8,964] 0
0/32: chown d0/f2 308134/1763236 0
0/33: do_aio_rw - xfsctl(XFS_IOC_DIOINFO) d0/f2[285 2 308134 1763236 320 1457078] return 25, fallback to stat()
0/33: awrite - io_getevents failed -4 <<<<
0/34: dwrite - xfsctl(XFS_IOC_DIOINFO) d0/f3[285 2 308134 1763236 320 1457078] return 25, fallback to stat()
Note the 0/33, when the data csum mismatch triggered, it always fail
with -4 (-EINTR).
It looks like with lucky enough concurrency, we can get to the following
situation inside fsstress:
Process A | Process B
-----------------------------------+---------------------------------------
do_aio_rw() |
|- io_sumit(); |
|- io_get_events(); |
| Returned -EINTR, but IO hasn't |
| finished. |
`- free(buf); | malloc();
| Got the same memory of @buf from
| thread A.
| Modify the memory
| Now the buffer is changed while
| still under IO
This is the typical buffer modification during direct IO, which is going
to cause csum mismatch for btrfs, and btrfs properly detects it.
This is the direct cause of the problem.
The root cause is that, io_uring would use signals to handle
submission/completion of IOs.
Thus io_uring operations would interrupt AIO operations, thus causing
the above problem.
[FIX]
To fix the problem, we can just retry io_getevents() so that we can
properly wait for the IO.
This prevents us to modify the IO buffer before writeback really
finishes.
With this fixes, I can no longer reproduce the data corruption.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The "sort -R" is slower than "shuf" even with the full output because
"sort -R" actually sort them to group the identical keys.
$ time bash -c "seq 1000000 | shuf >/dev/null"
bash -c "seq 1000000 | shuf >/dev/null" 0.18s user 0.03s system 104% cpu 0.196 total
$ time bash -c "seq 1000000 | sort -R >/dev/null"
bash -c "seq 1000000 | sort -R >/dev/null" 19.61s user 0.03s system 99% cpu 19.739 total
Since the "find"'s outputs never be identical, we can just use "shuf" to
optimize the selection.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Use _random_file() helper to choose a random file in a directory.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
Currently, we use "ls ... | sort -R | head -n1" (or tail) to choose a
random file in a directory.It sorts the files with "ls", sort it randomly
and pick the first line, which wastes the "ls" sort.
Also, using "sort -R | head -n1" is inefficient. For example, in a
directory with 1000000 files, it takes more than 15 seconds to pick a file.
$ time bash -c "ls -U | sort -R | head -n 1 >/dev/null"
bash -c "ls -U | sort -R | head -n 1 >/dev/null" 15.38s user 0.14s system 99% cpu 15.536 total
$ time bash -c "ls -U | shuf -n 1 >/dev/null"
bash -c "ls -U | shuf -n 1 >/dev/null" 0.30s user 0.12s system 138% cpu 0.306 total
So, we should just use "ls -U" and "shuf -n 1" to choose a random file.
Introduce _random_file() helper to do it properly.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
btrfs supports creating nesting subvolumes however snapshots are not
recurive. When a snapshot is taken of a volume which contains a subvolume
the subvolume is replaced with a stub subvolume which has the same name and
uses inode number 2. This test validates that the stub volume copies
permissions of the original volume.
Signed-off-by: Lee Trager <lee@trager.us>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The mount option "discard=async" is not meant to be used on the zoned mode.
Skip it from the test.
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The '-f' parameter is fsck.ext# specific, where it's documented to:
Force checking even if filesystem is marked clean
_repair_test_fs() is only called on _check_test_fs() failure, so
dropping the parameter should be possible without changing ext#
behaviour.
Doing so fixes _repair_test_fs() on exfat, where fsck.exfat doesn't
support '-f'.
Signed-off-by: David Disseldorp <ddiss@suse.de>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The truncate command in those tests use an uninitialized variable i.
in kdevops, i must contain some leftover, so we get errors like:
/data/fstests-install/xfstests/tests/generic/298: line 45: /dev/loop12):
syntax error: operand expected (error token is "/dev/loop12)")
Apparently, noone including the author of the tests knows why this
truncate command is in the test, so remove the wrong truncate command.
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
commit 60054d51 ("check: fix excluded tests are only expunged in the
first iteration") change to use exclude_tests array instead of file.
The check if a test is in expunge file was using grep -q $TEST_ID FILE
so it was checking if the test was a non-exact match to one of the
lines, for a common example: "generic/001 # exclude this test" would be
a match to test generic/001.
The commit regressed this example, because the new code checks for exact
match of [ "generic/001" == "generic/001 " ]. Change the code to match a
regular expression to deal with this case and any other suffix correctly.
NOTE that the original code would have matched test generic/100 with lines
like "generic/1000" when we get to 4 digit seqnum, so the regular
expression does an exact match to the first word of the line.
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
1. Add missing options and wrap the cli example line.
2. Cleanup and also add missing "-K" operation for options description
part.
Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Zorro Lang <zlang@redhat.com>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|
|
The locker was waiting for sem_otime on sem0 to became non-zero after
incrementing sem0 himself. So sem_otime was never 0 at the time of
checking it, so the check was redundant/wrong.
This patch:
- moves the increment of sem1 to the lock-tester site
- lock-setter waits for that sem1 event, for which this patch replaces
the wait loop on sem_otime with GETVAL loop, adding a small sleep
- increment of sem0 to 2 moved past that sem1 event. That sem0 event
is currently not used/waited.
This guarantees that the lock-setter is working only after lock-getter
is fully initialized.
CC: fstests@vger.kernel.org
CC: Murphy Zhou <xzhou@redhat.com>
CC: Jeff Layton <jlayton@kernel.org>
CC: Zorro Lang <zlang@redhat.com>
Signed-off-by: Stas Sergeev <stsp2@yandex.ru>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Zorro Lang <zlang@kernel.org>
|