So df in btrfs is tricky at best, and relying on it for accurate information is
not great, but it's the best way to verify this test. To get around btrfs being
inconsistent sometimes just use _within_tolerance to check our new df value to
make sure that our truncate did something. With this patch I no longer see
transient failures of this test. Thanks,
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Reviewed-by: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
ls -l will show the nblocks for the directory, and this made it into the golden
output for 314. The problem is nblocks is 0 for btrfs directories because we're
awesome, which makes us fail this test. So filter out the "total blah" line of
ls -l so btrfs can pass this test too. Thanks,
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Introduce generic test 315 to verify if the disk space is
released after truncating a preallocated file back to the
old smaller size. Before Linux-3.10, Btrfs/OCFS2 test
failed in this case.
The test file is fallocated with FALLOC_FL_KEEP_SIZE option.
Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Test generic/192 fails if noatime is set
generic/192
-delta1 - access time after sleep in-core: 40
-delta2 - access time after sleep on-disk: 40
+delta1 - access time after sleep in-core: 0
+delta2 - access time after sleep on-disk: 0
but it's pointless to test atime effects with noatime.
Signed-off-by: David Sterba <dsterba@suse.cz>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Tests if subdirectories created on the filesystem will properly inherit sgid bit
when this is set on the parent directory, once the process has the properly
permissions to create a subdirectory, this, should inherit parent's sgid bit if
this is set and irix_sgid_inherit sysctl is disabled.
V2: add missing source of "attr" file for _require_acls
V3: use _ls_l to filter out the selinux "."
renumber to 314 to make the merge easier
V4: fix 314.out to the correct output
Thanks to Sandeen who have written this patch
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Regression test for commit:
3972f26 btrfs: update timestamps on truncate()
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
generic/193 runs the test in $here - the root of the xfstests source
tree/installation. IOWs, it doesn't test the filesystem on either
the TEST_DIR or SCRATCH_MNT, and so it not testing the filesystem
we think it is testing. Bad. Fixing this is the majority of the
change - introducing $test_root and $test_user for the files with
different owners, and then redirecting error output and filtering
the output appropriately.
And then add checks that truncate clears the suid/sgid bits
appropriately, something that has never been tested on XFS (and
likely other filesystems) so will cause kernels between 3.1 and 3.9
to assert fail as Dave Jones has recently reported.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
In some configurations (e.g. 1 KB block size), ext4 can decide it is
better to zero out several blocks rather than splitting unwritten
extent. This changes results SEEK_HOLE / SEEK_DATA returns and thus the
test fails. Fix the problem by disabling the feature for this test.
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The generic/286 test tests SEEK_HOLE and SEEK_DATA, and is reasonably
fast. We should just run the test by default.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Use the built-in _test_mount function from xfstests so it will use
the correct mount options for xfstests. The script used a simple
umount-and-mount sequence, which caused a test failure on an XFS
filesystem that used both realtime and external log devices.
Signed-off-by: Michael L. Semon <mlsemon35@gmail.com>
Reviewed-by: Eric Sandeen <sandeen@rehat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
xfs was having issues with generic/311 because of caching issues. Make
_check_scratch_fs take an optional argument to use as the device to fsck.
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Now in tests/ there are some test cases whose mode is 0644. But they
should be 0755. So fix it.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Reviewed-by: Dave Chinner <david@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
supports seek data/hole operation or not. Here _require_seek_data_hole
is defined to do this work.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
After finished test, temporarily fio config file should be removed.
This commit tries to fix this problem in the following test cases:
- generic/299-300
- ext4/301-304
- shared/305
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
generic/233 attempts to direct output to tee, but instead of using a
pipe it uses an append operator. Hence it leaves a file named "tee"
in the root directory of the xfstests execution path. Just direct
the output to the $seqres.full file rather than trying to tee it
into the test output as well.
Reported-by: "Michael L. Semon" <mlsemon35@gmail.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The -F flag to xfs_io originally enabled it to operate on non-xfs
filesystems. This restriction was removed upstream in favor of
gracefully failing on the handful of operations that actually
required xfs, and the option was deprecated.
However, xfstests is still used on distros with older xfsprogs, and
so "xfs_io -F" was necessary throughout xfstests.
Simplify this by appending -F to XFS_IO_PROG when it's needed -
i.e. if we're using old xfsprogs on a non-xfs filesystem.
This will eliminate errors when new tests leave out the -F, and
if and when -F is finally removed, there will be one central
location in xfstests to update.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
In one place of test 306, we mistakenly used /dev/null and /dev/zero
instead of equivalent devices created on tested filesystem. So we were
not really testing the functionality we intended.
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
--
This test sets up a dm flakey target and then runs my fsync tester I've been
using to verify btrfs's fsync() is working properly. It will create a dm flakey
device, mount it, run my test, make the flakey device start dropping writes, and
then unmount the fs. Then we mount it back up and make sure the md5sums match
and then run fsck on the device to make sure we got a consistent fs. I used the
output from a run on BTRFS since it's the only one that passes this test
properly. I verified each test manually to make sure they were in fact valid
files. XFS and Ext4 both fail this test in one way or another.
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
[rjohnston@sgi.com changed syncfs() to sync() for older kernels]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
This test sets up a dm flakey target and then runs my fsync tester I've been
using to verify btrfs's fsync() is working properly. It will create a dm flakey
device, mount it, run my test, make the flakey device start dropping writes, and
then unmount the fs. Then we mount it back up and make sure the md5sums match
and then run fsck on the device to make sure we got a consistent fs. I used the
output from a run on BTRFS since it's the only one that passes this test
properly. I verified each test manually to make sure they were in fact valid
files. XFS and Ext4 both fail this test in one way or another.
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
[rjohnston@sgi.com changed syncfs() to sync() for older kernels]
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
older xfs_io refused to write to /dev/null because it's
not a file on an xfs filesystem. So add -F.
While we're at it, add more testcases:
* symlink on a RO fs pointing to a file on a RW fs.
* bind-mounted rw file on an RO fs.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
generic/131 attempts to kill processes that may no longer exist when
the test finishes and has a broken redirect for the error messages
and they end up in a file named "1" in the xfstests root instead of
/dev/null.
Not only that, the attempts to redirect stderr to stdout in the
middle of the test use incorrect redirect syntax, so they create an
empty file named "1" in the xfstests root...
IOWs, all the redirects in the test are broken. Fix them and clean
up the failure case to use the exit trap to trigger the cleanup
function....
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
generic/232 attempts to direct output to tee, but instead of using a
pipe it uses an append operator. Hence it leaves a file named "tee"
in the root directory of the xfstests execution path. Just direct
the output to the $seqres.full file rather than trying to tee it
into the test output as well.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Josef Bacik
Jie Liu
David Sterba
Carlos Maiolino
Eryu Guan
Dave Chinner
Jan Kara
Theodore Ts'o
Michael L. Semon
Zheng Liu
Eric Sandeen
Rich Johnston