We had a regression where we were not copying csums properly when balancing a
prealloc extent. Unfortunately the way this showed up the most was with the
csum simply missing, which doesn't result in an error to userspace. So I've
copied what generic/310 does and check dmesg for csum errors when the test
starts and then compare that count to the csum errors after the test finishes to
see if there was a problem. This approach caught the error without my fix, and
then passed fine with my fix in place but with the previous errors still in
dmesg. Thanks,
[rjohnston: changed test number to 13]
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Reviewed-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
With coreutils v8.16 the style of apostrophes changed from `word' to
'word'. This is breaking some tests which use the older form.
This commit introduces function changes the golden output of the
affected tests and introduces a filter for the older style output.
[dchinner: modified to use a global filter in check rather than
per-test filters]
[rjohnston: minor comment change]
Signed-off-by: Tomas Racek <tracek@redhat.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
We were actually testing this improperly, there was a bug in the set default
code so we weren't actually honoring the 0 subvolid properly. To fix this we
need to get the subvolid for the subvol we want to set as the default and use
that in the command. With this patch we now pass again with the fix for the 0
subvolid. Thanks,
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Tested-by: David Sterba <dsterba@suse.cz>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
_scratch_mkfs_sized requires an integer number of bytes
as input; if it's given something else, catch it and _notrun.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Put the code for obtaining the list of test into one place which makes
things more readable. It will also allow us to re-init the list in the
future if we need it.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Currently when no tests or test groups are specified xfstests will
silently test nothing. Interestingly enough when test groups to exclude
are specified the rest of the tests will be run.
This commit changes that to run all possible tests (for a given file
system) when no specific tests has been specified. This matches the old
xfstests behaviour.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The show_ops() output should come as part of the -f option
help.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Josef noticed that using /dev/zero to generate most of the test
data doesn't work if someone overrides the mount options to
enable compression. The test that performs a cancellation failed
because the replace operation was already finished when the
cancel request was executed.
Since /dev/urandom is too slow to generate multiple GB, the
way how the filesystem data is generated is completely changed
with this patch. Now /dev/urandom is used to generate one 1MB
file and this file is copied up to 2048 times. /dev/zero is no
longer used.
The runtime of the test is about the same as before. Compression
works now, online duplication will again cause issues, but
we don't have online duplication today.
Reported-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Reviewed-by: Jan Schmidt <list.xfs@jan-o-sch.net>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Otherwise it fails with ENOSPC on CRC enabled filesystems because
of the larger inode size.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The XFS implementation of _scratch_mkfs_sized ignores MKFS_OPTIONS
when a custom block size is set and so isn't testing things like
CRCs on such sized filesytsems. Fix this by ensuring we don't try to
override the block size is it is set in MKFS_OPTIONS. xfs/204 shows
this problem.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Because if it corrupts the filesystem it currently goes undetected.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Turns out btrfs-convert broke on July 3, and lo! we
do not have a regression test, and now we have one,
and there was much rejoicing.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The current code incorrectly gets block size information from $TEST_DEV
instead of from $TEST_DIR. This returns the block size of the filesystem
hosting the device file rather than that of the filesystem on $TEST_DEV.
Signed-off-by: chandan <chandan@linux.vnet.ibm.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The btrfs-progs tools changed the output:
- 100GiB instead of 100GB
xfstest btrfs/006 is one that failed due to this change.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
This test performs btrfs device replace tests with all possible profiles
(single/dup/mixed/raid0/raid1/raid10), one round with the '-r' option
to 'btrfs replace start' and one round without this option. The
cancelation is tested only once and with the dup/single profile for
metadata/data.
This test takes 181 seconds on my SSD equiped test box and 237s on
spinning disks. Almost all the time is spent when the filesystem is
populated with test data. The replace operation itself takes less than
a second for all the tests, except for the test that is marked as
'thorough' which will run for about 8 seconds on my test box.
The amount of tests done depends on the number of devices in the
SCRATCH_DEV_POOL. For full test coverage, at least 5 devices should
be available (e.g. 5 partitions). With less than 2 entries in
SCRATCH_DEV_POOL, the test is not executed.
The source and target devices for the replace operation are arbitrarily
chosen out of SCRATCH_DEV_POOl. Since the target device mustn't be
smaller than the source device, the requirement for this test is that
all devices have _exactly_ the same size. If this is not the case, the
test terminates with _notrun.
To check the filesystems after replacing a device, a scrub run is
performed, a btrfsck run, and finally the filesystem is remounted.
This commit depends on my other commit:
"xfstest: don't remove the two first devices from SCRATCH_DEV_POOL"
[rjohnston: renumbered to btrfs/011]
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
One problem was the output of "uniq -c" which added spaces depending
on the size of the count value (e.g. one space less for 10+ devices).
The second problem was that "btrfs fi show" was doing the same:
"devid %4llu size %s used %s path %s".
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Since common/config is executed twice, if SCRATCH_DEV_POOL is configured
via the environment, the current code removes the first device entry twice
which means that you lose the second device for the test.
The fix is to not remove anything from SCRATCH_DEV_POOL anymore.
That used to be done (I can only guess) to allow to pass the
SCRATCH_DEV_POOL as an argument to _scratch_mkfs. Since _scratch_mkfs adds
the SCRATCH_DEV, the pool mustn't contain that device anymore.
A new function _scratch_pool_mkfs is introduced that does the expected
thing.
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Add a test for testing all 3 (user, group and project) quotas together.
This is a modified version of xfstest 050.
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
This is to test whether snapshot-aware defrag can work well on partial extents.
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
This test failed for me with output from 'btrfs balance':
QA output created by 003
+Done, had to relocate 4 out of 4 chunks
+Done, had to relocate 5 out of 5 chunks
Silence is golden
Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Josef Bacik
Eric Sandeen
Tomas Racek
Eric Sandeen
Lukas Czerner
Stefan Behrens
Dave Chinner
chandan
Chandra Seetharaman
Liu Bo
Dwight Engen