So this test does lots of fallocate/truncate noise while doing aio
overwrites to try and exercise a deadlock found in ext4. Because it
runs so hard with ENOSPC it can sometimes cause truncate to fail on
btrfs. This is ok and doesn't affect the validity of the test, we
just need to catch the output so it doesn't cause the test to fail.
Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
The 'testfile' environment variable is initialized before the
xfstests environment is included into generic/313. TEST_DIR is not
defined at this point and causes the test to operate on the root.
Move the testfile initialization down after the general environment
is sourced.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
In changing the default log sizes in mkfs, the freespace
calculations in generic/204 are no longer valid and so it fails with
ENOSPC before ti has finished creating the necessary files.. Make
the test use a fixed log size of 5MB for XFS so that freespace
calculations remain valid and the test passes regardless of whether
we have a new or old mkfs binary.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Otherwise we end up with an ever-growing file for every test that is
run and that makes it hard to isolate failures.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Generic/314 can fail when the group write file mode bit for "subdir" does not
match that found in the golden output, as has been seen in ext4 regression
testing. It appears that the golden output for generic/314 was taken on a
system where the $qa_user's umask cleared that mode bit - most likely, where
the umask was 022. Depending upon the distro, it's not uncommon for a user's
default umask to have a different value, such as 002. When that's the case,
we get a false negative failure when the group write mode bit for "subdir" is
not cleared. This failure is unrelated to the value of the SGID mode bit
that is the object of this test.
We could either require that $qa_user's account be configured in advance with
a umask of 022, or explicitly set a umask value compatible with the golden
output when creating "subdir". The latter option is more robust.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
generic/240
Date: Wed, 11 Dec 2013 11:21:28 -0000
From: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
If we execute generic/240 on a fs which has its fs block size greater
than 64k (for example, NFS), this test will fail with:
io_submit failed: Invalid argument
This will happen because in src/aio-dio-regress/aiodio_sparse2.c this
expression
num_aio = filesize / step;
will set num_aio to 0 and this means that no io_prep_write() will happen
before calling io_submit().
Fixing filesize to be 8 * "fs block size".
Signed-off-by: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
While running xfstest 280, we occasionally got such error:
setquota: Cannot set quota for user 0 from kernel on
/dev/mapper/xfstests-disk1: No such device
setquota: Cannot write quota for 0 on /dev/mapper/xfstests-disk1: No such
device
setquota calls syscall quotactl, and the kernel will wait for the filesystem
to unfreeze and then performs command. Then kernel will double check if the
device is still mounted. If not, an ENODEV will be thrown.
While in the testcase, unfreeze and umount might be so close that the device
got umounted before quotactl is performed.
Reported-by: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
Signed-off-by: Guangyu Sun <guangyu.sun@oracle.com>
Reviewed-by: Eric Sandeen <sandeen@redaht.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Btrfs was screwing up rename+fsync, add some regression tests for
the various scenarios it was screwing up.
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Btrfs had some issues with fsync()'ing directories and fsync()'ing
after renames. These three new tests cover the 3 different issues
we were seeing. This breaks out the dmflakey stuff into a common
helper to be shared between generic/311 and this new test.
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
generic/273 factors the "space available" output from df into the
calculation for the size of the origin data set. Recent commit
bfdd1e72b3 xfstests: added -P option to $DF_PROG
... converted the use of 'df' to $DF_PROG. This implicitly adds the
-T parameter to add the fs type column, shifts the available space
column over by one and unintentionally causes 273 to look at "used
space" and create too small of a data set for a useful test.
Realign to the available space value.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
For historical reasons beyond my knowledge xfstests tries to abuse the
scratch device as test device for nfs and udf. Because not all test
have inherited the right usage of the _setup_testdir and _cleanup_testdir
helpers this leads to lots of unessecary test failures.
Remove the special casing, which gets nfs down to a minimal number of
failures.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Sugned-off-by: Dave Chinner <david@fromorbit.com>
This test is based on generic/273, a regression test for commit
9a3a5da xfs: check for stale inode before acquiring iflock on push
On unpatched kernel, rm processes would hang.
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Various tests opencode checks to find out the minimum support direct I/O
size. Replace those with a generic helper that handles network filesystems as
well. Also remove the Linux 2.4 workaround we had in once place.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Added -P option to $DF_PROG and changed the invocation of
'df' command in generic/{251,260,273,275} testcases
with $DF_PROG.
Otherwise the testcases will fail if the scratch
device has a long name (for example, if it's an LVM volume).
Because df outputs its usage stats with two lines:
/dev/mapper/xfstests-disk1
3030800 4608 2868908 1% /tmp/mnt/disk1
Signed-off-by: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Due to partially committed series (fd080d64b6)
generic/273 test uses '_no_of_online_cpus' function which is not defined.
Now it's safe to switch it to 'src/feature -o'.
Signed-off-by: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
Reviewed-by: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
On Wed, 30 Oct 2013 09:24:41 -0700
Christoph Hellwig <hch@infradead.org> wrote:
> On Wed, Oct 30, 2013 at 09:19:55AM -0700, Christoph Hellwig wrote:
> > On Mon, Oct 28, 2013 at 11:43:28AM -0400, Dwight Engen wrote:
> > > Hi Cristoph, on my system (where fsgqa is id 501) the one liner
> > > the test is running is:
> > >
> > > # ./src/nsexec -s -U -M "0 501 1000" -G "0 501 1000" ./src/lstat64
> > > Usage: lstat64 [-t] filename ...
> >
> > The id here is 1000 and the following works just fine:
> >
> > /src/nsexec -s -U -M "0 1000 1000" -G "0 1000 1000" ./src/lstat64
> > Usage: lstat64 [-t] filename ...
>
> But:
>
> ./src/nsexec -s -U -M "0 1000 1000" -G "0 501
> 1000" /root/xfstests/src/lstat64 execvp: Permission denied
>
>
> Which is probably due to:
> root@vm:~/xfstests# ls -ld ~
> drwx------ 6 root root 4096 Oct 30 16:24 /root
>
>
> Guess we need a relative path here?
Yep, that makes sense. I modeled this on 219 which was using
$here/src/lstat64 but didn't think about the fact that in my test fsgqa
might have traversal problems. I see plenty of other tests are using
relative paths so the following patch should (hopefully) fix 317 for you.
Thanks for tracking it down.
Signed-off-by: Dwight Engen <dwight.engen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
generic/317 and 318 need /proc/<pid>/[uid_map|gid_map], test fail on
older kernels without that support.
Add a _require_ugid_map() function and called by 317 and 318.
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
The content of /proc/cpuinfo file is platform-dependent.
So we can not use it reliably to check a number of available cpus.
It would be better to use sysfs interface, as _no_of_online_cpus() does.
Signed-off-by: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Btrfs will default to mixed block groups for 1 gigabyte file systems and
smaller, which means data and metadata share the same area. This makes
generic/274 fail for us because we cannot reserve enough metadata space to do
our writes. Bumping the scratch fs up to 2 gigabytes allows us to do our normal
metadata/data separation and allows us to pass this test. Thanks,
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
This test is motivated by an issue found by a btrfs user, addressed
and described by the following Linux kernel patch:
https://patchwork.kernel.org/patch/3046931/
The steps to reproduce the issue on btrfs are the following:
$ mkfs.btrfs -f /dev/loop0
$ mount /dev/loop0 /mnt
$ mkdir /mnt/acl
$ setfacl -d --set u::rwx,g::rwx,o::- /mnt/acl
$ getfacl /mnt/acl
user::rwx
group::rwx
other::r-x
default:user::rwx
default:group::rwx
default:other::---
$ mkdir /mnt/acl/dir1
$ getfacl /mnt/acl/dir1
user::rwx
group::rwx
other::---
After unmounting and mounting again the filesystem, getfacl returned the
expected default ACL for the subdirectory:
$ umount /mnt/acl
$ mount /dev/loop0 /mnt
$ getfacl /mnt/acl/dir1
user::rwx
group::rwx
other::---
default:user::rwx
default:group::rwx
default:other::---
This means that the underlying ACL xattr was persisted correctly but
the in memory representation of the inode had (incorrectly) a NULL ACL.
[rjohnston: renumbered test to 319]
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
test 294 is using the scratch device w/o mkfs-ing it first,
this runs the risk of following a test which completely
fills the fs, causing 294 to fail.
add "rm -f $seqres.full" as well, it was growing on every run.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Starting from util-linux v2.23 fstrim(1) reports trimmed bytes
differently, e.g.
new fstrim: /mnt/ext4: 9.7 GiB (10411118592 bytes) trimmed
old fstrim: /mnt/ext4: 10411118592 bytes were trimmed
generic/260 reports syntax error
+./tests/generic/260: line 111: [: 9.7: integer expression expected
+./tests/generic/260: line 121: [: 9.7: integer expression expected
+./tests/generic/260: line 183: [: 9.7: integer expression expected
Add a new filter called _filter_fstrim in common/filter and get the
correct trimmed bytes in generic/260, so the test passes with both old
and new fstrim.
Signed-off-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Josef Bacik
Brian Foster
Dave Chinner
Dave Chinner
Lukas Czerner
Eric Whitney
Stanislav Kholmanskikh
Guangyu Sun
Josef Bacik
Christoph Hellwig
Eryu Guan
Dwight Engen
Filipe David Borba Manana
Eric Sandeen