[LTP] TMPDIR minimal size requirement

[LTP] TMPDIR minimal size requirement

[Petr Vorel]

Hi all,

recent LTP change 3626124a4 ("fallocate06: Increase test loop device size to
1GB") [1] increased requirement on TMPDIR size for syscalls. That hits the size
of tmpfs we use in openSUSE Tumbleweed on /tmp. While we can workaround easily
this in openQA framework we use for LTP testing (use TMPDIR=/var/tmp for
fallocate06 or even for whole runtest syscalls), I wonder if we should have some
LTP upstream fix.

What comes to mi mind, we could use as TMPDIR default /var/tmp (only if it does
not exists, we would fallback to /tmp). We could also warn against tmpfs use.
Or at least document on README that although many tests need 300 MB for a block
device, some tests require more and suggest to use /var/tmp.

Also, Cyril asked for a patch with minimal device size per filesystem (IMHO
still would be done). At least for Btrfs and XFS some people complains 300
MB is not a real testing scenario, therefore if we implement it and use 1 GB
e.g. for Btrfs and XFS we will hit limits on "low resource" devices.

Kind regards,
Petr

[1] https://github.com/linux-test-project/ltp/commit/3626124a42adfe536af2abff63213fa1ccc63795

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] TMPDIR minimal size requirement

[Petr Vorel]

> Hi all,

> recent LTP change 3626124a4 ("fallocate06: Increase test loop device size to
> 1GB") [1] increased requirement on TMPDIR size for syscalls. That hits the size
> of tmpfs we use in openSUSE Tumbleweed on /tmp. While we can workaround easily
> this in openQA framework we use for LTP testing (use TMPDIR=/var/tmp for
> fallocate06 or even for whole runtest syscalls), I wonder if we should have some
> LTP upstream fix.

FYI I see the problem also on fallocate05 on bcachefs, although here it might be
a bug in fallocate05 or in bcachefs, because the test somehow works with ENOTSUP
(there is a code before which checks on ENOTSUP). And I was not able to
reproduce it on similar VM outside of our openQA test framework, it could be our
bug in openQA runner as well.

Kind regards,
Petr

fallocate05.c:136: TPASS: write()
tst_test.c:1701: TINFO: === Testing on bcachefs ===
tst_test.c:1117: TINFO: Formatting /dev/loop0 with bcachefs opts='' extra opts=''
tst_test.c:1131: TINFO: Mounting /dev/loop0 to /tmp/LTP_falBnQfVp/mntpoint fstyp=bcachefs flags=0
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file0 size 21710183
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file1 size 8070086
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file2 size 3971177
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file3 size 36915315
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file4 size 70310993
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file5 size 4807935
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file6 size 90739786
tst_fill_fs.c:36: TINFO: Creating file mntpoint/file7 size 76896492
tst_fill_fs.c:63: TINFO: write(): ENOSPC (28)
fallocate05.c:81: TPASS: write() wrote 131072 bytes
fallocate05.c:102: TINFO: fallocate()d 0 extra blocks on full FS
fallocate05.c:114: TPASS: fallocate() on full FS
fallocate05.c:130: TPASS: fallocate(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)
fallocate05.c:134: TFAIL: write(): ENOSPC (28)

> What comes to mi mind, we could use as TMPDIR default /var/tmp (only if it does
> not exists, we would fallback to /tmp). We could also warn against tmpfs use.
> Or at least document on README that although many tests need 300 MB for a block
> device, some tests require more and suggest to use /var/tmp.

> Also, Cyril asked for a patch with minimal device size per filesystem (IMHO
> still would be done). At least for Btrfs and XFS some people complains 300
> MB is not a real testing scenario, therefore if we implement it and use 1 GB
> e.g. for Btrfs and XFS we will hit limits on "low resource" devices.

> Kind regards,
> Petr

> [1] https://github.com/linux-test-project/ltp/commit/3626124a42adfe536af2abff63213fa1ccc63795

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp