Re: [PATCH] generic/805: mask unmount failures on cleanup

["Darrick J. Wong" <djwong@kernel.org>]

On Wed, Oct 14, 2026 at 10:59:45PM -0700, Christoph Hellwig wrote:
> On Wed, Oct 14, 2026 at 08:32:36PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > On (what's left of) my NFS-booted QA fleet I get test "failures" like
> > this:
> > 
> >  --- a/generic/805.out	2026-09-22 13:33:18.326820071 -0700
> >  +++ b/generic/805.out.bad	2026-10-14 18:58:34.438866321 -0700
> >  @@ -1,2 +1,4 @@
> >   QA output created by 805
> >   Silence is golden
> >  +umount.nfs: remote share not in 'host:dir' format
> >  +umount.nfs: /mnt/s: not mounted
> > 
> > This is caused by the _unmount call in _cleanup getting cranky about the
> > scratch mount not being mounted despite the -q flag.  I think this is
> > because umount(8) detects the underlying fs as NFS and delegates
> > entirely to umount.nfs, and that program whines.
> > 
> > So fix the test by redirecting the output.  We've already decided if the
> > test passes or not before we get to cleanup.
> 
> We really should not end up here with NFS.
> 
> Oh, this test is missing a
> 
> _require_block_device $SCRATCH_DEV
> 
> so that all the dm magic is skipped when on a non-bdev fs.

_require_dm_target already checks that scratch-dev is a bdev, but I
think you've misunderstood me.

The test vm boots off NFS to test XFS, so / and /mnt are NFS.
SCRATCH_DEV is XFS, so when SCRATCH_MNT is mounted, /mnt/s will be XFS.
After unmounting SCRATCH_MNT, /mnt/s will look like it's NFS, so that's
why umount does what it does:

# mkfs.xfs -f /dev/sda
# mount /dev/sda /mnt/s
# umount /mnt/s
# umount /mnt/s
umount.nfs: remote share not in 'host:dir' format
umount.nfs: /mnt/s: not mounted

(I'm not fstesting NFS, I left that to Chuck and Anna.)

--D