On Fri, Jun 14, 2024 at 01:44:30PM -0700, Darrick J. Wong wrote:
> On Fri, Jun 14, 2024 at 07:27:59PM +0100, Theodore Ts'o wrote:
> > On Thu, Jun 13, 2024 at 09:16:18PM -0700, Darrick J. Wong wrote:
> > > 
> > > Amusingly enough, I still have that patch (and generic/1220) in my
> > > fstests branch, and I haven't seen this problem happen on g/1220 in
> > > quite a while.
> > 
> > Remind me what your fstests git repo is again?
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=djwong-wtf
> 
> > The generic/750 test in my patch 2/2 that I sent out reproduces the
> > problem super-reliably, so long as fsstress actually issues the
> > io_uring reads and writes.  So if you have your patch applied which
> > suppresses io_uring from fstress by default, you might need to modify
> > the patch series to force the io_uring, at which point it quite nicely
> > demonstrates the fsstress ; umount problem.  (It sometimes requires
> > more rounds of fsstress ; umounts before it repro's on the xfs/4k, but
> > it repro's really nicely on ext4/4k).
> 
> Hm, your g/750 test mounts and unmounts in a loop, which might be why
> mine hasn't tripped yet.  I'll try applying it and report back.

Have you had chance to try applying this?  There has since been an
upstream generic/750, so I've renumbered this to be generic/760 in my
local.  It's still failing on my latest kernel, although it does seem
to be timing related.  Using kvm-xfstests, the xfs/4k configuration
wasn't failing if I enabled KASAN, but on my default install-kconfig
kernel configuration, the now-renumbered generic/760 fails for both
ext4/4k and xfs/4k (at least most of the time):

 [22:25:55]- output mismatch (see /results/xfs/results-4k/generic/760.out.bad)
    --- tests/generic/760.out   2024-07-11 10:44:31.000000000 -0400
    +++ /results/xfs/results-4k/generic/760.out.bad     2024-07-11 22:25:55.031883734 -0400
    @@ -1,2 +1,3 @@
     QA output created by 760
    -Silence is golden
    +umount: /vdc: target is busy.
    +failed to umount
    ...
    (Run 'diff -u /root/xfstests/tests/generic/760.out /results/xfs/results-4k/generic/760.out.bad'  to see the entire diff)

xfs/4k: 1 tests, 1 failures, 72 seconds
  generic/760  Failed   1s
  generic/760  Failed   3s
  generic/760  Pass     33s
  generic/760  Failed   3s
  generic/760  Pass     32s
Totals: 5 tests, 0 skipped, 3 failures, 0 errors, 72s

	  	   	      		  - Ted