Re: [PATCH 12/14] generic/204: break out of fs-filling loop early if we ENOSPC

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Thu, Nov 02, 2017 at 10:23:48PM +0800, Eryu Guan wrote:
> On Wed, Nov 01, 2017 at 02:47:42PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> > 
> > If filling up the filesystem causes us to hit ENOSPC earlier than we
> > thought we would (the sizing estimates become less and less accurate as
> > we add more metadata) then just bail out -- we're checking that the fs
> > is robust enough to cut us off before we actually run out of space for
> > writing metadata and crash the fs.
> > 
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> >  tests/generic/204 |   11 +++++++++--
> >  1 file changed, 9 insertions(+), 2 deletions(-)
> > 
> > 
> > diff --git a/tests/generic/204 b/tests/generic/204
> > index 4c203a2..1e2c1e1 100755
> > --- a/tests/generic/204
> > +++ b/tests/generic/204
> > @@ -82,8 +82,15 @@ echo files $files, resvblks $resv_blks >> $seqres.full
> >  _scratch_resvblks $resv_blks >> $seqres.full 2>&1
> >  
> >  for i in `seq 1 $files`; do
> > -    echo -n > $SCRATCH_MNT/$i
> > -    echo XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX > $SCRATCH_MNT/$i
> > +	# Open/truncate file, write stuff.  If we run out of space early,
> > +	# we can bail out of the loop.
> > +	out="$($XFS_IO_PROG \
> > +		-c "open -f -t $SCRATCH_MNT/$i" \
> > +		-c 'close' \
> > +		-c "open -f -t $SCRATCH_MNT/$i" \
> > +		-c 'pwrite -q -S 0x58 0 36' 2>&1 | _filter_scratch)"
> > +	echo "${out}" | grep -q 'No space left on device' && break
> 
> This doesn't look correct to me. This test is meant to catch spurious
> ENOSPC, it's designed to be a "delayed allocation ENOSPC test"[1] by
> "writing lots of single block files" and catch early ENOSPC. IMHO, this
> change ignores ENOSPC, which defeats the test purpose. And I don't quite
> understand the purpose of truncate open/close/truncate open/write
> sequence..
> 
> Also this significantly increases the test time, 4s -> ~120s for me.

Hmm, you're right, this one /is/ trying to test ENOSPC.  It would seem
that a better fix for this test would be to set space= based on the
available blocks count instead of hardcoding 97920000.

--D

> 
> Thanks,
> Eryu
> 
> [1] commit 143368a047ea ("xfstests: add test 204, a simple delayed
> allocation ENOSPC test")
> 
> > +	test -n "${out}" && echo "${out}"
> >  done
> >  
> >  # success, all done
> > 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html