Re: [PATCH] tests/xfs: test log recovery metadata LSN ordering

[Brian Foster <bfoster@redhat.com>]

On Mon, Aug 15, 2016 at 01:29:33PM +0800, Eryu Guan wrote:
> On Fri, Aug 12, 2016 at 08:13:30AM -0400, Brian Foster wrote:
> > XFS had a bug that lead to a possible out-of-order log recovery
> > situation (e.g., replay a stale modification from the log over more
> > recent metadata in destination buffer). This resulted in false
> > corruption reports during log recovery and thus mount failure.
> > 
> > This condition is caused by system crash or filesystem shutdown shortly
> > after a successful log recovery. Add a test to run a combined workload,
> > fs shutdown and log recovery loop known to reproduce the problem on
> > affected kernels.
> > 
> > Signed-off-by: Brian Foster <bfoster@redhat.com>
> > ---
> > 
> > This test reproduces the problem described and addressed in the
> > following patchset:
> > 
> >   http://oss.sgi.com/pipermail/xfs/2016-August/050840.html
> > 
> > It runs anywhere from 50-100s in the couple of environments I've tested
> > on so far and reproduces the problem for me with 100% reliability. Note
> > that the bug only affects crc=1 kernels.
> 
> Looks good overall, and tested with the above patchset applied and test
> passed without problems. Some minor issues inline
> 
> > 
> > Brian
> > 
> >  tests/xfs/999     | 87 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  tests/xfs/999.out |  2 ++
> >  tests/xfs/group   |  1 +
> >  3 files changed, 90 insertions(+)
> >  create mode 100755 tests/xfs/999
> >  create mode 100644 tests/xfs/999.out
> > 
> > diff --git a/tests/xfs/999 b/tests/xfs/999
> > new file mode 100755
> > index 0000000..f9dd7f7
> > --- /dev/null
> > +++ b/tests/xfs/999
> > @@ -0,0 +1,87 @@
> > +#! /bin/bash
> > +# FS QA Test No. 999
> > +#
> > +# Test XFS log recovery ordering on v5 superblock filesystems. XFS had a problem
> > +# where it would incorrectly replay older modifications from the log over more
> > +# recent versions of metadata due to failure to update metadata LSNs during log
> > +# recovery. This could result in false positive reports of corruption during log
> > +# recovery and permanent mount failure.
> > +#
> > +# To test this situation, run frequent shutdowns immediately after log recovery.
> > +# Ensure that log recovery does not recover stale modifications and cause
> > +# spurious corruption reports and/or mount failures.
> > +#
> > +#-----------------------------------------------------------------------
> > +# Copyright (c) 2016 Red Hat, Inc.  All Rights Reserved.
> > +#
> > +# This program is free software; you can redistribute it and/or
> > +# modify it under the terms of the GNU General Public License as
> > +# published by the Free Software Foundation.
> > +#
> > +# This program is distributed in the hope that it would be useful,
> > +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> > +# GNU General Public License for more details.
> > +#
> > +# You should have received a copy of the GNU General Public License
> > +# along with this program; if not, write the Free Software Foundation,
> > +# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
> > +#-----------------------------------------------------------------------
> > +#
> > +
> > +seq=`basename $0`
> > +seqres=$RESULT_DIR/$seq
> > +echo "QA output created by $seq"
> > +
> > +here=`pwd`
> > +tmp=/tmp/$$
> > +status=1	# failure is the default!
> > +trap "_cleanup; exit \$status" 0 1 2 3 15
> > +
> > +_cleanup()
> > +{
> > +	cd /
> > +	rm -f $tmp.*
> > +	killall -9 fsstress > /dev/null 2>&1
> 
> We need a '_require_command "$KILLALL_PROGA" killall' and use
> $KILLALL_PROG in the test.
> 

Ok.

> > +	_scratch_unmount > /dev/null 2>&1
> > +}
> > +
> > +# get standard environment, filters and checks
> > +. ./common/rc
> > +
> > +# Modify as appropriate.
> > +_supported_fs xfs
> 
> I'm wondering if this test can be made generic by adding a
> "_require_scratch_shutdown"? Like generic/042 to generic/051
> 

Hmm, probably. I'll give it a try, thanks!

Brian

> Thanks,
> Eryu