Re: [PATCH 2/2] xfs/288: skip repair -n when checking empty root leaf block behavior

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

On Thu, Jul 07, 2022 at 08:25:25PM +0800, Zorro Lang wrote:
> On Tue, Jul 05, 2022 at 03:02:34PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > Update this test to reflect the (once again) corrected behavior of the
> > xattr leaf block verifiers.
> > 
> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > ---
> >  tests/xfs/288 |   32 +++++++++++++-------------------
> >  1 file changed, 13 insertions(+), 19 deletions(-)
> > 
> > 
> > diff --git a/tests/xfs/288 b/tests/xfs/288
> > index e3d230e9..aa664a26 100755
> > --- a/tests/xfs/288
> > +++ b/tests/xfs/288
> > @@ -8,7 +8,7 @@
> >  # that leaf directly (as xfsprogs commit f714016).
> >  #
> >  . ./common/preamble
> > -_begin_fstest auto quick repair fuzzers
> > +_begin_fstest auto quick repair fuzzers attr
> >  
> >  # Import common functions.
> >  . ./common/filter
> > @@ -50,25 +50,19 @@ if [ "$count" != "0" ]; then
> >  	_notrun "xfs_db can't set attr hdr.count to 0"
> >  fi
> >  
> > -# make sure xfs_repair can find above corruption. If it can't, that
> > -# means we need to fix this bug on current xfs_repair
> > -_scratch_xfs_repair -n >> $seqres.full 2>&1
> 
> So we drop the `xfs_repair -n` test.

Yep.

> Will the latest xfs_repair fail or pass on that? I'm wondering what's the expect
> result of `xfs_repair -n` on a xfs with empty leaf? Should it report errors,
> or nothing wrong?

xfs_repair -n no longer fails on attr block 0 being an empty leaf block
since those are part of the ondisk format and are not a corruption.

xfs_repair (without the -n) will clear the attr fork since there aren't
any xattrs if attr block 0 is empty.

--D

> Thanks,
> Zorro
> 
> > -if [ $? -eq 0 ];then
> > -	_fail "xfs_repair can't find the corruption"
> > -else
> > -	# If xfs_repair can find this corruption, then this repair
> > -	# should junk above leaf attribute and fix this XFS.
> > -	_scratch_xfs_repair >> $seqres.full 2>&1
> > +# Check that xfs_repair discards the attr fork if block 0 is an empty leaf
> > +# block.  Empty leaf blocks at the start of the xattr data can be a byproduct
> > +# of a shutdown race, and hence are not a corruption.
> > +_scratch_xfs_repair >> $seqres.full 2>&1
> >  
> > -	# Old xfs_repair maybe find and fix this corruption by
> > -	# reset the first used heap value and the usedbytes cnt
> > -	# in ablock 0. That's not what we want. So check if
> > -	# xfs_repair has junked the whole ablock 0 by xfs_db.
> > -	_scratch_xfs_db -x -c "inode $inum" -c "ablock 0" | \
> > -		grep -q "no attribute data"
> > -	if [ $? -ne 0 ]; then
> > -		_fail "xfs_repair didn't junk the empty attr leaf"
> > -	fi
> > +# Old xfs_repair maybe find and fix this corruption by
> > +# reset the first used heap value and the usedbytes cnt
> > +# in ablock 0. That's not what we want. So check if
> > +# xfs_repair has junked the whole ablock 0 by xfs_db.
> > +_scratch_xfs_db -x -c "inode $inum" -c "ablock 0" | \
> > +	grep -q "no attribute data"
> > +if [ $? -ne 0 ]; then
> > +	_fail "xfs_repair didn't junk the empty attr leaf"
> >  fi
> >  
> >  echo "Silence is golden"
> > 
>