On Wed, Sep 24, 2014 at 02:47:05PM -0400, Brian Foster wrote: > XFS allocates extra indirect blocks for delayed allocation extents at > write time. When delalloc extents are split, the existing indirect block > reservation was historically divided up evenly among the new extents > even though the overall requirement for two extents could exceed the > requirement for the original. Repeated delalloc extent splits ultimately > leads to extents with 0 indirect blocks and in turn leads to assert > failures in XFS. > > Add a test to stress indirect block reservation for delayed allocation > extents. The test converts a single delalloc extent to many and operates > on the remaining extents to detect or trigger potential problems. > > Signed-off-by: Brian Foster <bfoster@xxxxxxxxxx> > --- > > Here's a simple reproducer for the indirect block reservation problem > called out here: > > http://oss.sgi.com/archives/xfs/2014-09/msg00337.html > > It reproduces the assert failures described therein: > > XFS: Assertion failed: startblockval(del.br_startblock) > 0, file: fs/xfs/libxfs/xfs_bmap.c, line: 5281 > > Note that this test also unintentionally fails on XFS. The test file > ends up zero-sized after the remount and thus hexdump doesn't produce > any output. This doesn't occur on ext4, I suspect due to the fact that > the range being zeroed is flushed beforehand, though I could be wrong > about that. Tested with ext4 and xfs, and ext4 passes/xfs fails the test as described here. Reviewed-by: Eryu Guan <eguan@xxxxxxxxxx> With one nitpick below.. > > In any event, this calls out a separate bug in XFS where if appending > data is chucked from cache by zero range before written back (eof is > page aligned), we lose the on-disk inode size update and the inode size > changes unexpectedly across the remount (assuming nothing else changes > the size, of course). > > Brian > > tests/generic/033 | 88 +++++++++++++++++++++++++++++++++++++++++++++++++++ > tests/generic/033.out | 4 +++ > tests/generic/group | 1 + > 3 files changed, 93 insertions(+) > create mode 100755 tests/generic/033 > create mode 100644 tests/generic/033.out > > diff --git a/tests/generic/033 b/tests/generic/033 > new file mode 100755 > index 0000000..41198b7 > --- /dev/null > +++ b/tests/generic/033 > @@ -0,0 +1,88 @@ > +#! /bin/bash > +# FS QA Test No. 033 > +# > +# This test stresses indirect block reservation for delayed allocation extents. > +# XFS reserves extra blocks for deferred allocation of delalloc extents. These > +# reserved blocks can be divided among more extents than anticipated if the > +# original extent for which the blocks were reserved is split into multiple > +# delalloc extents. If this scenario repeats, eventually some extents are left > +# without any indirect block reservation whatsoever. This leads to assert > +# failures and possibly other problems in XFS. > +# > +#----------------------------------------------------------------------- > +# Copyright (c) 2014 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.* > +} > + > +# get standard environment, filters and checks > +. ./common/rc > +. ./common/filter > + > +# real QA test starts here > +rm -f $seqres.full > + > +# Modify as appropriate. > +_supported_fs generic > +_supported_os Linux > +_require_scratch > +_require_xfs_io_command "fzero" > + > +_scratch_mkfs >/dev/null 2>&1 > +_scratch_mount > + > +file=$SCRATCH_MNT/file.$seq > +bytes=$((64 * 1024)) > + > +# create sequential delayed allocation > +$XFS_IO_PROG -f -c "pwrite 0 $bytes" $file | _filter_xfs_io \ > + >> $seqres.full 2>&1 The output of xfs_io is redirected to $seqres.full, so it's not necessary to be filtered, for debug purpose. And the following two xfs_io calls. Thanks, Eryu > + > +# Zero every other 4k range to split the larger delalloc extent into many more > +# smaller extents. Use zero instead of hole punch because the former does not > +# force writeback (and hence delalloc conversion). It can simply discard > +# delalloc blocks and convert the ranges to unwritten. > +endoff=$((bytes - 4096)) > +for i in $(seq 0 8192 $endoff); do > + $XFS_IO_PROG -c "fzero -k $i 4k" $file | _filter_xfs_io \ > + >> $seqres.full 2>&1 > +done > + > +# now zero the opposite set to remove remaining delalloc extents > +for i in $(seq 4096 8192 $endoff); do > + $XFS_IO_PROG -c "fzero -k $i 4k" $file | _filter_xfs_io \ > + >> $seqres.full 2>&1 > +done > + > +_scratch_remount > +hexdump $file > + > +status=0 > +exit > diff --git a/tests/generic/033.out b/tests/generic/033.out > new file mode 100644 > index 0000000..419d831 > --- /dev/null > +++ b/tests/generic/033.out > @@ -0,0 +1,4 @@ > +QA output created by 033 > +0000000 0000 0000 0000 0000 0000 0000 0000 0000 > +* > +0010000 > diff --git a/tests/generic/group b/tests/generic/group > index 8e0c22a..1227408 100644 > --- a/tests/generic/group > +++ b/tests/generic/group > @@ -32,6 +32,7 @@ > 027 auto enospc > 028 auto quick > 032 auto quick rw > +033 auto quick rw > 053 acl repair auto quick > 062 attr udf auto quick > 068 other auto freeze dangerous stress > -- > 1.8.3.1 > > _______________________________________________ > xfs mailing list > xfs@xxxxxxxxxxx > http://oss.sgi.com/mailman/listinfo/xfs _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
