generic/033: add xfs delalloc indirect block depletion reproducer
PATCH 1/1 generic/033: add xfs delalloc indirect block depletion reproducer
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.
---
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.
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
tests/generic/033 new file +88 -0
@@ -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 + +# 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
tests/generic/033.out new file +4 -0
@@ -0,0 +1,4 @@ +QA output created by 033 +0000000 0000 0000 0000 0000 0000 0000 0000 0000 +* +0010000
tests/generic/group +1 -0
@@ -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
Signature · 1 line
1.8.3.1
Quoted 22 lines · Brian Foster
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 <[email protected]>---
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:
Quoted 7 lines
XFS: Assertion failed:
startblockval(del.br_startblock)> 0, file:fs/xfs/libxfs/xfs_bmap.c, line: 5281Note 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.
With one nitpick below..
Quoted 78 lines
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.outtests/generic/033 new file +56
-0@@ -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 $seqres.full + +# Modify as appropriate. +_supported_fs generic +_supported_os Linux +_require_scratch +_require_xfs_io_command "fzero"
tests/generic/033 line 66
Quoted 10 lines
tests/generic/033 new file +10
-0+ +_scratch_mkfs >/dev/null 2>&1 +_scratch_mount + +file=$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 $seqres.full, so it's not necessary to be filtered, for debug purpose.
And the following two xfs_io calls.
Thanks,
Eryu
Quoted 51 lines
tests/generic/033 new file +22
-0+ +# Zero every other 4k range to split the larger delalloc extent into many more +# smaller extents. Use zero instead of hole punch because the former does not +# force writeback (and hence delalloc conversion). It can simply discard +# delalloc blocks and convert the ranges to unwritten. +endoff=$((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 +exittests/generic/033.out new file +4
-0@@ -0,0 +1,4 @@ +QA output created by 033 +0000000 0000 0000 0000 0000 0000 0000 0000 0000 +* +0010000tests/generic/group +1
-0@@ -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 stressSignature · 6 lines
1.8.3.1 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Quoted 130 lines · Eryu Guan
Quoted 22 lines · Brian Foster
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 <[email protected]>---
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:
Quoted 7 lines
XFS: Assertion failed:
startblockval(del.br_startblock)> 0, file:fs/xfs/libxfs/xfs_bmap.c, line: 5281Note 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 <[email protected]>With one nitpick below..
Quoted 78 lines
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.outtests/generic/033 new file +56
-0@@ -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 $seqres.full + +# Modify as appropriate. +_supported_fs generic +_supported_os Linux +_require_scratch +_require_xfs_io_command "fzero"tests/generic/033 line 66
Quoted 10 lines
tests/generic/033 new file +10
-0+ +_scratch_mkfs >/dev/null 2>&1 +_scratch_mount + +file=$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>&1The output of
xfs_iois redirected to $seqres.full, so it's not necessary to be filtered, for debug purpose.
Quoted 2 lines
And the following two
xfs_iocalls.
Indeed, I'll post v2. Thanks for the review.
Brian
Quoted 59 lines
Thanks,
EryuQuoted 51 lines
tests/generic/033 new file +22
-0+ +# Zero every other 4k range to split the larger delalloc extent into many more +# smaller extents. Use zero instead of hole punch because the former does not +# force writeback (and hence delalloc conversion). It can simply discard +# delalloc blocks and convert the ranges to unwritten. +endoff=$((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 +exittests/generic/033.out new file +4
-0@@ -0,0 +1,4 @@ +QA output created by 033 +0000000 0000 0000 0000 0000 0000 0000 0000 0000 +* +0010000tests/generic/group +1
-0@@ -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 stressSignature · 6 lines
1.8.3.1 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs