Stress btrfs' block group allocation and deallocation while running fstrim in parallel. Part of the goal is also to get data block groups deallocated so that new metadata block groups, using the same physical device space ranges, get allocated while fstrim is running. This caused several issues ranging from invalid memory accesses, kernel crashes, metadata or data corruption, free space cache inconsistencies, free space leaks and memory leaks.
Signed-off-by: Filipe Manana <fdman...@suse.com> --- V2: Addressed Dave's comments. tests/generic/038 | 152 ++++++++++++++++++++++++++++++++++++++++++++++++++ tests/generic/038.out | 2 + tests/generic/group | 1 + 3 files changed, 155 insertions(+) create mode 100755 tests/generic/038 create mode 100644 tests/generic/038.out diff --git a/tests/generic/038 b/tests/generic/038 new file mode 100755 index 0000000..217aa7a --- /dev/null +++ b/tests/generic/038 @@ -0,0 +1,152 @@ +#! /bin/bash +# FSQA Test No. 038 +# +# This test was motivated by btrfs issues, but it's generic enough as it +# doesn't use any btrfs specific features. +# +# Stress btrfs' block group allocation and deallocation while running fstrim in +# parallel. Part of the goal is also to get data block groups deallocated so +# that new metadata block groups, using the same physical device space ranges, +# get allocated while fstrim is running. This caused several issues ranging +# from invalid memory accesses, kernel crashes, metadata or data corruption, +# free space cache inconsistencies, free space leaks and memory leaks. +# +# These issues were fixed by the following btrfs linux kernel patches: +# +# Btrfs: fix invalid block group rbtree access after bg is removed +# Btrfs: fix crash caused by block group removal +# Btrfs: fix freeing used extents after removing empty block group +# Btrfs: fix race between fs trimming and block group remove/allocation +# Btrfs: fix race between writing free space cache and trimming +# Btrfs: make btrfs_abort_transaction consider existence of new block groups +# Btrfs: fix memory leak after block remove + trimming +# Btrfs: fix extent map leak on chunk allocation failure +# +# The issues were found on a qemu/kvm guest with 4 virtual CPUs, 4Gb of ram and +# scsi-hd devices with discard support enabled (that means hole punching in the +# disk's image file is performed by the host). +# +#----------------------------------------------------------------------- +# +# Copyright (C) 2014 SUSE Linux Products GmbH. All Rights Reserved. +# Author: Filipe Manana <fdman...@suse.com> +# +# 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" + +tmp=/tmp/$$ +status=1 # failure is the default! +trap "_cleanup; exit \$status" 0 1 2 3 15 + +_cleanup() +{ + rm -fr $tmp +} + +# get standard environment, filters and checks +. ./common/rc +. ./common/filter + +# real QA test starts here +_need_to_be_root +_supported_fs btrfs +_supported_os Linux +_require_scratch +_require_fstrim + +rm -f $seqres.full + +# Keep allocating and deallocating 1G of data space with the goal of creating +# and deleting 1 block group constantly. The intention is to race with the +# fstrim loop below. +fallocate_loop() +{ + local name=$1 + while true; do + $XFS_IO_PROG -f -c "falloc -k 0 1G" \ + $SCRATCH_MNT/$name &> /dev/null + sleep 3 + $XFS_IO_PROG -c "truncate 0" \ + $SCRATCH_MNT/$name &> /dev/null + sleep 3 + done +} + +trim_loop() +{ + while true; do + $FSTRIM_PROG $SCRATCH_MNT + done +} + +# Create a bunch of small files that get their single extent inlined in the +# btree, so that we consume a lot of metadata space and get a chance of a +# data block group getting deleted and reused for metadata later. Sometimes +# the creation of all these files succeeds other times we get ENOSPC failures +# at some point - this depends on how fast the btrfs' cleaner kthread is +# notified about empty block groups, how fast it deletes them and how fast +# the fallocate calls happen. So we don't really care if they all succeed or +# not, the goal is just to keep metadata space usage growing while data block +# groups are deleted. +create_files() +{ + local prefix=$1 + + for ((i = 1; i <= 400000; i++)); do + $XFS_IO_PROG -f -c "pwrite -S 0xaa 0 3900" \ + $SCRATCH_MNT/"${prefix}_$i" &> /dev/null + if [ $? -ne 0 ]; then + echo "Failed creating file ${prefix}_$i" >>$seqres.full + break + fi + done + +} + +_scratch_mkfs >>$seqres.full 2>&1 +_require_fs_space $SCRATCH_MNT $((10 * 1024 * 1024)) +_scratch_mount + +for ((i = 0; i < $((4 * $LOAD_FACTOR)); i++)); do + trim_loop & + trim_pids[$i]=$! +done + +for ((i = 0; i < $((1 * $LOAD_FACTOR)); i++)); do + fallocate_loop "falloc_file_$i" & + fallocate_pids[$i]=$! +done + +create_files "foobar" + +kill ${fallocate_pids[@]} +kill ${trim_pids[@]} +wait + +# Sleep a bit, otherwise umount fails often with EBUSY (TODO: investigate why). +sleep 3 + +# The fstests framework will now check for fs consistency with fsck. +# The trimming was racy and caused some btree nodes to get full of zeroes on +# disk, which obviously caused fs metadata corruption. The race often lead +# to missing free space entries in a block group's free space cache too. + +echo "Silence is golden" +status=0 +exit diff --git a/tests/generic/038.out b/tests/generic/038.out new file mode 100644 index 0000000..5e0f13e --- /dev/null +++ b/tests/generic/038.out @@ -0,0 +1,2 @@ +QA output created by 038 +Silence is golden diff --git a/tests/generic/group b/tests/generic/group index 9f355fc..1e89848 100644 --- a/tests/generic/group +++ b/tests/generic/group @@ -40,6 +40,7 @@ 035 auto quick 036 auto aio rw stress 037 metadata auto quick +038 auto stress 053 acl repair auto quick 062 attr udf auto quick 068 other auto freeze dangerous stress -- 2.1.3 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html