On Tue, Feb 16, 2021 at 8:45 PM Josef Bacik <josef@xxxxxxxxxxxxxx> wrote:
>
> This is a regression test for a problem where we would flip read only if
> we reflink'ed enough extents to generate key'ed references, and then got
> a hash collision with those references. This is a test for the fix
>
> btrfs: do not error out if the extent ref hash doesn't match
>
> and is relatively straightforward, simply generate such a file and
> watch for fireworks.
>
> Signed-off-by: Josef Bacik <josef@xxxxxxxxxxxxxx>
> ---
> tests/btrfs/231 | 81 +++++++++++++++++++++++++++++++++++++++++++++
> tests/btrfs/231.out | 11 ++++++
> tests/btrfs/group | 1 +
> 3 files changed, 93 insertions(+)
> create mode 100755 tests/btrfs/231
> create mode 100644 tests/btrfs/231.out
>
> diff --git a/tests/btrfs/231 b/tests/btrfs/231
> new file mode 100755
> index 00000000..b4787f9f
> --- /dev/null
> +++ b/tests/btrfs/231
> @@ -0,0 +1,81 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2021 Josef Bacik. All Rights Reserved.
> +#
> +# FS QA Test 231
> +#
> +# This is a regression test for a problem fixed by
> +#
> +# btrfs: do not error out if the extent ref hash doesn't match
> +#
> +# Simply generate a file with a lot of extent references, then reflink in a few
> +# offsets that will generate hash collisions, sync and validate we can still
> +# write to the file system.
> +
> +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
> +. ./common/reflink
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# real QA test starts here
> +_supported_fs btrfs
> +_require_test
> +_require_scratch_reflink
> +
> +_scratch_mkfs >>$seqres.full 2>&1
> +_scratch_mount
> +
> +FILE=$SCRATCH_MNT/file
> +
> +# Create a 1m extent to reflink
> +$XFS_IO_PROG -f -c "pwrite 0 1M" -c "fsync" $FILE | _filter_xfs_io
> +
> +# Generate a bunch of extent references so we're forced to use key'ed extent
> +# references.
> +offset=2
> +for i in {0..10000}
100 iterations is more than enough to reproduce the bug, and reduces
the test runtime from about 50 seconds down to 3 seconds here.
Even 50 iterations is enough.
> +do
> + $XFS_IO_PROG -c "reflink ${FILE} 0 ${offset}M 1M" $FILE \
> + > /dev/null 2>&1
We can remove 2>&1, so we end up failing if for some unexpected reason
a reflink fails.
Might help one day uncover some bug.
> + offset=$(( offset + 2 ))
> +done
> +
> +# Our key is
> +#
> +# key.objectid = bytenr
> +# key.type = BTRFS_EXTENT_DATA_REF_KEY
> +# key.offset = hash(tree, inode, offset)
> +#
> +# The tree id is 5, the inode is 257, and the reflink'ed offset is 0, the below
> +# offsets generate a hash collision with that offset. We only need two to
> +# collide, but if it's worth doing it's worth overdoing.
> +$XFS_IO_PROG -c "reflink ${FILE} 0 17999258914816 1M" $FILE | _filter_xfs_io
> +$XFS_IO_PROG -c "reflink ${FILE} 0 35998517829632 1M" $FILE | _filter_xfs_io
> +$XFS_IO_PROG -c "reflink ${FILE} 0 53752752058368 1M" $FILE | _filter_xfs_io
> +
> +# Sync to make sure this works, it'll error out if we abort the transaction, but
> +# write a file just to make sure
> +$BTRFS_UTIL_PROG filesystem sync $SCRATCH_MNT
> +$XFS_IO_PROG -f -c "pwrite 0 1M" $SCRATCH_MNT/write | _filter_xfs_io
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/btrfs/231.out b/tests/btrfs/231.out
> new file mode 100644
> index 00000000..f52c71ac
> --- /dev/null
> +++ b/tests/btrfs/231.out
> @@ -0,0 +1,11 @@
> +QA output created by 231
> +wrote 1048576/1048576 bytes at offset 0
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +linked 1048576/1048576 bytes at offset 17999258914816
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +linked 1048576/1048576 bytes at offset 35998517829632
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +linked 1048576/1048576 bytes at offset 53752752058368
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +wrote 1048576/1048576 bytes at offset 0
> +XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> diff --git a/tests/btrfs/group b/tests/btrfs/group
> index a7c65983..65cedf7f 100644
> --- a/tests/btrfs/group
> +++ b/tests/btrfs/group
> @@ -233,3 +233,4 @@
> 228 auto quick volume
> 229 auto quick send clone
> 230 auto quick qgroup limit
> +231 auto
We can also have it in the group "clone", and with 100 or 50
iterations the 'quick' group as well.
Other than that it looks good.
Thanks.
> --
> 2.26.2
>
--
Filipe David Manana,
“Whether you think you can, or you think you can't — you're right.”
