On Thu, Nov 09, 2017 at 04:32:42PM +0800, Eryu Guan wrote:
> From mmap(2) manpage, "a file is mapped in multiples of the page
> size. For a file that is not a multiple of the page size, the
> remaining memory is zeroed when mapped", this test is to test this
> behavior on truncate down.
>
> This is inspired by an XFS bug that truncate down fails to zero page
> cache beyond new EOF and causes stale data written to disk
> unexpectedly and a subsequent mmap sees non-zeros post EOF.
>
> Patch "xfs: truncate pagecache before writeback in
> xfs_setattr_size()" fixed the bug on XFS.
>
> Signed-off-by: Eryu Guan <eguan@xxxxxxxxxx>
> ---
> v2:
> - update comments and commit log to reflect the final fix of the bug
> - dump fsx error message to stdout on error, so it's easier to know
> what is failing
Ping on this test.
Thanks,
Eryu
>
> tests/generic/466 | 135 ++++++++++++++++++++++++++++++++++++++++++++++++++
> tests/generic/466.out | 9 ++++
> tests/generic/group | 1 +
> 3 files changed, 145 insertions(+)
> create mode 100755 tests/generic/466
> create mode 100644 tests/generic/466.out
>
> diff --git a/tests/generic/466 b/tests/generic/466
> new file mode 100755
> index 000000000000..435bd39a42cc
> --- /dev/null
> +++ b/tests/generic/466
> @@ -0,0 +1,135 @@
> +#! /bin/bash
> +# FS QA Test 466
> +#
> +# Test that mmap read doesn't see non-zero data past EOF on truncate down.
> +#
> +# This is inspired by an XFS bug that truncate down fails to zero page cache
> +# beyond new EOF and causes stale data written to disk unexpectedly and a
> +# subsequent mmap reads and sees non-zeros post EOF.
> +#
> +# Patch "xfs: truncate pagecache before writeback in xfs_setattr_size()" fixed
> +# the bug on XFS.
> +#
> +#-----------------------------------------------------------------------
> +# Copyright (c) 2017 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/$$
> +file=$TEST_DIR/$seq.fsx
> +status=1 # failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> + cd /
> + rm -f $file $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# real QA test starts here
> +_supported_fs generic
> +_supported_os Linux
> +_require_test
> +
> +run_fsx()
> +{
> + $here/ltp/fsx $2 --replay-ops $1 $file 2>&1 | tee -a $seqres.full >$tmp.fsx
> + if [ ${PIPESTATUS[0]} -ne 0 ]; then
> + cat $tmp.fsx
> + exit 1
> + fi
> +}
> +
> +# run fsx with and without fsync(2) after write to get more coverage
> +test_fsx()
> +{
> + echo "fsx --replay-ops ${1#*.}"
> + run_fsx $1
> +
> + echo "fsx -y --replay-ops ${1#*.}"
> + run_fsx $1 -y
> +}
> +
> +# simplified fsx operations that work on small & not blocksize-aligned offsets,
> +# so filesystems with small block size could reproduce too
> +cat >$tmp.fsxops.0 <<EOF
> +# create file with unwritten extent, KEEP_SIZE flag is required, otherwise page
> +# straddles new i_size in the writeback triggered by truncate, range [i_size,
> +# page_boundary] will be zeroed there, and bug won't be reproduced
> +fallocate 0x0 0x1000 0x0 keep_size
> +
> +# overwrite the unwritten extents with non-zeros, but extent will stay in
> +# unwritten till I/O completion
> +write 0x0 0x1000 0x0
> +
> +# truncate down the file, which should zero page cache beyong new EOF but a
> +# buggy kernel won't
> +truncate 0x0 0x10 0x1000
> +
> +# unmap the file and invalidate the pagecache of the block
> +punch_hole 0x0 0x10 0x10
> +
> +# mmap reads the whole block from disk, and fsx will check page range beyond
> +# EOF to make sure we only see zeros there
> +mapread 0x0 0x10 0x10
> +EOF
> +
> +# to get a bit more test coverage, try other operation combinations too
> +# same as fsxops.0, but skip punch_hole to keep the pagecache before mapread
> +cat >$tmp.fsxops.1 <<EOF
> +fallocate 0x0 0x1000 0x0 keep_size
> +write 0x0 0x1000 0x0
> +truncate 0x0 0x10 0x1000
> +mapread 0x0 0x10 0x10
> +EOF
> +
> +# same as fsxops.0, but fallocate without KEEP_SIZE flag
> +cat >$tmp.fsxops.2 <<EOF
> +fallocate 0x0 0x1000 0x0
> +write 0x0 0x1000 0x0
> +truncate 0x0 0x10 0x1000
> +punch_hole 0x0 0x10 0x10
> +mapread 0x0 0x10 0x10
> +EOF
> +
> +# this is from the original fsxops when bug was first hit
> +cat >$tmp.fsxops.3 <<EOF
> +fallocate 0x35870 0xa790 0x0 keep_size
> +write 0x2aa50 0xc37f 0x0
> +truncate 0x0 0x36dcd 0x36dcf
> +zero_range 0x35849 0x1584 0x36dcd
> +mapread 0x361c8 0xc05 0x36dcd
> +EOF
> +
> +for i in 0 1 2 3; do
> + test_fsx $tmp.fsxops.$i
> +done
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/generic/466.out b/tests/generic/466.out
> new file mode 100644
> index 000000000000..8430bb462ad1
> --- /dev/null
> +++ b/tests/generic/466.out
> @@ -0,0 +1,9 @@
> +QA output created by 466
> +fsx --replay-ops fsxops.0
> +fsx -y --replay-ops fsxops.0
> +fsx --replay-ops fsxops.1
> +fsx -y --replay-ops fsxops.1
> +fsx --replay-ops fsxops.2
> +fsx -y --replay-ops fsxops.2
> +fsx --replay-ops fsxops.3
> +fsx -y --replay-ops fsxops.3
> diff --git a/tests/generic/group b/tests/generic/group
> index fbe0a7f4a717..b71644008b55 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -468,3 +468,4 @@
> 463 auto quick clone dangerous
> 464 auto rw
> 465 auto rw quick aio
> +466 auto quick
> --
> 2.13.6
>
--
To unsubscribe from this list: send the line "unsubscribe fstests" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
