All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Darrick J. Wong" <djwong@kernel.org>
To: Shiyang Ruan <ruansy.fnst@fujitsu.com>
Cc: fstests@vger.kernel.org
Subject: Re: [PATCH v3 15/17] generic: add race test between reflink and mmap read
Date: Tue, 11 Jan 2022 10:55:16 -0800	[thread overview]
Message-ID: <20220111185516.GF31566@magnolia> (raw)
In-Reply-To: <20211214081914.2478122-16-ruansy.fnst@fujitsu.com>

On Tue, Dec 14, 2021 at 04:19:12PM +0800, Shiyang Ruan wrote:
> Test for races or FS corruption between reflink and mmap reading the
> target file. (MMAP version of generic/164,165)

Hi, now that this test has been running for a couple of weeks, I have
observed periodic false positives from this test:

QA output created by 670
Format and mount
Initialize files
Reflink and mmap reread the files!
00001000:  61 61 61 61 61 61 61 61 62 62 62 62 62 62 62 62 aaaabbbb
Finished reflinking

I suspect that if the _mread_range of file3 races with the page cache
invalidation that FICLONERANGE performs, it is possible that the mread
dump will contain a mix of 0x61 and 0x62.  Looking at mread_f, it looks
like it does a byte-at-a-time copy of the mmap...

	if (rflag) {
		for (tmp = length - 1, c = 0; tmp >= 0; tmp--, c = 1) {
			*bp = *(((char *)mapping->addr) + dumpoffset + tmp);
			cnt++;

...which is a sufficient window for the page cache mapping to get
invalidated such that the mread will block on the page fault until the
reflink operation finishes.

I think the solution here is to adjust the egrep regexp above to find
any line that does /not/ contain a's or b's, since (in principle) the
reflink could run fast enough that every byte read hits a pgae fault.
What do you think?

--D

> Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> ---
>  common/reflink        | 11 +++++++
>  tests/generic/913     | 72 +++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/913.out |  5 +++
>  3 files changed, 88 insertions(+)
>  create mode 100755 tests/generic/913
>  create mode 100644 tests/generic/913.out
> 
> diff --git a/common/reflink b/common/reflink
> index 68dbdedd..455260c6 100644
> --- a/common/reflink
> +++ b/common/reflink
> @@ -186,6 +186,17 @@ _read_range() {
>  	$XFS_IO_PROG $xfs_io_args -f -c "pread -q -v $offset $len" "$file" | cut -d ' ' -f '3-18'
>  }
>  
> +# Prints a range of a file as a hex dump
> +_mread_range() {
> +	file="$1"
> +	offset="$2"
> +	len="$3"
> +	xfs_io_args="$4"
> +
> +	$XFS_IO_PROG $xfs_io_args -f -c "mmap -rw 0 $((offset + len))" \
> +		-c "mread -v $offset $len" "$file" | cut -d ' ' -f '3-18'
> +}
> +
>  # Compare ranges of two files
>  _compare_range() {
>  	file1="$1"
> diff --git a/tests/generic/913 b/tests/generic/913
> new file mode 100755
> index 00000000..f709c36c
> --- /dev/null
> +++ b/tests/generic/913
> @@ -0,0 +1,72 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +#
> +# FS QA Test No. 913
> +#
> +# Test for races or FS corruption between reflink and mmap reading the
> +# target file. (MMAP version of generic/164,165)
> +#
> +. ./common/preamble
> +_begin_fstest auto clone
> +
> +_register_cleanup "_cleanup" BUS
> +
> +# Import common functions.
> +. ./common/filter
> +. ./common/reflink
> +
> +# real QA test starts here
> +_require_scratch_reflink
> +_require_cp_reflink
> +
> +echo "Format and mount"
> +_scratch_mkfs > $seqres.full 2>&1
> +_scratch_mount >> $seqres.full 2>&1
> +
> +testdir=$SCRATCH_MNT/test-$seq
> +finished_file=/tmp/finished
> +rm -rf $finished_file
> +mkdir $testdir
> +
> +loops=512
> +nr_loops=$((loops - 1))
> +blksz=65536
> +
> +echo "Initialize files"
> +echo >> $seqres.full
> +_pwrite_byte 0x61 0 $((loops * blksz)) $testdir/file1 >> $seqres.full
> +_pwrite_byte 0x62 0 $((loops * blksz)) $testdir/file2 >> $seqres.full
> +_cp_reflink $testdir/file1 $testdir/file3
> +_scratch_cycle_mount
> +
> +fbytes() {
> +	egrep -v '(61 61 61 61 61 61 61 61 61 61 61 61 61 61 61 61|62 62 62 62 62 62 62 62 62 62 62 62 62 62 62 62)'
> +}
> +
> +reader() {
> +	while [ ! -e $finished_file ]; do
> +		_mread_range $testdir/file3 0 $((loops * blksz)) | fbytes
> +	done
> +}
> +
> +echo "Reflink and mmap reread the files!"
> +reader &
> +for i in `seq 1 2`; do
> +	seq $nr_loops -1 0 | while read i; do
> +		_reflink_range  $testdir/file1 $((i * blksz)) \
> +				$testdir/file3 $((i * blksz)) $blksz >> $seqres.full
> +		[ $? -ne 0 ] && break
> +	done
> +	seq $nr_loops -1 0 | while read i; do
> +		_reflink_range  $testdir/file2 $((i * blksz)) \
> +				$testdir/file3 $((i * blksz)) $blksz >> $seqres.full
> +		[ $? -ne 0 ] && break
> +	done
> +done
> +echo "Finished reflinking"
> +touch $finished_file
> +wait
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/generic/913.out b/tests/generic/913.out
> new file mode 100644
> index 00000000..a34df6ce
> --- /dev/null
> +++ b/tests/generic/913.out
> @@ -0,0 +1,5 @@
> +QA output created by 913
> +Format and mount
> +Initialize files
> +Reflink and mmap reread the files!
> +Finished reflinking
> -- 
> 2.34.1
> 
> 
> 

  reply	other threads:[~2022-01-11 18:55 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-12-14  8:18 [PATCH v3 00/17] generic: add some mmap CoW tests Shiyang Ruan
2021-12-14  8:18 ` [PATCH v3 01/17] generic: add mmap CoW test for ranges of two shared files Shiyang Ruan
2021-12-14  8:18 ` [PATCH v3 02/17] generic: add mmap CoW test for regular&destination extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 03/17] generic: add mmap CoW test for unwritten&destination extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 04/17] generic: add mmap CoW test for holes&destination extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 05/17] common/rc: Introduce _require_scratch_delalloc() Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 06/17] generic: add mmap CoW test for delalloc&destination extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 07/17] generic: add mmap CoW test for mixed&destination extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 08/17] generic: add mmap CoW test for regular&source extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 09/17] generic: add mmap CoW test for unwritten&source extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 10/17] generic: add mmap CoW test for holes&source extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 11/17] generic: add mmap CoW test for delalloc&source extents Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 12/17] generic: add buffered-io CoW test for mixed&source extents Shiyang Ruan
2021-12-25 13:27   ` Eryu Guan
2021-12-14  8:19 ` [PATCH v3 13/17] generic: add direct-io " Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 14/17] generic: add mmap " Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 15/17] generic: add race test between reflink and mmap read Shiyang Ruan
2022-01-11 18:55   ` Darrick J. Wong [this message]
2022-01-12  2:38     ` Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 16/17] generic: add race test that mmap write to source of reflink Shiyang Ruan
2021-12-14  8:19 ` [PATCH v3 17/17] generic: add race test that mmap write to target " Shiyang Ruan
2021-12-19 16:04 ` [PATCH v3 00/17] generic: add some mmap CoW tests Eryu Guan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20220111185516.GF31566@magnolia \
    --to=djwong@kernel.org \
    --cc=fstests@vger.kernel.org \
    --cc=ruansy.fnst@fujitsu.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.