linux-xfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "Darrick J. Wong" <djwong@kernel.org>
To: Eryu Guan <eguan@linux.alibaba.com>
Cc: Eryu Guan <guan@eryu.me>,
	guaneryu@gmail.com, linux-xfs@vger.kernel.org,
	fstests@vger.kernel.org
Subject: Re: [PATCH 2/3] generic: test shutdowns of a nested filesystem
Date: Mon, 16 Aug 2021 21:16:16 -0700	[thread overview]
Message-ID: <20210817041616.GC12640@magnolia> (raw)
In-Reply-To: <20210817031649.GO60846@e18g06458.et15sqa>

On Tue, Aug 17, 2021 at 11:16:49AM +0800, Eryu Guan wrote:
> On Mon, Aug 16, 2021 at 09:35:24AM -0700, Darrick J. Wong wrote:
> > On Mon, Aug 16, 2021 at 12:28:20AM +0800, Eryu Guan wrote:
> > > On Tue, Jul 27, 2021 at 05:10:30PM -0700, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <djwong@kernel.org>
> > > > 
> > > > generic/475, but we're running fsstress on a disk image inside the
> > > > scratch filesystem
> > > > 
> > > > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > > > ---
> > > >  tests/generic/725     |  136 +++++++++++++++++++++++++++++++++++++++++++++++++
> > > >  tests/generic/725.out |    2 +
> > > >  2 files changed, 138 insertions(+)
> > > >  create mode 100755 tests/generic/725
> > > >  create mode 100644 tests/generic/725.out
> > > > 
> > > > 
> > > > diff --git a/tests/generic/725 b/tests/generic/725
> > > > new file mode 100755
> > > > index 00000000..f43bcb37
> > > > --- /dev/null
> > > > +++ b/tests/generic/725
> > > > @@ -0,0 +1,136 @@
> > > > +#! /bin/bash
> > > > +# SPDX-License-Identifier: GPL-2.0
> > > > +# Copyright (c) 2021 Oracle, Inc.  All Rights Reserved.
> > > > +#
> > > > +# FS QA Test No. 725
> > > > +#
> > > > +# Test nested log recovery with repeated (simulated) disk failures.  We kick
> > > > +# off fsstress on a loopback filesystem mounted on the scratch fs, then switch
> > > > +# out the underlying scratch device with dm-error to see what happens when the
> > > > +# disk goes down.  Having taken down both fses in this manner, remount them and
> > > > +# repeat.  This test simulates VM hosts crashing to try to shake out CoW bugs
> > > > +# in writeback on the host that cause VM guests to fail to recover.
> > > 
> > > It currently fails for me on btrfs, the loop mount failed in 2nd
> > > iteration, seems like a bug in btrfs.
> > 
> > Yep.  Until recently (aka the Big Xfs Log Recovery Bughunt of 2021) it
> > wouldn't pass xfs either. :/
> > 
> > > > +#
> > > > +. ./common/preamble
> > > > +_begin_fstest shutdown auto log metadata eio
> > > > +
> > > > +_cleanup()
> > > > +{
> > > > +	cd /
> > > > +	$KILLALL_PROG -9 fsstress > /dev/null 2>&1
> > > > +	wait
> > > > +	if [ -n "$loopmnt" ]; then
> > > > +		umount $loopmnt 2>/dev/null
> > > 
> > > $UMOUNT_PROG
> > > 
> > > > +		rm -r -f $loopmnt
> > > > +	fi
> > > > +	rm -f $tmp.*
> > > > +	_dmerror_unmount
> > > > +	_dmerror_cleanup
> > > > +}
> > > > +
> > > > +# Import common functions.
> > > > +. ./common/dmerror
> > > > +. ./common/reflink
> > > > +
> > > > +# Modify as appropriate.
> > > > +_supported_fs generic
> > > > +
> > > > +_require_scratch_reflink
> > > > +_require_cp_reflink
> > > > +_require_dm_target error
> > > > +_require_command "$KILLALL_PROG" "killall"
> > > > +
> > > > +echo "Silence is golden."
> > > > +
> > > > +_scratch_mkfs >> $seqres.full 2>&1
> > > > +_require_metadata_journaling $SCRATCH_DEV
> > > > +_dmerror_init
> > > > +_dmerror_mount
> > > > +
> > > > +# Create a fs image consuming 1/3 of the scratch fs
> > > > +scratch_freesp_bytes=$(stat -f -c '%a * %S' $SCRATCH_MNT | bc)
> > > 
> > > _get_available_space $SCRATCH_MNT ?
> > > 
> > > > +loopimg_bytes=$((scratch_freesp_bytes / 3))
> > > > +
> > > > +loopimg=$SCRATCH_MNT/testfs
> > > > +truncate -s $loopimg_bytes $loopimg
> > > > +_mkfs_dev $loopimg
> > > > +
> > > > +loopmnt=$tmp.mount
> > > > +mkdir -p $loopmnt
> > > > +
> > > > +scratch_aliveflag=$tmp.runsnap
> > > > +snap_aliveflag=$tmp.snapping
> > > > +
> > > > +snap_loop_fs() {
> > > > +	touch "$snap_aliveflag"
> > > > +	while [ -e "$scratch_aliveflag" ]; do
> > > > +		rm -f $loopimg.a
> > > > +		_cp_reflink $loopimg $loopimg.a
> > > > +		sleep 1
> > > > +	done
> > > > +	rm -f "$snap_aliveflag"
> > > > +}
> > > > +
> > > > +fsstress=($FSSTRESS_PROG $FSSTRESS_AVOID -d "$loopmnt" -n 999999 -p "$((LOAD_FACTOR * 4))")
> > > > +
> > > > +for i in $(seq 1 $((25 * TIME_FACTOR)) ); do
> > > > +	touch $scratch_aliveflag
> > > > +	snap_loop_fs >> $seqres.full 2>&1 &
> > > > +
> > > > +	if ! _mount $loopimg $loopmnt -o loop; then
> > > > +		rm -f $scratch_aliveflag
> > > > +		_fail "loop mount failed"
> > > 
> > > I found it a bit easier to debug if print $i here.
> > 
> > Ok, I'll change it to "loop $i mount failed".
> > 
> > > > +		break
> > > > +	fi
> > > > +
> > > > +	("${fsstress[@]}" >> $seqres.full &) > /dev/null 2>&1
> > > > +
> > > > +	# purposely include 0 second sleeps to test shutdown immediately after
> > > > +	# recovery
> > > > +	sleep $((RANDOM % (3 * TIME_FACTOR) ))
> > > > +	rm -f $scratch_aliveflag
> > > > +
> > > > +	# This test aims to simulate sudden disk failure, which means that we
> > > > +	# do not want to quiesce the filesystem or otherwise give it a chance
> > > > +	# to flush its logs.  Therefore we want to call dmsetup with the
> > > > +	# --nolockfs parameter; to make this happen we must call the load
> > > > +	# error table helper *without* 'lockfs'.
> > > > +	_dmerror_load_error_table
> > > > +
> > > > +	ps -e | grep fsstress > /dev/null 2>&1
> > > > +	while [ $? -eq 0 ]; do
> > > > +		$KILLALL_PROG -9 fsstress > /dev/null 2>&1
> > > > +		wait > /dev/null 2>&1
> > > > +		ps -e | grep fsstress > /dev/null 2>&1
> > > > +	done
> > > > +	for ((i = 0; i < 10; i++)); do
> > > > +		test -e "$snap_aliveflag" || break
> > > > +		sleep 1
> > > > +	done
> > > > +
> > > > +	# Mount again to replay log after loading working table, so we have a
> > > > +	# consistent XFS after test.
> > > 
> > > This is a generic test, fix the XFS specific comments?
> > 
> > Oops.  "...a consistent fs after test."
> > 
> > > > +	$UMOUNT_PROG $loopmnt
> > > > +	_dmerror_unmount || _fail "unmount failed"
> > > > +	_dmerror_load_working_table
> > > > +	if ! _dmerror_mount; then
> > > > +		dmsetup table | tee -a /dev/ttyprintk
> > > > +		lsblk | tee -a /dev/ttyprintk
> > > > +		$XFS_METADUMP_PROG -a -g -o $DMERROR_DEV $seqres.dmfail.md
> > > 
> > > Above logs all should go to $seqres.full ?
> > 
> > Oops, yeah.  I'll remove them since I was only using them to check the
> > system state.
> > 
> > > And $XFS_METADUMP_PROG is not suitable for a generic test.
> > 
> > I'll create _metadump_dev so that this at least works for the two
> > filesystems for which we have dump creation helpers (ext* and xfs).
> 
> Sounds great!
> 
> > 
> > > > +		_fail "mount failed"
> > > > +	fi
> > > > +done
> > > > +
> > > > +# Make sure the fs image file is ok
> > > > +if [ -f "$loopimg" ]; then
> > > > +	if _mount $loopimg $loopmnt -o loop; then
> > > > +		$UMOUNT_PROG $loopmnt &> /dev/null
> > > > +	else
> > > > +		echo "final loop mount failed"
> > > > +	fi
> > > > +	_check_xfs_filesystem $loopimg none none
> > > 
> > > Same here, use _check_scratch_fs?
> > 
> > $loopimg is a file within the scratch fs.
> 
> _check_scratch_fs can take dev as argument, and default to $SCRATCH_DEV,
> I think that works in this case?

It could be made to work with a large enough crowbar, but that's
seriously overkill because $loopimg is a file *within* the scratch
filesystem.  The $loopimg fs gets formatted without the
SCRATCH_LOGDEV/SCRATCH_RTDEV options (because it is not itself the
scratch filesystem), which means that in order to (ab)use
_check_scratch_fs to do the same thing as _check_xfs_filesystem, you
have to exclude those options.  So yes, this:

	SCRATCH_RTDEV= SCRATCH_LOGDEV= _check_scratch_fs $loopimg

is the equivalent of this:

	_check_xfs_filesystem $loopimg none none

But the first is longer and pointless.

--D

> Thanks,
> Eryu

  reply	other threads:[~2021-08-17  4:16 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-28  0:10 [PATCHSET 0/3] fstests: exercise code refactored in 5.14 Darrick J. Wong
2021-07-28  0:10 ` [PATCH 1/3] generic: test xattr operations only Darrick J. Wong
2021-08-12  5:34   ` Zorro Lang
2021-08-12 17:04     ` Darrick J. Wong
2021-08-15 15:46       ` Eryu Guan
2021-07-28  0:10 ` [PATCH 2/3] generic: test shutdowns of a nested filesystem Darrick J. Wong
2021-08-12  5:44   ` Zorro Lang
2021-08-12 17:07     ` Darrick J. Wong
2021-08-13 14:52       ` Zorro Lang
2021-08-15 16:28   ` Eryu Guan
2021-08-16 16:35     ` Darrick J. Wong
2021-08-17  3:16       ` Eryu Guan
2021-08-17  4:16         ` Darrick J. Wong [this message]
2021-08-17 15:54           ` Darrick J. Wong
2021-07-28  0:10 ` [PATCH 3/3] xfs: test regression in shrink when the new EOFS splits a sparse inode cluster Darrick J. Wong
2021-08-12  6:01   ` Zorro Lang
2021-09-01  0:11 [PATCHSET v2 0/3] fstests: exercise code refactored in 5.14 Darrick J. Wong
2021-09-01  0:11 ` [PATCH 2/3] generic: test shutdowns of a nested filesystem Darrick J. Wong

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=20210817041616.GC12640@magnolia \
    --to=djwong@kernel.org \
    --cc=eguan@linux.alibaba.com \
    --cc=fstests@vger.kernel.org \
    --cc=guan@eryu.me \
    --cc=guaneryu@gmail.com \
    --cc=linux-xfs@vger.kernel.org \
    --subject='Re: [PATCH 2/3] generic: test shutdowns of a nested filesystem' \
    /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

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).