On Tue, Feb 12, 2019 at 09:11:42PM -0800, Darrick J. Wong wrote:
> On Tue, Feb 12, 2019 at 08:51:19PM -0800, Darrick J. Wong wrote:
> > On Tue, Feb 12, 2019 at 09:04:36AM -0500, Brian Foster wrote:
> > > On Mon, Feb 11, 2019 at 06:17:54PM -0800, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> > > >
> > > > Create a test (+ helper program) that opens as many unlinked files as it
> > > > possibly can on the scratch filesystem, then closes all the files at
> > > > once to stress-test unlinked file cleanup. Add an xfs-specific test to
> > > > make sure that the fallback code doesn't bitrot.
> > > >
> > > > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> > > > ---
> > > > src/Makefile | 2 -
> > > > src/tmpfile.c | 127 +++++++++++++++++++++++++++++++++++++++++++++++++
> > > > tests/generic/710 | 65 +++++++++++++++++++++++++
> > > > tests/generic/710.out | 2 +
> > > > tests/generic/711 | 73 ++++++++++++++++++++++++++++
> > > > tests/generic/711.out | 2 +
> > > > tests/generic/group | 2 +
> > > > tests/xfs/736 | 71 +++++++++++++++++++++++++++
> > > > tests/xfs/736.out | 2 +
> > > > tests/xfs/737 | 79 ++++++++++++++++++++++++++++++
> > > > tests/xfs/737.out | 2 +
> > > > tests/xfs/group | 2 +
> > > > 12 files changed, 428 insertions(+), 1 deletion(-)
> > > > create mode 100644 src/tmpfile.c
> > > > create mode 100755 tests/generic/710
> > > > create mode 100644 tests/generic/710.out
> > > > create mode 100755 tests/generic/711
> > > > create mode 100644 tests/generic/711.out
> > > > create mode 100755 tests/xfs/736
> > > > create mode 100644 tests/xfs/736.out
> > > > create mode 100755 tests/xfs/737
> > > > create mode 100644 tests/xfs/737.out
> > > >
> > > >
...
> >
> > > I'm wondering if we should have a log recovery test as well, btw.
> >
> > Yes. I'll turn g/710 and x/736 into the log recovery tests.
> >
> > (Oh wow flood of asserts this is going to take a while to straighten
> > out)
>
> NFI what this is about, but here's the crash:
>
> XFS: Assertion failed: VFS_I(ip)->i_nlink == 0, file: fs/xfs/xfs_log_recover.c, line: 5072
Heh, we somehow end up with (what looks like) a linked inode on the
unlinked list..?
Brian
> WARNING: CPU: 0 PID: 2936 at fs/xfs/xfs_message.c:104 assfail+0x27/0x2a [xfs]
> Modules linked in: xfs libcrc32c bfq dax_pmem nd_pmem device_dax sch_fq_codel ip_tables x_tables nfsv4 af_packet
> CPU: 0 PID: 2936 Comm: mount Tainted: G W 4.20.0-rc6-djw #rc6
> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-1ubuntu1 04/01/2014
> RIP: 0010:assfail+0x27/0x2a [xfs]
> Code: 0f 0b c3 0f 1f 44 00 00 48 89 f1 41 89 d0 48 c7 c6 50 aa 34 a0 48 89 fa 31 ff e8 6d f9 ff ff 80 3d 9c 58 0a 00 00 74 02 0f 0b <0f> 0b c3 48 8b b3 a8 02 00 00 48 c7 c7 c8 ae 34 a0 c6 05 08 54 0a
> RSP: 0018:ffffc90003607c90 EFLAGS: 00010246
> RAX: 0000000000000000 RBX: ffff88802f763a80 RCX: 0000000000000000
> RDX: 00000000ffffffc0 RSI: 000000000000000a RDI: ffffffffa033ce6f
> RBP: ffff88802f763c50 R08: 0000000000000000 R09: 0000000000000000
> R10: 000000000000000a R11: f000000000000000 R12: 0000000000000000
> R13: 0000000000000000 R14: ffff88803d6b5000 R15: 0000000000000000
> FS: 00007f67fe2ea080(0000) GS:ffff88803e800000(0000) knlGS:0000000000000000
> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 000055ac0c1ec1d8 CR3: 000000003d47e006 CR4: 00000000001606b0
> Call Trace:
> xlog_recover_process_one_iunlink+0x15e/0x170 [xfs]
> xlog_recover_process_iunlinks.isra.50+0x7c/0xc0 [xfs]
> xlog_recover_finish+0x33/0xa0 [xfs]
> xfs_log_mount_finish+0x5f/0x100 [xfs]
> xfs_mountfs+0x568/0x990 [xfs]
> ? xfs_mru_cache_create+0x172/0x1d0 [xfs]
> xfs_fs_fill_super+0x4d1/0x6d0 [xfs]
> ? xfs_test_remount_options+0x60/0x60 [xfs]
> mount_bdev+0x17f/0x1b0
> mount_fs+0x15/0x7d
> vfs_kern_mount.part.43+0x54/0x160
> do_mount+0x1d2/0xd90
> ? memdup_user+0x4b/0x70
> ksys_mount+0xba/0xd0
> __x64_sys_mount+0x21/0x30
> do_syscall_64+0x50/0x160
> entry_SYSCALL_64_after_hwframe+0x49/0xbe
> RIP: 0033:0x7f67fdbad3ca
> Code: 48 8b 0d c1 8a 2c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 49 89 ca b8 a5 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 8e 8a 2c 00 f7 d8 64 89 01 48
> RSP: 002b:00007fff05981278 EFLAGS: 00000206 ORIG_RAX: 00000000000000a5
> RAX: ffffffffffffffda RBX: 000055ac0c1e4a40 RCX: 00007f67fdbad3ca
> RDX: 000055ac0c1ebd90 RSI: 000055ac0c1e4c40 RDI: 000055ac0c1e4c20
> RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
> R10: 00000000c0ed0000 R11: 0000000000000206 R12: 000055ac0c1e4c20
> R13: 000055ac0c1ebd90 R14: 0000000000000000 R15: 00007f67fe0ce8a4
> irq event stamp: 60960
> hardirqs last enabled at (60959): [<ffffffff810cac15>] console_unlock+0x435/0x5e0
> hardirqs last disabled at (60960): [<ffffffff81001ba0>] trace_hardirqs_off_thunk+0x1a/0x1c
> softirqs last enabled at (60956): [<ffffffff81a003a8>] __do_softirq+0x3a8/0x4bd
> softirqs last disabled at (60943): [<ffffffff8106166c>] irq_exit+0xbc/0xe0
> ---[ end trace f0ce4aa6d3ed581f ]---
>
> <over and over and over>
>
> XFS (pmem4): xfs_do_force_shutdown(0x8) called from line 368 of file fs/xfs/xfs_trans.c. Return address = ffffffffa02edf93
> XFS (pmem4): Corruption of in-memory data detected. Shutting down filesystem
> XFS (pmem4): Please unmount the filesystem and rectify the problem(s)
> XFS (pmem4): xfs_imap_to_bp: xfs_trans_read_buf() returned error -5.
> XFS (pmem4): xlog_recover_clear_agi_bucket: failed to clear agi 0. Continuing.
>
> <over and over and over>
>
> Reproduces on scsi disk too. Nightnight.
>
> --D
>
> > > > +for i in $(seq 1 $nr_cpus); do
> > > > + wait
> > > > +done
> > >
> > > Can't we just pass the pids forked by the loop above? Though the manpage
> > > says wait should wait for all child pids as it is, so perhaps the loop
> > > is unnecessary?
> >
> > Oh, I did not know that. Thanks for the review!
> >
> > > Brian
> > >
> > > > +
> > > > +# Unmount to prove that we can clean it all
> > > > +echo umount >> $seqres.full
> > > > +before=$(date +%s)
> > > > +_scratch_unmount
> > > > +after=$(date +%s)
> > > > +echo "Unmount took $((after - before))s." >> $seqres.full
> > > > +
> > > > +# Mount so that we can run the usual checks
> > > > +echo silence is golden
> > > > +_scratch_mount
> > > > +status=0
> > > > +exit
> > > > diff --git a/tests/generic/711.out b/tests/generic/711.out
> > > > new file mode 100644
> > > > index 00000000..cbbe36e9
> > > > --- /dev/null
> > > > +++ b/tests/generic/711.out
> > > > @@ -0,0 +1,2 @@
> > > > +QA output created by 711
> > > > +silence is golden
> > > > diff --git a/tests/generic/group b/tests/generic/group
> > > > index f56eb475..26999ca1 100644
> > > > --- a/tests/generic/group
> > > > +++ b/tests/generic/group
> > > > @@ -529,3 +529,5 @@
> > > > 524 auto quick
> > > > 525 auto quick rw
> > > > 709 auto quick
> > > > +710 auto quick unlink
> > > > +711 auto quick unlink
> > > > diff --git a/tests/xfs/736 b/tests/xfs/736
> > > > new file mode 100755
> > > > index 00000000..e33de0ae
> > > > --- /dev/null
> > > > +++ b/tests/xfs/736
> > > > @@ -0,0 +1,71 @@
> > > > +#! /bin/bash
> > > > +# SPDX-License-Identifier: GPL-2.0+
> > > > +# Copyright (c) 2019 Oracle, Inc. All Rights Reserved.
> > > > +#
> > > > +# FS QA Test No. 736
> > > > +#
> > > > +# Stress test creating a lot of unlinked O_TMPFILE files and closing them
> > > > +# all at once, checking that we don't blow up the filesystem. This is sort
> > > > +# of a performance test for the xfs unlinked inode backref patchset, but it
> > > > +# applies to most other filesystems.
> > > > +#
> > > > +# Here we force the use of the slow iunlink bucket walk code in a single
> > > > +# threaded situation.
> > > > +#
> > > > +seq=`basename $0`
> > > > +seqres=$RESULT_DIR/$seq
> > > > +echo "QA output created by $seq"
> > > > +tmp=/tmp/$$
> > > > +status=1 # failure is the default!
> > > > +testfile=$TEST_DIR/$seq.txt
> > > > +trap "_cleanup; exit \$status" 0 1 2 3 15
> > > > +
> > > > +_cleanup()
> > > > +{
> > > > + cd /
> > > > + rm -f $tmp.*
> > > > +}
> > > > +
> > > > +# get standard environment, filters and checks
> > > > +. ./common/rc
> > > > +. ./common/attr
> > > > +. ./common/filter
> > > > +. ./common/inject
> > > > +
> > > > +# real QA test starts here
> > > > +_supported_fs generic
> > > > +_supported_os Linux
> > > > +_require_xfs_io_error_injection "iunlink_fallback"
> > > > +_require_scratch
> > > > +
> > > > +rm -f $seqres.full
> > > > +_scratch_mkfs >> $seqres.full 2>&1
> > > > +_scratch_mount
> > > > +
> > > > +# Set ULIMIT_NOFILE to min(file-max, 30000 files per LOAD_FACTOR)
> > > > +# so that this test doesn't take forever or OOM the box
> > > > +max_files=$((30000 * LOAD_FACTOR))
> > > > +max_allowable_files=$(cat /proc/sys/fs/file-max)
> > > > +test $max_files -gt $max_allowable_files && max_files=$max_allowable_files
> > > > +ulimit -n $max_files
> > > > +
> > > > +# Force xfs to use the iunlinked fallback 50% of the time
> > > > +_scratch_inject_error "iunlink_fallback" "2"
> > > > +
> > > > +# Open a lot of unlinked files
> > > > +echo create >> $seqres.full
> > > > +program=$PWD/src/tmpfile
> > > > +(cd $SCRATCH_MNT ; $program >> $seqres.full)
> > > > +
> > > > +# Unmount to prove that we can clean it all
> > > > +echo umount >> $seqres.full
> > > > +before=$(date +%s)
> > > > +_scratch_unmount
> > > > +after=$(date +%s)
> > > > +echo "Unmount took $((after - before))s." >> $seqres.full
> > > > +
> > > > +# Mount so that we can run the usual checks
> > > > +echo silence is golden
> > > > +_scratch_mount
> > > > +status=0
> > > > +exit
> > > > diff --git a/tests/xfs/736.out b/tests/xfs/736.out
> > > > new file mode 100644
> > > > index 00000000..0258a248
> > > > --- /dev/null
> > > > +++ b/tests/xfs/736.out
> > > > @@ -0,0 +1,2 @@
> > > > +QA output created by 736
> > > > +silence is golden
> > > > diff --git a/tests/xfs/737 b/tests/xfs/737
> > > > new file mode 100755
> > > > index 00000000..47e65607
> > > > --- /dev/null
> > > > +++ b/tests/xfs/737
> > > > @@ -0,0 +1,79 @@
> > > > +#! /bin/bash
> > > > +# SPDX-License-Identifier: GPL-2.0+
> > > > +# Copyright (c) 2019 Oracle, Inc. All Rights Reserved.
> > > > +#
> > > > +# FS QA Test No. 737
> > > > +#
> > > > +# Stress test creating a lot of unlinked O_TMPFILE files and closing them
> > > > +# all at once, checking that we don't blow up the filesystem. This is sort
> > > > +# of a performance test for the xfs unlinked inode backref patchset, but it
> > > > +# applies to most other filesystems.
> > > > +#
> > > > +# Here we force the use of the slow iunlink bucket walk code, using every
> > > > +# CPU possible.
> > > > +#
> > > > +seq=`basename $0`
> > > > +seqres=$RESULT_DIR/$seq
> > > > +echo "QA output created by $seq"
> > > > +tmp=/tmp/$$
> > > > +status=1 # failure is the default!
> > > > +testfile=$TEST_DIR/$seq.txt
> > > > +trap "_cleanup; exit \$status" 0 1 2 3 15
> > > > +
> > > > +_cleanup()
> > > > +{
> > > > + cd /
> > > > + rm -f $tmp.*
> > > > +}
> > > > +
> > > > +# get standard environment, filters and checks
> > > > +. ./common/rc
> > > > +. ./common/attr
> > > > +. ./common/filter
> > > > +. ./common/inject
> > > > +
> > > > +# real QA test starts here
> > > > +_supported_fs generic
> > > > +_supported_os Linux
> > > > +_require_xfs_io_error_injection "iunlink_fallback"
> > > > +_require_scratch
> > > > +
> > > > +rm -f $seqres.full
> > > > +_scratch_mkfs >> $seqres.full 2>&1
> > > > +_scratch_mount
> > > > +
> > > > +# Load up all the CPUs, two threads per CPU.
> > > > +nr_cpus=$(( $(getconf _NPROCESSORS_ONLN) * 2 ))
> > > > +
> > > > +# Set ULIMIT_NOFILE to min(file-max, 30000 files per cpu per LOAD_FACTOR)
> > > > +# so that this test doesn't take forever or OOM the box
> > > > +max_files=$((30000 * LOAD_FACTOR))
> > > > +max_allowable_files=$(( $(cat /proc/sys/fs/file-max) / nr_cpus ))
> > > > +test $max_files -gt $max_allowable_files && max_files=$max_allowable_files
> > > > +ulimit -n $max_files
> > > > +
> > > > +# Force xfs to use the iunlinked fallback 50% of the time
> > > > +_scratch_inject_error "iunlink_fallback" "2"
> > > > +
> > > > +# Open a lot of unlinked files
> > > > +echo create >> $seqres.full
> > > > +program=$PWD/src/tmpfile
> > > > +for i in $(seq 1 $nr_cpus); do
> > > > + (mkdir $SCRATCH_MNT/$i ; cd $SCRATCH_MNT/$i ; $program >> $seqres.full) &
> > > > +done
> > > > +for i in $(seq 1 $nr_cpus); do
> > > > + wait
> > > > +done
> > > > +
> > > > +# Unmount to prove that we can clean it all
> > > > +echo umount >> $seqres.full
> > > > +before=$(date +%s)
> > > > +_scratch_unmount
> > > > +after=$(date +%s)
> > > > +echo "Unmount took $((after - before))s." >> $seqres.full
> > > > +
> > > > +# Mount so that we can run the usual checks
> > > > +echo silence is golden
> > > > +_scratch_mount
> > > > +status=0
> > > > +exit
> > > > diff --git a/tests/xfs/737.out b/tests/xfs/737.out
> > > > new file mode 100644
> > > > index 00000000..bdc4966d
> > > > --- /dev/null
> > > > +++ b/tests/xfs/737.out
> > > > @@ -0,0 +1,2 @@
> > > > +QA output created by 737
> > > > +silence is golden
> > > > diff --git a/tests/xfs/group b/tests/xfs/group
> > > > index 7b7d69f1..d3189cd5 100644
> > > > --- a/tests/xfs/group
> > > > +++ b/tests/xfs/group
> > > > @@ -497,3 +497,5 @@
> > > > 497 dangerous_fuzzers dangerous_scrub dangerous_online_repair
> > > > 498 dangerous_fuzzers dangerous_norepair
> > > > 499 auto quick
> > > > +736 auto quick unlink
> > > > +737 auto quick unlink
> > > >
