Re: [PATCH 5/8] xfs: regression test for allocsp handing out stale disk contents

[Zorro Lang <zlang@redhat.com>]

On Tue, Jan 11, 2022 at 01:50:35PM -0800, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@kernel.org>
> 
> Add a regression test to check that XFS_IOC_ALLOCSP isn't handing out
> stale disk blocks for preallocation.
> 
> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> ---

Hi Darrick,

This case is easily failed on some multi-striped storage[1]. The full output
as [2], the out.bad output as [3].

Thanks,
Zorro

[1]
FSTYP         -- xfs (non-debug)
PLATFORM      -- Linux/x86_64 xxx-xxxxx-xx 5.14.0-xxx #1 SMP PREEMPT Fri Jan 14 09:24:44 UTC 2022
MKFS_OPTIONS  -- -f -m crc=1,finobt=1,rmapbt=0,reflink=1,bigtime=1,inobtcount=1 -i sparse=1 /dev/sda4
MOUNT_OPTIONS -- -o context=system_u:object_r:root_t:s0 /dev/sda4 /mnt/xfstests/scratch

xfs/832	_check_xfs_filesystem: filesystem on /dev/sda4 is inconsistent (r)
(see /var/lib/xfstests/results//xfs/832.full for details)
- output mismatch (see /var/lib/xfstests/results//xfs/832.out.bad)
    --- tests/xfs/832.out	2022-01-15 22:18:35.175245791 -0500
    +++ /var/lib/xfstests/results//xfs/832.out.bad	2022-01-15 22:20:46.630697627 -0500
    @@ -1,2 +1,3 @@
     QA output created by 832
    +ioctl: No space left on device
     Silence is golden
    ...
    (Run 'diff -u /var/lib/xfstests/tests/xfs/832.out /var/lib/xfstests/results//xfs/832.out.bad'  to see the entire diff)
Ran: xfs/832
Failures: xfs/832
Failed 1 of 1 tests

[2]
wrote 33554432/33554432 bytes at offset 0
32 MiB, 4 ops; 0.1368 sec (233.848 MiB/sec and 29.2310 ops/sec)
meta-data=/dev/sda4              isize=512    agcount=1, agsize=4112 blks
         =                       sectsz=512   attr=2, projid32bit=1
         =                       crc=1        finobt=1, sparse=1, rmapbt=0
         =                       reflink=1    bigtime=1 inobtcount=1
data     =                       bsize=4096   blocks=4112, imaxpct=25
         =                       sunit=16     swidth=32 blks
naming   =version 2              bsize=4096   ascii-ci=0, ftype=1
log      =internal log           bsize=4096   blocks=1872, version=2
         =                       sectsz=512   sunit=16 blks, lazy-count=1
realtime =none                   extsz=4096   blocks=0, rtextents=0
Setting up 4096 runs for block size 4096
_check_xfs_filesystem: filesystem on /dev/sda4 is inconsistent (r)
*** xfs_repair -n output ***
Phase 1 - find and verify superblock...
Only one AG detected - cannot validate filesystem geometry.
Use the -o force_geometry option to proceed.
*** end xfs_repair output
...

[3]
QA output created by 832
ioctl: No space left on device
Silence is golden


>  .gitignore        |    1 
>  src/Makefile      |    2 -
>  src/allocstale.c  |  117 +++++++++++++++++++++++++++++++++++++++++++++++++++++
>  tests/xfs/832     |   56 +++++++++++++++++++++++++
>  tests/xfs/832.out |    2 +
>  5 files changed, 177 insertions(+), 1 deletion(-)
>  create mode 100644 src/allocstale.c
>  create mode 100755 tests/xfs/832
>  create mode 100644 tests/xfs/832.out
> 
> 
> diff --git a/.gitignore b/.gitignore
> index 65b93307..ba0c572b 100644
> --- a/.gitignore
> +++ b/.gitignore
> @@ -56,6 +56,7 @@ tags
>  # src/ binaries
>  /src/af_unix
>  /src/alloc
> +/src/allocstale
>  /src/append_reader
>  /src/append_writer
>  /src/attr_replace_test
> diff --git a/src/Makefile b/src/Makefile
> index 1737ed0e..111ce1d9 100644
> --- a/src/Makefile
> +++ b/src/Makefile
> @@ -18,7 +18,7 @@ TARGETS = dirstress fill fill2 getpagesize holes lstat64 \
>  	t_ext4_dax_journal_corruption t_ext4_dax_inline_corruption \
>  	t_ofd_locks t_mmap_collision mmap-write-concurrent \
>  	t_get_file_time t_create_short_dirs t_create_long_dirs t_enospc \
> -	t_mmap_writev_overlap checkpoint_journal mmap-rw-fault
> +	t_mmap_writev_overlap checkpoint_journal mmap-rw-fault allocstale
>  
>  LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \
>  	preallo_rw_pattern_writer ftrunc trunc fs_perms testx looptest \
> diff --git a/src/allocstale.c b/src/allocstale.c
> new file mode 100644
> index 00000000..6253fe4c
> --- /dev/null
> +++ b/src/allocstale.c
> @@ -0,0 +1,117 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2022 Oracle.  All Rights Reserved.
> + * Author: Darrick J. Wong <djwong@kernel.org>
> + *
> + * Test program to try to trip over XFS_IOC_ALLOCSP mapping stale disk blocks
> + * into a file.
> + */
> +#include <xfs/xfs.h>
> +#include <stdlib.h>
> +#include <stdio.h>
> +#include <unistd.h>
> +#include <sys/types.h>
> +#include <sys/stat.h>
> +#include <fcntl.h>
> +#include <errno.h>
> +#include <unistd.h>
> +#include <string.h>
> +
> +#ifndef XFS_IOC_ALLOCSP
> +# define XFS_IOC_ALLOCSP	_IOW ('X', 10, struct xfs_flock64)
> +#endif
> +
> +int
> +main(
> +	int		argc,
> +	char		*argv[])
> +{
> +	struct stat	sb;
> +	char		*buf, *zeroes;
> +	unsigned long	i;
> +	unsigned long	iterations;
> +	int		fd, ret;
> +
> +	if (argc != 3) {
> +		fprintf(stderr, "Usage: %s filename iterations\n", argv[0]);
> +		return 1;
> +	}
> +
> +	errno = 0;
> +	iterations = strtoul(argv[2], NULL, 0);
> +	if (errno) {
> +		perror(argv[2]);
> +		return 1;
> +	}
> +
> +	fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0600);
> +	if (fd < 0) {
> +		perror(argv[1]);
> +		return 1;
> +	}
> +
> +	ret = fstat(fd, &sb);
> +	if (ret) {
> +		perror(argv[1]);
> +		return 1;
> +	}
> +
> +	buf = malloc(sb.st_blksize);
> +	if (!buf) {
> +		perror("pread buffer");
> +		return 1;
> +	}
> +
> +	zeroes = calloc(1, sb.st_blksize);
> +	if (!zeroes) {
> +		perror("zeroes buffer");
> +		return 1;
> +	}
> +
> +	for (i = 1; i <= iterations; i++) {
> +		struct xfs_flock64	arg = { };
> +		ssize_t			read_bytes;
> +		off_t			offset = sb.st_blksize * i;
> +
> +		/* Ensure the last block of the file is a hole... */
> +		ret = ftruncate(fd, offset - 1);
> +		if (ret) {
> +			perror("truncate");
> +			return 1;
> +		}
> +
> +		/*
> +		 * ...then use ALLOCSP to allocate the last block in the file.
> +		 * An unpatched kernel neglects to mark the new mapping
> +		 * unwritten or to zero the ondisk block, so...
> +		 */
> +		arg.l_whence = SEEK_SET;
> +		arg.l_start = offset;
> +		ret = ioctl(fd, XFS_IOC_ALLOCSP, &arg);
> +		if (ret < 0) {
> +			perror("ioctl");
> +			return 1;
> +		}
> +
> +		/* ... we can read old disk contents here. */
> +		read_bytes = pread(fd, buf, sb.st_blksize,
> +						offset - sb.st_blksize);
> +		if (read_bytes < 0) {
> +			perror(argv[1]);
> +			return 1;
> +		}
> +		if (read_bytes != sb.st_blksize) {
> +			fprintf(stderr, "%s: short read of %zd bytes\n",
> +					argv[1], read_bytes);
> +			return 1;
> +		}
> +
> +		if (memcmp(zeroes, buf, sb.st_blksize) != 0) {
> +			fprintf(stderr, "%s: found junk near offset %zd!\n",
> +					argv[1], offset - sb.st_blksize);
> +			return 2;
> +		}
> +	}
> +
> +	return 0;
> +}
> diff --git a/tests/xfs/832 b/tests/xfs/832
> new file mode 100755
> index 00000000..3820ff8c
> --- /dev/null
> +++ b/tests/xfs/832
> @@ -0,0 +1,56 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2022 Oracle.  All Rights Reserved.
> +#
> +# FS QA Test 832
> +#
> +# Regression test for commit:
> +#
> +# 983d8e60f508 ("xfs: map unwritten blocks in XFS_IOC_{ALLOC,FREE}SP just like fallocate")
> +#
> +. ./common/preamble
> +_begin_fstest auto quick prealloc
> +
> +# Import common functions.
> +. ./common/filter
> +
> +# real QA test starts here
> +
> +# Modify as appropriate.
> +_supported_fs xfs
> +_require_test
> +_require_scratch
> +
> +size_mb=32
> +# Write a known pattern to the disk so that we can detect stale disk blocks
> +# being mapped into the file.  In the test author's experience, the bug will
> +# reproduce within the first 500KB's worth of ALLOCSP calls, so running up
> +# to 16MB should suffice.
> +$XFS_IO_PROG -d -c "pwrite -S 0x58 -b 8m 0 ${size_mb}m" $SCRATCH_DEV > $seqres.full
> +MKFS_OPTIONS="-K $MKFS_OPTIONS" _scratch_mkfs_sized $((size_mb * 1048576)) >> $seqres.full
> +
> +_scratch_mount
> +
> +# Force the file to be created on the data device, which we pre-initialized
> +# with a known pattern.  The bug exists in the generic bmap code, so the choice
> +# of backing device does not matter, and ignoring the rt device gets us out of
> +# needing to detect things like rt extent size.
> +_xfs_force_bdev data $SCRATCH_MNT
> +testfile=$SCRATCH_MNT/a
> +
> +# Allow the test program to expand the file to consume half the free space.
> +blksz=$(_get_file_block_size $SCRATCH_MNT)
> +iterations=$(( (size_mb / 2) * 1048576 / blksz))
> +echo "Setting up $iterations runs for block size $blksz" >> $seqres.full
> +
> +# Run reproducer program and dump file contents if we see stale data.  Full
> +# details are in the source for the C program, but in a nutshell we run ALLOCSP
> +# one block at a time to see if it'll give us blocks full of 'X'es.
> +$here/src/allocstale $testfile $iterations
> +res=$?
> +test $res -eq 2 && od -tx1 -Ad -c $testfile
> +
> +# success, all done
> +echo Silence is golden
> +status=0
> +exit
> diff --git a/tests/xfs/832.out b/tests/xfs/832.out
> new file mode 100644
> index 00000000..bb8a6c12
> --- /dev/null
> +++ b/tests/xfs/832.out
> @@ -0,0 +1,2 @@
> +QA output created by 832
> +Silence is golden
>