It's possible for post-eof blocks to end up being used for direct I/O
writes. dio write performs an upfront unwritten extent allocation, sends
the dio and then updates the inode size (if necessary) on write
completion. If a file release occurs while a file extending dio write is
in flight, it is possible to mistake the post-eof blocks for speculative
preallocation and incorrectly truncate them from the inode. This means
that the resulting dio write completion can discover a hole and allocate
new blocks rather than perform unwritten extent conversion.
A kernel warning can be reproduced by generic/299 on XFS:
XFS: Assertion failed: tp->t_blk_res_used <= tp->t_blk_res, \
file: fs/xfs//xfs_trans.c, line: 309
The root cause is that xfs_free_eofblocks() uses i_size to truncate
post-eof blocks from the inode, but async, file extending direct writes
do not update i_size until write completion, long after inode locks are
dropped. Therefore, xfs_free_eofblocks() effectively truncates the inode
to the incorrect size.
Besides reproduce above kernel warning, the verification of written
data is an important distinction between this test and generic/299.
For cover this filesystem corruption testing, write this new case to
check data integrality manually, not only depend on a kernel warning.
To increase the test stress of aio-dio-eof-race, add two arguments to
this source code to change the file size will be written.
Signed-off-by: Zorro Lang <zlang@xxxxxxxxxx>
---
Hi,
V3 did below changes:
1) add "feature" require
2) add more comments for filling fs operation
3) remove unnecessary od command which is used to dump stale data
4) change a comment in aio-dio-eof-race.c, due to I've changed that code
Thanks reviewing from Eryu and Brian,
Zorro
src/aio-dio-regress/aio-dio-eof-race.c | 76 ++++++++++++++++++++++-------
tests/generic/999 | 87 ++++++++++++++++++++++++++++++++++
tests/generic/999.out | 2 +
tests/generic/group | 1 +
4 files changed, 148 insertions(+), 18 deletions(-)
create mode 100755 tests/generic/999
create mode 100644 tests/generic/999.out
diff --git a/src/aio-dio-regress/aio-dio-eof-race.c b/src/aio-dio-regress/aio-dio-eof-race.c
index 79fd928..bd71b03 100644
--- a/src/aio-dio-regress/aio-dio-eof-race.c
+++ b/src/aio-dio-regress/aio-dio-eof-race.c
@@ -29,11 +29,21 @@
#include <libaio.h>
-/* Sized to allow 4 x 512 AIOs */
-#define BUF_SIZE 2048
+unsigned long buf_size = 0;
+unsigned long size_MB = 0;
#define IO_PATTERN 0xab
void
+usage(char *progname)
+{
+ fprintf(stderr, "usage: %s [-s filesize] [-b bufsize] filename\n"
+ "\t-s filesize: specify the minimum file size"
+ "\t-b bufsize: buffer size",
+ progname);
+ exit(1);
+}
+
+void
dump_buffer(
void *buf,
off64_t offset,
@@ -83,24 +93,54 @@ int main(int argc, char *argv[])
struct iocb *iocbs[] = { &iocb1, &iocb2, &iocb3, &iocb4 };
void *buf;
struct stat statbuf;
- char cmp_buf[BUF_SIZE];
int fd, err = 0;
off_t eof;
+ int c;
+ char *cmp_buf = NULL;
+ char *filename = NULL;
+
+ while ((c = getopt(argc, argv, "s:b:")) != -1) {
+ char *endp;
+
+ switch (c) {
+ case 's': /* XXX MB size will be extended */
+ size_MB = strtol(optarg, &endp, 0);
+ break;
+ case 'b': /* buffer size */
+ buf_size = strtol(optarg, &endp, 0);
+ break;
+ default:
+ usage(argv[0]);
+ }
+ }
+
+ if (size_MB == 0) /* default size is 8MB */
+ size_MB = 8;
+ if (buf_size < 2048) /* default minimum buffer size is 2048 bytes */
+ buf_size = 2048;
+
+ if (optind == argc - 1)
+ filename = argv[optind];
+ else
+ usage(argv[0]);
+
+
- fd = open(argv[1], O_DIRECT | O_CREAT | O_TRUNC | O_RDWR, 0600);
+ fd = open(filename, O_DIRECT | O_CREAT | O_TRUNC | O_RDWR, 0600);
if (fd == -1) {
perror("open");
return 1;
}
- err = posix_memalign(&buf, getpagesize(), BUF_SIZE);
+ err = posix_memalign(&buf, getpagesize(), buf_size);
if (err) {
fprintf(stderr, "error %s during %s\n",
strerror(err),
"posix_memalign");
return 1;
}
- memset(cmp_buf, IO_PATTERN, BUF_SIZE);
+ cmp_buf = malloc(buf_size);
+ memset(cmp_buf, IO_PATTERN, buf_size);
err = io_setup(4, &ctx);
if (err) {
@@ -112,9 +152,9 @@ int main(int argc, char *argv[])
eof = 0;
- /* Keep extending until 8MB (fairly arbitrary) */
- while (eof < 8 * 1024 * 1024) {
- memset(buf, IO_PATTERN, BUF_SIZE);
+ /* Keep extending until size_MB */
+ while (eof < size_MB * 1024 * 1024) {
+ memset(buf, IO_PATTERN, buf_size);
fstat(fd, &statbuf);
eof = statbuf.st_size;
@@ -125,10 +165,10 @@ int main(int argc, char *argv[])
* management and stale block zeroing for races and can lead to
* data corruption when not handled properly.
*/
- io_prep_pwrite(&iocb1, fd, buf, BUF_SIZE/4, eof + 0*BUF_SIZE/4);
- io_prep_pwrite(&iocb2, fd, buf, BUF_SIZE/4, eof + 1*BUF_SIZE/4);
- io_prep_pwrite(&iocb3, fd, buf, BUF_SIZE/4, eof + 2*BUF_SIZE/4);
- io_prep_pwrite(&iocb4, fd, buf, BUF_SIZE/4, eof + 3*BUF_SIZE/4);
+ io_prep_pwrite(&iocb1, fd, buf, buf_size/4, eof + 0*buf_size/4);
+ io_prep_pwrite(&iocb2, fd, buf, buf_size/4, eof + 1*buf_size/4);
+ io_prep_pwrite(&iocb3, fd, buf, buf_size/4, eof + 2*buf_size/4);
+ io_prep_pwrite(&iocb4, fd, buf, buf_size/4, eof + 3*buf_size/4);
err = io_submit(ctx, 4, iocbs);
if (err != 4) {
@@ -150,20 +190,20 @@ int main(int argc, char *argv[])
* And then read it back.
*
* Using pread to keep it simple, but AIO has the same effect.
- * eof is the prior eof; we just wrote BUF_SIZE more.
+ * eof is the prior eof; we just wrote buf_size more.
*/
- if (pread(fd, buf, BUF_SIZE, eof) != BUF_SIZE) {
+ if (pread(fd, buf, buf_size, eof) != buf_size) {
perror("pread");
return 1;
}
/*
* We launched 4 AIOs which, stitched together, should write
- * a seamless BUF_SIZE worth of IO_PATTERN to the last block.
+ * a seamless buf_size worth of IO_PATTERN to the last block.
*/
- if (memcmp(buf, cmp_buf, BUF_SIZE)) {
+ if (memcmp(buf, cmp_buf, buf_size)) {
printf("corruption while extending from %ld\n", eof);
- dump_buffer(buf, 0, BUF_SIZE);
+ dump_buffer(buf, 0, buf_size);
return 1;
}
}
diff --git a/tests/generic/999 b/tests/generic/999
new file mode 100755
index 0000000..09bcc5f
--- /dev/null
+++ b/tests/generic/999
@@ -0,0 +1,87 @@
+#! /bin/bash
+# FS QA Test No. 999
+#
+# Try to trigger a race of free eofblocks and file extending dio writes.
+# A known bug of XFS has been fixed by "e4229d6 xfs: fix eofblocks race
+# with file extending async dio writes"
+#
+#-----------------------------------------------------------------------
+# 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/$$
+status=1 # failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+ cd /
+ rm -f $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
+
+# Modify as appropriate.
+_supported_fs generic
+_supported_os Linux
+_require_scratch
+_require_test_program "feature"
+_require_aiodio aio-dio-eof-race
+
+# limit the filesystem size, to save the time of filling filesystem
+_scratch_mkfs_sized $((256 * 1024 * 1024)) >>$seqres.full 2>&1
+_scratch_mount
+
+# try to write more bytes than filesystem size to fill the filesystem,
+# then remove all these data. If we still can find these stale data in
+# a file' eofblock, then it's a bug
+$XFS_IO_PROG -f -c "pwrite -S 0x55 0 $((256 * 1024 * 1024 * 2))" \
+ $SCRATCH_MNT/fillfs-$seq 2>/dev/null
+rm -f $SCRATCH_MNT/fillfs-$seq
+
+# open & close the file frequently, to trigger xfs_free_eofblocks
+while true; do
+ $XFS_IO_PROG -f -c open $SCRATCH_MNT/tst-aio-dio-eof-race.$seq \
+ >/dev/null 2>&1
+done &
+open_close_pid=$!
+
+nr_cpu=`$here/src/feature -o`
+fsize=$((nr_cpu * 10))
+if [ $fsize -gt 200 ]; then
+ fsize=200
+fi
+# start a background aio writer, which does several extending loops
+# internally and check data integrality
+$AIO_TEST -s $fsize -b 65536 $SCRATCH_MNT/tst-aio-dio-eof-race.$seq
+status=$?
+
+kill $open_close_pid
+wait $open_close_pid
+exit
diff --git a/tests/generic/999.out b/tests/generic/999.out
new file mode 100644
index 0000000..6362b86
--- /dev/null
+++ b/tests/generic/999.out
@@ -0,0 +1,2 @@
+QA output created by 999
+Success, all done.
diff --git a/tests/generic/group b/tests/generic/group
index 6d6e4f6..aa99a05 100644
--- a/tests/generic/group
+++ b/tests/generic/group
@@ -428,3 +428,4 @@
423 auto quick
424 auto quick
425 auto quick attr
+999 auto quick aio rw
--
2.7.4
--
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
