On Tue, Apr 25, 2017 at 02:51:06PM -0600, Ross Zwisler wrote:
> This adds a regression test for the following kernel patch:
>
> dax: fix data corruption due to stale mmap reads
>
> The above patch fixes an issue where users of DAX can suffer data
> corruption from stale mmap reads via the following sequence:
>
> - open an mmap over a 2MiB hole
>
> - read from a 2MiB hole, faulting in a 2MiB zero page
>
> - write to the hole with write(3p). The write succeeds but we incorrectly
> leave the 2MiB zero page mapping intact.
>
> - via the mmap, read the data that was just written. Since the zero page
> mapping is still intact we read back zeroes instead of the new data.
>
> Signed-off-by: Ross Zwisler <ross.zwisler@xxxxxxxxxxxxxxx>
I reproduced it now with 4.10 kernel, I think it was because I enalbed
KASAN on my 4.11-rc8 kernel. And the special mkfs options for ext4/xfs
are not needed either.
> ---
> .gitignore | 1 +
> src/Makefile | 2 +-
> src/t_dax_stale_pmd.c | 59 +++++++++++++++++++++++++++++++++++++++++++++
> tests/generic/427 | 67 +++++++++++++++++++++++++++++++++++++++++++++++++++
> tests/generic/427.out | 2 ++
> tests/generic/group | 1 +
> 6 files changed, 131 insertions(+), 1 deletion(-)
> create mode 100644 src/t_dax_stale_pmd.c
> create mode 100755 tests/generic/427
> create mode 100644 tests/generic/427.out
>
> diff --git a/.gitignore b/.gitignore
> index ded4a61..9664dc9 100644
> --- a/.gitignore
> +++ b/.gitignore
> @@ -134,6 +134,7 @@
> /src/renameat2
> /src/t_rename_overwrite
> /src/t_mmap_dio
> +/src/t_dax_stale_pmd
>
> # dmapi/ binaries
> /dmapi/src/common/cmd/read_invis
> diff --git a/src/Makefile b/src/Makefile
> index abfd873..7e22b50 100644
> --- a/src/Makefile
> +++ b/src/Makefile
> @@ -12,7 +12,7 @@ TARGETS = dirstress fill fill2 getpagesize holes lstat64 \
> godown resvtest writemod makeextents itrash rename \
> multi_open_unlink dmiperf unwritten_sync genhashnames t_holes \
> t_mmap_writev t_truncate_cmtime dirhash_collide t_rename_overwrite \
> - holetest t_truncate_self t_mmap_dio af_unix
> + holetest t_truncate_self t_mmap_dio af_unix t_dax_stale_pmd
>
> 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/t_dax_stale_pmd.c b/src/t_dax_stale_pmd.c
> new file mode 100644
> index 0000000..59fbbe1
> --- /dev/null
> +++ b/src/t_dax_stale_pmd.c
> @@ -0,0 +1,59 @@
> +#include <errno.h>
> +#include <fcntl.h>
> +#include <libgen.h>
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <sys/mman.h>
> +#include <sys/stat.h>
> +#include <sys/types.h>
> +#include <unistd.h>
> +
> +#define MiB(a) ((a)*1024*1024)
> +
> +void err_exit(char *op)
> +{
> + fprintf(stderr, "%s: %s\n", op, strerror(errno));
> + exit(1);
> +}
> +
> +int main(int argc, char *argv[])
> +{
> + volatile int a __attribute__((__unused__));
> + char *buffer = "HELLO WORLD!";
> + char *data;
> + int fd;
> +
> + if (argc < 2) {
> + printf("Usage: %s <pmem file>\n", basename(argv[0]));
> + exit(0);
> + }
> +
> + fd = open(argv[1], O_RDWR);
> + if (fd < 0)
> + err_exit("fd");
> +
> + data = mmap(NULL, MiB(2), PROT_READ, MAP_SHARED, fd, MiB(2));
> +
> + /*
> + * This faults in a 2MiB zero page to satisfy the read.
> + * 'a' is volatile so this read doesn't get optimized out.
> + */
> + a = data[0];
> +
> + pwrite(fd, buffer, strlen(buffer), MiB(2));
> +
> + /*
> + * Try and use the mmap to read back the data we just wrote with
> + * pwrite(). If the kernel bug is present the mapping from the 2MiB
> + * zero page will still be intact, and we'll read back zeros instead.
> + */
> + if (strncmp(buffer, data, strlen(buffer))) {
> + fprintf(stderr, "strncmp mismatch: '%s' vs '%s'\n", buffer,
> + data);
> + exit(1);
> + }
> +
> + close(fd);
> + return 0;
> +}
> diff --git a/tests/generic/427 b/tests/generic/427
> new file mode 100755
> index 0000000..6e265a1
> --- /dev/null
> +++ b/tests/generic/427
> @@ -0,0 +1,67 @@
> +#! /bin/bash
> +# FS QA Test 427
> +#
> +# This is a regression test for kernel patch:
> +# dax: fix data corruption due to stale mmap reads
> +# created by Ross Zwisler <ross.zwisler@xxxxxxxxxxxxxxx>
> +#
> +#-----------------------------------------------------------------------
> +# Copyright (c) 2017 Intel Corporation. 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
> +
> +# Modify as appropriate.
> +_supported_fs generic
> +_supported_os Linux
> +_require_test_program "t_dax_stale_pmd"
> +_require_xfs_io_command "falloc"
I'm wondering if falloc is really needed? If not, this test could be run
with ext2/3 too. See below.
> +_require_user
This is not needed anymore.
> +
> +# real QA test starts here
> +
> +# ensure we have no pre-existing block allocations, so we get a hole
> +rm -f $TEST_DIR/testfile
> +$XFS_IO_PROG -f -c "falloc 0 4M" $TEST_DIR/testfile >> $seqres.full 2>&1
I found that 'xfs_io -fc "truncate 4M" $TEST_DIR/testfile' works too,
from the comments in test and kernel patch, if I understand correctly,
we only need to mmap un-allocated blocks, right?
If truncate(2) works too, I think we can move truncate operation to the
t_dax_stale_pmd program too, because the whole truncate && mmap && read
sequence are logically together, this also avoids the confusion on why
testfile is in 4M size.
Thanks,
Eryu
> +chmod 0644 $TEST_DIR/testfile
> +
> +src/t_dax_stale_pmd $TEST_DIR/testfile
> +
> +# success, all done
> +echo "Silence is golden"
> +status=0
> +exit
> diff --git a/tests/generic/427.out b/tests/generic/427.out
> new file mode 100644
> index 0000000..61295e5
> --- /dev/null
> +++ b/tests/generic/427.out
> @@ -0,0 +1,2 @@
> +QA output created by 427
> +Silence is golden
> diff --git a/tests/generic/group b/tests/generic/group
> index f29009c..06f6e9d 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -429,3 +429,4 @@
> 424 auto quick
> 425 auto quick attr
> 426 auto quick exportfs
> +427 auto quick
> --
> 2.9.3
>
