[PATCH 1/2] xfs: fix incorrect argument count check

[PATCH 1/2] xfs: fix incorrect argument count check

[Ross Zwisler]

t_mmap_dio.c actually requires 4 arguments, not 3 as the current check
enforces:

usage: t_mmap_dio <src file> <dest file> <size> <msg>
open src(No such file or directory) len 0 (null)

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Fixes: 456581661b4d ("xfs: test per-inode DAX flag by IO")
---
 src/t_mmap_dio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/t_mmap_dio.c b/src/t_mmap_dio.c
index 69b9ca8..6c8ca1a 100644
--- a/src/t_mmap_dio.c
+++ b/src/t_mmap_dio.c
@@ -39,7 +39,7 @@ int main(int argc, char **argv)
 	char *dfile;
 	unsigned long len, opt;
 
-	if (argc < 4)
+	if (argc < 5)
 		usage(basename(argv[0]));
 
 	while ((opt = getopt(argc, argv, "b")) != -1)
-- 
2.9.3

[PATCH 2/2] dax: add regression test for stale mmap reads

[Ross Zwisler]

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@linux.intel.com>
---
 .gitignore            |  1 +
 src/Makefile          |  2 +-
 src/t_dax_stale_pmd.c | 56 ++++++++++++++++++++++++++++++++++++++++++
 tests/generic/427     | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/generic/427.out |  2 ++
 tests/generic/group   |  1 +
 6 files changed, 129 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..d0016eb
--- /dev/null
+++ b/src/t_dax_stale_pmd.c
@@ -0,0 +1,56 @@
+#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)))
+		err_exit("strncmp mismatch!");
+
+	close(fd);
+	return 0;
+}
diff --git a/tests/generic/427 b/tests/generic/427
new file mode 100755
index 0000000..baf1099
--- /dev/null
+++ b/tests/generic/427
@@ -0,0 +1,68 @@
+#! /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@linux.intel.com>
+#
+#-----------------------------------------------------------------------
+# 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 $seqres.full
+
+# Modify as appropriate.
+_supported_fs generic
+_supported_os Linux
+_require_scratch_dax
+_require_test_program "t_dax_stale_pmd"
+_require_user
+
+# real QA test starts here
+_scratch_mkfs >>$seqres.full 2>&1
+_scratch_mount "-o dax"
+
+$XFS_IO_PROG -f -c "falloc 0 4M" $SCRATCH_MNT/testfile >> $seqres.full 2>&1
+chmod 0644 $SCRATCH_MNT/testfile
+chown $qa_user $SCRATCH_MNT/testfile
+
+_user_do "src/t_dax_stale_pmd $SCRATCH_MNT/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

Re: [PATCH 2/2] dax: add regression test for stale mmap reads

[Eryu Guan]

On Mon, Apr 24, 2017 at 11:49:32AM -0600, Ross Zwisler wrote:
> This adds a regression test for the following kernel patch:
> 
>   dax: fix data corruption due to stale mmap reads
> 

Seems that this patch hasn't been merged into linus tree, thus 4.11-rc8
kernel should fail this test, but it passed for me, tested with 4.11-rc8
kernel on both ext4 and xfs, with both brd devices and pmem devices
created from "memmap=10G!5G memmap=15G!15G" kernel boot command line.
Did I miss anything?

# ./check -s ext4_pmem_4k generic/427
SECTION       -- ext4_pmem_4k
RECREATING    -- ext4 on /dev/pmem0
FSTYP         -- ext4
PLATFORM      -- Linux/x86_64 hp-dl360g9-15 4.11.0-rc8.kasan
MKFS_OPTIONS  -- -b 4096 /dev/pmem1
MOUNT_OPTIONS -- -o acl,user_xattr -o context=system_u:object_r:root_t:s0 /dev/pmem1 /scratch

generic/427 1s ... 1s
Ran: generic/427
Passed all 1 tests

Some comments inline.

> 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@linux.intel.com>
> ---
>  .gitignore            |  1 +
>  src/Makefile          |  2 +-
>  src/t_dax_stale_pmd.c | 56 ++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/427     | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/427.out |  2 ++
>  tests/generic/group   |  1 +
>  6 files changed, 129 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..d0016eb
> --- /dev/null
> +++ b/src/t_dax_stale_pmd.c
> @@ -0,0 +1,56 @@
> +#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");
                         ^^^^ Nitpick, the "op" should be "open"?
> +
> +	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)))
> +		err_exit("strncmp mismatch!");

strncmp doesn't set errno, this err_exit message might be confusing:
"strncmp mismatch!: Success"

> +
> +	close(fd);
> +	return 0;
> +}
> diff --git a/tests/generic/427 b/tests/generic/427
> new file mode 100755
> index 0000000..baf1099
> --- /dev/null
> +++ b/tests/generic/427
> @@ -0,0 +1,68 @@
> +#! /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@linux.intel.com>
> +#
> +#-----------------------------------------------------------------------
> +# 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_scratch_dax

I don't think dax is a requirement here, this test could run on normal
block device without "-o dax" option too. It won't hurt to run with more
test configurations. And test on nvdimm device with dax mount option
could be one of the test configs, e.g.

TEST_DEV=/dev/pmem0
SCRATCH_DEV=/dev/pmem1
MOUNT_OPTIONS="-o dax"
...

> +_require_test_program "t_dax_stale_pmd"
> +_require_user

_require_xfs_io_command "falloc"

So test _notrun on ext2/3.

> +
> +# real QA test starts here
> +_scratch_mkfs >>$seqres.full 2>&1
> +_scratch_mount "-o dax"

Same here, dax is not required.

> +
> +$XFS_IO_PROG -f -c "falloc 0 4M" $SCRATCH_MNT/testfile >> $seqres.full 2>&1
> +chmod 0644 $SCRATCH_MNT/testfile
> +chown $qa_user $SCRATCH_MNT/testfile

Any specific reason to use $qa_user to run this test? Comments would be
great.

Thanks,
Eryu

> +
> +_user_do "src/t_dax_stale_pmd $SCRATCH_MNT/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
> 

Re: [PATCH 2/2] dax: add regression test for stale mmap reads

[Ross Zwisler]

On Tue, Apr 25, 2017 at 07:27:39PM +0800, Eryu Guan wrote:
> On Mon, Apr 24, 2017 at 11:49:32AM -0600, Ross Zwisler wrote:
> > This adds a regression test for the following kernel patch:
> > 
> >   dax: fix data corruption due to stale mmap reads
> > 
> 
> Seems that this patch hasn't been merged into linus tree, thus 4.11-rc8
> kernel should fail this test, but it passed for me, tested with 4.11-rc8
> kernel on both ext4 and xfs, with both brd devices and pmem devices
> created from "memmap=10G!5G memmap=15G!15G" kernel boot command line.
> Did I miss anything?
> 
> # ./check -s ext4_pmem_4k generic/427
> SECTION       -- ext4_pmem_4k

Ooh, I didn't add this 'ext4_pmem_4k' section goodness, and it's not present
in the xfstests/master that I was using.  Do you have patches to add that?

> RECREATING    -- ext4 on /dev/pmem0
> FSTYP         -- ext4
> PLATFORM      -- Linux/x86_64 hp-dl360g9-15 4.11.0-rc8.kasan
> MKFS_OPTIONS  -- -b 4096 /dev/pmem1
> MOUNT_OPTIONS -- -o acl,user_xattr -o context=system_u:object_r:root_t:s0 /dev/pmem1 /scratch
> 
> generic/427 1s ... 1s
> Ran: generic/427
> Passed all 1 tests

Your memmap params look fine.  I tested with BRD and PMEM, and with EXT4 and
XFS, and all combinations failed for me as expected with v4.11-rc8.

One issue could have been that the test file already existed when the test was
run.  I wasn't removing it between runs earlier, but I've fixed that for v2.

Another issue I guess could have been that the hole that we got back from the
filesystem was smaller than 2MiB?  Can you try running v2 (which I'll post in
a second) against a TEST_DEV made with one of the following:

ext4: mkfs.ext4 -b 4096 -E stride=512 -F $TEST_DEV
xfs:  mkfs.xfs -f -d su=2m,sw=1 $TEST_DEV

This helps us get 2MiB sized and aligned allocations so we can fault in PMDs,
but I'm not sure whether or not it would matter for holes.

> Some comments inline.
> 
> > 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@linux.intel.com>
> > ---
> >  .gitignore            |  1 +
> >  src/Makefile          |  2 +-
> >  src/t_dax_stale_pmd.c | 56 ++++++++++++++++++++++++++++++++++++++++++
> >  tests/generic/427     | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++
> >  tests/generic/427.out |  2 ++
> >  tests/generic/group   |  1 +
> >  6 files changed, 129 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..d0016eb
> > --- /dev/null
> > +++ b/src/t_dax_stale_pmd.c
> > @@ -0,0 +1,56 @@
> > +#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");
>                          ^^^^ Nitpick, the "op" should be "open"?
> > +
> > +	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)))
> > +		err_exit("strncmp mismatch!");
> 
> strncmp doesn't set errno, this err_exit message might be confusing:
> "strncmp mismatch!: Success"

Ah, thanks, fixed in v2.

> > +
> > +	close(fd);
> > +	return 0;
> > +}
> > diff --git a/tests/generic/427 b/tests/generic/427
> > new file mode 100755
> > index 0000000..baf1099
> > --- /dev/null
> > +++ b/tests/generic/427
> > @@ -0,0 +1,68 @@
> > +#! /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@linux.intel.com>
> > +#
> > +#-----------------------------------------------------------------------
> > +# 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_scratch_dax
> 
> I don't think dax is a requirement here, this test could run on normal
> block device without "-o dax" option too. It won't hurt to run with more
> test configurations. And test on nvdimm device with dax mount option
> could be one of the test configs, e.g.
> 
> TEST_DEV=/dev/pmem0
> SCRATCH_DEV=/dev/pmem1
> MOUNT_OPTIONS="-o dax"
> ...

Yep, agreed, fixed in v2.

> > +_require_test_program "t_dax_stale_pmd"
> > +_require_user
> 
> _require_xfs_io_command "falloc"
> 
> So test _notrun on ext2/3.

Fixed in v2.

> > +
> > +# real QA test starts here
> > +_scratch_mkfs >>$seqres.full 2>&1
> > +_scratch_mount "-o dax"
> 
> Same here, dax is not required.

Fixed in v2.

> 
> > +
> > +$XFS_IO_PROG -f -c "falloc 0 4M" $SCRATCH_MNT/testfile >> $seqres.full 2>&1
> > +chmod 0644 $SCRATCH_MNT/testfile
> > +chown $qa_user $SCRATCH_MNT/testfile
> 
> Any specific reason to use $qa_user to run this test? Comments would be
> great.

Nope, just cargo-culting my way through my first xfstest. :)  I've removed
this for v2.

> Thanks,
> Eryu

Thanks for the review!

> > +
> > +_user_do "src/t_dax_stale_pmd $SCRATCH_MNT/testfile"
> > +
> > +# success, all done
> > +echo "Silence is golden"
> > +status=0
> > +exit
> > diff --git a/tests/generic/427.out b/tests/generic/427.out

Re: [PATCH 2/2] dax: add regression test for stale mmap reads

[Eryu Guan]

On Tue, Apr 25, 2017 at 02:39:11PM -0600, Ross Zwisler wrote:
> On Tue, Apr 25, 2017 at 07:27:39PM +0800, Eryu Guan wrote:
> > On Mon, Apr 24, 2017 at 11:49:32AM -0600, Ross Zwisler wrote:
> > > This adds a regression test for the following kernel patch:
> > > 
> > >   dax: fix data corruption due to stale mmap reads
> > > 
> > 
> > Seems that this patch hasn't been merged into linus tree, thus 4.11-rc8
> > kernel should fail this test, but it passed for me, tested with 4.11-rc8
> > kernel on both ext4 and xfs, with both brd devices and pmem devices
> > created from "memmap=10G!5G memmap=15G!15G" kernel boot command line.
> > Did I miss anything?
> > 
> > # ./check -s ext4_pmem_4k generic/427
> > SECTION       -- ext4_pmem_4k
> 
> Ooh, I didn't add this 'ext4_pmem_4k' section goodness, and it's not present
> in the xfstests/master that I was using.  Do you have patches to add that?

That's one of my config sections, it's all user-defined, not committed
to fstests repo :)

You can take a look at README.config-sections for more details. Here is
my local.config file for your reference

[default]
TEST_DEV=/dev/pmem0
SCRATCH_DEV=/dev/pmem1
TEST_DIR=/mnt
SCRATCH_MNT=/scratch

RECREATE_TEST_DEV=true

[xfs_pmem_4k]
FSTYP=xfs
MKFS_OPTIONS="-f -m crc=1 -b size=4k"

[ext4_pmem_4k]
FSTYP=ext4
MKFS_OPTIONS="-b 4096"

> 
> > RECREATING    -- ext4 on /dev/pmem0
> > FSTYP         -- ext4
> > PLATFORM      -- Linux/x86_64 hp-dl360g9-15 4.11.0-rc8.kasan
> > MKFS_OPTIONS  -- -b 4096 /dev/pmem1
> > MOUNT_OPTIONS -- -o acl,user_xattr -o context=system_u:object_r:root_t:s0 /dev/pmem1 /scratch
> > 
> > generic/427 1s ... 1s
> > Ran: generic/427
> > Passed all 1 tests
> 
> Your memmap params look fine.  I tested with BRD and PMEM, and with EXT4 and
> XFS, and all combinations failed for me as expected with v4.11-rc8.
> 
> One issue could have been that the test file already existed when the test was
> run.  I wasn't removing it between runs earlier, but I've fixed that for v2.
> 
> Another issue I guess could have been that the hole that we got back from the
> filesystem was smaller than 2MiB?  Can you try running v2 (which I'll post in
> a second) against a TEST_DEV made with one of the following:
> 
> ext4: mkfs.ext4 -b 4096 -E stride=512 -F $TEST_DEV
> xfs:  mkfs.xfs -f -d su=2m,sw=1 $TEST_DEV
> 
> This helps us get 2MiB sized and aligned allocations so we can fault in PMDs,
> but I'm not sure whether or not it would matter for holes.

I guess that's the point to reproduce the failure, I'll confirm with v2
patches. If these non-default & not widely tested mkfs options are
required to reproduce it, I think we can specify them in the test, as
extra mkfs options to _scratch_mkfs, as what generic/413 does.

> 
> > Some comments inline.
> > 
> > > 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@linux.intel.com>
> > > ---
> > >  .gitignore            |  1 +
> > >  src/Makefile          |  2 +-
> > >  src/t_dax_stale_pmd.c | 56 ++++++++++++++++++++++++++++++++++++++++++
> > >  tests/generic/427     | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++
> > >  tests/generic/427.out |  2 ++
> > >  tests/generic/group   |  1 +
> > >  6 files changed, 129 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..d0016eb
> > > --- /dev/null
> > > +++ b/src/t_dax_stale_pmd.c
> > > @@ -0,0 +1,56 @@
> > > +#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");
> >                          ^^^^ Nitpick, the "op" should be "open"?
> > > +
> > > +	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)))
> > > +		err_exit("strncmp mismatch!");
> > 
> > strncmp doesn't set errno, this err_exit message might be confusing:
> > "strncmp mismatch!: Success"
> 
> Ah, thanks, fixed in v2.
> 
> > > +
> > > +	close(fd);
> > > +	return 0;
> > > +}
> > > diff --git a/tests/generic/427 b/tests/generic/427
> > > new file mode 100755
> > > index 0000000..baf1099
> > > --- /dev/null
> > > +++ b/tests/generic/427
> > > @@ -0,0 +1,68 @@
> > > +#! /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@linux.intel.com>
> > > +#
> > > +#-----------------------------------------------------------------------
> > > +# 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_scratch_dax
> > 
> > I don't think dax is a requirement here, this test could run on normal
> > block device without "-o dax" option too. It won't hurt to run with more
> > test configurations. And test on nvdimm device with dax mount option
> > could be one of the test configs, e.g.
> > 
> > TEST_DEV=/dev/pmem0
> > SCRATCH_DEV=/dev/pmem1
> > MOUNT_OPTIONS="-o dax"
> > ...
> 
> Yep, agreed, fixed in v2.

Then perhaps the test program should be renamed? As no dax is required.
How about t_mmap_stale_pmd?

> 
> > > +_require_test_program "t_dax_stale_pmd"
> > > +_require_user
> > 
> > _require_xfs_io_command "falloc"
> > 
> > So test _notrun on ext2/3.
> 
> Fixed in v2.
> 
> > > +
> > > +# real QA test starts here
> > > +_scratch_mkfs >>$seqres.full 2>&1
> > > +_scratch_mount "-o dax"
> > 
> > Same here, dax is not required.
> 
> Fixed in v2.
> 
> > 
> > > +
> > > +$XFS_IO_PROG -f -c "falloc 0 4M" $SCRATCH_MNT/testfile >> $seqres.full 2>&1
> > > +chmod 0644 $SCRATCH_MNT/testfile
> > > +chown $qa_user $SCRATCH_MNT/testfile
> > 
> > Any specific reason to use $qa_user to run this test? Comments would be
> > great.
> 
> Nope, just cargo-culting my way through my first xfstest. :)  I've removed
> this for v2.

I think it's in a pretty good shape for "first fstests test" :)

> 
> > Thanks,
> > Eryu
> 
> Thanks for the review!

Thanks for adding new test!

Eryu

> 
> > > +
> > > +_user_do "src/t_dax_stale_pmd $SCRATCH_MNT/testfile"
> > > +
> > > +# success, all done
> > > +echo "Silence is golden"
> > > +status=0
> > > +exit
> > > diff --git a/tests/generic/427.out b/tests/generic/427.out
> --
> To unsubscribe from this list: send the line "unsubscribe fstests" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html