| Message ID | 20171025204704.3382-1-ross.zwisler@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Oct 25, 2017 at 02:47:04PM -0600, Ross Zwisler wrote: > Add a test that exercises DAX's new MAP_SYNC flag. > > This test creates a file and writes to it via an mmap(), but never syncs > via fsync/msync. This process is tracked via dm-log-writes, then replayed. > > If MAP_SYNC is working the dm-log-writes replay will show the test file > with 1MiB of on-media block allocations. This is because each allocating > page fault included an implicit metadata sync. If MAP_SYNC isn't working > (which you can test by fiddling with the parameters to mmap()) the file > will be smaller or missing entirely. > > Note that dm-log-writes doesn't track the data that we write via the > mmap(), so we can't do any data integrity checking. We can only verify > that the metadata writes for the page faults happened. > > Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> ..... > --- /dev/null > +++ b/src/t_map_sync.c > @@ -0,0 +1,92 @@ > +#include <errno.h> > +#include <fcntl.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) > + > +/* > + * These two defines were added to the kernel via commits entitled > + * "mm: Define MAP_SYNC and VM_SYNC flags" and > + * "mm: introduce MAP_SHARED_VALIDATE, a mechanism to safely define new mmap > + * flags", respectively. > + */ > +#ifndef MAP_SYNC > +#define MAP_SYNC 0x80000 > +#endif > + > +#ifndef MAP_SHARED_VALIDATE > +#define MAP_SHARED_VALIDATE 0x3 > +#endif Autoconf rules for detecting supported functionality, please... > + > +void err_exit(char *op) > +{ > + fprintf(stderr, "%s: %s\n", op, strerror(errno)); > + exit(1); > +} > + > +void mark_log(char *logwrites_name, char *mark_name) > +{ > + char command[256]; > + > + snprintf(command, 256, "dmsetup message %s 0 mark %s", > + logwrites_name, mark_name); > + > + if (system(command)) > + err_exit("mark_log"); > +} > + > +int main(int argc, char *argv[]) > +{ > + int page_size = getpagesize(); > + int len = MiB(1); > + int i, fd, err; > + char *data; > + > + if (argc < 4) { > + printf("Usage: %s <file> <logwrites_name> <mark_name>\n", > + basename(argv[0])); > + exit(0); > + } > + > + fd = open(argv[1], O_RDWR|O_CREAT|O_TRUNC, S_IRUSR|S_IWUSR); > + if (fd < 0) > + err_exit("fd"); > + > + ftruncate(fd, len); > + > + data = mmap(NULL, len, PROT_READ|PROT_WRITE, > + MAP_SHARED_VALIDATE|MAP_SYNC, fd, 0); > + if (data == MAP_FAILED) > + err_exit("mmap"); As I say to all these sorts of one-off test prgrams: please add the new MAP_SYNC flag to xfs_io rather than writing a one-off test program to set it and write some data. And if we're going to be adding special custom tests just because we need to insert dm-log marks, add that functionality to xfs_io, too. That way we can create complex custom dm logwrite tests without needing one-off test programs for them all... > +#! /bin/bash > +# FS QA Test No. 466 > +# > +# Use md_log_writes to verify that MAP_SYNC actually syncs metadata during dm_log_writes? > +# We should see $SCRATCH_MNT/test as having 1MiB in block allocations > +du -sh $SCRATCH_MNT/test | _filter_scratch | _filter_spaces Perhaps stat -c %b $SCRATCH_MNT/test ? Cheers, Dave.
On Wed, Oct 25, 2017 at 11:47 PM, Ross Zwisler <ross.zwisler@linux.intel.com> wrote: > Add a test that exercises DAX's new MAP_SYNC flag. > > This test creates a file and writes to it via an mmap(), but never syncs > via fsync/msync. This process is tracked via dm-log-writes, then replayed. > > If MAP_SYNC is working the dm-log-writes replay will show the test file > with 1MiB of on-media block allocations. This is because each allocating > page fault included an implicit metadata sync. If MAP_SYNC isn't working > (which you can test by fiddling with the parameters to mmap()) the file > will be smaller or missing entirely. > > Note that dm-log-writes doesn't track the data that we write via the > mmap(), so we can't do any data integrity checking. We can only verify > that the metadata writes for the page faults happened. > > Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> > > --- > > Changes since v2: > - Fixed _require_log_writes() so that DAX will be disallowed if the > version of the dm-log-writes target is older than v1.1.0. (Amir) It seems like your kernel patch bumped the version to 1.0.1... > > --- > .gitignore | 1 + > common/dmlogwrites | 4 ++- > common/rc | 6 ++-- > src/Makefile | 3 +- > src/t_map_sync.c | 92 +++++++++++++++++++++++++++++++++++++++++++++++++++ > tests/generic/466 | 76 ++++++++++++++++++++++++++++++++++++++++++ > tests/generic/466.out | 3 ++ > tests/generic/group | 1 + > 8 files changed, 182 insertions(+), 4 deletions(-) > create mode 100644 src/t_map_sync.c > create mode 100755 tests/generic/466 > create mode 100644 tests/generic/466.out > > diff --git a/.gitignore b/.gitignore > index 2014c08..9fc0695 100644 > --- a/.gitignore > +++ b/.gitignore > @@ -119,6 +119,7 @@ > /src/t_getcwd > /src/t_holes > /src/t_immutable > +/src/t_map_sync > /src/t_mmap_cow_race > /src/t_mmap_dio > /src/t_mmap_fallocate > diff --git a/common/dmlogwrites b/common/dmlogwrites > index 247c744..71d008d 100644 > --- a/common/dmlogwrites > +++ b/common/dmlogwrites > @@ -23,8 +23,10 @@ _require_log_writes() > [ -z "$LOGWRITES_DEV" -o ! -b "$LOGWRITES_DEV" ] && \ > _notrun "This test requires a valid \$LOGWRITES_DEV" > > - _exclude_scratch_mount_option dax > _require_dm_target log-writes > + if [[ ${DMTARGET_VER[0]} == 1 && ${DMTARGET_VER[1]} < 1 ]]; then > + _exclude_scratch_mount_option dax > + fi IMO, this would be better as: local z=0 _scratch_has_mount_option dax && z=1 _require_dm_target log-writes 1 0 $z Or something like that
On Thu, Oct 26, 2017 at 08:56:38AM +1100, Dave Chinner wrote: > On Wed, Oct 25, 2017 at 02:47:04PM -0600, Ross Zwisler wrote: > > Add a test that exercises DAX's new MAP_SYNC flag. > > > > This test creates a file and writes to it via an mmap(), but never syncs > > via fsync/msync. This process is tracked via dm-log-writes, then replayed. > > > > If MAP_SYNC is working the dm-log-writes replay will show the test file > > with 1MiB of on-media block allocations. This is because each allocating > > page fault included an implicit metadata sync. If MAP_SYNC isn't working > > (which you can test by fiddling with the parameters to mmap()) the file > > will be smaller or missing entirely. > > > > Note that dm-log-writes doesn't track the data that we write via the > > mmap(), so we can't do any data integrity checking. We can only verify > > that the metadata writes for the page faults happened. > > > > Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> > ..... > > --- /dev/null > > +++ b/src/t_map_sync.c > > @@ -0,0 +1,92 @@ > > +#include <errno.h> > > +#include <fcntl.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) > > + > > +/* > > + * These two defines were added to the kernel via commits entitled > > + * "mm: Define MAP_SYNC and VM_SYNC flags" and > > + * "mm: introduce MAP_SHARED_VALIDATE, a mechanism to safely define new mmap > > + * flags", respectively. > > + */ > > +#ifndef MAP_SYNC > > +#define MAP_SYNC 0x80000 > > +#endif > > + > > +#ifndef MAP_SHARED_VALIDATE > > +#define MAP_SHARED_VALIDATE 0x3 > > +#endif > > Autoconf rules for detecting supported functionality, please... Yep, that's better. As you've suggested down below I'm adding this functionality to xfs_io instead, and I've added autoconf rules there. > > + > > +void err_exit(char *op) > > +{ > > + fprintf(stderr, "%s: %s\n", op, strerror(errno)); > > + exit(1); > > +} > > + > > +void mark_log(char *logwrites_name, char *mark_name) > > +{ > > + char command[256]; > > + > > + snprintf(command, 256, "dmsetup message %s 0 mark %s", > > + logwrites_name, mark_name); > > + > > + if (system(command)) > > + err_exit("mark_log"); > > +} > > + > > +int main(int argc, char *argv[]) > > +{ > > + int page_size = getpagesize(); > > + int len = MiB(1); > > + int i, fd, err; > > + char *data; > > + > > + if (argc < 4) { > > + printf("Usage: %s <file> <logwrites_name> <mark_name>\n", > > + basename(argv[0])); > > + exit(0); > > + } > > + > > + fd = open(argv[1], O_RDWR|O_CREAT|O_TRUNC, S_IRUSR|S_IWUSR); > > + if (fd < 0) > > + err_exit("fd"); > > + > > + ftruncate(fd, len); > > + > > + data = mmap(NULL, len, PROT_READ|PROT_WRITE, > > + MAP_SHARED_VALIDATE|MAP_SYNC, fd, 0); > > + if (data == MAP_FAILED) > > + err_exit("mmap"); > > As I say to all these sorts of one-off test prgrams: please add the > new MAP_SYNC flag to xfs_io rather than writing a one-off > test program to set it and write some data. > > And if we're going to be adding special custom tests just because > we need to insert dm-log marks, add that functionality to xfs_io, > too. > > That way we can create complex custom dm logwrite tests without > needing one-off test programs for them all... Yep, that was a better path. I've got things working - need to clean up and I'll send out soon. > > +#! /bin/bash > > +# FS QA Test No. 466 > > +# > > +# Use md_log_writes to verify that MAP_SYNC actually syncs metadata during > > dm_log_writes? Fixed. > > +# We should see $SCRATCH_MNT/test as having 1MiB in block allocations > > +du -sh $SCRATCH_MNT/test | _filter_scratch | _filter_spaces > > Perhaps stat -c %b $SCRATCH_MNT/test ? Maybe, but doesn't the output of 'stat -c %b' depend on the block size the filesystem is using? I think to use stat I'd have to check both %b and %B, and account for different block sizes, or do some shell math. I think it may be easier to just use du. Thank you for the review, sorry for the delayed response.
On Thu, Nov 16, 2017 at 02:28:15PM -0700, Ross Zwisler wrote: > On Thu, Oct 26, 2017 at 08:56:38AM +1100, Dave Chinner wrote: > > Perhaps stat -c %b $SCRATCH_MNT/test ? > > Maybe, but doesn't the output of 'stat -c %b' depend on the block size the > filesystem is using? I think to use stat I'd have to check both %b and %B, > and account for different block sizes, or do some shell math. I think it may > be easier to just use du. Ah, never mind, I think you meant 'stat -c %s' - that's cleaner, will switch.
On Thu, Oct 26, 2017 at 07:59:39AM +0300, Amir Goldstein wrote: > On Wed, Oct 25, 2017 at 11:47 PM, Ross Zwisler > <ross.zwisler@linux.intel.com> wrote: > > Add a test that exercises DAX's new MAP_SYNC flag. > > > > This test creates a file and writes to it via an mmap(), but never syncs > > via fsync/msync. This process is tracked via dm-log-writes, then replayed. > > > > If MAP_SYNC is working the dm-log-writes replay will show the test file > > with 1MiB of on-media block allocations. This is because each allocating > > page fault included an implicit metadata sync. If MAP_SYNC isn't working > > (which you can test by fiddling with the parameters to mmap()) the file > > will be smaller or missing entirely. > > > > Note that dm-log-writes doesn't track the data that we write via the > > mmap(), so we can't do any data integrity checking. We can only verify > > that the metadata writes for the page faults happened. > > > > Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> > > > > --- > > > > Changes since v2: > > - Fixed _require_log_writes() so that DAX will be disallowed if the > > version of the dm-log-writes target is older than v1.1.0. (Amir) > > It seems like your kernel patch bumped the version to 1.0.1... Ah, yep, that's the version number that I submitted but Mike changed it to v1.1.0 in his tree. Here's the patch that was merged for v4.15-rc1: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers?id=98d82f48f1983ceef5c8d2f6c87bfee2918790ee > > > > --- > > .gitignore | 1 + > > common/dmlogwrites | 4 ++- > > common/rc | 6 ++-- > > src/Makefile | 3 +- > > src/t_map_sync.c | 92 +++++++++++++++++++++++++++++++++++++++++++++++++++ > > tests/generic/466 | 76 ++++++++++++++++++++++++++++++++++++++++++ > > tests/generic/466.out | 3 ++ > > tests/generic/group | 1 + > > 8 files changed, 182 insertions(+), 4 deletions(-) > > create mode 100644 src/t_map_sync.c > > create mode 100755 tests/generic/466 > > create mode 100644 tests/generic/466.out > > > > diff --git a/.gitignore b/.gitignore > > index 2014c08..9fc0695 100644 > > --- a/.gitignore > > +++ b/.gitignore > > @@ -119,6 +119,7 @@ > > /src/t_getcwd > > /src/t_holes > > /src/t_immutable > > +/src/t_map_sync > > /src/t_mmap_cow_race > > /src/t_mmap_dio > > /src/t_mmap_fallocate > > diff --git a/common/dmlogwrites b/common/dmlogwrites > > index 247c744..71d008d 100644 > > --- a/common/dmlogwrites > > +++ b/common/dmlogwrites > > @@ -23,8 +23,10 @@ _require_log_writes() > > [ -z "$LOGWRITES_DEV" -o ! -b "$LOGWRITES_DEV" ] && \ > > _notrun "This test requires a valid \$LOGWRITES_DEV" > > > > - _exclude_scratch_mount_option dax > > _require_dm_target log-writes > > + if [[ ${DMTARGET_VER[0]} == 1 && ${DMTARGET_VER[1]} < 1 ]]; then > > + _exclude_scratch_mount_option dax > > + fi > > IMO, this would be better as: > > local z=0 > _scratch_has_mount_option dax && z=1 > _require_dm_target log-writes 1 0 $z > > Or something like that Yep, this is nicer. Fixed. Thanks for the review.
diff --git a/.gitignore b/.gitignore index 2014c08..9fc0695 100644 --- a/.gitignore +++ b/.gitignore @@ -119,6 +119,7 @@ /src/t_getcwd /src/t_holes /src/t_immutable +/src/t_map_sync /src/t_mmap_cow_race /src/t_mmap_dio /src/t_mmap_fallocate diff --git a/common/dmlogwrites b/common/dmlogwrites index 247c744..71d008d 100644 --- a/common/dmlogwrites +++ b/common/dmlogwrites @@ -23,8 +23,10 @@ _require_log_writes() [ -z "$LOGWRITES_DEV" -o ! -b "$LOGWRITES_DEV" ] && \ _notrun "This test requires a valid \$LOGWRITES_DEV" - _exclude_scratch_mount_option dax _require_dm_target log-writes + if [[ ${DMTARGET_VER[0]} == 1 && ${DMTARGET_VER[1]} < 1 ]]; then + _exclude_scratch_mount_option dax + fi _require_test_program "log-writes/replay-log" } diff --git a/common/rc b/common/rc index e2a8229..df438c3 100644 --- a/common/rc +++ b/common/rc @@ -1792,8 +1792,10 @@ _require_dm_target() modprobe dm-$_target >/dev/null 2>&1 - $DMSETUP_PROG targets 2>&1 | grep -q ^$_target - if [ $? -ne 0 ]; then + _version=$($DMSETUP_PROG targets 2>&1 | grep ^$_target) + if [[ $_version =~ .*v([0-9]+).([0-9]+).([0-9]+) ]]; then + DMTARGET_VER=(${BASH_REMATCH[@]:1}) + else _notrun "This test requires dm $_target support" fi } diff --git a/src/Makefile b/src/Makefile index 3eb25b1..af7e7e9 100644 --- a/src/Makefile +++ b/src/Makefile @@ -13,7 +13,8 @@ TARGETS = dirstress fill fill2 getpagesize holes lstat64 \ 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 t_mmap_stale_pmd \ - t_mmap_cow_race t_mmap_fallocate fsync-err t_mmap_write_ro + t_mmap_cow_race t_mmap_fallocate fsync-err t_mmap_write_ro \ + t_map_sync 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_map_sync.c b/src/t_map_sync.c new file mode 100644 index 0000000..29b546a --- /dev/null +++ b/src/t_map_sync.c @@ -0,0 +1,92 @@ +#include <errno.h> +#include <fcntl.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) + +/* + * These two defines were added to the kernel via commits entitled + * "mm: Define MAP_SYNC and VM_SYNC flags" and + * "mm: introduce MAP_SHARED_VALIDATE, a mechanism to safely define new mmap + * flags", respectively. + */ +#ifndef MAP_SYNC +#define MAP_SYNC 0x80000 +#endif + +#ifndef MAP_SHARED_VALIDATE +#define MAP_SHARED_VALIDATE 0x3 +#endif + +void err_exit(char *op) +{ + fprintf(stderr, "%s: %s\n", op, strerror(errno)); + exit(1); +} + +void mark_log(char *logwrites_name, char *mark_name) +{ + char command[256]; + + snprintf(command, 256, "dmsetup message %s 0 mark %s", + logwrites_name, mark_name); + + if (system(command)) + err_exit("mark_log"); +} + +int main(int argc, char *argv[]) +{ + int page_size = getpagesize(); + int len = MiB(1); + int i, fd, err; + char *data; + + if (argc < 4) { + printf("Usage: %s <file> <logwrites_name> <mark_name>\n", + basename(argv[0])); + exit(0); + } + + fd = open(argv[1], O_RDWR|O_CREAT|O_TRUNC, S_IRUSR|S_IWUSR); + if (fd < 0) + err_exit("fd"); + + ftruncate(fd, len); + + data = mmap(NULL, len, PROT_READ|PROT_WRITE, + MAP_SHARED_VALIDATE|MAP_SYNC, fd, 0); + if (data == MAP_FAILED) + err_exit("mmap"); + + /* + * We intentionally don't sync 'fd' manually. If MAP_SYNC is working + * these allocating page faults will cause the filesystem to sync its + * metadata so that when we replay the dm-log-writes log the test file + * will have 1 MiB worth of block allocations. + * + * dm-log-writes doesn't track the data that we write via the mmap(), + * so we can't check that, we can only verify that the metadata writes + * happened. + */ + for (i = 0; i < len; i+=page_size) + data[i] = 0xff; + + mark_log(argv[2], argv[3]); + + err = munmap(data, len); + if (err < 0) + err_exit("munmap"); + + err = close(fd); + if (err < 0) + err_exit("close"); + + return 0; +} diff --git a/tests/generic/466 b/tests/generic/466 new file mode 100755 index 0000000..592213f --- /dev/null +++ b/tests/generic/466 @@ -0,0 +1,76 @@ +#! /bin/bash +# FS QA Test No. 466 +# +# Use md_log_writes to verify that MAP_SYNC actually syncs metadata during +# page faults. +# +#----------------------------------------------------------------------- +# 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` +status=1 # failure is the default! +trap "_cleanup; exit \$status" 0 1 2 3 15 + +_cleanup() +{ + _log_writes_cleanup +} + +# get standard environment, filters and checks +. ./common/rc +. ./common/filter +. ./common/dmlogwrites + +# remove previous $seqres.full before test +rm -f $seqres.full + +# real QA test starts here +_supported_fs generic +_supported_os Linux +_require_log_writes +_require_scratch_dax +_require_test_program "t_map_sync" + +_log_writes_init +_log_writes_mkfs >> $seqres.full 2>&1 +_log_writes_mount -o dax + +src/t_map_sync $SCRATCH_MNT/test $LOGWRITES_NAME pre_unmap + +# Unmount the scratch dir and tear down the log writes target +_log_writes_unmount +_log_writes_remove +_check_scratch_fs + +# check pre-unmap state +_log_writes_replay_log pre_unmap +_scratch_mount + +# We should see $SCRATCH_MNT/test as having 1MiB in block allocations +du -sh $SCRATCH_MNT/test | _filter_scratch | _filter_spaces + +_scratch_unmount +_check_scratch_fs + +echo "Silence is golden" +status=0 +exit diff --git a/tests/generic/466.out b/tests/generic/466.out new file mode 100644 index 0000000..4c54d7a --- /dev/null +++ b/tests/generic/466.out @@ -0,0 +1,3 @@ +QA output created by 466 +1.0M SCRATCH_MNT/test +Silence is golden diff --git a/tests/generic/group b/tests/generic/group index fbe0a7f..65328c6 100644 --- a/tests/generic/group +++ b/tests/generic/group @@ -468,3 +468,4 @@ 463 auto quick clone dangerous 464 auto rw 465 auto rw quick aio +466 auto quick dax
Add a test that exercises DAX's new MAP_SYNC flag. This test creates a file and writes to it via an mmap(), but never syncs via fsync/msync. This process is tracked via dm-log-writes, then replayed. If MAP_SYNC is working the dm-log-writes replay will show the test file with 1MiB of on-media block allocations. This is because each allocating page fault included an implicit metadata sync. If MAP_SYNC isn't working (which you can test by fiddling with the parameters to mmap()) the file will be smaller or missing entirely. Note that dm-log-writes doesn't track the data that we write via the mmap(), so we can't do any data integrity checking. We can only verify that the metadata writes for the page faults happened. Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com> --- Changes since v2: - Fixed _require_log_writes() so that DAX will be disallowed if the version of the dm-log-writes target is older than v1.1.0. (Amir) --- .gitignore | 1 + common/dmlogwrites | 4 ++- common/rc | 6 ++-- src/Makefile | 3 +- src/t_map_sync.c | 92 +++++++++++++++++++++++++++++++++++++++++++++++++++ tests/generic/466 | 76 ++++++++++++++++++++++++++++++++++++++++++ tests/generic/466.out | 3 ++ tests/generic/group | 1 + 8 files changed, 182 insertions(+), 4 deletions(-) create mode 100644 src/t_map_sync.c create mode 100755 tests/generic/466 create mode 100644 tests/generic/466.out