linux-stable.git/drivers/md/dm.c, branch linux-6.4.y

dm: use op specific max_sectors when splitting abnormal io

2023-06-15T16:47:16+00:00

Split abnormal IO in terms of the corresponding operation specific
max_sectors (max_discard_sectors, max_secure_erase_sectors or
max_write_zeroes_sectors).

This fixes a significant dm-thinp discard performance regression that
was introduced with commit e2dd8aca2d76 ("dm bio prison v1: improve
concurrent IO performance"). Relative to discard: max_discard_sectors
is used instead of max_sectors; which fixes excessive discard splitting
(e.g. max_sectors=128K vs max_discard_sectors=64M).

Tested by discarding an 1 Petabyte dm-thin device:
lvcreate -V 1125899906842624B -T test/pool -n thin
time blkdiscard /dev/test/thin

Before this fix (splitting discards every 128K): ~116m
 After this fix (splitting discards every 64M) : 0m33.460s

Reported-by: Zorro Lang 
Fixes: 06961c487a33 ("dm: split discards further if target sets max_discard_granularity")
Requires: 13f6facf3fae ("dm: allow targets to require splitting WRITE_ZEROES and SECURE_ERASE")
Fixes: e2dd8aca2d76 ("dm bio prison v1: improve concurrent IO performance")
Signed-off-by: Mike Snitzer

dm: don't lock fs when the map is NULL during suspend or resume

2023-06-15T16:46:53+00:00

As described in commit 38d11da522aa ("dm: don't lock fs when the map is
NULL in process of resume"), a deadlock may be triggered between
do_resume() and do_mount().

This commit preserves the fix from commit 38d11da522aa but moves it to
where it also serves to fix a similar deadlock between do_suspend()
and do_mount().  It does so, if the active map is NULL, by clearing
DM_SUSPEND_LOCKFS_FLAG in dm_suspend() which is called by both
do_suspend() and do_resume().

Fixes: 38d11da522aa ("dm: don't lock fs when the map is NULL in process of resume")
Signed-off-by: Li Lingfeng 
Signed-off-by: Mike Snitzer

dm: unexport dm_get_queue_limits()

2023-04-14T16:32:20+00:00

There are no dm_get_queue_limits() callers outside of DM core and
there shouldn't be.

Also, remove its BUG_ON(!atomic_read(&md->holders)) to micro-optimize
__process_abnormal_io().

Signed-off-by: Mike Snitzer

dm: allow targets to require splitting WRITE_ZEROES and SECURE_ERASE

2023-04-14T16:07:27+00:00

Introduce max_write_zeroes_granularity and
max_secure_erase_granularity flags in the dm_target struct.

If a target sets these then DM core will split IO of these operation
types accordingly (in terms of max_write_zeroes_sectors and
max_secure_erase_sectors respectively).

Signed-off-by: Mike Snitzer

dm: split discards further if target sets max_discard_granularity

2023-03-30T19:57:50+00:00

The block core (bio_split_discard) will already split discards based
on the 'discard_granularity' and 'max_discard_sectors' queue_limits.
But the DM thin target also needs to ensure that it doesn't receive a
discard that spans a 'max_discard_sectors' boundary.

Introduce a dm_target 'max_discard_granularity' flag that if set will
cause DM core to split discard bios relative to 'max_discard_sectors'.
This treats 'discard_granularity' as a "min_discard_granularity" and
'max_discard_sectors' as a "max_discard_granularity".

Requested-by: Joe Thornber 
Signed-off-by: Mike Snitzer

dm: fix __send_duplicate_bios() to always allow for splitting IO

2023-03-30T19:54:32+00:00

Commit 7dd76d1feec70 ("dm: improve bio splitting and associated IO
accounting") only called setup_split_accounting() from
__send_duplicate_bios() if a single bio were being issued. But the case
where duplicate bios are issued must call it too.

Otherwise the bio won't be split and resubmitted (via recursion through
block core back to DM) to submit the later portions of a bio (which may
map to an entirely different target).

For example, when discarding an entire DM striped device with the
following DM table:
 vg-lvol0: 0 159744 striped 2 128 7:0 2048 7:1 2048
 vg-lvol0: 159744 45056 striped 2 128 7:2 2048 7:3 2048

Before (broken, discards the first striped target's devices twice):
 device-mapper: striped: target_stripe=0, bdev=7:0, start=2048 len=79872
 device-mapper: striped: target_stripe=1, bdev=7:1, start=2048 len=79872
 device-mapper: striped: target_stripe=0, bdev=7:0, start=2049 len=22528
 device-mapper: striped: target_stripe=1, bdev=7:1, start=2048 len=22528

After (works as expected):
 device-mapper: striped: target_stripe=0, bdev=7:0, start=2048 len=79872
 device-mapper: striped: target_stripe=1, bdev=7:1, start=2048 len=79872
 device-mapper: striped: target_stripe=0, bdev=7:2, start=2048 len=22528
 device-mapper: striped: target_stripe=1, bdev=7:3, start=2048 len=22528

Fixes: 7dd76d1feec70 ("dm: improve bio splitting and associated IO accounting")
Cc: stable@vger.kernel.org
Reported-by: Orange Kao 
Signed-off-by: Mike Snitzer

dm: fix improper splitting for abnormal bios

2023-03-30T19:54:32+00:00

"Abnormal" bios include discards, write zeroes and secure erase. By no
longer passing the calculated 'len' pointer, commit 7dd06a2548b2 ("dm:
allow dm_accept_partial_bio() for dm_io without duplicate bios") took a
senseless approach to disallowing dm_accept_partial_bio() from working
for duplicate bios processed using __send_duplicate_bios().

It inadvertently and incorrectly stopped the use of 'len' when
initializing a target's io (in alloc_tio). As such the resulting tio
could address more area of a device than it should.

For example, when discarding an entire DM striped device with the
following DM table:
 vg-lvol0: 0 159744 striped 2 128 7:0 2048 7:1 2048
 vg-lvol0: 159744 45056 striped 2 128 7:2 2048 7:3 2048

Before this fix:

 device-mapper: striped: target_stripe=0, bdev=7:0, start=2048 len=102400
 blkdiscard: attempt to access beyond end of device
 loop0: rw=2051, sector=2048, nr_sectors = 102400 limit=81920

 device-mapper: striped: target_stripe=1, bdev=7:1, start=2048 len=102400
 blkdiscard: attempt to access beyond end of device
 loop1: rw=2051, sector=2048, nr_sectors = 102400 limit=81920

After this fix;

 device-mapper: striped: target_stripe=0, bdev=7:0, start=2048 len=79872
 device-mapper: striped: target_stripe=1, bdev=7:1, start=2048 len=79872

Fixes: 7dd06a2548b2 ("dm: allow dm_accept_partial_bio() for dm_io without duplicate bios")
Cc: stable@vger.kernel.org
Reported-by: Orange Kao 
Signed-off-by: Mike Snitzer

Merge tag 'for-6.3/dm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

2023-03-24T21:20:48+00:00

Pull device mapper fixes from Mike Snitzer:

 - Fix DM thin to work as a swap device by using 'limit_swap_bios' DM
   target flag (initially added to allow swap to dm-crypt) to throttle
   the amount of outstanding swap bios.

 - Fix DM crypt soft lockup warnings by calling cond_resched() from the
   cpu intensive loop in dmcrypt_write().

 - Fix DM crypt to not access an uninitialized tasklet. This fix allows
   for consistent handling of IO completion, by _not_ needlessly punting
   to a workqueue when tasklets are not needed.

 - Fix DM core's alloc_dev() initialization for DM stats to check for
   and propagate alloc_percpu() failure.

* tag 'for-6.3/dm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm stats: check for and propagate alloc_percpu failure
  dm crypt: avoid accessing uninitialized tasklet
  dm crypt: add cond_resched() to dmcrypt_write()
  dm thin: fix deadlock when swapping to thin device

dm stats: check for and propagate alloc_percpu failure

2023-03-16T17:37:06+00:00

Check alloc_precpu()'s return value and return an error from
dm_stats_init() if it fails. Update alloc_dev() to fail if
dm_stats_init() does.

Otherwise, a NULL pointer dereference will occur in dm_stats_cleanup()
even if dm-stats isn't being actively used.

Fixes: fd2ed4d25270 ("dm: add statistics support")
Cc: stable@vger.kernel.org
Signed-off-by: Jiasheng Jiang 
Signed-off-by: Mike Snitzer

block: count 'ios' and 'sectors' when io is done for bio-based device

2023-03-15T15:25:04+00:00

While using iostat for raid, I observed very strange 'await'
occasionally, and turns out it's due to that 'ios' and 'sectors' is
counted in bdev_start_io_acct(), while 'nsecs' is counted in
bdev_end_io_acct(). I'm not sure why they are ccounted like that
but I think this behaviour is obviously wrong because user will get
wrong disk stats.

Fix the problem by counting 'ios' and 'sectors' when io is done, like
what rq-based device does.

Fixes: 394ffa503bc4 ("blk: introduce generic io stat accounting help function")
Signed-off-by: Yu Kuai 
Reviewed-by: Christoph Hellwig 
Link: https://lore.kernel.org/r/20230223091226.1135678-1-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe