md: add a mddev_trace_remap helper

Add a helper to trace bio remapping that hides some argument
dereferences and the check for a DM-mapped MD device.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed--by: Song Liu <song@kernel.org>
Tested-by: Song Liu <song@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240303140150.5435-2-hch@lst.de

Merge branch 'dmraid-fix-6.9' into md-6.9

This is the second half of fixes for dmraid. The first half is available
at [1].

This set contains fixes:
 - reshape can start unexpected, cause data corruption, patch 1,5,6;
 - deadlocks that reshape concurrent with IO, patch 8;
 - a lockdep warning, patch 9;

For all the dmraid related tests in lvm2 suite, there is no new
regressions compared against 6.6 kernels (which is good baseline before
recent regressions).

[1] https://lore.kernel.org/all/CAPhsuW7u1UKHCDOBDhD7DzOVtkGemDz_QnJ4DUq_kSN-Q3G66Q@mail.gmail.com/

* dmraid-fix-6.9:
  dm-raid: fix lockdep waring in "pers->hot_add_disk"
  dm-raid456, md/raid456: fix a deadlock for dm-raid456 while io concurrent with reshape
  dm-raid: add a new helper prepare_suspend() in md_personality
  md/dm-raid: don't call md_reap_sync_thread() directly
  dm-raid: really frozen sync_thread during suspend
  md: add a new helper reshape_interrupted()
  md: export helper md_is_rdwr()
  md: export helpers to stop sync_thread
  md: don't clear MD_RECOVERY_FROZEN for new dm-raid until resume

dm-raid: fix lockdep waring in "pers->hot_add_disk"

The lockdep assert is added by commit a448af25becf ("md/raid10: remove
rcu protection to access rdev from conf") in print_conf(). And I didn't
notice that dm-raid is calling "pers->hot_add_disk" without holding
'reconfig_mutex'.

"pers->hot_add_disk" read and write many fields that is protected by
'reconfig_mutex', and raid_resume() already grab the lock in other
contex. Hence fix this problem by protecting "pers->host_add_disk"
with the lock.

Fixes: 9092c02d9435 ("DM RAID: Add ability to restore transiently failed devices on resume")
Fixes: a448af25becf ("md/raid10: remove rcu protection to access rdev from conf")
Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-10-yukuai1@huaweicloud.com

dm-raid456, md/raid456: fix a deadlock for dm-raid456 while io concurrent with reshape

For raid456, if reshape is still in progress, then IO across reshape
position will wait for reshape to make progress. However, for dm-raid,
in following cases reshape will never make progress hence IO will hang:

1) the array is read-only;
2) MD_RECOVERY_WAIT is set;
3) MD_RECOVERY_FROZEN is set;

After commit c467e97f079f ("md/raid6: use valid sector values to determine
if an I/O should wait on the reshape") fix the problem that IO across
reshape position doesn't wait for reshape, the dm-raid test
shell/lvconvert-raid-reshape.sh start to hang:

[root@fedora ~]# cat /proc/979/stack
[<0>] wait_woken+0x7d/0x90
[<0>] raid5_make_request+0x929/0x1d70 [raid456]
[<0>] md_handle_request+0xc2/0x3b0 [md_mod]
[<0>] raid_map+0x2c/0x50 [dm_raid]
[<0>] __map_bio+0x251/0x380 [dm_mod]
[<0>] dm_submit_bio+0x1f0/0x760 [dm_mod]
[<0>] __submit_bio+0xc2/0x1c0
[<0>] submit_bio_noacct_nocheck+0x17f/0x450
[<0>] submit_bio_noacct+0x2bc/0x780
[<0>] submit_bio+0x70/0xc0
[<0>] mpage_readahead+0x169/0x1f0
[<0>] blkdev_readahead+0x18/0x30
[<0>] read_pages+0x7c/0x3b0
[<0>] page_cache_ra_unbounded+0x1ab/0x280
[<0>] force_page_cache_ra+0x9e/0x130
[<0>] page_cache_sync_ra+0x3b/0x110
[<0>] filemap_get_pages+0x143/0xa30
[<0>] filemap_read+0xdc/0x4b0
[<0>] blkdev_read_iter+0x75/0x200
[<0>] vfs_read+0x272/0x460
[<0>] ksys_read+0x7a/0x170
[<0>] __x64_sys_read+0x1c/0x30
[<0>] do_syscall_64+0xc6/0x230
[<0>] entry_SYSCALL_64_after_hwframe+0x6c/0x74

This is because reshape can't make progress.

For md/raid, the problem doesn't exist because register new sync_thread
doesn't rely on the IO to be done any more:

1) If array is read-only, it can switch to read-write by ioctl/sysfs;
2) md/raid never set MD_RECOVERY_WAIT;
3) If MD_RECOVERY_FROZEN is set, mddev_suspend() doesn't hold
   'reconfig_mutex', hence it can be cleared and reshape can continue by
   sysfs api 'sync_action'.

However, I'm not sure yet how to avoid the problem in dm-raid yet. This
patch on the one hand make sure raid_message() can't change
sync_thread() through raid_message() after presuspend(), on the other
hand detect the above 3 cases before wait for IO do be done in
dm_suspend(), and let dm-raid requeue those IO.

Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-9-yukuai1@huaweicloud.com

dm-raid: add a new helper prepare_suspend() in md_personality

There are no functional changes for now, prepare to fix a deadlock for
dm-raid456.

Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-8-yukuai1@huaweicloud.com

md/dm-raid: don't call md_reap_sync_thread() directly

Currently md_reap_sync_thread() is called from raid_message() directly
without holding 'reconfig_mutex', this is definitely unsafe because
md_reap_sync_thread() can change many fields that is protected by
'reconfig_mutex'.

However, hold 'reconfig_mutex' here is still problematic because this
will cause deadlock, for example, commit 130443d60b1b ("md: refactor
idle/frozen_sync_thread() to fix deadlock").

Fix this problem by using stop_sync_thread() to unregister sync_thread,
like md/raid did.

Fixes: be83651f0050 ("DM RAID: Add message/status support for changing sync action")
Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-7-yukuai1@huaweicloud.com

dm-raid: really frozen sync_thread during suspend

1) commit f52f5c71f3d4 ("md: fix stopping sync thread") remove
   MD_RECOVERY_FROZEN from __md_stop_writes() and doesn't realize that
   dm-raid relies on __md_stop_writes() to frozen sync_thread
   indirectly. Fix this problem by adding MD_RECOVERY_FROZEN in
   md_stop_writes(), and since stop_sync_thread() is only used for
   dm-raid in this case, also move stop_sync_thread() to
   md_stop_writes().
2) The flag MD_RECOVERY_FROZEN doesn't mean that sync thread is frozen,
   it only prevent new sync_thread to start, and it can't stop the
   running sync thread; In order to frozen sync_thread, after seting the
   flag, stop_sync_thread() should be used.
3) The flag MD_RECOVERY_FROZEN doesn't mean that writes are stopped, use
   it as condition for md_stop_writes() in raid_postsuspend() doesn't
   look correct. Consider that reentrant stop_sync_thread() do nothing,
   always call md_stop_writes() in raid_postsuspend().
4) raid_message can set/clear the flag MD_RECOVERY_FROZEN at anytime,
   and if MD_RECOVERY_FROZEN is cleared while the array is suspended,
   new sync_thread can start unexpected. Fix this by disallow
   raid_message() to change sync_thread status during suspend.

Note that after commit f52f5c71f3d4 ("md: fix stopping sync thread"), the
test shell/lvconvert-raid-reshape.sh start to hang in stop_sync_thread(),
and with previous fixes, the test won't hang there anymore, however, the
test will still fail and complain that ext4 is corrupted. And with this
patch, the test won't hang due to stop_sync_thread() or fail due to ext4
is corrupted anymore. However, there is still a deadlock related to
dm-raid456 that will be fixed in following patches.

Reported-by: Mikulas Patocka <mpatocka@redhat.com>
Closes: https://lore.kernel.org/all/e5e8afe2-e9a8-49a2-5ab0-958d4065c55e@redhat.com/
Fixes: 1af2048a3e87 ("dm raid: fix deadlock caused by premature md_stop_writes()")
Fixes: 9dbd1aa3a81c ("dm raid: add reshaping support to the target")
Fixes: f52f5c71f3d4 ("md: fix stopping sync thread")
Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-6-yukuai1@huaweicloud.com

md: add a new helper reshape_interrupted()

The helper will be used for dm-raid456 later to detect the case that
reshape can't make progress.

Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-5-yukuai1@huaweicloud.com

md: export helper md_is_rdwr()

There are no functional changes for now, prepare to fix a deadlock for
dm-raid456.

Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-4-yukuai1@huaweicloud.com

md: export helpers to stop sync_thread

Add new helpers:

  void md_idle_sync_thread(struct mddev *mddev);
  void md_frozen_sync_thread(struct mddev *mddev);
  void md_unfrozen_sync_thread(struct mddev *mddev);

The helpers will be used in dm-raid in later patches to fix regressions
and prevent calling md_reap_sync_thread() directly.

Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-3-yukuai1@huaweicloud.com

md: don't clear MD_RECOVERY_FROZEN for new dm-raid until resume

After commit 9dbd1aa3a81c ("dm raid: add reshaping support to the
target") raid_ctr() will set MD_RECOVERY_FROZEN before md_run() and
expect to keep array frozen until resume. However, md_run() will clear
the flag by setting mddev->recovery to 0.

Before commit 1baae052cccd ("md: Don't ignore suspended array in
md_check_recovery()"), dm-raid actually relied on suspending to prevent
starting new sync_thread.

Fix this problem by keeping 'MD_RECOVERY_FROZEN' for dm-raid in
md_run().

Fixes: 1baae052cccd ("md: Don't ignore suspended array in md_check_recovery()")
Fixes: 9dbd1aa3a81c ("dm raid: add reshaping support to the target")
Cc: stable@vger.kernel.org # v6.7+
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Mike Snitzer <snitzer@kernel.org>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240305072306.2562024-2-yukuai1@huaweicloud.com

Revert "Revert "md/raid5: Wait for MD_SB_CHANGE_PENDING in raid5d""

This reverts commit bed9e27baf52a09b7ba2a3714f1e24e17ced386d.

The original set [1][2] was expected to undo a suboptimal fix in [2], and
replace it with a better fix [1]. However, as reported by Dan Moulding [2]
causes an issue with raid5 with journal device.

Revert [2] for now to close the issue. We will follow up on another issue
reported by Juxiao Bi, as [2] is expected to fix it. We believe this is a
good trade-off, because the latter issue happens less freqently.

In the meanwhile, we will NOT revert [1], as it contains the right logic.

[1] commit d6e035aad6c0 ("md: bypass block throttle for superblock update")
[2] commit bed9e27baf52 ("Revert "md/raid5: Wait for MD_SB_CHANGE_PENDING in raid5d"")

Reported-by: Dan Moulding <dan@danm.net>
Closes: https://lore.kernel.org/linux-raid/20240123005700.9302-1-dan@danm.net/
Fixes: bed9e27baf52 ("Revert "md/raid5: Wait for MD_SB_CHANGE_PENDING in raid5d"")
Cc: stable@vger.kernel.org # v5.19+
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20240125082131.788600-1-song@kernel.org

nbd: use the atomic queue limits API in nbd_set_size

Use queue_limits_start_update / queue_limits_commit_update to update
all the limits in one go and with proper sanity checking.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240229143846.1047223-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nbd: freeze the queue for queue limits updates

nbd currently updates the logical and physical block sizes as well
as the discard_sectors on a live queue.  Freeze the queue first to
make sure there are not commands in flight that can see torn or
inconsistent limits.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240229143846.1047223-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nbd: don't clear discard_sectors in nbd_config_put

nbd_config_put currently clears discard_sectors when unusing a device.
This is pretty odd behavior and different from the sector size
configuration which is simply left in places and then reconfigured when
nbd_set_size is as part of configuring the device.  Change nbd_set_size
to clear discard_sectors if discard is not supported so that all the
queue limits changes are handled in one place.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240229143846.1047223-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

pktcdvd: don't set max_hw_sectors on the underlying device

pktcdvd sets max_hw_sectors on the queue of the underlying device that
it doesn't own (and doesn't reset it ever) since the driver was merged.
This can create all kinds of problems as the underlying driver doesn't
even know about it changing the limit.

As the state purpose is to not create I/Os larger than a single frame,
and pktcdvd never builds bios larger than that, just set REQ_NOMERGE
on the bios it submits so that largers I/Os never get built.

Note: I don't have packet writing hardware, so this is compile tested
only.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240229144408.1047967-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

dm: use queue_limits_set

Use queue_limits_set which validates the limits and takes care of
updating the readahead settings instead of directly assigning them to
the queue.  For that make sure all limits are actually updated before
the assignment.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Snitzer <snitzer@kernel.org>
Link: https://lore.kernel.org/r/20240228225653.947152-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: add a queue_limits_stack_bdev helper

Add a small wrapper around blk_stack_limits that allows passing a bdev
for the bottom device and prints an error in case of misaligned
device. The name fits into the new queue limits API and the intent is
to eventually replace disk_stack_limits.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240228225653.947152-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: add a queue_limits_set helper

Add a small wrapper around queue_limits_commit_update for stacking
drivers that don't want to update existing limits, but set an
entirely new set.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240228225653.947152-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'md-6.9-20240301' of https://git./linux/kernel/git/song/md into for-6.9/block

Pull MD updates from Song:

"The major changes are:

 1. Refactor raid1 read_balance, by Yu Kuai and Paul Luse.
 2. Clean up and fix for md_ioctl, by Li Nan.
 3. Other small fixes, by Gui-Dong Han and Heming Zhao."

* tag 'md-6.9-20240301' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md: (22 commits)
  md/raid1: factor out helpers to choose the best rdev from read_balance()
  md/raid1: factor out the code to manage sequential IO
  md/raid1: factor out choose_bb_rdev() from read_balance()
  md/raid1: factor out choose_slow_rdev() from read_balance()
  md/raid1: factor out read_first_rdev() from read_balance()
  md/raid1-10: factor out a new helper raid1_should_read_first()
  md/raid1-10: add a helper raid1_check_read_range()
  md/raid1: fix choose next idle in read_balance()
  md/raid1: record nonrot rdevs while adding/removing rdevs to conf
  md/raid1: factor out helpers to add rdev to conf
  md: add a new helper rdev_has_badblock()
  md/raid5: fix atomicity violation in raid5_cache_count
  md/md-bitmap: fix incorrect usage for sb_index
  md: check mddev->pers before calling md_set_readonly()
  md: clean up openers check in do_md_stop() and md_set_readonly()
  md: sync blockdev before stopping raid or setting readonly
  md: factor out a helper to sync mddev
  md: Don't clear MD_CLOSING when the raid is about to stop
  md: return directly before setting did_set_md_closing
  md: clean up invalid BUG_ON in md_ioctl
  ...

Merge branch 'raid1-read_balance' into md-6.9

From: Yu Kuai <yukuai3@huawei.com>
Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
The original idea is that Paul want to optimize raid1 read
performance([1]), however, we think that the original code for
read_balance() is quite complex, and we don't want to add more
complexity. Hence we decide to refactor read_balance() first, to make
code cleaner and easier for follow up.

Before this patchset, read_balance() has many local variables and many
branches, it want to consider all the scenarios in one iteration. The
idea of this patch is to divide them into 4 different steps:

1) If resync is in progress, find the first usable disk, patch 5;
Otherwise:
2) Loop through all disks and skipping slow disks and disks with bad
blocks, choose the best disk, patch 10. If no disk is found:
3) Look for disks with bad blocks and choose the one with most number of
sectors, patch 8. If no disk is found:
4) Choose first found slow disk with no bad blocks, or slow disk with
most number of sectors, patch 7.

Note that step 3) and step 4) are super code path, and performance
should not be considered.

And after this patchset, we'll continue to optimize read_balance for
step 2), specifically how to choose the best rdev to read.

[1] https://lore.kernel.org/all/20240102125115.129261-1-paul.e.luse@linux.intel.com/

Yu Kuai (11):
  md: add a new helper rdev_has_badblock()
  md/raid1: factor out helpers to add rdev to conf
  md/raid1: record nonrot rdevs while adding/removing rdevs to conf
  md/raid1: fix choose next idle in read_balance()
  md/raid1-10: add a helper raid1_check_read_range()
  md/raid1-10: factor out a new helper raid1_should_read_first()
  md/raid1: factor out read_first_rdev() from read_balance()
  md/raid1: factor out choose_slow_rdev() from read_balance()
  md/raid1: factor out choose_bb_rdev() from read_balance()
  md/raid1: factor out the code to manage sequential IO
  md/raid1: factor out helpers to choose the best rdev from
    read_balance()

md/raid1: factor out helpers to choose the best rdev from read_balance()

The way that best rdev is chosen:

1) If the read is sequential from one rdev:
 - if rdev is rotational, use this rdev;
 - if rdev is non-rotational, use this rdev until total read length
   exceed disk opt io size;

2) If the read is not sequential:
 - if there is idle disk, use it, otherwise:
 - if the array has non-rotational disk, choose the rdev with minimal
   inflight IO;
 - if all the underlaying disks are rotational disk, choose the rdev
   with closest IO;

There are no functional changes, just to make code cleaner and prepare
for following refactor.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-12-yukuai1@huaweicloud.com

md/raid1: factor out the code to manage sequential IO

There is no functional change for now, make read_balance() cleaner and
prepare to fix problems and refactor the handler of sequential IO.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-11-yukuai1@huaweicloud.com

md/raid1: factor out choose_bb_rdev() from read_balance()

read_balance() is hard to understand because there are too many status
and branches, and it's overlong.

This patch factor out the case to read the rdev with bad blocks from
read_balance(), there are no functional changes.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-10-yukuai1@huaweicloud.com

md/raid1: factor out choose_slow_rdev() from read_balance()

read_balance() is hard to understand because there are too many status
and branches, and it's overlong.

This patch factor out the case to read the slow rdev from
read_balance(), there are no functional changes.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-9-yukuai1@huaweicloud.com

md/raid1: factor out read_first_rdev() from read_balance()

read_balance() is hard to understand because there are too many status
and branches, and it's overlong.

This patch factor out the case to read the first rdev from
read_balance(), there are no functional changes.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-8-yukuai1@huaweicloud.com

md/raid1-10: factor out a new helper raid1_should_read_first()

If resync is in progress, read_balance() should find the first usable
disk, otherwise, data could be inconsistent after resync is done. raid1
and raid10 implement the same checking, hence factor out the checking
to make code cleaner.

Noted that raid1 is using 'mddev->recovery_cp', which is updated after
all resync IO is done, while raid10 is using 'conf->next_resync', which
is inaccurate because raid10 update it before submitting resync IO.
Fortunately, raid10 read IO can't concurrent with resync IO, hence there
is no problem. And this patch also switch raid10 to use
'mddev->recovery_cp'.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-7-yukuai1@huaweicloud.com

md/raid1-10: add a helper raid1_check_read_range()

The checking and handler of bad blocks appear many timers during
read_balance() in raid1 and raid10. This helper will be used in later
patches to simplify read_balance() a lot.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-6-yukuai1@huaweicloud.com

md/raid1: fix choose next idle in read_balance()

Commit 12cee5a8a29e ("md/raid1: prevent merging too large request") add
the case choose next idle in read_balance():

read_balance:
 for_each_rdev
  if(next_seq_sect == this_sector || dist == 0)
  -> sequential reads
   best_disk = disk;
   if (...)
    choose_next_idle = 1
    continue;

 for_each_rdev
 -> iterate next rdev
  if (pending == 0)
   best_disk = disk;
   -> choose the next idle disk
   break;

  if (choose_next_idle)
   -> keep using this rdev if there are no other idle disk
   contine

However, commit 2e52d449bcec ("md/raid1: add failfast handling for reads.")
remove the code:

-               /* If device is idle, use it */
-               if (pending == 0) {
-                       best_disk = disk;
-                       break;
-               }

Hence choose next idle will never work now, fix this problem by
following:

1) don't set best_disk in this case, read_balance() will choose the best
   disk after iterating all the disks;
2) add 'pending' so that other idle disk will be chosen;
3) add a new local variable 'sequential_disk' to record the disk, and if
   there is no other idle disk, 'sequential_disk' will be chosen;

Fixes: 2e52d449bcec ("md/raid1: add failfast handling for reads.")
Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-5-yukuai1@huaweicloud.com

md/raid1: record nonrot rdevs while adding/removing rdevs to conf

For raid1, each read will iterate all the rdevs from conf and check if
any rdev is non-rotational, then choose rdev with minimal IO inflight
if so, or rdev with closest distance otherwise.

Disk nonrot info can be changed through sysfs entry:

/sys/block/[disk_name]/queue/rotational

However, consider that this should only be used for testing, and user
really shouldn't do this in real life. Record the number of non-rotational
disks in conf, to avoid checking each rdev in IO fast path and simplify
read_balance() a little bit.

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-4-yukuai1@huaweicloud.com

md/raid1: factor out helpers to add rdev to conf

There are no functional changes, just make code cleaner and prepare to
record disk non-rotational information while adding and removing rdev to
conf

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-3-yukuai1@huaweicloud.com

md: add a new helper rdev_has_badblock()

The current api is_badblock() must pass in 'first_bad' and
'bad_sectors', however, many caller just want to know if there are
badblocks or not, and these caller must define two local variable that
will never be used.

Add a new helper rdev_has_badblock() that will only return if there are
badblocks or not, remove unnecessary local variables and replace
is_badblock() with the new helper in many places.

There are no functional changes, and the new helper will also be used
later to refactor read_balance().

Co-developed-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Paul Luse <paul.e.luse@linux.intel.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240229095714.926789-2-yukuai1@huaweicloud.com

ublk: add UBLK_CMD_DEL_DEV_ASYNC

The current command UBLK_CMD_DEL_DEV won't return until the device is
released, this way looks more reliable, but makes userspace more
difficult to implement, especially about orders: unmap command
buffer(which holds one ublkc reference), ublkc close,
io_uring_file_unregister, ublkb close.

Add UBLK_CMD_DEL_DEV_ASYNC so that device deletion won't wait release,
then userspace needn't worry about the above order. Actually both loop
and nbd is deleted in this async way.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20240223075539.89945-3-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: improve getting & putting ublk device

Firstly convert get_device() and put_device() into ublk_get_device()
and ublk_put_device().

Secondly annotate ublk_get_device() & ublk_put_device() as noinline
for trace, especially it is often to trigger device deletion hang
when incorrect order is used on ublkc mmap, ublkc close,
io_uring_sqe_unregister_file, ublkb close.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20240223075539.89945-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-mq: don't change nr_hw_queues and nr_maps for kdump kernel

For most of ARCHs, 'nr_cpus=1' is passed for kdump kernel, so
nr_hw_queues for each mapping is supposed to be 1 already.

More importantly, this way may cause trouble for driver, because blk-mq and
driver see different queue mapping since driver should setup hardware
queue setting before calling into allocating blk-mq tagset.

So not overriding nr_hw_queues and nr_maps for kdump kernel.

Cc: Wen Xiong <wenxiong@us.ibm.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240228040857.306483-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

md/raid5: fix atomicity violation in raid5_cache_count

In raid5_cache_count():
    if (conf->max_nr_stripes < conf->min_nr_stripes)
        return 0;
    return conf->max_nr_stripes - conf->min_nr_stripes;
The current check is ineffective, as the values could change immediately
after being checked.

In raid5_set_cache_size():
    ...
    conf->min_nr_stripes = size;
    ...
    while (size > conf->max_nr_stripes)
        conf->min_nr_stripes = conf->max_nr_stripes;
    ...

Due to intermediate value updates in raid5_set_cache_size(), concurrent
execution of raid5_cache_count() and raid5_set_cache_size() may lead to
inconsistent reads of conf->max_nr_stripes and conf->min_nr_stripes.
The current checks are ineffective as values could change immediately
after being checked, raising the risk of conf->min_nr_stripes exceeding
conf->max_nr_stripes and potentially causing an integer overflow.

This possible bug is found by an experimental static analysis tool
developed by our team. This tool analyzes the locking APIs to extract
function pairs that can be concurrently executed, and then analyzes the
instructions in the paired functions to identify possible concurrency bugs
including data races and atomicity violations. The above possible bug is
reported when our tool analyzes the source code of Linux 6.2.

To resolve this issue, it is suggested to introduce local variables
'min_stripes' and 'max_stripes' in raid5_cache_count() to ensure the
values remain stable throughout the check. Adding locks in
raid5_cache_count() fails to resolve atomicity violations, as
raid5_set_cache_size() may hold intermediate values of
conf->min_nr_stripes while unlocked. With this patch applied, our tool no
longer reports the bug, with the kernel configuration allyesconfig for
x86_64. Due to the lack of associated hardware, we cannot test the patch
in runtime testing, and just verify it according to the code logic.

Fixes: edbe83ab4c27 ("md/raid5: allow the stripe_cache to grow and shrink.")
Cc: stable@vger.kernel.org
Signed-off-by: Gui-Dong Han <2045gemini@gmail.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240112071017.16313-1-2045gemini@gmail.com
Signed-off-by: Song Liu <song@kernel.org>

ubd: open the backing files in ubd_add

Opening the backing device only when the block device is opened is
a bit weird as no one configures block devices to not use them.
Opend them at add time, close them at remove time and remove the
now superflous opened counter as remove can simply check for
disk_openers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: remove the queue pointer in struct ubd

No need for it now, everything goes through the gendisk.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: move set_disk_ro to ubd_add

No need to delay this until open time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: move setting the variable queue limits to ubd_add

No reason to delay this until open time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: move setting the nonrot flag to ubd_add

No reason to delay this until open time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: remove ubd_disk_register

Fold it into the only caller to remove lots of references to the
global ubd_devs array.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: remove the ubd_gendisk array

And add a disk pointer to the ubd structure instead to keep all
the per-device information together.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Richard Weinberger <richard@nod.at>
Link: https://lore.kernel.org/r/20240222072417.3773131-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

xen-blkfront: atomically update queue limits

Pass the initial queue limits to blk_mq_alloc_disk and use the
blkif_set_queue_limits API to update the limits on reconnect.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20240221125845.3610668-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

xen-blkfront: don't redundantly set max_sements in blkif_recover

blkif_set_queue_limits already sets the max_sements limits, so don't do
it a second time.  Also remove a comment about a long fixe bug in
blk_mq_update_nr_hw_queues.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20240221125845.3610668-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

xen-blkfront: rely on the default discard granularity

The block layer now sets the discard granularity to the physical
block size default.  Take advantage of that in xen-blkfront and only
set the discard granularity if explicitly specified.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20240221125845.3610668-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

xen-blkfront: set max_discard/secure erase limits to UINT_MAX

Currently xen-blkfront set the max discard limit to the capacity of
the device, which is suboptimal when the capacity changes.  Just set
it to UINT_MAX, which has the same effect and is simpler.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20240221125845.3610668-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

md/md-bitmap: fix incorrect usage for sb_index

Commit d7038f951828 ("md-bitmap: don't use ->index for pages backing the
bitmap file") removed page->index from bitmap code, but left wrong code
logic for clustered-md. current code never set slot offset for cluster
nodes, will sometimes cause crash in clustered env.

Call trace (partly):
 md_bitmap_file_set_bit+0x110/0x1d8 [md_mod]
 md_bitmap_startwrite+0x13c/0x240 [md_mod]
 raid1_make_request+0x6b0/0x1c08 [raid1]
 md_handle_request+0x1dc/0x368 [md_mod]
 md_submit_bio+0x80/0xf8 [md_mod]
 __submit_bio+0x178/0x300
 submit_bio_noacct_nocheck+0x11c/0x338
 submit_bio_noacct+0x134/0x614
 submit_bio+0x28/0xdc
 submit_bh_wbc+0x130/0x1cc
 submit_bh+0x1c/0x28

Fixes: d7038f951828 ("md-bitmap: don't use ->index for pages backing the bitmap file")
Cc: stable@vger.kernel.org # v6.6+
Signed-off-by: Heming Zhao <heming.zhao@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240223121128.28985-1-heming.zhao@suse.com

md: check mddev->pers before calling md_set_readonly()

If 'mddev->pers' is NULL, there is nothing to do in md_set_readonly().
Except for md_ioctl(), the other two callers of md_set_readonly() have
already checked 'mddev->pers'. To simplify the code, move the check of
'mddev->pers' to the caller.

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-10-linan666@huaweicloud.com

md: clean up openers check in do_md_stop() and md_set_readonly()

Before stopping or setting readonly, mddev_set_closing_and_sync_blockdev()
is always called to check the openers. So no longer need to check it again
in do_md_stop() and md_set_readonly(). Clean it up.

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-9-linan666@huaweicloud.com

md: sync blockdev before stopping raid or setting readonly

Commit a05b7ea03d72 ("md: avoid crash when stopping md array races
with closing other open fds.") added sync_block before stopping raid and
setting readonly. Later in commit 260fa034ef7a ("md: avoid deadlock when
dirty buffers during md_stop.") it is moved to ioctl. array_state_store()
was ignored. Add sync blockdev to array_state_store() now.

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-8-linan666@huaweicloud.com

md: factor out a helper to sync mddev

There are no functional changes, prepare to sync mddev in
array_state_store().

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-7-linan666@huaweicloud.com

md: Don't clear MD_CLOSING when the raid is about to stop

The raid should not be opened anymore when it is about to be stopped.
However, other processes can open it again if the flag MD_CLOSING is
cleared before exiting. From now on, this flag will not be cleared when
the raid will be stopped.

Fixes: 065e519e71b2 ("md: MD_CLOSING needs to be cleared after called md_set_readonly or do_md_stop")
Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-6-linan666@huaweicloud.com

md: return directly before setting did_set_md_closing

There is nothing to do at 'out' before setting 'did_set_md_closing'
in md_ioctl(). Return directly, and it will help us to remove
'did_set_md_closing' later.

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-5-linan666@huaweicloud.com

md: clean up invalid BUG_ON in md_ioctl

'disk->private_data' is set to mddev in md_alloc() and never set to NULL,
and users need to open mddev before submitting ioctl. So mddev must not
have been freed during ioctl, and there is no need to check mddev here.
Clean up it.

Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-4-linan666@huaweicloud.com

md: changed the switch of RAID_VERSION to if

There is only one case of this 'switch'. Change it to 'if'.

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-3-linan666@huaweicloud.com

md: merge the check of capabilities into md_ioctl_valid()

There is no functional change. Just to make code cleaner.

Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20240226031444.3606764-2-linan666@huaweicloud.com

bdev: remove SLAB_MEM_SPREAD flag usage

The SLAB_MEM_SPREAD flag is already a no-op as of 6.8-rc1, remove
its usage so we can delete it from slab. No functional change.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Link: https://lore.kernel.org/r/20240224134646.829105-1-chengming.zhou@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/blk-mq: Don't complete locally if capacities are different

The logic in blk_mq_complete_need_ipi() assumes SMP systems where all
CPUs have equal compute capacities and only LLC cache can make
a different on perceived performance. But this assumption falls apart on
HMP systems where LLC is shared, but the CPUs have different capacities.
Staying local then can have a big performance impact if the IO request
was done from a CPU with higher capacity but the interrupt is serviced
on a lower capacity CPU.

Use the new cpus_equal_capacity() function to check if we need to send
an IPI.

Without the patch I see the BLOCK softirq always running on little cores
(where the hardirq is serviced). With it I can see it running on all
cores.

This was noticed after the topology change [1] where now on a big.LITTLE
we truly get that the LLC is shared between all cores where as in the
past it was being misrepresented for historical reasons. The logic
exposed a missing dependency on capacities for such systems where there
can be a big performance difference between the CPUs.

This of course introduced a noticeable change in behavior depending on
how the topology is presented. Leading to regressions in some workloads
as the performance of the BLOCK softirq on littles can be noticeably
worse on some platforms.

Worth noting that we could have checked for capacities being greater
than or equal instead for equality. This will lead to favouring higher
performance always. But opted for equality instead to match the
performance of the requester without making an assumption that can lead
to power trade-offs which these systems tend to be sensitive about. If
the requester would like to run faster, it's better to rely on the
scheduler to give the IO requester via some facility to run on a faster
core; and then if the interrupt triggered on a CPU with different
capacity we'll make sure to match the performance the requester is
supposed to run at.

[1] https://lpc.events/event/16/contributions/1342/attachments/962/1883/LPC-2022-Android-MC-Phantom-Domains.pdf

Signed-off-by: Qais Yousef <qyousef@layalina.io>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20240223155749.2958009-3-qyousef@layalina.io
Signed-off-by: Jens Axboe <axboe@kernel.dk>

sched: Add a new function to compare if two cpus have the same capacity

The new helper function is needed to help blk-mq check if it needs to
dispatch the softirq on another CPU to match the performance level the
IO requester is running at. This is important on HMP systems where not
all CPUs have the same compute capacity.

Signed-off-by: Qais Yousef <qyousef@layalina.io>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20240223155749.2958009-2-qyousef@layalina.io
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-lib: check for kill signal

Some of these block operations can access a significant capacity and
take longer than the user expected. A user may change their mind about
wanting to run that command and attempt to kill the process and do
something else with their device. But since the task is uninterruptable,
they have to wait for it to finish, which could be many hours.

Check for a fatal signal at each iteration so the user doesn't have to
wait for their regretted operation to complete naturally.

Reported-by: Conrad Meyer <conradmeyer@meta.com>
Tested-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://lore.kernel.org/r/20240223155910.3622666-5-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: io wait hang check helper

This is the same in two places, and another will be added soon. Create a
helper for it.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://lore.kernel.org/r/20240223155910.3622666-4-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: cleanup __blkdev_issue_write_zeroes

Use min to calculate the next number of sectors like everyone else.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://lore.kernel.org/r/20240223155910.3622666-3-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: blkdev_issue_secure_erase loop style

Use consistent coding style in this file. All the other loops for the
same purpose use "while (nr_sects)", so they win.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://lore.kernel.org/r/20240223155910.3622666-2-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: fix deadlock between bd_link_disk_holder and partition scan

'open_mutex' of gendisk is used to protect open/close block devices. But
in bd_link_disk_holder(), it is used to protect the creation of symlink
between holding disk and slave bdev, which introduces some issues.

When bd_link_disk_holder() is called, the driver is usually in the process
of initialization/modification and may suspend submitting io. At this
time, any io hold 'open_mutex', such as scanning partitions, can cause
deadlocks. For example, in raid:

T1                              T2
bdev_open_by_dev
 lock open_mutex [1]
 ...
  efi_partition
  ...
   md_submit_bio
md_ioctl mddev_syspend
  -> suspend all io
 md_add_new_disk
  bind_rdev_to_array
   bd_link_disk_holder
    try lock open_mutex [2]
    md_handle_request
     -> wait mddev_resume

T1 scan partition, T2 add a new device to raid. T1 waits for T2 to resume
mddev, but T2 waits for open_mutex held by T1. Deadlock occurs.

Fix it by introducing a local mutex 'blk_holder_mutex' to replace
'open_mutex'.

Fixes: 1b0a2d950ee2 ("md: use new apis to suspend array for ioctls involed array reconfiguration")
Reported-by: mgperkow@gmail.com
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=218459
Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20240221090122.1281868-1-linan666@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Do not include rbtree.h in blk-zoned.c

The block zone code does not use RB-tree. So remove the include of
linux/rbtree.h as it is not needed.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20240222131724.1803520-2-dlemoal@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Clear zone limits for a non-zoned stacked queue

Device mapper may create a non-zoned mapped device out of a zoned device
(e.g., the dm-zoned target). In such case, some queue limit such as the
max_zone_append_sectors and zone_write_granularity endup being non zero
values for a block device that is not zoned. Avoid this by clearing
these limits in blk_stack_limits() when the stacked zoned limit is
false.

Fixes: 3093a479727b ("block: inherit the zoned characteristics in blk_stack_limits")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20240222131724.1803520-1-dlemoal@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: Delete nullb.{queue_depth, nr_queues}

Since commit 8b631f9cf0b8 ("null_blk: remove the bio based I/O path"),
struct nullb members queue_depth and nr_queues are only ever written, so
delete them.

With that, null_exit_hctx() can also be deleted.

Signed-off-by: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240222083420.6026-1-john.g.garry@oracle.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

pktcdvd: set queue limits at disk allocation time

Remove pkt_init_queue and just pass the two parameters directly to
blk_alloc_disk.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240222073647.3776769-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

pktcdvd: stop setting q->queuedata

The two users can get the private data from the gendisk with one less
pointer dereference, and we can drop the useless q parameter from
pkt_make_request_write.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240222073647.3776769-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: fix virt_boundary handling in blk_validate_limits

Don't set the default max_segment_size value when a virt_boundary is
used.

Fixes: d690cb8ae14b ("block: add an API to atomically update queue limits")
Reported-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Link: https://lore.kernel.org/r/20240221125010.3609444-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_mq_alloc_disk instead of
setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Tested-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20240220093248.3290292-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: remove null_gendisk_register

null_gendisk_register isn't a very useful abstraction given that it
doesn't even allocate the gendisk.  Merge it into the only caller
instead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Tested-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20240220093248.3290292-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: refactor tag_set setup

Move the tagset initialization out of null_add_dev into a new
null_setup_tagset helper, and move the shared vs local differences
out of null_init_tag_set into the callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Tested-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20240220093248.3290292-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: initialize the tag_set timeout in null_init_tag_set

Otherwise it will be reset to the always same value when initializing a
device using the shared tag_set.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Tested-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20240220093248.3290292-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: remove the bio based I/O path

The bio based I/O path complicates null_blk and also make various
data structures, including the per-command one way bigger than
required for the main request based interface.   As the bio-based
path is mostly used by stacking drivers and simple memory based
drivers, and brd is a good example driver for the latter there is
no need to have a bio based path in null_blk.  Remove the path
to simplify the driver and make future block layer API changes
simpler by not having to deal with the complex two API setup in
null_blk.

Note that the queue_mode field in struct nullb_device is kept as
that is simpler than having two different places to check the
value and fully open coding the debugfs helpers as the existing
ones won't work without a named struct member.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Tested-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20240220093248.3290292-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

mmc: pass queue_limits to blk_mq_alloc_disk

Pass the queue limit set at initialization time directly to
blk_mq_alloc_disk instead of updating it right after the allocation.

This requires refactoring the code a bit so that what was mmc_setup_queue
before also allocates the gendisk now and actually sets all limits.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Link: https://lore.kernel.org/r/20240215070300.2200308-18-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: pass queue_limits to blk_mq_alloc_disk

Pass the limits ublk imposes directly to blk_mq_alloc_disk instead of
setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-17-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

scm_blk: pass queue_limits to blk_mq_alloc_disk

Pass the few limits scm_block imposes directly to blk_mq_alloc_disk
instead of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-16-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubiblock: pass queue_limits to blk_mq_alloc_disk

Pass the few limits ubiblock imposes directly to blk_mq_alloc_disk
instead of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com>
Link: https://lore.kernel.org/r/20240215070300.2200308-15-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

mtd_blkdevs: pass queue_limits to blk_mq_alloc_disk

Pass the few limits mtd_blkdevs imposes directly to blk_mq_alloc_disk
instead of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-14-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

mspro_block: pass queue_limits to blk_mq_alloc_disk

Pass the few limits mspro_block imposes directly to blk_mq_alloc_disk
instead of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-13-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ms_block: pass queue_limits to blk_mq_alloc_disk

Pass the few limits ms_block imposes directly to blk_mq_alloc_disk
instead of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-12-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

gdrom: pass queue_limits to blk_mq_alloc_disk

Pass the few limits gdrom imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-11-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

sunvdc: pass queue_limits to blk_mq_alloc_disk

Pass the few limits sunvdc imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-10-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

rnbd-clt: pass queue_limits to blk_mq_alloc_disk

Pass the limits rnbd-clt imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

While at it don't set an explicit number of discard segments, as 1 is
the default (which most drivers rely on).

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20240215070300.2200308-9-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

rbd: pass queue_limits to blk_mq_alloc_disk

Pass the limits rbd imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ps3disk: pass queue_limits to blk_mq_alloc_disk

Pass the few limits ps3disk imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nbd: pass queue_limits to blk_mq_alloc_disk

Pass the few limits nbd imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

mtip: pass queue_limits to blk_mq_alloc_disk

Pass the few limits mtip imposes directly to blk_mq_alloc_disk instead
of setting them one at a time and drop the pointless setting of a io_min
that is equal to the physical block size.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: pass queue_limits to blk_mq_alloc_disk

Pass the few limits floppy imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Denis Efremov <efremov@linux.com>
Link: https://lore.kernel.org/r/20240215070300.2200308-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

aoe: pass queue_limits to blk_mq_alloc_disk

Pass the few limits aoe imposes directly to blk_mq_alloc_disk instead
of setting them one at a time and improve the way the default
max_hw_sectors is initialized while we're at it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ubd: pass queue_limits to blk_mq_alloc_disk

Pass the few limits ubd imposes directly to blk_mq_alloc_disk instead
of setting them one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240215070300.2200308-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

dcssblk: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-10-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

pmem: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-9-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

btt: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

zram: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

n64cart: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

brd: pass queue_limits to blk_mq_alloc_disk

Pass the queue limits directly to blk_alloc_disk instead of setting them
one at a time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Link: https://lore.kernel.org/r/20240215071055.2201424-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>