nvme: host: hwmon: constify pointers to hwmon_channel_info

Statically allocated array of pointed to hwmon_channel_info can be made
const for safety.

Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Keith Busch <kbusch@kernel.org>

nvmet-tcp: pass iov_len instead of sg->length to bvec_set_page()

iov_len is the valid data length, so pass iov_len instead of sg->length to
bvec_set_page().

Fixes: 5bfaba275ae6 ("nvmet-tcp: don't map pages which can't come from HIGHMEM")
Signed-off-by: Rakshana Sridhar <rakshanas@chelsio.com>
Signed-off-by: Varun Prakash <varun@chelsio.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>

ublk: zoned: support REQ_OP_ZONE_RESET_ALL

There isn't any reason to not support REQ_OP_ZONE_RESET_ALL given everything
is actually handled in userspace, not mention it is pretty easy to support
RESET_ALL.

So enable REQ_OP_ZONE_RESET_ALL and let userspace handle it.

Verified by 'tools/zbc_reset_zone -all /dev/ublkb0' in libzbc[1] with
libublk-rs based ublk-zoned target prototype[2], follows command line
for creating ublk-zoned:

cargo run --example zoned -- add -1 1024 # add $dev_id $DEV_SIZE

[1] https://github.com/westerndigitalcorporation/libzbc
[2] https://github.com/ming1/libublk-rs/tree/zoned.v2

Cc: Niklas Cassel <Niklas.Cassel@wdc.com>
Cc: Damien Le Moal <dlemoal@kernel.org>
Cc: Andreas Hindborg <a.hindborg@samsung.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230810124326.321472-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'md-next-20230817' of https://git./linux/kernel/git/song/md into for-6.6/block

Pull MD changes from Song:

"1. Fix perf regression for raid0 large sequential writes, by Jan Kara.
 2. Fix split bio iostat for raid0, by David Jeffery.
 3. Various raid1 fixes, by Heinz Mauelshagen and Xueshi Hu."

* tag 'md-next-20230817' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md: raid0: account for split bio in iostat accounting
  md/raid0: Fix performance regression for large sequential writes
  md/raid0: Factor out helper for mapping and submitting a bio
  md raid1: allow writebehind to work on any leg device set WriteMostly
  md/raid1: hold the barrier until handle_read_error() finishes
  md/raid1: free the r1bio before waiting for blocked rdev
  md/raid1: call free_r1bio() before allow_barrier() in raid_end_bio_io()

md: raid0: account for split bio in iostat accounting

When a bio is split by md raid0, the newly created bio will not be tracked
by md for I/O accounting. Only the portion of I/O still assigned to the
original bio which was reduced by the split will be accounted for. This
results in md iostat data sometimes showing I/O values far below the actual
amount of data being sent through md.

md_account_bio() needs to be called for all bio generated by the bio split.

A simple example of the issue was generated using a raid0 device on partitions
to the same device. Since all raid0 I/O then goes to one device, it makes it
easy to see a gap between the md device and its sd storage. Reading an lvm
device on top of the md device, the iostat output (some 0 columns and extra
devices removed to make the data more compact) was:

Device             tps    kB_read/s    kB_wrtn/s    kB_dscd/s    kB_read
md2               0.00         0.00         0.00         0.00          0
sde               0.00         0.00         0.00         0.00          0
md2            1364.00    411496.00         0.00         0.00     411496
sde            1734.00    646144.00         0.00         0.00     646144
md2            1699.00    510680.00         0.00         0.00     510680
sde            2155.00    802784.00         0.00         0.00     802784
md2             803.00    241480.00         0.00         0.00     241480
sde            1016.00    377888.00         0.00         0.00     377888
md2               0.00         0.00         0.00         0.00          0
sde               0.00         0.00         0.00         0.00          0

I/O was generated doing large direct I/O reads (12M) with dd to a linear
lvm volume on top of the 4 leg raid0 device.

The md2 reads were showing as roughly 2/3 of the reads to the sde device
containing all of md2's raid partitions. The sum of reads to sde was
1826816 kB, which was the expected amount as it was the amount read by
dd. With the patch, the total reads from md will match the reads from
sde and be consistent with the amount of I/O generated.

Fixes: 10764815ff47 ("md: add io accounting for raid0 and raid5")
Signed-off-by: David Jeffery <djeffery@redhat.com>
Tested-by: Laurence Oberman <loberman@redhat.com>
Reviewed-by: Laurence Oberman <loberman@redhat.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230816181433.13289-1-djeffery@redhat.com

md/raid0: Fix performance regression for large sequential writes

Commit f00d7c85be9e ("md/raid0: fix up bio splitting.") among other
things changed how bio that needs to be split is submitted. Before this
commit, we have split the bio, mapped and submitted each part. After
this commit, we map only the first part of the split bio and submit the
second part unmapped. Due to bio sorting in __submit_bio_noacct() this
results in the following request ordering:

  9,0   18     1181     0.525037895 15995  Q  WS 1479315464 + 63392

  Split off chunk-sized (1024 sectors) request:

  9,0   18     1182     0.629019647 15995  X  WS 1479315464 / 1479316488

  Request is unaligned to the chunk so it's split in
  raid0_make_request().  This is the first part mapped and punted to
  bio_list:

  8,0   18     7053     0.629020455 15995  A  WS 739921928 + 1016 <- (9,0) 1479315464

  Now raid0_make_request() returns, second part is postponed on
  bio_list. __submit_bio_noacct() resorts the bio_list, mapped request
  is submitted to the underlying device:

  8,0   18     7054     0.629022782 15995  G  WS 739921928 + 1016

  Now we take another request from the bio_list which is the remainder
  of the original huge request. Split off another chunk-sized bit from
  it and the situation repeats:

  9,0   18     1183     0.629024499 15995  X  WS 1479316488 / 1479317512
  8,16  18     6998     0.629025110 15995  A  WS 739921928 + 1016 <- (9,0) 1479316488
  8,16  18     6999     0.629026728 15995  G  WS 739921928 + 1016
  ...
  9,0   18     1184     0.629032940 15995  X  WS 1479317512 / 1479318536 [libnetacq-write]
  8,0   18     7059     0.629033294 15995  A  WS 739922952 + 1016 <- (9,0) 1479317512
  8,0   18     7060     0.629033902 15995  G  WS 739922952 + 1016
  ...

  This repeats until we consume the whole original huge request. Now we
  finally get to processing the second parts of the split off requests
  (in reverse order):

  8,16  18     7181     0.629161384 15995  A  WS 739952640 + 8 <- (9,0) 1479377920
  8,0   18     7239     0.629162140 15995  A  WS 739952640 + 8 <- (9,0) 1479376896
  8,16  18     7186     0.629163881 15995  A  WS 739951616 + 8 <- (9,0) 1479375872
  8,0   18     7242     0.629164421 15995  A  WS 739951616 + 8 <- (9,0) 1479374848
  ...

I guess it is obvious that this IO pattern is extremely inefficient way
to perform sequential IO. It also makes bio_list to grow to rather long
lengths.

Change raid0_make_request() to map both parts of the split bio. Since we
know we are provided with at most chunk-sized bios, we will always need
to split the incoming bio at most once.

Fixes: f00d7c85be9e ("md/raid0: fix up bio splitting.")
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230814092720.3931-2-jack@suse.cz
Signed-off-by: Song Liu <song@kernel.org>

md/raid0: Factor out helper for mapping and submitting a bio

Factor out helper function for mapping and submitting a bio out of
raid0_make_request(). We will use it later for submitting both parts of
a split bio.

Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230814092720.3931-1-jack@suse.cz
Signed-off-by: Song Liu <song@kernel.org>

md raid1: allow writebehind to work on any leg device set WriteMostly

As the WriteMostly flag can be set on any component device of a RAID1
array, remove the constraint that it only works if set on the first one.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Tested-by: Xiao Ni <xni@redhat.com>
Link: https://lore.kernel.org/r/2a9592bf3340f34bf588eec984b23ee219f3985e.1692013451.git.heinzm@redhat.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid1: hold the barrier until handle_read_error() finishes

handle_read_error() will call allow_barrier() to match the former barrier
raising. However, it should put the allow_barrier() at the end to avoid a
concurrent raid reshape.

Fixes: 689389a06ce7 ("md/raid1: simplify handle_read_error().")
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xueshi Hu <xueshi.hu@smartx.com>
Link: https://lore.kernel.org/r/20230814135356.1113639-4-xueshi.hu@smartx.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid1: free the r1bio before waiting for blocked rdev

Raid1 reshape will change mempool and r1conf::raid_disks which are
needed to free r1bio. allow_barrier() make a concurrent raid1_reshape()
possible. So, free the in-flight r1bio before waiting blocked rdev.

Fixes: 6bfe0b499082 ("md: support blocking writes to an array on device failure")
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xueshi Hu <xueshi.hu@smartx.com>
Link: https://lore.kernel.org/r/20230814135356.1113639-3-xueshi.hu@smartx.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid1: call free_r1bio() before allow_barrier() in raid_end_bio_io()

After allow_barrier, a concurrent raid1_reshape() will replace old mempool
and r1conf::raid_disks. Move allow_barrier() to the end of raid_end_bio_io(),
so that r1bio can be freed safely.

Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Xueshi Hu <xueshi.hu@smartx.com>
Link: https://lore.kernel.org/r/20230814135356.1113639-2-xueshi.hu@smartx.com
Signed-off-by: Song Liu <song@kernel.org>

blk-cgroup: Fix NULL deref caused by blkg_policy_data being installed before init

blk-iocost sometimes causes the following crash:

  BUG: kernel NULL pointer dereference, address: 00000000000000e0
  ...
  RIP: 0010:_raw_spin_lock+0x17/0x30
  Code: be 01 02 00 00 e8 79 38 39 ff 31 d2 89 d0 5d c3 0f 1f 00 0f 1f 44 00 00 55 48 89 e5 65 ff 05 48 d0 34 7e b9 01 00 00 00 31 c0 <f0> 0f b1 0f 75 02 5d c3 89 c6 e8 ea 04 00 00 5d c3 0f 1f 84 00 00
  RSP: 0018:ffffc900023b3d40 EFLAGS: 00010046
  RAX: 0000000000000000 RBX: 00000000000000e0 RCX: 0000000000000001
  RDX: ffffc900023b3d20 RSI: ffffc900023b3cf0 RDI: 00000000000000e0
  RBP: ffffc900023b3d40 R08: ffffc900023b3c10 R09: 0000000000000003
  R10: 0000000000000064 R11: 000000000000000a R12: ffff888102337000
  R13: fffffffffffffff2 R14: ffff88810af408c8 R15: ffff8881070c3600
  FS:  00007faaaf364fc0(0000) GS:ffff88842fdc0000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00000000000000e0 CR3: 00000001097b1000 CR4: 0000000000350ea0
  Call Trace:
   <TASK>
   ioc_weight_write+0x13d/0x410
   cgroup_file_write+0x7a/0x130
   kernfs_fop_write_iter+0xf5/0x170
   vfs_write+0x298/0x370
   ksys_write+0x5f/0xb0
   __x64_sys_write+0x1b/0x20
   do_syscall_64+0x3d/0x80
   entry_SYSCALL_64_after_hwframe+0x46/0xb0

This happens because iocg->ioc is NULL. The field is initialized by
ioc_pd_init() and never cleared. The NULL deref is caused by
blkcg_activate_policy() installing blkg_policy_data before initializing it.

blkcg_activate_policy() was doing the following:

1. Allocate pd's for all existing blkg's and install them in blkg->pd[].
2. Initialize all pd's.
3. Online all pd's.

blkcg_activate_policy() only grabs the queue_lock and may release and
re-acquire the lock as allocation may need to sleep. ioc_weight_write()
grabs blkcg->lock and iterates all its blkg's. The two can race and if
ioc_weight_write() runs during #1 or between #1 and #2, it can encounter a
pd which is not initialized yet, leading to crash.

The crash can be reproduced with the following script:

  #!/bin/bash

  echo +io > /sys/fs/cgroup/cgroup.subtree_control
  systemd-run --unit touch-sda --scope dd if=/dev/sda of=/dev/null bs=1M count=1 iflag=direct
  echo 100 > /sys/fs/cgroup/system.slice/io.weight
  bash -c "echo '8:0 enable=1' > /sys/fs/cgroup/io.cost.qos" &
  sleep .2
  echo 100 > /sys/fs/cgroup/system.slice/io.weight

with the following patch applied:

> diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
> index fc49be622e05..38d671d5e10c 100644
> --- a/block/blk-cgroup.c
> +++ b/block/blk-cgroup.c
> @@ -1553,6 +1553,12 @@ int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol)
>  pd->online = false;
>  }
>
> +       if (system_state == SYSTEM_RUNNING) {
> +               spin_unlock_irq(&q->queue_lock);
> +               ssleep(1);
> +               spin_lock_irq(&q->queue_lock);
> +       }
> +
>  /* all allocated, init in the same order */
>  if (pol->pd_init_fn)
>  list_for_each_entry_reverse(blkg, &q->blkg_list, q_node)

I don't see a reason why all pd's should be allocated, initialized and
onlined together. The only ordering requirement is that parent blkgs to be
initialized and onlined before children, which is guaranteed from the
walking order. Let's fix the bug by allocating, initializing and onlining pd
for each blkg and holding blkcg->lock over initialization and onlining. This
ensures that an installed blkg is always fully initialized and onlined
removing the the race window.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Breno Leitao <leitao@debian.org>
Fixes: 9d179b865449 ("blkcg: Fix multiple bugs in blkcg_activate_policy()")
Link: https://lore.kernel.org/r/ZN0p5_W-Q9mAHBVY@slm.duckdns.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

drivers/rnbd: restore sysfs interface to rnbd-client

Commit 137380c0ec40 renamed 'rnbd-client' to 'rnbd_client', this changed
sysfs interface to /sys/devices/virtual/rnbd_client/ctl/map_device
from /sys/devices/virtual/rnbd-client/ctl/map_device.

CC: Ivan Orlov <ivan.orlov0322@gmail.com>
CC: "Md. Haris Iqbal" <haris.iqbal@ionos.com>
CC: Jack Wang <jinpu.wang@ionos.com>
Fixes: 137380c0ec40 ("block/rnbd: make all 'class' structures const")
Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20230816022210.2501228-1-lizhijian@fujitsu.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'md-next-20230814-resend' of https://git./linux/kernel/git/song/md into for-6.6/block

Pull MD fixes from Song:

"1. raid6test build fixes, by WANG Xuerui
 2. Various non-urgent fixes."

* tag 'md-next-20230814-resend' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md/raid5-cache: fix null-ptr-deref for r5l_flush_stripe_to_raid()
  raid6: test: only check for Altivec if building on powerpc hosts
  raid6: test: make sure all intermediate and artifact files are .gitignored
  raid6: test: cosmetic cleanups for the test Makefile
  raid6: guard the tables.c include of <linux/export.h> with __KERNEL__
  raid6: remove the <linux/export.h> include from recov.c
  md: Hold mddev->reconfig_mutex when trying to get mddev->sync_thread
  md/raid10: fix a 'conf->barrier' leakage in raid10_takeover()
  md: raid1: fix potential OOB in raid1_remove_disk()
  md/raid5-cache: fix a deadlock in r5l_exit_log()

md/raid5-cache: fix null-ptr-deref for r5l_flush_stripe_to_raid()

r5l_flush_stripe_to_raid() will check if the list 'flushing_ios' is
empty, and then submit 'flush_bio', however, r5l_log_flush_endio()
is clearing the list first and then clear the bio, which will cause
null-ptr-deref:

T1: submit flush io
raid5d
 handle_active_stripes
  r5l_flush_stripe_to_raid
   // list is empty
   // add 'io_end_ios' to the list
   bio_init
   submit_bio
   // io1

T2: io1 is done
r5l_log_flush_endio
 list_splice_tail_init
 // clear the list
T3: submit new flush io
...
r5l_flush_stripe_to_raid
 // list is empty
 // add 'io_end_ios' to the list
 bio_init
 bio_uninit
 // clear bio->bi_blkg
 submit_bio
 // null-ptr-deref

Fix this problem by clearing bio before clearing the list in
r5l_log_flush_endio().

Fixes: 0dd00cba99c3 ("raid5-cache: fully initialize flush_bio when needed")
Reported-and-tested-by: Corey Hickey <bugfood-ml@fatooh.org>
Closes: https://lore.kernel.org/all/cddd7213-3dfd-4ab7-a3ac-edd54d74a626@fatooh.org/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>

raid6: test: only check for Altivec if building on powerpc hosts

Altivec is only available for powerpc hosts, so only check for its
availability when the host is powerpc, to avoid error messages being
shown on architectures other than x86, arm or powerpc.

Signed-off-by: WANG Xuerui <git@xen0n.name>
Link: https://lore.kernel.org/r/20230731104911.411964-6-kernel@xen0n.name
Signed-off-by: Song Liu <song@kernel.org>

raid6: test: make sure all intermediate and artifact files are .gitignored

Currently when the raid6test utility is built, the resulting binary and
an int.uc file are not being ignored, which can get inadvertently
committed as a result when one works on the raid6 code. Ignore them to
make `git status` clean at all times.

Signed-off-by: WANG Xuerui <git@xen0n.name>
Link: https://lore.kernel.org/r/20230731104911.411964-5-kernel@xen0n.name
Signed-off-by: Song Liu <song@kernel.org>

raid6: test: cosmetic cleanups for the test Makefile

Use tabs/spaces consistently: hard tabs for marking recipe lines only,
spaces for everything else.

Also, the OPTFLAGS declaration actually included the tabs preceding the
line comment, making compiler invocation lines unnecessarily long. As
the entire block of declarations are meant for ad-hoc customization
(otherwise they would probably make use of `?=` instead of `=`), move
the "Adjust as desired" comment above the block too to fix the long
invocation lines.

Signed-off-by: WANG Xuerui <git@xen0n.name>
Link: https://lore.kernel.org/r/20230731104911.411964-4-kernel@xen0n.name
Signed-off-by: Song Liu <song@kernel.org>

raid6: guard the tables.c include of <linux/export.h> with __KERNEL__

The export directives for the tables are already emitted with __KERNEL__
guards, but the <linux/export.h> include is not, causing errors when
building the raid6test program. Guard this include too to fix the
raid6test build.

Signed-off-by: WANG Xuerui <git@xen0n.name>
Link: https://lore.kernel.org/r/20230731104911.411964-3-kernel@xen0n.name
Signed-off-by: Song Liu <song@kernel.org>

raid6: remove the <linux/export.h> include from recov.c

There is no exported symbol left in recov.c, so the include is now
unnecessary, and breaks the raid6test build. Remove it.

Signed-off-by: WANG Xuerui <git@xen0n.name>
Link: https://lore.kernel.org/r/20230731104911.411964-2-kernel@xen0n.name
Signed-off-by: Song Liu <song@kernel.org>

md: Hold mddev->reconfig_mutex when trying to get mddev->sync_thread

Commit ba9d9f1a707f ("Revert "md: unlock mddev before reap sync_thread in
action_store"") removed the scenario of calling md_unregister_thread()
without holding mddev->reconfig_mutex, so add a lock holding check before
acquiring mddev->sync_thread by passing mdev to md_unregister_thread().

Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230803071711.2546560-1-lilingfeng@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: fix a 'conf->barrier' leakage in raid10_takeover()

After commit b39f35ebe86d ("md: don't quiesce in mddev_suspend()"),
'conf->barrier' will be leaked in the case that raid10 takeover raid0:

level_store
 pers->takeover -> raid10_takeover
  raid10_takeover_raid0
   WRITE_ONCE(conf->barrier, 1)

mddev_suspend
// still raid0
mddev->pers = pers
// switch to raid10
mddev_resume
// resume without suspend

After the above commit, mddev_resume() will not decrease 'conf->barrier'
that is set in raid10_takeover_raid0().

Fix this problem by not setting 'conf->barrier' in raid10_takeover_raid0().

By the way, this problem is found while I'm trying to make
mddev_suspend/resume() to be independent from raid personalities. raid10
is the only personality to use reference count in the quiesce() callback
and this problem is only related to raid10.

Fixes: b39f35ebe86d ("md: don't quiesce in mddev_suspend()")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Link: https://lore.kernel.org/r/20230731022800.1424902-1-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md: raid1: fix potential OOB in raid1_remove_disk()

If rddev->raid_disk is greater than mddev->raid_disks, there will be
an out-of-bounds in raid1_remove_disk(). We have already found
similar reports as follows:

1) commit d17f744e883b ("md-raid10: fix KASAN warning")
2) commit 1ebc2cec0b7d ("dm raid: fix KASAN warning in raid5_remove_disk")

Fix this bug by checking whether the "number" variable is
valid.

Signed-off-by: Zhang Shurong <zhang_shurong@foxmail.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/tencent_0D24426FAC6A21B69AC0C03CE4143A508F09@qq.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid5-cache: fix a deadlock in r5l_exit_log()

Commit b13015af94cf ("md/raid5-cache: Clear conf->log after finishing
work") introduce a new problem:

// caller hold reconfig_mutex
r5l_exit_log
 flush_work(&log->disable_writeback_work)
r5c_disable_writeback_async
 wait_event
  /*
   * conf->log is not NULL, and mddev_trylock()
   * will fail, wait_event() can never pass.
   */
 conf->log = NULL

Fix this problem by setting 'config->log' to NULL before wake_up() as it
used to be, so that wait_event() from r5c_disable_writeback_async() can
exist. In the meantime, move forward md_unregister_thread() so that
null-ptr-deref this commit fixed can still be fixed.

Fixes: b13015af94cf ("md/raid5-cache: Clear conf->log after finishing work")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230708091727.1417894-1-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

ublk: Switch to memdup_user_nul() helper

Use memdup_user_nul() helper instead of open-coding
to simplify the code.

Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230815114815.1551171-1-ruanjinjie@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: uapi: Fix compilation errors using ioprio.h with C++

The use of the "class" argument name in the ioprio_value() inline
function in include/uapi/linux/ioprio.h confuses C++ compilers
resulting in compilation errors such as:

/usr/include/linux/ioprio.h:110:43: error: expected primary-expression before ‘int’
  110 | static __always_inline __u16 ioprio_value(int class, int level, int hint)
      |                                           ^~~

for user C++ programs including linux/ioprio.h.

Avoid these errors by renaming the arguments of the ioprio_value()
function to prioclass, priolevel and priohint. For consistency, the
arguments of the IOPRIO_PRIO_VALUE() and IOPRIO_PRIO_VALUE_HINT() macros
are also renamed in the same manner.

Reported-by: Igor Pylypiv <ipylypiv@google.com>
Fixes: 01584c1e2337 ("scsi: block: Improve ioprio value validity checks")
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Tested-by: Igor Pylypiv <ipylypiv@google.com>
Link: https://lore.kernel.org/r/20230814215833.259286-1-dlemoal@kernel.org
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Bring back zero_fill_bio_iter

This reverts 6f822e1b5d9dda3d20e87365de138046e3baa03a - this helper is
used by bcachefs.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Link: https://lore.kernel.org/r/20230813182636.2966159-4-kent.overstreet@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Allow bio_iov_iter_get_pages() with bio->bi_bdev unset

bio_iov_iter_get_pages() trims the IO based on the block size of the
block device the IO will be issued to.

However, bcachefs is a multi device filesystem; when we're creating the
bio we don't yet know which block device the bio will be submitted to -
we have to handle the alignment checks elsewhere.

Thus this is needed to avoid a null ptr deref.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Link: https://lore.kernel.org/r/20230813182636.2966159-3-kent.overstreet@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Add some exports for bcachefs

 - bio_set_pages_dirty(), bio_check_pages_dirty() - dio path
 - blk_status_to_str() - error messages
 - bio_add_folio() - this should definitely be exported for everyone,
   it's the modern version of bio_add_page()

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Cc: linux-block@vger.kernel.org
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Link: https://lore.kernel.org/r/20230813182636.2966159-2-kent.overstreet@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: fix 'warn: variable dereferenced before check 'req'' from Smatch

The added check of 'req_op(req) == REQ_OP_ZONE_APPEND' should have been
done after the request is confirmed as valid.

Actually here, the request should always been true, so add one
WARN_ON_ONCE(!req), meantime move the zone_append check after
checking the request.

Cc: Andreas Hindborg <a.hindborg@samsung.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Fixes: 29802d7ca33b ("ublk: enable zoned storage support")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230811135216.420404-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: fix bad lockdep annotation in blk-iolatency

A previous commit added a lockdep annotation, but botched it. Use the
right type.

Fixes: 4eb44d10766a ("block: remove init_mutex and open-code blk_iolatency_try_init")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

swim3: mark swim3_init() static

This is the module init function, which by definition is used only
locally, so mark it static to avoid a warning:

drivers/block/swim3.c:1280:5: error: no previous prototype for 'swim3_init' [-Werror=missing-prototypes]

Reviewed-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: remove init_mutex and open-code blk_iolatency_try_init

Commit a13696b83da4 ("blk-iolatency: Make initialization lazy") adds
a mutex named "init_mutex" in blk_iolatency_try_init for the race
condition of initializing RQ_QOS_LATENCY.
Now a new lock has been add to struct request_queue by commit a13bd91be223
("block/rq_qos: protect rq_qos apis with a new lock"). And it has been
held in blkg_conf_open_bdev before calling blk_iolatency_init.
So it's not necessary to keep init_mutex in blk_iolatency_try_init, just
remove it.

Since init_mutex has been removed, blk_iolatency_try_init can be
open-coded back to iolatency_set_limit() like ioc_qos_write().

Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com>
Reviewed-by: Michal Koutný <mkoutny@suse.com>
Link: https://lore.kernel.org/r/20230810035111.2236335-1-lilingfeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: Fix signedness bug returning warning

There are two warnings reported by smatch:

drivers/block/ublk_drv.c:445 ublk_setup_iod_zoned() warn:
signedness bug returning '(-95)'
drivers/block/ublk_drv.c:963 ublk_setup_iod() warn:
signedness bug returning '(-5)'

The type of "blk_status_t" is either be a u32 or u8, but this two
functions return a negative value when not supported or failed. Use
the error code of the blk module to fix these warnings.

Fixes: 29802d7ca33b ("ublk: enable zoned storage support")
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Closes: https://lore.kernel.org/r/202308100201.TCRhgdvN-lkp@intel.com/
Signed-off-by: Li Zetao <lizetao1@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230810084836.3535322-1-lizetao1@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bio-integrity: create multi-page bvecs in bio_integrity_add_page()

In general, the bvec data structure consists of one for physically
continuous pages. But, in the bvec configuration for bip, physically
continuous integrity pages are composed of each bvec.

Allow bio_integrity_add_page() to create multi-page bvecs, just like
the bio payloads. This simplifies adding larger payloads, and fixes
support for non-tiny workloads with nvme, which stopped using
scatterlist for metadata a while ago.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Fixes: 783b94bd9250 ("nvme-pci: do not build a scatterlist to map metadata")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jinyoung Choi <j-young.choi@samsung.com>
Tested-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230803025202epcms2p82f57cbfe32195da38c776377b55aed59@epcms2p8
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bio-integrity: cleanup adding integrity pages to bip's bvec.

bio_integrity_add_page() returns the add length if successful, else 0,
just as bio_add_page.  Simply check return value checking in
bio_integrity_prep to not deal with a > 0 but < len case that can't
happen.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jinyoung Choi <j-young.choi@samsung.com>
Tested-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230803025058epcms2p5a4d0db5da2ad967668932d463661c633@epcms2p5
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bio-integrity: update the payload size in bio_integrity_add_page()

Previously, the bip's bi_size has been set before an integrity pages
were added. If a problem occurs in the process of adding pages for
bip, the bi_size mismatch problem must be dealt with.

When the page is successfully added to bvec, the bi_size is updated.

The parts affected by the change were also contained in this commit.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jinyoung Choi <j-young.choi@samsung.com>
Tested-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230803024956epcms2p38186a17392706650c582d38ef3dbcd32@epcms2p3
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: make bvec_try_merge_hw_page() non-static

This will be used for multi-page configuration for integrity payload.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jinyoung Choi <j-young.choi@samsung.com>
Tested-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230803024827epcms2p838d9e9131492c86a159fff25d195658f@epcms2p8
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/mq-deadline: use correct way to throttling write requests

The original formula was inaccurate:
dd->async_depth = max(1UL, 3 * q->nr_requests / 4);

For write requests, when we assign a tags from sched_tags,
data->shallow_depth will be passed to sbitmap_find_bit,
see the following code:

nr = sbitmap_find_bit_in_word(&sb->map[index],
min_t (unsigned int,
__map_depth(sb, index),
depth),
alloc_hint, wrap);

The smaller of data->shallow_depth and __map_depth(sb, index)
will be used as the maximum range when allocating bits.

For a mmc device (one hw queue, deadline I/O scheduler):
q->nr_requests = sched_tags = 128, so according to the previous
calculation method, dd->async_depth = data->shallow_depth = 96,
and the platform is 64bits with 8 cpus, sched_tags.bitmap_tags.sb.shift=5,
sb.maps[]=32/32/32/32, 32 is smaller than 96, whether it is a read or
a write I/O, tags can be allocated to the maximum range each time,
which has not throttling effect.

In addition, refer to the methods of bfg/kyber I/O scheduler,
limit ratiois are calculated base on sched_tags.bitmap_tags.sb.shift.

This patch can throttle write requests really.

Fixes: 07757588e507 ("block/mq-deadline: Reserve 25% of scheduler tags for synchronous requests")

Signed-off-by: Zhiguo Niu <zhiguo.niu@unisoc.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/1691061162-22898-1-git-send-email-zhiguo.niu@unisoc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: enable zoned storage support

Add zoned storage support to ublk: report_zones and operations:
 - REQ_OP_ZONE_OPEN
 - REQ_OP_ZONE_CLOSE
 - REQ_OP_ZONE_FINISH
 - REQ_OP_ZONE_RESET
 - REQ_OP_ZONE_APPEND

The zone append feature uses the `addr` field of `struct ublksrv_io_cmd` to
communicate ALBA back to the kernel. Therefore ublk must be used with the
user copy feature (UBLK_F_USER_COPY) for zoned storage support to be
available. Without this feature, ublk will not allow zoned storage support.

Signed-off-by: Andreas Hindborg <a.hindborg@samsung.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230804114610.179530-4-nmi@metaspace.dk
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: move check for empty address field on command submission

In preparation for zoned storage support, move the check for empty `addr`
field into the command handler case statement. Note that the check makes no
sense for `UBLK_IO_NEED_GET_DATA` because the `addr` field must always be
set for this command.

Signed-off-by: Andreas Hindborg <a.hindborg@samsung.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230804114610.179530-3-nmi@metaspace.dk
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk: add helper to check if device supports user copy

This will be used by ublk zoned storage support.

Signed-off-by: Andreas Hindborg <a.hindborg@samsung.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230804114610.179530-2-nmi@metaspace.dk
Signed-off-by: Jens Axboe <axboe@kernel.dk>

iocost_monitor: improve it by adding iocg wait_ms

The iocg can have three throttled metrics: wait, debt, delay. This patch
add missing wait_ms to IocgStat to show the latest wait_ms of iocg.

As we are here, group iocg usage percents "inflt%" and "usage%" together,
and group iocg throttled metrics "wait", "debt" and "delay" together.

Effect after changes:

nvme0n1 RUN  per=50.0ms cur_per=177105.713:v1053528.587 busy= +0 vrate=135.00%:270.00% params=ssd_dfl(CQ)
                          active    weight      hweight% inflt% usage%    wait    debt   delay
InterfererGroup0             *   100/  100  54.28/  9.09   0.34  24.07    0.00    0.00    0.00
interfered                   *    84/ 1000  45.72/ 90.91   0.48  41.09    0.00    0.00    0.00

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230804065039.8885-3-chengming.zhou@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

iocost_monitor: print vrate inuse along with base_vrate

The real vrate iocost inuse is not base_vrate, but the atomic vtime_rate.
We need iocost_monitor tool to display this real vrate that iocost use,
to check if the boosted compensated vrate is normal.

Effect after change:

nvme0n1 RUN  per=50.0ms cur_per=172116.580:v1040587.433 busy= +0 \
vrate=135.00%:270.00% params=ssd_dfl(CQ)
                ^
                |
         this is real vrate inuse

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Link: https://lore.kernel.org/r/20230804065039.8885-2-chengming.zhou@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

iocost_monitor: fix kernel queue kobj changes

When I use iocost_monitor on nvme0n1, this error shows up:
"Could not find ioc for nvme0n1"

There is no kobj in struct queue in recent kernel, it seems that the commit
2bd85221a625 ("block: untangle request_queue refcounting from sysfs")
move the queue kobj to struct gendisk.

Fix it by using mq_kobj which is at the same level with queue kobj.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Link: https://lore.kernel.org/r/20230804065039.8885-1-chengming.zhou@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/Kconfig: Fix compile error for romfs

There are some compile errors reported by kernel test robot:

arm-linux-gnueabi-ld: fs/romfs/storage.o: in function `romfs_dev_read':
storage.c:(.text+0x64): undefined reference to `__brelse'
arm-linux-gnueabi-ld: storage.c:(.text+0x9c): undefined reference to `__bread_gfp'
arm-linux-gnueabi-ld: fs/romfs/storage.o: in function `romfs_dev_strnlen':
storage.c:(.text+0x128): undefined reference to `__brelse'
arm-linux-gnueabi-ld: storage.c:(.text+0x16c): undefined reference to `__bread_gfp'
arm-linux-gnueabi-ld: fs/romfs/storage.o: in function `romfs_dev_strcmp':
storage.c:(.text+0x22c): undefined reference to `__bread_gfp'
arm-linux-gnueabi-ld: storage.c:(.text+0x27c): undefined reference to `__brelse'
arm-linux-gnueabi-ld: storage.c:(.text+0x2a8): undefined reference to `__bread_gfp'
arm-linux-gnueabi-ld: storage.c:(.text+0x2bc): undefined reference to `__brelse'
arm-linux-gnueabi-ld: storage.c:(.text+0x2d4): undefined reference to `__brelse'
arm-linux-gnueabi-ld: storage.c:(.text+0x2f4): undefined reference to `__brelse'
arm-linux-gnueabi-ld: storage.c:(.text+0x304): undefined reference to `__brelse'

The reason for the problem is that the commit
"925c86a19bac" ("fs: add CONFIG_BUFFER_HEAD") has added a new config
"CONFIG_BUFFER_HEAD" that controls building the buffer_head code, and
romfs needs to use the buffer_head API, but no corresponding config has
beed added. Select the config "CONFIG_BUFFER_HEAD" in romfs Kconfig to
resolve the problem.

Fixes: 925c86a19bac ("fs: add CONFIG_BUFFER_HEAD")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202308031810.pQzGmR1v-lkp@intel.com/
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Tested-by: Li Zetao <lizetao1@huawei.com>
Signed-off-by: Li Zetao <lizetao1@huawei.com>
[axboe: fold in Christoph's incremental]
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs: add CONFIG_BUFFER_HEAD

Add a new config option that controls building the buffer_head code, and
select it from all file systems and stacking drivers that need it.

For the block device nodes and alternative iomap based buffered I/O path
is provided when buffer_head support is not enabled, and iomap needs a
a small tweak to define the IOMAP_F_BUFFER_HEAD flag to 0 to not call
into the buffer_head code when it doesn't exist.

Otherwise this is just Kconfig and ifdef changes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230801172201.1923299-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: use iomap for writes to block devices

Use iomap in buffer_head compat mode to write to block devices.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Reviewed-by: Pankaj Raghav <p.raghav@samsung.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Link: https://lore.kernel.org/r/20230801172201.1923299-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: stop setting ->direct_IO

Direct I/O on block devices now nevers goes through aops->direct_IO.
Stop setting it and set the FMODE_CAN_ODIRECT in ->open instead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230801172201.1923299-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: open code __generic_file_write_iter for blkdev writes

Open code __generic_file_write_iter to remove the indirect call into
->direct_IO and to prepare using the iomap based write code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Link: https://lore.kernel.org/r/20230801172201.1923299-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs: rename and move block_page_mkwrite_return

block_page_mkwrite_return is neither block nor mkwrite specific, and
should not be under CONFIG_BLOCK.  Move it to mm.h and rename it to
vmf_fs_error.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Link: https://lore.kernel.org/r/20230801172201.1923299-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs: remove emergency_thaw_bdev

Fold emergency_thaw_bdev into it's only caller, to prepare for buffer.c
to be built only when buffer_head support is enabled.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Link: https://lore.kernel.org/r/20230801172201.1923299-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'md-next-20230729' of https://git./linux/kernel/git/song/md into for-6.6/block

Pull MD updates from Song:

"1. Deprecate bitmap file support, by Christoph Hellwig;
 2. Fix deadlock with md sync thread, by Yu Kuai;
 3. Refactor md io accounting, by Yu Kuai;
 4. Various non-urgent fixes by Li Nan, Yu Kuai, and Jack Wang."

* tag 'md-next-20230729' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md: (36 commits)
  md/md-bitmap: hold 'reconfig_mutex' in backlog_store()
  md/md-bitmap: remove unnecessary local variable in backlog_store()
  md/raid10: use dereference_rdev_and_rrdev() to get devices
  md/raid10: factor out dereference_rdev_and_rrdev()
  md/raid10: check replacement and rdev to prevent submit the same io twice
  md/raid1: Avoid lock contention from wake_up()
  md: restore 'noio_flag' for the last mddev_resume()
  md: don't quiesce in mddev_suspend()
  md: remove redundant check in fix_read_error()
  md/raid10: optimize fix_read_error
  md/raid1: prioritize adding disk to 'removed' mirror
  md/md-faulty: enable io accounting
  md/md-linear: enable io accounting
  md/md-multipath: enable io accounting
  md/raid10: switch to use md_account_bio() for io accounting
  md/raid1: switch to use md_account_bio() for io accounting
  raid5: fix missing io accounting in raid5_align_endio()
  md: also clone new io if io accounting is disabled
  md: move initialization and destruction of 'io_acct_set' to md.c
  md: deprecate bitmap file support
  ...

md/md-bitmap: hold 'reconfig_mutex' in backlog_store()

Several reasons why 'reconfig_mutex' should be held:

1) rdev_for_each() is not safe to be called without the lock, because
   rdev can be removed concurrently.
2) mddev_destroy_serial_pool() and mddev_create_serial_pool() should not
   be called concurrently.
3) mddev_suspend() from mddev_destroy/create_serial_pool() should be
   protected by the lock.

Fixes: 10c92fca636e ("md-bitmap: create and destroy wb_info_pool with the change of backlog")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230706083727.608914-3-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/md-bitmap: remove unnecessary local variable in backlog_store()

Local variable is definied first in the beginning of backlog_store(),
there is no need to define it again.

Fixes: 8c13ab115b57 ("md/bitmap: don't set max_write_behind if there is no write mostly device")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230706083727.608914-2-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: use dereference_rdev_and_rrdev() to get devices

Commit 2ae6aaf76912 ("md/raid10: fix io loss while replacement replace
rdev") reads replacement first to prevent io loss. However, there are same
issue in wait_blocked_dev() and raid10_handle_discard(), too. Fix it by
using dereference_rdev_and_rrdev() to get devices.

Fixes: d30588b2731f ("md/raid10: improve raid10 discard request")
Fixes: f2e7e269a752 ("md/raid10: pull the code that wait for blocked dev into one function")
Signed-off-by: Li Nan <linan122@huawei.com>
Link: https://lore.kernel.org/r/20230701080529.2684932-4-linan666@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: factor out dereference_rdev_and_rrdev()

Factor out a helper to get 'rdev' and 'replacement' from config->mirrors.
Just to make code cleaner and prepare to fix the bug of io loss while
'replacement' replace 'rdev'.

There is no functional change.

Signed-off-by: Li Nan <linan122@huawei.com>
Link: https://lore.kernel.org/r/20230701080529.2684932-3-linan666@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: check replacement and rdev to prevent submit the same io twice

After commit 4ca40c2ce099 ("md/raid10: Allow replacement device to be
replace old drive."), 'rdev' and 'replacement' could appear to be
identical. There are already checks for that in wait_blocked_dev() and
raid10_write_request(). Add check for raid10_handle_discard() now.

Signed-off-by: Li Nan <linan122@huawei.com>
Link: https://lore.kernel.org/r/20230701080529.2684932-2-linan666@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid1: Avoid lock contention from wake_up()

wake_up is called unconditionally in a few paths such as make_request(),
which cause lock contention under high concurrency workload like below
    raid1_end_write_request
     wake_up
      __wake_up_common_lock
       spin_lock_irqsave

Improve performance by only call wake_up() if waitqueue is not empty

Fio test script:

[global]
name=random reads and writes
ioengine=libaio
direct=1
readwrite=randrw
rwmixread=70
iodepth=64
buffered=0
filename=/dev/md0
size=1G
runtime=30
time_based
randrepeat=0
norandommap
refill_buffers
ramp_time=10
bs=4k
numjobs=400
group_reporting=1
[job1]

Test result with 2 ramdisk in raid1 on a Intel Broadwell 56 cores server.

Before this patch       With this patch
READ BW=4621MB/s BW=7337MB/s
WRITE BW=1980MB/s BW=3144MB/s

The patch is inspired by Yu Kuai's change for raid10:
https://lore.kernel.org/r/20230621105728.1268542-1-yukuai1@huaweicloud.com

Cc: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230705113227.148494-1-jinpu.wang@ionos.com
Signed-off-by: Song Liu <song@kernel.org>

md: restore 'noio_flag' for the last mddev_resume()

memalloc_noio_save() is called for the first mddev_suspend(), and
repeated mddev_suspend() only increase 'suspended'. However,
memalloc_noio_restore() is also called for the first mddev_resume(),
which means that memory reclaim will be enabled before the last
mddev_resume() is called, while the array is still suspended.

Fix this problem by restore 'noio_flag' for the last mddev_resume().

Fixes: 78f57ef9d50a ("md: use memalloc scope APIs in mddev_suspend()/mddev_resume()")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230628012931.88911-3-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md: don't quiesce in mddev_suspend()

Some levels doesn't implement "pers->quiesce", for example
raid0_quiesce() is empty, and now that all levels will drop 'active_io'
until io is done, wait for 'active_io' to be 0 is enough to make sure all
normal io is done, and percpu_ref_kill() for 'active_io' will make sure
no new normal io can be dispatched. There is no need to call
"pers->quiesce" anymore from mddev_suspend().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230628012931.88911-2-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md: remove redundant check in fix_read_error()

In fix_read_error(), 'success' will be checked immediately after assigning
it, if it is set to 1 then the loop will break. Checking it again in
condition of loop is redundant. Clean it up.

Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230623173236.2513554-3-linan666@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: optimize fix_read_error

We dereference r10_bio->read_slot too many times in fix_read_error().
Optimize it by using a variable to store read_slot.

Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230623173236.2513554-2-linan666@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/raid1: prioritize adding disk to 'removed' mirror

New disk should be added to "removed" position first instead of to be a
replacement. Commit 6090368abcb4 ("md/raid10: prioritize adding disk to
'removed' mirror") has fixed this issue for raid10. Fix it for raid1 now.

Signed-off-by: Li Nan <linan122@huawei.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20230627014332.3810102-1-linan666@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>

md/md-faulty: enable io accounting

use md_account_bio() to enable io accounting, also make sure
mddev_suspend() will wait for all io to be done.

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/20230621165110.1498313-9-yukuai1@huaweicloud.com

md/md-linear: enable io accounting

use md_account_bio() to enable io accounting, also make sure
mddev_suspend() will wait for all io to be done.

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/20230621165110.1498313-8-yukuai1@huaweicloud.com

md/md-multipath: enable io accounting

use md_account_bio() to enable io accounting, also make sure
mddev_suspend() will wait for all io to be done.

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/20230621165110.1498313-7-yukuai1@huaweicloud.com

md/raid10: switch to use md_account_bio() for io accounting

Make sure that 'active_io' will represent inflight io instead of io that
is dispatching, and io accounting from all levels will be consistent.

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/20230621165110.1498313-6-yukuai1@huaweicloud.com

md/raid1: switch to use md_account_bio() for io accounting

Two problems can be fixed this way:

1) 'active_io' will represent inflight io instead of io that is
dispatching.

2) If io accounting is enabled or disabled while io is still inflight,
bio_start_io_acct() and bio_end_io_acct() is not balanced and io
inflight counter will be leaked.

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/20230621165110.1498313-5-yukuai1@huaweicloud.com

raid5: fix missing io accounting in raid5_align_endio()

Io will only be accounted as done from raid5_align_endio() if the io
succeeded, and io inflight counter will be leaked if such io failed.

Fix this problem by switching to use md_account_bio() for io accounting.

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/20230621165110.1498313-4-yukuai1@huaweicloud.com

md: also clone new io if io accounting is disabled

Currently, 'active_io' is grabbed before make_reqeust() is called, and
it's dropped immediately make_reqeust() returns. Hence 'active_io'
actually means io is dispatching, not io is inflight.

For raid0 and raid456 that io accounting is enabled, 'active_io' will
also be grabbed when bio is cloned for io accounting, and this 'active_io'
is dropped until io is done.

Always clone new bio so that 'active_io' will mean that io is inflight,
raid1 and raid10 will switch to use this method in later patches.

Now that bio will be cloned even if io accounting is disabled, also
rename related structure from '*_acct_*' to '*_clone_*'.

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/20230621165110.1498313-3-yukuai1@huaweicloud.com

md: move initialization and destruction of 'io_acct_set' to md.c

'io_acct_set' is only used for raid0 and raid456, prepare to use it for
raid1 and raid10, so that io accounting from different levels can be
consistent.

By the way, follow up patches will also use this io clone mechanism to
make sure 'active_io' represents in flight io, not io that is dispatching,
so that mddev_suspend will wait for io to be done as designed.

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/20230621165110.1498313-2-yukuai1@huaweicloud.com

md: deprecate bitmap file support

The support for bitmaps on files is a very bad idea abusing various kernel
APIs, and fundamentally requires the file to not be on the actual array
without a way to check that this is actually the case.  Add a deprecation
warning to see if we might be able to eventually drop it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-12-hch@lst.de

md: make bitmap file support optional

The support for write intent bitmaps in files on an external files in md
is a hot mess that abuses ->bmap to map file offsets into physical device
objects, and also abuses buffer_heads in a creative way.

Make this code optional so that MD can be built into future kernels
without buffer_head support, and so that we can eventually deprecate it.

Note this does not affect the internal bitmap support, which has none of
the problems.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-11-hch@lst.de

md-bitmap: don't use ->index for pages backing the bitmap file

The md driver allocates pages for storing the bitmap file data, which
are not page cache pages, and then stores the page granularity file
offset in page->index, which is a field that isn't really valid except
for page cache pages.

Use a separate index for the superblock, and use the scheme used at
read size to recalculate the index for the bitmap pages instead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-10-hch@lst.de

md-bitmap: account for mddev->bitmap_info.offset in read_sb_page

Diretly apply mddev->bitmap_info.offset to the sector number to read
instead of doing that in both callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-9-hch@lst.de

md-bitmap: cleanup read_sb_page

Convert read_sb_page to the normal kernel coding style, calculate the
target sector only once, and add a local iosize variable to make the call
to sync_page_io more readable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-8-hch@lst.de

md-bitmap: refactor md_bitmap_init_from_disk

Split the confusing loop in md_bitmap_init_from_disk that iterates over
all chunks but also needs to read and map the pages into three separate
loops: one that iterates over the pages to read them, a second optional
one to iterate over the pages to mark them invalid if the bitmaps are
out of date, and a final one that actually iterates over the chunks.

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202306160552.smw0qbmb-lkp@intel.com/
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-7-hch@lst.de

md-bitmap: rename read_page to read_file_page

Make the difference to read_sb_page clear.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-6-hch@lst.de

md-bitmap: split file writes into a separate helper

Split the file write code out of write_page into a separate helper.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-5-hch@lst.de

md-bitmap: use %pD to print the file name in md_bitmap_file_kick

Don't bother allocating an extra buffer in the I/O failure handler and
instead use the printk built-in format to print the last 4 path name
components.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-4-hch@lst.de

md-bitmap: initialize variables at declaration time in md_bitmap_file_unmap

Just a small tidyup to prepare for bigger changes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-3-hch@lst.de

md-bitmap: set BITMAP_WRITE_ERROR in write_sb_page

Set BITMAP_WRITE_ERROR directly in write_sb_page instead of propagating
the error to the caller and setting it there.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230615064840.629492-2-hch@lst.de

md: enhance checking in md_check_recovery()

For md_check_recovery():

1) if 'MD_RECOVERY_RUNING' is not set, register new sync_thread.
2) if 'MD_RECOVERY_RUNING' is set:
 a) if 'MD_RECOVERY_DONE' is not set, don't do anything, wait for
   md_do_sync() to be done.
 b) if 'MD_RECOVERY_DONE' is set, unregister sync_thread. Current code
   expects that sync_thread is not NULL, otherwise new sync_thread will
   be registered, which will corrupt the array.

Make sure md_check_recovery() won't register new sync_thread if
'MD_RECOVERY_RUNING' is still set, and a new WARN_ON_ONCE() is added for
the above corruption,

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/20230529132037.2124527-7-yukuai1@huaweicloud.com

md: wake up 'resync_wait' at last in md_reap_sync_thread()

md_reap_sync_thread() is just replaced with wait_event(resync_wait, ...)
from action_store(), just make sure action_store() will still wait for
everything to be done in md_reap_sync_thread().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewd-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230529132037.2124527-6-yukuai1@huaweicloud.com

md: refactor idle/frozen_sync_thread() to fix deadlock

Our test found a following deadlock in raid10:

1) Issue a normal write, and such write failed:

  raid10_end_write_request
   set_bit(R10BIO_WriteError, &r10_bio->state)
   one_write_done
    reschedule_retry

  // later from md thread
  raid10d
   handle_write_completed
    list_add(&r10_bio->retry_list, &conf->bio_end_io_list)

  // later from md thread
  raid10d
   if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
    list_move(conf->bio_end_io_list.prev, &tmp)
    r10_bio = list_first_entry(&tmp, struct r10bio, retry_list)
    raid_end_bio_io(r10_bio)

Dependency chain 1: normal io is waiting for updating superblock

2) Trigger a recovery:

  raid10_sync_request
   raise_barrier

Dependency chain 2: sync thread is waiting for normal io

3) echo idle/frozen to sync_action:

  action_store
   mddev_lock
    md_unregister_thread
     kthread_stop

Dependency chain 3: drop 'reconfig_mutex' is waiting for sync thread

4) md thread can't update superblock:

  raid10d
   md_check_recovery
    if (mddev_trylock(mddev))
     md_update_sb

Dependency chain 4: update superblock is waiting for 'reconfig_mutex'

Hence cyclic dependency exist, in order to fix the problem, we must
break one of them. Dependency 1 and 2 can't be broken because they are
foundation design. Dependency 4 may be possible if it can be guaranteed
that no io can be inflight, however, this requires a new mechanism which
seems complex. Dependency 3 is a good choice, because idle/frozen only
requires sync thread to finish, which can be done asynchronously that is
already implemented, and 'reconfig_mutex' is not needed anymore.

This patch switch 'idle' and 'frozen' to wait sync thread to be done
asynchronously, and this patch also add a sequence counter to record how
many times sync thread is done, so that 'idle' won't keep waiting on new
started sync thread.

Noted that raid456 has similiar deadlock([1]), and it's verified[2] this
deadlock can be fixed by this patch as well.

[1] https://lore.kernel.org/linux-raid/5ed54ffc-ce82-bf66-4eff-390cb23bc1ac@molgen.mpg.de/T/#t
[2] https://lore.kernel.org/linux-raid/e9067438-d713-f5f3-0d3d-9e6b0e9efa0e@huaweicloud.com/

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230529132037.2124527-5-yukuai1@huaweicloud.com

md: add a mutex to synchronize idle and frozen in action_store()

Currently, for idle and frozen, action_store will hold 'reconfig_mutex'
and call md_reap_sync_thread() to stop sync thread, however, this will
cause deadlock (explained in the next patch). In order to fix the
problem, following patch will release 'reconfig_mutex' and wait on
'resync_wait', like md_set_readonly() and do_md_stop() does.

Consider that action_store() will set/clear 'MD_RECOVERY_FROZEN'
unconditionally, which might cause unexpected problems, for example,
frozen just set 'MD_RECOVERY_FROZEN' and is still in progress, while
'idle' clear 'MD_RECOVERY_FROZEN' and new sync thread is started, which
might starve in progress frozen. A mutex is added to synchronize idle
and frozen from action_store().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230529132037.2124527-4-yukuai1@huaweicloud.com

md: refactor action_store() for 'idle' and 'frozen'

Prepare to handle 'idle' and 'frozen' differently to fix a deadlock, there
are no functional changes except that MD_RECOVERY_RUNNING is checked
again after 'reconfig_mutex' is held.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230529132037.2124527-3-yukuai1@huaweicloud.com

Revert "md: unlock mddev before reap sync_thread in action_store"

This reverts commit 9dfbdafda3b34e262e43e786077bab8e476a89d1.

Because it will introduce a defect that sync_thread can be running while
MD_RECOVERY_RUNNING is cleared, which will cause some unexpected problems,
for example:

list_add corruption. prev->next should be next (ffff0001ac1daba0), but was ffff0000ce1a02a0. (prev=ffff0000ce1a02a0).
Call trace:
 __list_add_valid+0xfc/0x140
 insert_work+0x78/0x1a0
 __queue_work+0x500/0xcf4
 queue_work_on+0xe8/0x12c
 md_check_recovery+0xa34/0xf30
 raid10d+0xb8/0x900 [raid10]
 md_thread+0x16c/0x2cc
 kthread+0x1a4/0x1ec
 ret_from_fork+0x10/0x18

This is because work is requeued while it's still inside workqueue:

t1: t2:
action_store
 mddev_lock
  if (mddev->sync_thread)
   mddev_unlock
   md_unregister_thread
   // first sync_thread is done
md_check_recovery
 mddev_try_lock
 /*
  * once MD_RECOVERY_DONE is set, new sync_thread
  * can start.
  */
 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery)
 INIT_WORK(&mddev->del_work, md_start_sync)
 queue_work(md_misc_wq, &mddev->del_work)
  test_and_set_bit(WORK_STRUCT_PENDING_BIT, ...)
  // set pending bit
  insert_work
   list_add_tail
 mddev_unlock
   mddev_lock_nointr
   md_reap_sync_thread
   // MD_RECOVERY_RUNNING is cleared
 mddev_unlock

t3:

// before queued work started from t2
md_check_recovery
 // MD_RECOVERY_RUNNING is not set, a new sync_thread can be started
 INIT_WORK(&mddev->del_work, md_start_sync)
  work->data = 0
  // work pending bit is cleared
 queue_work(md_misc_wq, &mddev->del_work)
  insert_work
   list_add_tail
   // list is corrupted

The above commit is reverted to fix the problem, the deadlock this
commit tries to fix will be fixed in following patches.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230529132037.2124527-2-yukuai1@huaweicloud.com

block: cleanup bio_integrity_prep

If a problem occurs in the process of creating an integrity payload, the
status of bio is always BLK_STS_RESOURCE.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jinyoung Choi <j-young.choi@samsung.com>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230725051839epcms2p8e4d20ad6c51326ad032e8406f59d0aaa@epcms2p8
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Improve performance for BLK_MQ_F_BLOCKING drivers

blk_mq_run_queue() runs the queue asynchronously if BLK_MQ_F_BLOCKING
has been set. This is suboptimal since running the queue asynchronously
is slower than running the queue synchronously. This patch modifies
blk_mq_run_queue() as follows if BLK_MQ_F_BLOCKING has been set:
- Run the queue synchronously if it is allowed to sleep.
- Run the queue asynchronously if it is not allowed to sleep.
Additionally, blk_mq_run_hw_queue(hctx, false) calls are modified into
blk_mq_run_hw_queue(hctx, hctx->flags & BLK_MQ_F_BLOCKING) if the caller
may be invoked from atomic context.

The following caller chains have been reviewed:

blk_mq_run_hw_queue(hctx, false)
  blk_mq_get_tag()      /* may sleep, hence the functions it calls may also sleep */
  blk_execute_rq()             /* may sleep */
  blk_mq_run_hw_queues(q, async=false)
    blk_freeze_queue_start()   /* may sleep */
    blk_mq_requeue_work()      /* may sleep */
    scsi_kick_queue()
      scsi_requeue_run_queue() /* may sleep */
      scsi_run_host_queues()
        scsi_ioctl_reset()     /* may sleep */
  blk_mq_insert_requests(hctx, ctx, list, run_queue_async=false)
    blk_mq_dispatch_plug_list(plug, from_sched=false)
      blk_mq_flush_plug_list(plug, from_schedule=false)
        __blk_flush_plug(plug, from_schedule=false)
blk_add_rq_to_plug()
  blk_mq_submit_bio()  /* may sleep if REQ_NOWAIT has not been set */
  blk_mq_plug_issue_direct()
    blk_mq_flush_plug_list()   /* see above */
  blk_mq_dispatch_plug_list(plug, from_sched=false)
    blk_mq_flush_plug_list()   /* see above */
  blk_mq_try_issue_directly()
    blk_mq_submit_bio()        /* may sleep if REQ_NOWAIT has not been set */
  blk_mq_try_issue_list_directly(hctx, list)
    blk_mq_insert_requests() /* see above */

Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20230721172731.955724-4-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

scsi: Remove a blk_mq_run_hw_queues() call

blk_mq_kick_requeue_list() calls blk_mq_run_hw_queues() asynchronously.
Leave out the direct blk_mq_run_hw_queues() call. This patch causes
scsi_run_queue() to call blk_mq_run_hw_queues() asynchronously instead
of synchronously. Since scsi_run_queue() is not called from the hot I/O
submission path, this patch does not affect the hot path.

This patch prepares for allowing blk_mq_run_hw_queue() to sleep if
BLK_MQ_F_BLOCKING has been set. scsi_run_queue() may be called from
atomic context and must not sleep. Hence the removal of the
blk_mq_run_hw_queues(q, false) call. See also scsi_unblock_requests().

Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230721172731.955724-3-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

scsi: Inline scsi_kick_queue()

Inline scsi_kick_queue() to prepare for modifying the second argument
passed to blk_mq_run_hw_queues().

Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230721172731.955724-2-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: don't pass a bio to bio_try_merge_hw_seg

There is no good reason to pass the bio to bio_try_merge_hw_seg.  Just
pass the current bvec and rename the function to bvec_try_merge_hw_page.
This will allow reusing this function for supporting multi-page integrity
payload bvecs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Link: https://lore.kernel.org/r/20230724165433.117645-9-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: move the bi_size update out of __bio_try_merge_page

The update of bi_size is the only thing in __bio_try_merge_page that
needs a bio.  Move it to the callers, and merge __bio_try_merge_page
and page_is_mergeable into a single bvec_try_merge_page that only takes
the current bvec instead of a full bio.  This will allow reusing this
function for supporting multi-page integrity payload bvecs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Link: https://lore.kernel.org/r/20230724165433.117645-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: downgrade a bio_full call in bio_add_page

bio_add_page already checks that there is space in bi_size a little
earlier.  So after we failed to add to an existing segment, just check
that there is another one available instead of duplicating the bi_size
check.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Link: https://lore.kernel.org/r/20230724165433.117645-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: move the bi_size overflow check in __bio_try_merge_page

Checking for availability in bi_size in a function that attempts to
merge into an existing segment is a bit odd, as the limit also applies
when adding a new segment.  This code works fine as we always call
__bio_try_merge_page, but contributes to sub-optimal calling conventions
and doesn't lead to clear code.

Move it to two of the callers instead, the third one already has a more
strict check that includes max_hw_segments anyway.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230724165433.117645-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: move the bi_vcnt check out of __bio_try_merge_page

Move the bi_vcnt out of __bio_try_merge_page and into the two callers
that don't already have it in preparation for additional changes to
__bio_try_merge_page.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230724165433.117645-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: move the BIO_CLONED checks out of __bio_try_merge_page

__bio_try_merge_page is a way too low-level helper to assert that the
bio is not cloned.  Move the check into bio_add_page and
bio_iov_iter_get_pages instead, which are the high level entry points
that should enforce this variant.  bio_add_hw_page already this
check, coverig the third (indirect) caller of __bio_try_merge_page.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230724165433.117645-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: use SECTOR_SHIFT bio_add_hw_page

Use the SECTOR_SHIFT magic constant instead of the magic number.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinyoung Choi <j-young.choi@samsung.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20230724165433.117645-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>