nvmet: add helpers to set the result field for connect commands

The code to set the result field for the admin and I/O connect commands
is not only verbose and duplicated, but also violates the aliasing
rules as it accesses both the u16 and u32 members in the union.

Add a little helper to sort all that out.

Fixes: db1312dd9548 ("nvmet: implement basic In-Band Authentication")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.de>

nvme: improve the NVME_CONNECT_AUTHREQ* definitions

Mark them as unsigned so that we don't need extra casts, and define
them relative to cdword0 instead of requiring extra shifts.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.de>

nvmet-auth: don't try to cancel a non-initialized work_struct

Currently blktests nvme/002 trips up debugobjects if CONFIG_NVME_AUTH is
enabled, but authentication is not on a queue.  This is because
nvmet_auth_sq_free cancels sq->auth_expired_work unconditionaly, while
auth_expired_work is only ever initialized if authentication is enabled
for a given controller.

Fix this by calling most of what is nvmet_init_auth unconditionally
when initializing the SQ, and just do the setting of the result
field in the connect command handler.

Fixes: db1312dd9548 ("nvmet: implement basic In-Band Authentication")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.de>

nvmet-tcp: remove nvmet_tcp_finish_cmd

There is only a single call-site of nvmet_tcp_finish_cmd(), this
becomes redundant. Remove nvmet_tcp_finish_cmd() and use the original
function body instead.

Suggested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: add bounds check on Transfer Tag

ttag is used as an index to get cmd in nvmet_tcp_handle_h2c_data_pdu(),
add a bounds check to avoid out-of-bounds access.

Signed-off-by: Varun Prakash <varun@chelsio.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: handle ICReq PDU received in NVMET_TCP_Q_LIVE state

As per NVMe/TCP transport specification ICReq PDU is the first PDU received
by the controller and controller should receive only one ICReq PDU.

If controller receives more than one ICReq PDU then this can be considered
as fatal error.

nvmet-tcp driver does not check for ICReq PDU opcode if queue state is
NVMET_TCP_Q_LIVE. In LIVE state ICReq PDU is treated as CapsuleCmd PDU,
this can result in abnormal behavior.

Add a check for ICReq PDU in nvmet_tcp_done_recv_pdu() to fix this issue.

Signed-off-by: Varun Prakash <varun@chelsio.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: fix NULL pointer dereference during release

nvmet-tcp frees CMD buffers in nvmet_tcp_uninit_data_in_cmds(),
and waits the inflight IO requests in nvmet_sq_destroy(). During wait
the inflight IO requests, the callback nvmet_tcp_queue_response()
is called from backend after IO complete, this leads a typical
Use-After-Free issue like this:

 BUG: kernel NULL pointer dereference, address: 0000000000000008
 #PF: supervisor read access in kernel mode
 #PF: error_code(0x0000) - not-present page
 PGD 107f80067 P4D 107f80067 PUD 10789e067 PMD 0
 Oops: 0000 [#1] PREEMPT SMP NOPTI
 CPU: 1 PID: 123 Comm: kworker/1:1H Kdump: loaded Tainted: G            E      6.0.0-rc2.bm.1-amd64 #15
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
 Workqueue: nvmet_tcp_wq nvmet_tcp_io_work [nvmet_tcp]
 RIP: 0010:shash_ahash_digest+0x2b/0x110
 Code: 1f 44 00 00 41 57 41 56 41 55 41 54 55 48 89 fd 53 48 89 f3 48 83 ec 08 44 8b 67 30 45 85 e4 74 1c 48 8b 57 38 b8 00 10 00 00 <44> 8b 7a 08 44 29 f8 39 42 0c 0f 46 42 0c 41 39 c4 76 43 48 8b 03
 RSP: 0018:ffffc9000051bdd8 EFLAGS: 00010206
 RAX: 0000000000001000 RBX: ffff888100ab5470 RCX: 0000000000000000
 RDX: 0000000000000000 RSI: ffff888100ab5470 RDI: ffff888100ab5420
 RBP: ffff888100ab5420 R08: ffff8881024d08c8 R09: ffff888103e1b4b8
 R10: 8080808080808080 R11: 0000000000000000 R12: 0000000000001000
 R13: 0000000000000000 R14: ffff88813412bd4c R15: ffff8881024d0800
 FS:  0000000000000000(0000) GS:ffff88883fa40000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000008 CR3: 0000000104b48000 CR4: 0000000000350ee0
 Call Trace:
  <TASK>
  nvmet_tcp_io_work+0xa52/0xb52 [nvmet_tcp]
  ? __switch_to+0x106/0x420
  process_one_work+0x1ae/0x380
  ? process_one_work+0x380/0x380
  worker_thread+0x30/0x360
  ? process_one_work+0x380/0x380
  kthread+0xe6/0x110
  ? kthread_complete_and_exit+0x20/0x20
  ret_from_fork+0x1f/0x30

Separate nvmet_tcp_uninit_data_in_cmds() into two steps:

  uninit data in cmds                  <- new step 1
  nvmet_sq_destroy();
  cancel_work_sync(&queue->io_work);
  free CMD buffers                     <- new step 2

Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: report the actual number of tagset maps

We've been reporting 2 maps regardless of whether the module parameter
asked for anything beyond the default queues. A consequence of this
means that blk-mq will reinitialize the all the hardware contexts and io
schedulers on every controller reset when the mapping is exactly the
same as before. This unnecessary overhead is adding several milliseconds
on a reset for environments that don't need it. Report the actual number
of mappings in use.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: set min_align_mask before calculating max_hw_sectors

If swiotlb is force enabled dma_max_mapping_size ends up calling
swiotlb_max_mapping_size which takes into account the min align mask for
the device.  Set the min align mask for nvme driver before calling
dma_max_mapping_size while calculating max hw sectors.

Signed-off-by: Rishabh Bhatnagar <risbhat@amazon.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: send a rediscover uevent when a persistent discovery controller reconnects

When a discovery controller is disconnected, no AENs will arrive to
notify the host about discovery log change events.

In order to solve this, send a uevent notification when a
persistent discovery controller reconnects. We add a new ctrl
flag NVME_CTRL_STARTED_ONCE that will be set on the first
start, and consecutive calls will find it set, and send the
event to userspace if the controller is a discovery controller.

Upon the event reception, userspace will re-read the discovery
log page and will act upon changes as it sees fit.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: enumerate controller flags

We expect to grow a few of these flags for various purposes
so make them a proper enumeration.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: ensure subsystem reset is single threaded

The subsystem reset writes to a register, so we have to ensure the
device state is capable of handling that otherwise the driver may access
unmapped registers. Use the state machine to ensure the subsystem reset
doesn't try to write registers on a device already undergoing this type
of reset.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=214771
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: restrict management ioctls to admin

The passthrough commands already have this restriction, but the other
operations do not. Require the same capabilities for all users as all of
these operations, which include resets and rescans, can be disruptive.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: copy firmware_rev on each init

The firmware revision can change on after a reset so copy the most
recent info each time instead of just the first time, otherwise the
sysfs firmware_rev entry may contain stale data.

Reported-by: Jeff Lien <jeff.lien@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Chao Leng <lengchao@huawei.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: handle effects after freeing the request

If a reset occurs after the scan work attempts to issue a command, the
reset may quisce the admin queue, which blocks the scan work's command
from dispatching. The scan work will not be able to complete while the
queue is quiesced.

Meanwhile, the reset work will cancel all outstanding admin tags and
wait until all requests have transitioned to idle, which includes the
passthrough request. But the passthrough request won't be set to idle
until after the scan_work flushes, so we're deadlocked.

Fix this by handling the end effects after the request has been freed.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=216354
Reported-by: Jonathan Derrick <Jonathan.Derrick@solidigm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chao Leng <lengchao@huawei.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

blk-cgroup: pass a gendisk to the blkg allocation helpers

Prepare for storing the blkcg information in the gendisk instead of
the request_queue.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-18-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: pass a gendisk to blkcg_schedule_throttle

Pass the gendisk to blkcg_schedule_throttle as part of moving the
blk-cgroup infrastructure to be gendisk based.  Remove the unused
!BLK_CGROUP stub while we're at it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-17-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: pass a gendisk to blkg_destroy_all

Pass the gendisk to blkg_destroy_all as part of moving the blk-cgroup
infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-16-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-throttle: pass a gendisk to blk_throtl_cancel_bios

Pass the gendisk to blk_throtl_cancel_bios as part of moving the
blk-cgroup infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-15-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-throttle: pass a gendisk to blk_throtl_register_queue

Pass the gendisk to blk_throtl_register_queue as part of moving the
blk-cgroup infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-14-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-throttle: pass a gendisk to blk_throtl_init and blk_throtl_exit

Pass the gendisk to blk_throtl_init and blk_throtl_exit as part of moving
the blk-cgroup infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-13-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-iocost: cleanup ioc_qos_write

Use a local disk variable instead of retrieving the disk and
request_queue over and over by various means.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-12-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-iocost: pass a gendisk to blk_iocost_init

Pass the gendisk to blk_iocost_init as part of moving the blk-cgroup
infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-11-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-iocost: simplify ioc_name

Just directly dereference the disk name instead of going through multiple
hoops to find the same value.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-10-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-iolatency: pass a gendisk to blk_iolatency_init

Pass the gendisk to blk_iolatency_init as part of moving the blk-cgroup
infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-9-hch@lst.de
[axboe: missed inline for blk_iolatency_init() and !CONFIG_BLK_CGROUP_IOLATENCY]
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-ioprio: pass a gendisk to blk_ioprio_init and blk_ioprio_exit

Pass the gendisk to blk_ioprio_init and blk_ioprio_exit as part of moving
the blk-cgroup infrastructure to be gendisk based.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: pass a gendisk to blkcg_init_queue and blkcg_exit_queue

Pass the gendisk to blkcg_init_disk and blkcg_exit_disk as part of moving
the blk-cgroup infrastructure to be gendisk based.  Also remove the
rather pointless kerneldoc comments for these internal functions with a
single caller each.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: remove blkg_lookup_check

The combinations of an error check with an ERR_PTR return and a lookup
with a NULL return leads to ugly handling of the return values in the
callers.  Just open coding the check and the lookup is much simpler.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: cleanup the blkg_lookup family of functions

Add a fully inlined blkg_lookup as the extra two checks aren't going
to generated a lot more code vs the call to the slowpath routine, and
open code the hint update in the two callers that care.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: remove open coded blkg_lookup instances

Use blkg_lookup instead of open coding it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: remove blk_queue_root_blkg

Just open code it in the only caller and drop the unused !BLK_CGROUP
stub.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-cgroup: fix error unwinding in blkcg_init_queue

When blk_throtl_init fails, we need to call blk_ioprio_exit.  Switch to
proper goto based unwinding to fix this.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andreas Herrmann <aherrmann@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921180501.1539876-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-mq: don't redirect completion for hctx withs only one ctx mapping

High-performance NVMe devices usually support a large hw queues, which
ensures a 1:1 mapping of hctx and ctx. In this case there will be no
remote request, so we don't need to care about it.

Signed-off-by: Liu Song <liusong@linux.alibaba.com>
Link: https://lore.kernel.org/r/1663731123-81536-1-git-send-email-liusong@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-throttle: improve bypassing bios checkings

"tg->has_rules" is extended to "tg->has_rules_iops/bps", thus bios that
don't need to be throttled can be checked accurately.

With this patch, bio will be throttled if:

1) Bio is read/write, and corresponding read/write iops limit exist.
2) If corresponding doesn't exist, corresponding bps limit exist and
bio is not throttled before.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921095309.1481289-3-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-throttle: remove THROTL_TG_HAS_IOPS_LIMIT

Currently, "tg->has_rules" and "tg->flags & THROTL_TG_HAS_IOPS_LIMIT"
both try to bypass bios that don't need to be throttled, however, they are
a little redundant and both not perfect:

1) "tg->has_rules" only distinguish read and write, but not iops and bps
   limit.
2) "tg->flags & THROTL_TG_HAS_IOPS_LIMIT" only check if iops limit
   exist, read and write is not distinguished, and bps limit is not
   checked.

tg->has_rules will extended to distinguish bps and iops in the following
patch. There is no need to keep the flag.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20220921095309.1481289-2-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk_drv: add START_USER_RECOVERY and END_USER_RECOVERY support

START_USER_RECOVERY and END_USER_RECOVERY are two new control commands
to support user recovery feature.

After a crash, user should send START_USER_RECOVERY, it will:
(1) check if (a)current ublk_device is UBLK_S_DEV_QUIESCED which was
    set by quiesce_work and (b)chardev is released
(2) reinit all ubqs, including:
    (a) put the task_struct and reset ->ubq_daemon to NULL.
    (b) reset all ublk_io.
(3) reset ub->mm to NULL.

Then, user should start a new process and send FETCH_REQ on each
ubq_daemon.

Finally, user should send END_USER_RECOVERY, it will:
(1) wait for all new ubq_daemons getting ready.
(2) update ublksrv_pid
(3) unquiesce the request queue and expect incoming ublk_queue_rq()
(4) convert ub's state to UBLK_S_DEV_LIVE

Note: we can handle STOP_DEV between START_USER_RECOVERY and
END_USER_RECOVERY. This is helpful to users who cannot start new process
after sending START_USER_RECOVERY ctrl-cmd.

Signed-off-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220923153919.44078-7-ZiyangZhang@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk_drv: support UBLK_F_USER_RECOVERY_REISSUE

UBLK_F_USER_RECOVERY_REISSUE implies that:
With a dying ubq_daemon, ublk_drv let monitor_work requeues rq issued to
userspace(ublksrv) before the ubq_daemon is dying.

UBLK_F_USER_RECOVERY_REISSUE is designed for backends which:
(1) tolerate double-write since ublk_drv may issue the same rq
    twice.
(2) does not let frontend users get I/O error, such as read-only FS
    and VM backend.

Signed-off-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220923153919.44078-6-ZiyangZhang@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk_drv: consider recovery feature in aborting mechanism

With USER_RECOVERY feature enabled, the monitor_work schedules
quiesce_work after finding a dying ubq_daemon. The monitor_work
should also abort all rqs issued to userspace before the ubq_daemon is
dying. The quiesce_work's job is to:
(1) quiesce request queue.
(2) check if there is any INFLIGHT rq. If so, we retry until all these
    rqs are requeued and become IDLE. These rqs should be requeued by
ublk_queue_rq(), task work, io_uring fallback wq or monitor_work.
(3) complete all ioucmds by calling io_uring_cmd_done(). We are safe to
    do so because no ioucmd can be referenced now.
(5) set ub's state to UBLK_S_DEV_QUIESCED, which means we are ready for
    recovery. This state is exposed to userspace by GET_DEV_INFO.

The driver can always handle STOP_DEV and cleanup everything no matter
ub's state is LIVE or QUIESCED. After ub's state is UBLK_S_DEV_QUIESCED,
user can recover with new process.

Note: we do not change the default behavior with reocvery feature
disabled. monitor_work still schedules stop_work and abort inflight
rqs. And finally ublk_device is released.

Signed-off-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220923153919.44078-5-ZiyangZhang@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk_drv: requeue rqs with recovery feature enabled

With recovery feature enabled, in ublk_queue_rq or task work
(in exit_task_work or fallback wq), we requeue rqs instead of
ending(aborting) them. Besides, No matter recovery feature is enabled
or disabled, we schedule monitor_work immediately.

Signed-off-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220923153919.44078-4-ZiyangZhang@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk_drv: define macros for recovery feature and check them

Define some macros for recovery feature.

UBLK_S_DEV_QUIESCED implies that ublk_device is quiesced
and is ready for recovery. This state can be observed by userspace.

UBLK_F_USER_RECOVERY implies that:
(1) ublk_drv enables recovery feature. It won't let monitor_work to
    automatically abort rqs and release the device.
(2) With a dying ubq_daemon, ublk_drv ends(aborts) rqs issued to
    userspace(ublksrv) before crash.
(3) With a dying ubq_daemon, in task work and ublk_queue_rq(),
    ublk_drv requeues rqs.

Signed-off-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220923153919.44078-3-ZiyangZhang@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

ublk_drv: check 'current' instead of 'ubq_daemon'

This check is not atomic. So with recovery feature, ubq_daemon may be
modified simultaneously by recovery task. Instead, check 'current' is
safe here because 'current' never changes.

Also add comment explaining this check, which is really important for
understanding recovery feature.

Signed-off-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220923153919.44078-2-ZiyangZhang@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge branch 'md-next' of https://git./linux/kernel/git/song/md into for-6.1/block

Pull MD updates and fixes from Song:

"1. Various raid5 fix and clean up, by Logan Gunthorpe and David Sloan.
 2. Raid10 performance optimization, by Yu Kuai."

* 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md: Fix spelling mistake in comments of r5l_log
  md/raid5: Wait for MD_SB_CHANGE_PENDING in raid5d
  md/raid10: convert resync_lock to use seqlock
  md/raid10: fix improper BUG_ON() in raise_barrier()
  md/raid10: prevent unnecessary calls to wake_up() in fast path
  md/raid10: don't modify 'nr_waitng' in wait_barrier() for the case nowait
  md/raid10: factor out code from wait_barrier() to stop_waiting_barrier()
  md: Remove extra mddev_get() in md_seq_start()
  md/raid5: Remove unnecessary bio_put() in raid5_read_one_chunk()
  md/raid5: Ensure stripe_fill happens on non-read IO with journal
  md/raid5: Don't read ->active_stripes if it's not needed
  md/raid5: Cleanup prototype of raid5_get_active_stripe()
  md/raid5: Drop extern on function declarations in raid5.h
  md/raid5: Refactor raid5_get_active_stripe()
  md: Replace snprintf with scnprintf
  md/raid10: fix compile warning
  md/raid5: Fix spelling mistakes in comments

md: Fix spelling mistake in comments of r5l_log

Fix spelling of dones't in comments.

Signed-off-by: Zhou nan <zhounan@nfschina.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Wait for MD_SB_CHANGE_PENDING in raid5d

A complicated deadlock exists when using the journal and an elevated
group_thrtead_cnt. It was found with loop devices, but its not clear
whether it can be seen with real disks. The deadlock can occur simply
by writing data with an fio script.

When the deadlock occurs, multiple threads will hang in different ways:

 1) The group threads will hang in the blk-wbt code with bios waiting to
    be submitted to the block layer:

        io_schedule+0x70/0xb0
        rq_qos_wait+0x153/0x210
        wbt_wait+0x115/0x1b0
        io_schedule+0x70/0xb0
        rq_qos_wait+0x153/0x210
        wbt_wait+0x115/0x1b0
        __rq_qos_throttle+0x38/0x60
        blk_mq_submit_bio+0x589/0xcd0
        wbt_wait+0x115/0x1b0
        __rq_qos_throttle+0x38/0x60
        blk_mq_submit_bio+0x589/0xcd0
        __submit_bio+0xe6/0x100
        submit_bio_noacct_nocheck+0x42e/0x470
        submit_bio_noacct+0x4c2/0xbb0
        ops_run_io+0x46b/0x1a30
        handle_stripe+0xcd3/0x36b0
        handle_active_stripes.constprop.0+0x6f6/0xa60
        raid5_do_work+0x177/0x330

    Or:
        io_schedule+0x70/0xb0
        rq_qos_wait+0x153/0x210
        wbt_wait+0x115/0x1b0
        __rq_qos_throttle+0x38/0x60
        blk_mq_submit_bio+0x589/0xcd0
        __submit_bio+0xe6/0x100
        submit_bio_noacct_nocheck+0x42e/0x470
        submit_bio_noacct+0x4c2/0xbb0
        flush_deferred_bios+0x136/0x170
        raid5_do_work+0x262/0x330

 2) The r5l_reclaim thread will hang in the same way, submitting a
    bio to the block layer:

        io_schedule+0x70/0xb0
        rq_qos_wait+0x153/0x210
        wbt_wait+0x115/0x1b0
        __rq_qos_throttle+0x38/0x60
        blk_mq_submit_bio+0x589/0xcd0
        __submit_bio+0xe6/0x100
        submit_bio_noacct_nocheck+0x42e/0x470
        submit_bio_noacct+0x4c2/0xbb0
        submit_bio+0x3f/0xf0
        md_super_write+0x12f/0x1b0
        md_update_sb.part.0+0x7c6/0xff0
        md_update_sb+0x30/0x60
        r5l_do_reclaim+0x4f9/0x5e0
        r5l_reclaim_thread+0x69/0x30b

    However, before hanging, the MD_SB_CHANGE_PENDING flag will be
    set for sb_flags in r5l_write_super_and_discard_space(). This
    flag will never be cleared because the submit_bio() call never
    returns.

 3) Due to the MD_SB_CHANGE_PENDING flag being set, handle_stripe()
    will do no processing on any pending stripes and re-set
    STRIPE_HANDLE. This will cause the raid5d thread to enter an
    infinite loop, constantly trying to handle the same stripes
    stuck in the queue.

    The raid5d thread has a blk_plug that holds a number of bios
    that are also stuck waiting seeing the thread is in a loop
    that never schedules. These bios have been accounted for by
    blk-wbt thus preventing the other threads above from
    continuing when they try to submit bios. --Deadlock.

To fix this, add the same wait_event() that is used in raid5_do_work()
to raid5d() such that if MD_SB_CHANGE_PENDING is set, the thread will
schedule and wait until the flag is cleared. The schedule action will
flush the plug which will allow the r5l_reclaim thread to continue,
thus preventing the deadlock.

However, md_check_recovery() calls can also clear MD_SB_CHANGE_PENDING
from the same thread and can thus deadlock if the thread is put to
sleep. So avoid waiting if md_check_recovery() is being called in the
loop.

It's not clear when the deadlock was introduced, but the similar
wait_event() call in raid5_do_work() was added in 2017 by this
commit:

    16d997b78b15 ("md/raid5: simplfy delaying of writes while metadata
                   is updated.")

Link: https://lore.kernel.org/r/7f3b87b6-b52a-f737-51d7-a4eec5c44112@deltatee.com
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

Merge branch 'md-next-raid10-optimize' into md-next

This patchset tries to avoid that two locks are held unconditionally
in hot path.

Test environment:

Architecture:
aarch64 Huawei KUNPENG 920
x86 Intel(R) Xeon(R) Platinum 8380

Raid10 initialize:
mdadm --create /dev/md0 --level 10 --bitmap none --raid-devices 4 \
    /dev/nvme0n1 /dev/nvme1n1 /dev/nvme2n1 /dev/nvme3n1

Test cmd:
(task set -c 0-15) fio -name=0 -ioengine=libaio -direct=1 -\
    group_reporting=1 -randseed=2022 -rwmixread=70 -refill_buffers \
    -filename=/dev/md0 -numjobs=16 -runtime=60s -bs=4k -iodepth=256 \
    -rw=randread

Test result:

aarch64:
before this patchset:           3.2 GiB/s
bind node before this patchset: 6.9 Gib/s
after this patchset:            7.9 Gib/s
bind node after this patchset:  8.0 Gib/s

x86:(bind node is not tested yet)
before this patchset: 7.0 GiB/s
after this patchset : 9.3 GiB/s

Please noted that in the test machine, memory access latency is very bad
across nodes compare to local node in aarch64, which is why bandwidth
while bind node is much better.

md/raid10: convert resync_lock to use seqlock

Currently, wait_barrier() will hold 'resync_lock' to read 'conf->barrier',
and io can't be dispatched until 'barrier' is dropped.

Since holding the 'barrier' is not common, convert 'resync_lock' to use
seqlock so that holding lock can be avoided in fast path.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-and-Tested-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: fix improper BUG_ON() in raise_barrier()

'conf->barrier' is protected by 'conf->resync_lock', reading
'conf->barrier' without holding the lock is wrong.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Acked-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: prevent unnecessary calls to wake_up() in fast path

Currently, wake_up() is called unconditionally in fast path such as
raid10_make_request(), which will cause lock contention under high
concurrency:

raid10_make_request
 wake_up
  __wake_up_common_lock
   spin_lock_irqsave

Improve performance by only call wake_up() if waitqueue is not empty
in allow_barrier() and raid10_make_request().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Acked-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: don't modify 'nr_waitng' in wait_barrier() for the case nowait

For the case nowait in wait_barrier(), there is no point to increase
nr_waiting and then decrease it.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Acked-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: factor out code from wait_barrier() to stop_waiting_barrier()

Currently the nasty condition in wait_barrier() is hard to read. This
patch factors out the condition into a function.

There are no functional changes.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Acked-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md: Remove extra mddev_get() in md_seq_start()

A regression is seen where mddev devices stay permanently after they
are stopped due to an elevated reference count.

This was tracked down to an extra mddev_get() in md_seq_start().

It only happened rarely because most of the time the md_seq_start()
is called with a zero offset. The path with an extra mddev_get() only
happens when it starts with a non-zero offset.

The commit noted below changed an mddev_get() to check its success
but inadvertently left the original call in. Remove the extra call.

Fixes: 12a6caf27324 ("md: only delete entries from all_mddevs when the disk is freed")
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Guoqing Jiang <Guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Remove unnecessary bio_put() in raid5_read_one_chunk()

When running chunk-sized reads on disks with badblocks duplicate bio
free/puts are observed:

   =============================================================================
   BUG bio-200 (Not tainted): Object already free
   -----------------------------------------------------------------------------
   Allocated in mempool_alloc_slab+0x17/0x20 age=3 cpu=2 pid=7504
    __slab_alloc.constprop.0+0x5a/0xb0
    kmem_cache_alloc+0x31e/0x330
    mempool_alloc_slab+0x17/0x20
    mempool_alloc+0x100/0x2b0
    bio_alloc_bioset+0x181/0x460
    do_mpage_readpage+0x776/0xd00
    mpage_readahead+0x166/0x320
    blkdev_readahead+0x15/0x20
    read_pages+0x13f/0x5f0
    page_cache_ra_unbounded+0x18d/0x220
    force_page_cache_ra+0x181/0x1c0
    page_cache_sync_ra+0x65/0xb0
    filemap_get_pages+0x1df/0xaf0
    filemap_read+0x1e1/0x700
    blkdev_read_iter+0x1e5/0x330
    vfs_read+0x42a/0x570
   Freed in mempool_free_slab+0x17/0x20 age=3 cpu=2 pid=7504
    kmem_cache_free+0x46d/0x490
    mempool_free_slab+0x17/0x20
    mempool_free+0x66/0x190
    bio_free+0x78/0x90
    bio_put+0x100/0x1a0
    raid5_make_request+0x2259/0x2450
    md_handle_request+0x402/0x600
    md_submit_bio+0xd9/0x120
    __submit_bio+0x11f/0x1b0
    submit_bio_noacct_nocheck+0x204/0x480
    submit_bio_noacct+0x32e/0xc70
    submit_bio+0x98/0x1a0
    mpage_readahead+0x250/0x320
    blkdev_readahead+0x15/0x20
    read_pages+0x13f/0x5f0
    page_cache_ra_unbounded+0x18d/0x220
   Slab 0xffffea000481b600 objects=21 used=0 fp=0xffff8881206d8940 flags=0x17ffffc0010201(locked|slab|head|node=0|zone=2|lastcpupid=0x1fffff)
   CPU: 0 PID: 34525 Comm: kworker/u24:2 Not tainted 6.0.0-rc2-localyes-265166-gf11c5343fa3f #143
   Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-1ubuntu1.1 04/01/2014
   Workqueue: raid5wq raid5_do_work
   Call Trace:
    <TASK>
    dump_stack_lvl+0x5a/0x78
    dump_stack+0x10/0x16
    print_trailer+0x158/0x165
    object_err+0x35/0x50
    free_debug_processing.cold+0xb7/0xbe
    __slab_free+0x1ae/0x330
    kmem_cache_free+0x46d/0x490
    mempool_free_slab+0x17/0x20
    mempool_free+0x66/0x190
    bio_free+0x78/0x90
    bio_put+0x100/0x1a0
    mpage_end_io+0x36/0x150
    bio_endio+0x2fd/0x360
    md_end_io_acct+0x7e/0x90
    bio_endio+0x2fd/0x360
    handle_failed_stripe+0x960/0xb80
    handle_stripe+0x1348/0x3760
    handle_active_stripes.constprop.0+0x72a/0xaf0
    raid5_do_work+0x177/0x330
    process_one_work+0x616/0xb20
    worker_thread+0x2bd/0x6f0
    kthread+0x179/0x1b0
    ret_from_fork+0x22/0x30
    </TASK>

The double free is caused by an unnecessary bio_put() in the
if(is_badblock(...)) error path in raid5_read_one_chunk().

The error path was moved ahead of bio_alloc_clone() in c82aa1b76787c
("md/raid5: move checking badblock before clone bio in
raid5_read_one_chunk"). The previous code checked and freed align_bio
which required a bio_put. After the move that is no longer needed as
raid_bio is returned to the control of the common io path which
performs its own endio resulting in a double free on bad device blocks.

Fixes: c82aa1b76787c ("md/raid5: move checking badblock before clone bio in raid5_read_one_chunk")
Signed-off-by: David Sloan <david.sloan@eideticom.com>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Guoqing Jiang <Guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Ensure stripe_fill happens on non-read IO with journal

When doing degrade/recover tests using the journal a kernel BUG
is hit at drivers/md/raid5.c:4381 in handle_parity_checks5():

  BUG_ON(!test_bit(R5_UPTODATE, &dev->flags));

This was found to occur because handle_stripe_fill() was skipped
for stripes in the journal due to a condition in that function.
Thus blocks were not fetched and R5_UPTODATE was not set when
the code reached handle_parity_checks5().

To fix this, don't skip handle_stripe_fill() unless the stripe is
for read.

Fixes: 07e83364845e ("md/r5cache: shift complex rmw from read path to write path")
Link: https://lore.kernel.org/linux-raid/e05c4239-41a9-d2f7-3cfa-4aa9d2cea8c1@deltatee.com/
Suggested-by: Song Liu <song@kernel.org>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Don't read ->active_stripes if it's not needed

The atomic_read() is not needed in many cases so only do
the read after the first checks are done.

Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Cleanup prototype of raid5_get_active_stripe()

Drop the three bools in the prototype of raid5_get_active_stripe()
and replace them with a flags parameter.

At the same time, drop the distinction with __raid5_get_active_stripe().

Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Drop extern on function declarations in raid5.h

externs should not be used in function declarations, so clean those
up.

Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Refactor raid5_get_active_stripe()

Refactor raid5_get_active_stripe() without the gotos with an
explicit infinite loop and some additional nesting.

Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Song Liu <song@kernel.org>

md: Replace snprintf with scnprintf

Current code produces a warning as shown below when total characters
in the constituent block device names plus the slashes exceeds 200.
snprintf() returns the number of characters generated from the given
input, which could cause the expression “200 – len” to wrap around
to a large positive number. Fix this by using scnprintf() instead,
which returns the actual number of characters written into the buffer.

[ 1513.267938] ------------[ cut here ]------------
[ 1513.267943] WARNING: CPU: 15 PID: 37247 at <snip>/lib/vsprintf.c:2509 vsnprintf+0x2c8/0x510
[ 1513.267944] Modules linked in:  <snip>
[ 1513.267969] CPU: 15 PID: 37247 Comm: mdadm Not tainted 5.4.0-1085-azure #90~18.04.1-Ubuntu
[ 1513.267969] Hardware name: Microsoft Corporation Virtual Machine/Virtual Machine, BIOS Hyper-V UEFI Release v4.1 05/09/2022
[ 1513.267971] RIP: 0010:vsnprintf+0x2c8/0x510
<-snip->
[ 1513.267982] Call Trace:
[ 1513.267986]  snprintf+0x45/0x70
[ 1513.267990]  ? disk_name+0x71/0xa0
[ 1513.267993]  dump_zones+0x114/0x240 [raid0]
[ 1513.267996]  ? _cond_resched+0x19/0x40
[ 1513.267998]  raid0_run+0x19e/0x270 [raid0]
[ 1513.268000]  md_run+0x5e0/0xc50
[ 1513.268003]  ? security_capable+0x3f/0x60
[ 1513.268005]  do_md_run+0x19/0x110
[ 1513.268006]  md_ioctl+0x195e/0x1f90
[ 1513.268007]  blkdev_ioctl+0x91f/0x9f0
[ 1513.268010]  block_ioctl+0x3d/0x50
[ 1513.268012]  do_vfs_ioctl+0xa9/0x640
[ 1513.268014]  ? __fput+0x162/0x260
[ 1513.268016]  ksys_ioctl+0x75/0x80
[ 1513.268017]  __x64_sys_ioctl+0x1a/0x20
[ 1513.268019]  do_syscall_64+0x5e/0x200
[ 1513.268021]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: 766038846e875 ("md/raid0: replace printk() with pr_*()")
Reviewed-by: Michael Kelley <mikelley@microsoft.com>
Acked-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Saurabh Sengar <ssengar@linux.microsoft.com>
Signed-off-by: Song Liu <song@kernel.org>

md/raid10: fix compile warning

With W=1, compiler complains.

drivers/md/raid10.c:1983: warning: bad line:

Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Song Liu <song@kernel.org>

md/raid5: Fix spelling mistakes in comments

Fix spelling of 'waitting' in comments.

Signed-off-by: XU pengfei <xupengfei@nfschina.com>
Signed-off-by: Song Liu <song@kernel.org>

block/blk-rq-qos: delete useless enmu RQ_QOS_IOPRIO

Since blk-ioprio handing was converted from a rqos policy to a direct call,
RQ_QOS_IOPRIO is not used anymore, just delete it.

Signed-off-by: Li Jinlin <lijinlin3@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://lore.kernel.org/r/20220916023241.32926-1-lijinlin3@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: aoe: use DEFINE_SHOW_ATTRIBUTE to simplify aoe_debugfs

Use DEFINE_SHOW_ATTRIBUTE helper macro to simplify the code.

Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Link: https://lore.kernel.org/r/20220915023424.3198940-1-liushixin2@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: move from strlcpy with unused retval to strscpy

Follow the advice of the below link and prefer 'strscpy' in this
subsystem. Conversion is 1:1 because the return value is not used.
Generated by a coccinelle script.

Link: https://lore.kernel.org/r/CAHk-=wgfRnXz0W3D37d01q3JFkr_i_uTL=V6A6G1oUZcprmknw@mail.gmail.com/
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Acked-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Acked-by: Geoff Levand <geoff@infradead.org>
Link: https://lore.kernel.org/r/20220818205958.6552-1-wsa+renesas@sang-engineering.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/drbd: remove useless comments in receive_DataReply()

All implementations of req->collision, _req_may_be_done and
drbd_fail_pending_reads have been removed, so remove the comments
in receive_DataReply() that provide no useful information.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Acked-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://lore.kernel.org/r/20220920015216.782190-3-cuigaosheng1@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

drbd: remove orphan _req_may_be_done() declaration

The _req_may_be_done() has been removed by
commit 6870ca6d463e ("drbd: factor out master_bio completion
and drbd_request destruction paths"), so remove the orphan
declaration.

Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
Acked-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://lore.kernel.org/r/20220920015216.782190-2-cuigaosheng1@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-wbt: call rq_qos_add() after wb_normal is initialized

Our test found a problem that wbt inflight counter is negative, which
will cause io hang(noted that this problem doesn't exist in mainline):

t1: device create t2: issue io
add_disk
 blk_register_queue
  wbt_enable_default
   wbt_init
    rq_qos_add
    // wb_normal is still 0
/*
 * in mainline, disk can't be opened before
 * bdev_add(), however, in old kernels, disk
 * can be opened before blk_register_queue().
 */
blkdev_issue_flush
                        // disk size is 0, however, it's not checked
                         submit_bio_wait
                          submit_bio
                           blk_mq_submit_bio
                            rq_qos_throttle
                             wbt_wait
      bio_to_wbt_flags
                               rwb_enabled
       // wb_normal is 0, inflight is not increased

    wbt_queue_depth_changed(&rwb->rqos);
     wbt_update_limits
     // wb_normal is initialized
                            rq_qos_track
                             wbt_track
                              rq->wbt_flags |= bio_to_wbt_flags(rwb, bio);
      // wb_normal is not 0，wbt_flags will be set
t3: io completion
blk_mq_free_request
 rq_qos_done
  wbt_done
   wbt_is_tracked
   // return true
   __wbt_done
    wbt_rqw_done
     atomic_dec_return(&rqw->inflight);
     // inflight is decreased

commit 8235b5c1e8c1 ("block: call bdev_add later in device_add_disk") can
avoid this problem, however it's better to fix this problem in wbt:

1) Lower kernel can't backport this patch due to lots of refactor.
2) Root cause is that wbt call rq_qos_add() before wb_normal is
initialized.

Fixes: e34cbd307477 ("blk-wbt: add general throttling mechanism")
Cc: <stable@vger.kernel.org>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20220913105749.3086243-1-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

rnbd-srv: remove struct rnbd_dev

Given that rnbd_srv_sess_dev already has an open_flags member, there
is no need for the rnbd_dev indirection as a simple block_device pointer
works just as well.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20220909131509.3263924-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

rnbd-srv: remove rnbd_dev_{open,close}

These can be trivially open coded in the callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20220909131509.3263924-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

rnbd-srv: remove rnbd_endio

Fold rnbd_endio into the only caller.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20220909131509.3263924-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

rnbd-srv: simplify rnbd_srv_fill_msg_open_rsp

Remove all the wrappers and just get the information directly from
the block device, or where no such helpers exist the request_queue.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Acked-by: Jack Wang <jinpu.wang@ionos.com>
Link: https://lore.kernel.org/r/20220909131509.3263924-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Fix the enum blk_eh_timer_return documentation

The documentation of the blk_eh_timer_return enumeration values does not
reflect correctly how e.g. the SCSI core uses these values. Fix the
documentation.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Damien Le Moal <damien.lemoal@wdc.com>
Cc: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Fixes: 88b0cfad2888 ("block: document the blk_eh_timer_return values")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Link: https://lore.kernel.org/r/20220920200626.3422296-1-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: add device ping attribute

Add a function to check if a device is accessible.
This makes mostly sense for copy pair secondary devices but it will work
for all devices.

The sysfs attribute ping is a write only attribute and will issue a NOP
CCW to the device.
In case of success it will return zero. If the device is not accessible
it will return an error code.

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-8-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: suppress generic error messages for PPRC secondary devices

Suppress generic command reject messages and dump of sense data for
Peer-To-Peer-Remote-Copy (PPRC) secondary errors.
If IO is issued on a PPRC secondary device, a specific
error message is printed instead.

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-7-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: add ioctl to perform a swap of the drivers copy pair

The newly defined ioctl BIODASDCOPYPAIRSWAP takes a structure that
specifies a copy pair that should be swapped. It will call the device
discipline function to perform the swap operation.

The structure looks as followed:

struct dasd_copypair_swap_data_t {
       char primary[20];
       char secondary[20];
       __u8 reserved[64];
};

where primary is the old primary device that will be replaced by the
secondary device. The old primary will become a secondary device
afterwards.

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-6-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: add copy pair swap capability

In case of errors or misbehaviour of the primary device a controlled
failover to one of the configured secondary devices needs to be
performed.

The swap processing stops I/O on the primary device, all requests are
re-queued to the blocklayer queue, the entries in the copy relation are
swapped and finally the link to the blockdevice is moved from primary to
secondary dasd device.
After this, the secondary becomes the new primary device and I/O is
restarted on that device.

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-5-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: add copy pair setup

A copy relation that is configured on the storage server side needs to be
enabled separately in the device driver. A sysfs interface is created
that allows userspace tooling to control such setup.

The following sysfs entries are added to store and read copy relation
information:

copy_pair
    - Add/Delete a copy pair relation to the DASD device driver
    - Query all previously added copy pair relations
copy_role
    - Query the copy pair role of the device

To add a copy pair to the DASD device driver it has to be specified
through the sysfs attribute copy_pair. Only one secondary device can be
specified at a time together with the primary device. Both, secondary
and primary can be used equally to define the copy pair.
The secondary devices have to be offline when adding the copy relation.
The primary device needs to be specified first followed by the comma
separated secondary device.
Read from the copy_pair attribute to get the current setup and write
"clear" to the attribute to delete any existing setup.

Example:
$ echo 0.0.9700,0.0.9740 > /sys/bus/ccw/devices/0.0.9700/copy_pair
$ cat /sys/bus/ccw/devices/0.0.9700/copy_pair
0.0.9700,0.0.9740

During device online processing the required data will be read from the
storage server and the information will be compared to the setup
requested through the copy_pair attribute. The registration of the
primary and secondary device will be handled accordingly.
A blockdevice is only allocated for copy relation primary devices.

To query the copy role of a device read from the copy_role sysfs
attribute. Possible values are primary, secondary, and none.

Example:
$ cat /sys/bus/ccw/devices/0.0.9700/copy_role
primary

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-4-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: add query PPRC function

Add function to query the Peer-to-Peer-Remote-Copy (PPRC) state of a
device by reading the related structure through a read subsystem data call.

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-3-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

s390/dasd: put block allocation in separate function

Put block allocation into a separate function to put some copy pair logic
in it in a later patch.

Signed-off-by: Stefan Haberland <sth@linux.ibm.com>
Reviewed-by: Jan Hoeppner <hoeppner@linux.ibm.com>
Link: https://lore.kernel.org/r/20220920192616.808070-2-sth@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-iocost: Remove unnecessary (void*) conversions

The key pointer is void and hence does not need an explicit cast.

Signed-off-by: Li zeming <zeming@nfschina.com>
Link: https://lore.kernel.org/r/20220919012825.2936-1-zeming@nfschina.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: remove PSI accounting from the bio layer

PSI accounting is now done by the VM code, where it should have been
since the beginning.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: https://lore.kernel.org/r/20220915094200.139713-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

erofs: add manual PSI accounting for the compressed address space

erofs uses an additional address space for compressed data read from disk
in addition to the one directly associated with the inode.  Reading into
the lower address space is open coded using add_to_page_cache_lru instead
of using the filemap.c helper for page allocation micro-optimizations,
which means it is not covered by the MM PSI annotations for ->read_folio
and ->readahead, so add manual ones instead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220915094200.139713-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

btrfs: add manual PSI accounting for compressed reads

btrfs compressed reads try to always read the entire compressed chunk,
even if only a subset is requested.  Currently this is covered by the
magic PSI accounting underneath submit_bio, but that is about to go
away. Instead add manual psi_memstall_{enter,leave} annotations.

Note that for readahead this really should be using readahead_expand,
but the additionals reads are also done for plain ->read_folio where
readahead_expand can't work, so this overall logic is left as-is for
now.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: David Sterba <dsterba@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: https://lore.kernel.org/r/20220915094200.139713-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

sched/psi: export psi_memstall_{enter,leave}

To properly account for all refaults from file system logic, file systems
need to call psi_memstall_enter directly, so export it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: https://lore.kernel.org/r/20220915094200.139713-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

mm: add PSI accounting around ->read_folio and ->readahead calls

PSI tries to account for the cost of bringing back in pages discarded by
the MM LRU management.  Currently the prime place for that is hooked into
the bio submission path, which is a rather bad place:

 - it does not actually account I/O for non-block file systems, of which
   we have many
 - it adds overhead and a layering violation to the block layer

Add the accounting into the two places in the core MM code that read
pages into an address space by calling into ->read_folio and ->readahead
so that the entire file system operations are covered, to broaden
the coverage and allow removing the accounting in the block layer going
forward.

As psi_memstall_enter can deal with nested calls this will not lead to
double accounting even while the bio annotations are still present.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Link: https://lore.kernel.org/r/20220915094200.139713-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: fix comment typo in submit_bio of block-core.c.

This patch fix a comment typo in block-core.c.

Signed-off-by: Ping-Xiang Chen <p.x.chen@uci.edu>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220914074237.31621-1-p.x.chen@uci.edu
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'nvme-6.1-2022-09-20' of git://git.infradead.org/nvme into for-6.1/block

Pull NVMe updates from Christoph:

"nvme updates for Linux 6.1

 - handle number of queue changes in the TCP and RDMA drivers
   (Daniel Wagner)
 - allow changing the number of queues in nvmet (Daniel Wagner)
 - also consider host_iface when checking ip options (Daniel Wagner)
 - don't map pages which can't come from HIGHMEM (Fabio M. De Francesco)
 - avoid unnecessary flush bios in nvmet (Guixin Liu)
 - shrink and better pack the nvme_iod structure (Keith Busch)
 - add comment for unaligned "fake" nqn (Linjun Bao)
 - print actual source IP address through sysfs "address" attr
   (Martin Belanger)
 - various cleanups (Jackie Liu, Wolfram Sang, Genjian Zhang)"

* tag 'nvme-6.1-2022-09-20' of git://git.infradead.org/nvme:
  nvme-tcp: print actual source IP address through sysfs "address" attr
  nvmet-tcp: don't map pages which can't come from HIGHMEM
  nvme-pci: move iod dma_len fill gaps
  nvme-pci: iod npages fits in s8
  nvme-pci: iod's 'aborted' is a bool
  nvme-pci: remove nvme_queue from nvme_iod
  nvme: consider also host_iface when checking ip options
  nvme-rdma: handle number of queue changes
  nvme-tcp: handle number of queue changes
  nvmet: expose max queues to configfs
  nvmet: avoid unnecessary flush bio
  nvmet-auth: remove redundant parameters req
  nvmet-auth: clean up with done_kfree
  nvme-auth: remove the redundant req->cqe->result.u16 assignment operation
  nvme: move from strlcpy with unused retval to strscpy
  nvme: add comment for unaligned "fake" nqn

bcache: fix set_at_max_writeback_rate() for multiple attached devices

Inside set_at_max_writeback_rate() the calculation in following if()
check is wrong,
if (atomic_inc_return(&c->idle_counter) <
    atomic_read(&c->attached_dev_nr) * 6)

Because each attached backing device has its own writeback thread
running and increasing c->idle_counter, the counter increates much
faster than expected. The correct calculation should be,
(counter / dev_nr) < dev_nr * 6
which equals to,
counter < dev_nr * dev_nr * 6

This patch fixes the above mistake with correct calculation, and helper
routine idle_counter_exceeded() is added to make code be more clear.

Reported-by: Mingzhe Zou <mingzhe.zou@easystack.cn>
Signed-off-by: Coly Li <colyli@suse.de>
Acked-by: Mingzhe Zou <mingzhe.zou@easystack.cn>
Link: https://lore.kernel.org/r/20220919161647.81238-6-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache:: fix repeated words in comments

Delete the redundant word 'we'.

Signed-off-by: Jilin Yuan <yuanjilin@cdjrlc.com>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20220919161647.81238-5-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: bset: Fix comment typos

Remove the redundant word `by`, correct the typo `creaated`.

CC: Kent Overstreet <kent.overstreet@gmail.com>
CC: linux-bcache@vger.kernel.org
Signed-off-by: Jules Maselbas <jmaselbas@kalray.eu>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20220919161647.81238-4-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: remove unused bch_mark_cache_readahead function def in stats.h

This is a cleanup for commit 1616a4c2ab1a ("bcache: remove bcache device
self-defined readahead")', currently no user for
bch_mark_cache_readahead() since that commit.

Signed-off-by: Lin Feng <linf@wangsu.com>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20220919161647.81238-3-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: remove unnecessary flush_workqueue

All pending works will be drained by destroy_workqueue(), no need to call
flush_workqueue() explicitly.

Signed-off-by: Li Lei <lilei@szsandstone.com>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20220919161647.81238-2-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme-tcp: print actual source IP address through sysfs "address" attr

TCP transport relies on the routing table to determine which source
address and interface to use when making a connection. Currently, there
is no way to tell from userspace where a connection was made. This
patch exposes the actual source address using a new field named
"src_addr=" in the "address" attribute.

This is needed to diagnose and identify connectivity issues. With the
source address we can infer the interface associated with each
connection.

This was tested with nvme-cli 2.0 to verify it does not have any
adverse effect. The new "src_addr=" field will simply be displayed in
the output of the "list-subsys" or "list -v" commands as shown here.

$ nvme list-subsys
nvme-subsys0 - NQN=nqn.2014-08.org.nvmexpress.discovery
\
 +- nvme0 tcp traddr=192.168.56.1,trsvcid=8009,src_addr=192.168.56.101 live

Signed-off-by: Martin Belanger <martin.belanger@dell.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: don't map pages which can't come from HIGHMEM

kmap() is being deprecated in favor of kmap_local_page().[1]

There are two main problems with kmap(): (1) It comes with an overhead as
mapping space is restricted and protected by a global lock for
synchronization and (2) it also requires global TLB invalidation when the
kmap’s pool wraps and it might block when the mapping space is fully
utilized until a slot becomes available.

The pages which will be mapped are allocated in nvmet_tcp_map_data(),
using the GFP_KERNEL flag. This assures that they cannot come from
HIGHMEM. This imply that a straight page_address() can replace the kmap()
of sg_page(sg) in nvmet_tcp_map_pdu_iovec(). As a side effect, we might
also delete the field "nr_mapped" from struct "nvmet_tcp_cmd" because,
after removing the kmap() calls, there would be no longer any need of it.

In addition, there is no reason to use a kvec for the command receive
data buffers iovec, use a bio_vec instead and let iov_iter handle the
buffer mapping and data copy.

Test with blktests on a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
HIGHMEM64GB enabled.

[1] "[PATCH] checkpatch: Add kmap and kmap_atomic to the deprecated
list" https://lore.kernel.org/all/20220813220034.806698-1-ira.weiny@intel.com/

Cc: Chaitanya Kulkarni <chaitanyak@nvidia.com>
Cc: Keith Busch <kbusch@kernel.org>
Suggested-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com>
Suggested-by: Christoph Hellwig <hch@lst.de>
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
[sagi: added bio_vec plus minor naming changes]
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: move iod dma_len fill gaps

The 32-bit field, dma_len, packs better in the iod struct above the
dma_addr_t on 64-bit systems.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: iod npages fits in s8

The largest allowed transfer is 4MB, which can use at most 1025 PRPs.
Each PRP is 8 bytes, so the maximum number of 4k nvme pages needed for
the iod_list is 3, which fits in an 's8' type.

While modifying this field, change the name to "nr_allocations" to
better represent that this is referring to the number of units allocated
from a dma_pool.

Also introduce a BUILD_BUG_ON to ensure we never accidently increase the
largest transfer limit beyond 127 chained prp lists.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: iod's 'aborted' is a bool

It's only true or false, so make this a bool to reflect that and save
some space in nvme_iod.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: remove nvme_queue from nvme_iod

We can get the nvme_queue from the req just as easily, so remove the
duplicate path to the same structure to save some space.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: consider also host_iface when checking ip options

It's perfectly fine to use the same traddr and trsvcid more than once
as long we use different host interface. This is used in setups where
the host has more than one interface but the target exposes only one
traddr/trsvcid combination.

Use the same acceptance rules for host_iface as we have for
host_traddr.

Signed-off-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Chao Leng <lengchao@huawei.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-rdma: handle number of queue changes

On reconnect, the number of queues might have changed.

In the case where we have more queues available than previously we try
to access queues which are not initialized yet.

The other case where we have less queues than previously, the
connection attempt will fail because the target doesn't support the
old number of queues and we end up in a reconnect loop.

Thus, only start queues which are currently present in the tagset
limited by the number of available queues. Then we update the tagset
and we can start any new queue.

Signed-off-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-tcp: handle number of queue changes

On reconnect, the number of queues might have changed.

In the case where we have more queues available than previously we try
to access queues which are not initialized yet.

The other case where we have less queues than previously, the
connection attempt will fail because the target doesn't support the
old number of queues and we end up in a reconnect loop.

Thus, only start queues which are currently present in the tagset
limited by the number of available queues. Then we update the tagset
and we can start any new queue.

Signed-off-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>