block/rnbd-clt: Move add_disk(dev->gd) to rnbd_clt_setup_gen_disk

It makes more sense to add gendisk in rnbd_clt_setup_gen_disk, instead
of do it in rnbd_clt_map_device.

Signed-off-by: Guoqing Jiang <guoqing.jiang@gmx.com>
Reviewed-by: Danil Kipnis <danil.kipnis@cloud.ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210419073722.15351-6-gi-oh.kim@ionos.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/rnbd-clt: Remove some arguments from rnbd_client_setup_device

Remove them since both sess and idx can be dereferenced from dev. And
sess is not used in the function.

Signed-off-by: Guoqing Jiang <guoqing.jiang@gmx.com>
Reviewed-by: Danil Kipnis <danil.kipnis@cloud.ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210419073722.15351-5-gi-oh.kim@ionos.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/rnbd-clt: Remove some arguments from insert_dev_if_not_exists_devpath

Remove 'pathname' and 'sess' since we can dereference it from 'dev'.

Signed-off-by: Guoqing Jiang <guoqing.jiang@gmx.com>
Reviewed-by: Danil Kipnis <danil.kipnis@cloud.ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210419073722.15351-4-gi-oh.kim@ionos.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Documentation/sysfs-block-rnbd: Add descriptions for remap_device and resize

Two sysfs entries, remap_device and resize, are missing.

Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Link: https://lore.kernel.org/r/20210419073722.15351-3-gi-oh.kim@ionos.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

MAINTAINERS: Change maintainer for rnbd module

Danil steps down, Haris will take over.
Also update email address to ionos.com, the old
cloud.ionos.com will still work for some time.

Signed-off-by: Danil Kipnis <danil.kipnis@cloud.ionos.com>
Acked-by: Md Haris Iqbal <haris.iqbal@cloud.ionos.com>
Signed-off-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Signed-off-by: Gioh Kim <gi-oh.kim@cloud.ionos.com>
Link: https://lore.kernel.org/r/20210419073722.15351-2-gi-oh.kim@ionos.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: remove redundant assignment to variable st

The variable st is being assigned a value that is never read and
it is being updated later with a new value. The initialization is
redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Denis Efremov <efremov@linux.com>
Acked-by: Willy Tarreau <w@1wt.eu>
Link: https://lore.kernel.org/r/20210415130020.1959951-1-colin.king@canonical.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: cleanups: remove FLOPPY_SILENT_DCL_CLEAR undef

FLOPPY_SILENT_DCL_CLEAR is not defined anywhere and comes from pre-git
era. Just drop this undef. There is FD_SILENT_DCL_CLEAR which is really
used.

Signed-off-by: Denis Efremov <efremov@linux.com>
Link: https://lore.kernel.org/r/20210416083449.72700-6-efremov@linux.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: cleanups: use memcpy() to copy reply_buffer

Use memcpy() in raw_cmd_done() to copy reply_buffer instead
of a for loop.

Signed-off-by: Denis Efremov <efremov@linux.com>
Link: https://lore.kernel.org/r/20210416083449.72700-5-efremov@linux.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: cleanups: use memset() to zero reply_buffer

Use memset() to zero reply buffer in raw_cmd_copyin() instead
of a for loop.

Signed-off-by: Denis Efremov <efremov@linux.com>
Link: https://lore.kernel.org/r/20210416083449.72700-4-efremov@linux.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: cleanups: use ST0 as reply_buffer index 0

Use ST0 as 0 index for reply_buffer array. get_fdc_version() is the only
function that uses index 0 directly instead of the ST0 define.

Signed-off-by: Denis Efremov <efremov@linux.com>
Link: https://lore.kernel.org/r/20210416083449.72700-3-efremov@linux.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: cleanups: remove trailing whitespaces

Cleanup trailing whitespaces as checkpatch.pl suggests.

Signed-off-by: Denis Efremov <efremov@linux.com>
Link: https://lore.kernel.org/r/20210416083449.72700-2-efremov@linux.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge branch 'md-next' of https://git./linux/kernel/git/song/md into for-5.13/drivers

Pull MD updates from Song:

"1. mddev_find_or_alloc() clean up, from Christoph.
 2. Fix NULL pointer deref with external bitmap, from Sudhakar."

* 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md/bitmap: wait for external bitmap writes to complete during tear down
  md: do not return existing mddevs from mddev_find_or_alloc
  md: refactor mddev_find_or_alloc
  md: factor out a mddev_alloc_unit helper from mddev_find

Merge tag 'nvme-5.13-2021-04-15' of git://git.infradead.org/nvme into for-5.13/drivers

Pull NVMe updates from Christoph:

"nvme updates for Linux 5.13

 - refactor the ioctl code
 - fix a segmentation fault during io parsing error in nvmet-tcp
   (Elad Grupi)
 - fix NULL derefence in nvme_ctrl_fast_io_fail_tmo_show/store
   (Gopal Tiwari)
 - properly respect the sgl_threshold flag in nvme-pci (Niklas Cassel)
 - misc cleanups (Niklas Cassel, Amit Engel, Minwoo Im, Colin Ian King)"

* tag 'nvme-5.13-2021-04-15' of git://git.infradead.org/nvme:
  nvme: fix NULL derefence in nvme_ctrl_fast_io_fail_tmo_show/store
  nvme: let namespace probing continue for unsupported features
  nvme: factor out nvme_ns_open and nvme_ns_release helpers
  nvme: move nvme_ns_head_ops to multipath.c
  nvme: factor out a nvme_tryget_ns_head helper
  nvme: move the ioctl code to a separate file
  nvme: don't bother to look up a namespace for controller ioctls
  nvme: simplify block device ioctl handling for the !multipath case
  nvme: simplify the compat ioctl handling
  nvme: factor out a nvme_ns_ioctl helper
  nvme: pass a user pointer to nvme_nvm_ioctl
  nvme: cleanup setting the disk name
  nvme: add a nvme_ns_head_multipath helper
  nvme: remove single trailing whitespace
  nvme-multipath: remove single trailing whitespace
  nvme-pci: remove single trailing whitespace
  nvme-pci: don't simple map sgl when sgls are disabled
  nvmet: fix a spelling mistake "nubmer" -> "number"
  nvmet-fc: simplify nvmet_fc_alloc_hostport
  nvmet-tcp: fix a segmentation fault during io parsing error

md/bitmap: wait for external bitmap writes to complete during tear down

NULL pointer dereference was observed in super_written() when it tries
to access the mddev structure.

[The below stack trace is from an older kernel, but the problem described
in this patch applies to the mainline kernel.]

[ 1194.474861] task: ffff8fdd20858000 task.stack: ffffb99d40790000
[ 1194.488000] RIP: 0010:super_written+0x29/0xe1
[ 1194.499688] RSP: 0018:ffff8ffb7fcc3c78 EFLAGS: 00010046
[ 1194.512477] RAX: 0000000000000000 RBX: ffff8ffb7bf4a000 RCX: ffff8ffb78991048
[ 1194.527325] RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff8ffb56b8a200
[ 1194.542576] RBP: ffff8ffb7fcc3c90 R08: 000000000000000b R09: 0000000000000000
[ 1194.558001] R10: ffff8ffb56b8a298 R11: 0000000000000000 R12: ffff8ffb56b8a200
[ 1194.573070] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
[ 1194.588117] FS:  0000000000000000(0000) GS:ffff8ffb7fcc0000(0000) knlGS:0000000000000000
[ 1194.604264] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1194.617375] CR2: 00000000000002b8 CR3: 00000021e040a002 CR4: 00000000007606e0
[ 1194.632327] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 1194.647865] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 1194.663316] PKRU: 55555554
[ 1194.674090] Call Trace:
[ 1194.683735]  <IRQ>
[ 1194.692948]  bio_endio+0xae/0x135
[ 1194.703580]  blk_update_request+0xad/0x2fa
[ 1194.714990]  blk_update_bidi_request+0x20/0x72
[ 1194.726578]  __blk_end_bidi_request+0x2c/0x4d
[ 1194.738373]  __blk_end_request_all+0x31/0x49
[ 1194.749344]  blk_flush_complete_seq+0x377/0x383
[ 1194.761550]  flush_end_io+0x1dd/0x2a7
[ 1194.772910]  blk_finish_request+0x9f/0x13c
[ 1194.784544]  scsi_end_request+0x180/0x25c
[ 1194.796149]  scsi_io_completion+0xc8/0x610
[ 1194.807503]  scsi_finish_command+0xdc/0x125
[ 1194.818897]  scsi_softirq_done+0x81/0xde
[ 1194.830062]  blk_done_softirq+0xa4/0xcc
[ 1194.841008]  __do_softirq+0xd9/0x29f
[ 1194.851257]  irq_exit+0xe6/0xeb
[ 1194.861290]  do_IRQ+0x59/0xe3
[ 1194.871060]  common_interrupt+0x1c6/0x382
[ 1194.881988]  </IRQ>
[ 1194.890646] RIP: 0010:cpuidle_enter_state+0xdd/0x2a5
[ 1194.902532] RSP: 0018:ffffb99d40793e68 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff43
[ 1194.917317] RAX: ffff8ffb7fce27c0 RBX: ffff8ffb7fced800 RCX: 000000000000001f
[ 1194.932056] RDX: 0000000000000000 RSI: 0000000000000004 RDI: 0000000000000000
[ 1194.946428] RBP: ffffb99d40793ea0 R08: 0000000000000004 R09: 0000000000002ed2
[ 1194.960508] R10: 0000000000002664 R11: 0000000000000018 R12: 0000000000000003
[ 1194.974454] R13: 000000000000000b R14: ffffffff925715a0 R15: 0000011610120d5a
[ 1194.988607]  ? cpuidle_enter_state+0xcc/0x2a5
[ 1194.999077]  cpuidle_enter+0x17/0x19
[ 1195.008395]  call_cpuidle+0x23/0x3a
[ 1195.017718]  do_idle+0x172/0x1d5
[ 1195.026358]  cpu_startup_entry+0x73/0x75
[ 1195.035769]  start_secondary+0x1b9/0x20b
[ 1195.044894]  secondary_startup_64+0xa5/0xa5
[ 1195.084921] RIP: super_written+0x29/0xe1 RSP: ffff8ffb7fcc3c78
[ 1195.096354] CR2: 00000000000002b8

bio in the above stack is a bitmap write whose completion is invoked after
the tear down sequence sets the mddev structure to NULL in rdev.

During tear down, there is an attempt to flush the bitmap writes, but for
external bitmaps, there is no explicit wait for all the bitmap writes to
complete. For instance, md_bitmap_flush() is called to flush the bitmap
writes, but the last call to md_bitmap_daemon_work() in md_bitmap_flush()
could generate new bitmap writes for which there is no explicit wait to
complete those writes. The call to md_bitmap_update_sb() will return
simply for external bitmaps and the follow-up call to md_update_sb() is
conditional and may not get called for external bitmaps. This results in a
kernel panic when the completion routine, super_written() is called which
tries to reference mddev in the rdev that has been set to
NULL(in unbind_rdev_from_array() by tear down sequence).

The solution is to call md_super_wait() for external bitmaps after the
last call to md_bitmap_daemon_work() in md_bitmap_flush() to ensure there
are no pending bitmap writes before proceeding with the tear down.

Cc: stable@vger.kernel.org
Signed-off-by: Sudhakar Panneerselvam <sudhakar.panneerselvam@oracle.com>
Reviewed-by: Zhao Heming <heming.zhao@suse.com>
Signed-off-by: Song Liu <song@kernel.org>

md: do not return existing mddevs from mddev_find_or_alloc

Instead of returning an existing mddev, just for it to be discarded
later directly return -EEXIST.  Rename the function to mddev_alloc now
that it doesn't find an existing mddev.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>

md: refactor mddev_find_or_alloc

Allocate the new mddev first speculatively, which greatly simplifies
the code flow.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>

md: factor out a mddev_alloc_unit helper from mddev_find

Split out a self contained helper to find a free minor for the md
"unit" number.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>

nvme: fix NULL derefence in nvme_ctrl_fast_io_fail_tmo_show/store

Adding entry for dev_attr_fast_io_fail_tmo to avoid the kernel crash
while reading and writing the fast_io_fail_tmo.

Fixes: 09fbed636382 (nvme: export fast_io_fail_tmo to sysfs)
Signed-off-by: Gopal Tiwari <gtiwari@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: let namespace probing continue for unsupported features

Instead of failing to scan the namespace entirely when unsupported
features are detected, just mark the gendisk hidden but allow other
access like the upcoming per-namespace character device.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>

nvme: factor out nvme_ns_open and nvme_ns_release helpers

These will be reused for the per-namespace character devices.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>

nvme: move nvme_ns_head_ops to multipath.c

Move the multipath block_device_operations to multipath.c, where they
belong.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>

nvme: factor out a nvme_tryget_ns_head helper

Add a helper to avoid opencoding ns_head->ref manipulations.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>

nvme: move the ioctl code to a separate file

Split out the ioctl code from core.c into a new file.  Also update
copyrights while we're at it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>

nvme: don't bother to look up a namespace for controller ioctls

Don't bother to look up a namespace just to drop if after retreiving the
controller for the multipath case.  Just look up a live controller for
the subsystem directly.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>

nvme: simplify block device ioctl handling for the !multipath case

Only use the existing ioctl handler for the multipath case, and add a
simpler one that reverts to the pre-multipath case for not shared
use case.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>

nvme: simplify the compat ioctl handling

Don't bother defining a separate compat_ioctl handler, and just handle
the NVME_IOCTL_SUBMIT_IO32 case inline.  Also only defined it for those
ABIs (currently just i386 vs x86_64) that are affected.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>

nvme: factor out a nvme_ns_ioctl helper

Factor out a helper for the namespace based ioctls.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>

nvme: pass a user pointer to nvme_nvm_ioctl

Pass the proper user pointer instead of the not all that useful integer
representation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>

nvme: cleanup setting the disk name

Return false from nvme_set_disk_name and let the caller set the
non-multipath name instead of duplicating the naming information in two
places.  Also remove the pointless local variables for the disk name
and flags and the not needed ctrl argument.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>

nvme: add a nvme_ns_head_multipath helper

Move the multipath gendisk out of #ifdef CONFIG_NVME_MULTIPATH and add
a new nvme_ns_head_multipath that uses it to check if a ns_head has
a multipath device associated with it.

Signed-off-by: Minwoo Im <minwoo.im.dev@gmail.com>
[hch: added the IS_ENABLED, converted a few existing users]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>

nvme: remove single trailing whitespace

There is a single trailing whitespace in core.c.
Since this is just a single whitespace, the chances of this affecting
backports to stable should be quite low, so let's just remove it.

Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-multipath: remove single trailing whitespace

There is a single trailing whitespace in multipath.c.
Since this is just a single whitespace, the chances of this affecting
backports to stable should be quite low, so let's just remove it.

Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: remove single trailing whitespace

There is a single trailing whitespace in pci.c.
Since this is just a single whitespace, the chances of this affecting
backports to stable should be quite low, so let's just remove it.

Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: don't simple map sgl when sgls are disabled

According to the module parameter description for sgl_threshold,
a value of 0 means that SGLs are disabled.

If SGLs are disabled, we should respect that, even for the case
where the request is made up of a single physical segment.

Fixes: 297910571f08 ("nvme-pci: optimize mapping single segment requests using SGLs")
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: fix a spelling mistake "nubmer" -> "number"

There is a spelling mistake in a pr_err error message. Fix it.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-fc: simplify nvmet_fc_alloc_hostport

Once a host is already created, avoid allocate additional hostports that
will be thrown away. add an helper function to handle host search.

Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Amit Engel <amit.engel@dell.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: fix a segmentation fault during io parsing error

In case there is an io that contains inline data and it goes to
parsing error flow, command response will free command and iov
before clearing the data on the socket buffer.
This will delay the command response until receive flow is completed.

Fixes: 872d26a391da ("nvmet-tcp: add NVMe over TCP target driver")
Signed-off-by: Elad Grupi <elad.grupi@dell.com>
Signed-off-by: Hou Pu <houpu.main@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

lightnvm: deprecated OCSSD support and schedule it for removal in Linux 5.15

Lightnvm was an innovative idea to expose more low-level control over SSDs.
But it failed to get properly standardized and remains a non-standarized
extension to NVMe that requires vendor specific quirks for a few now mostly
obsolete SSD devices.  The standardized ZNS command set for NVMe has take
over a lot of the approaches and allows for fully standardized operation.

Remove the Linux code to support open channel SSDs as the few production
deployments of the above mentioned SSDs are using userspace driver stacks
instead of the fairly limited Linux support.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Link: https://lore.kernel.org/r/20210413105257.159260-5-matias.bjorling@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: remove duplicate include in lightnvm.h

'linux/blkdev.h' and 'uapi/linux/lightnvm.h' included in 'lightnvm.h'
is duplicated.It is also included in the 5th and 7th line.

Signed-off-by: Zhang Yunkai <zhang.yunkai@zte.com.cn>
Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Link: https://lore.kernel.org/r/20210413105257.159260-4-matias.bjorling@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: return the correct return value

When memdup_user returns an error, memdup_user has two different return
values, use PTR_ERR to get the correct return value.

Signed-off-by: Tian Tao <tiantao6@hisilicon.com>
Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Link: https://lore.kernel.org/r/20210413105257.159260-3-matias.bjorling@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: use kobj_to_dev()

This fixs coccicheck warning:

drivers/nvme//host/lightnvm.c:1243:60-61: WARNING opportunity for
kobj_to_dev()

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Link: https://lore.kernel.org/r/20210413105257.159260-2-matias.bjorling@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: remove the -ERESTARTSYS handling in blkdev_get_by_dev

Now that md has been cleaned up we can get rid of this hack.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: add option for managing virtual boundary

This will enable changing the virtual boundary of null blk devices. For
now, null blk devices didn't have any restriction on the scatter/gather
elements received from the block layer. Add a module parameter and a
configfs option that will control the virtual boundary. This will
enable testing the efficiency of the block layer bounce buffer in case
a suitable application will send discontiguous IO to the given device.

Initial testing with patched FIO showed the following results (64 jobs,
128 iodepth, 1 nullb device):
IO size      READ (virt=false)   READ (virt=true)   Write (virt=false)  Write (virt=true)
----------  ------------------- -----------------  ------------------- -------------------
 1k            10.7M                8482k               10.8M              8471k
 2k            10.4M                8266k               10.4M              8271k
 4k            10.4M                8274k               10.3M              8226k
 8k            10.2M                8131k               9800k              7933k
 16k           9567k                7764k               8081k              6828k
 32k           8865k                7309k               5570k              5153k
 64k           7695k                6586k               2682k              2617k
 128k          5346k                5489k               1320k              1296k

Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
Link: https://lore.kernel.org/r/20210412095523.278632-1-mgurtovoy@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

gdrom: fix compilation error

Use the right name for the struct request variable that removes the
following compilation error :-

make --silent --keep-going --jobs=8
O=/home/tuxbuild/.cache/tuxmake/builds/1/tmp ARCH=sh
CROSS_COMPILE=sh4-linux-gnu- 'CC=sccache sh4-linux-gnu-gcc'
'HOSTCC=sccache gcc'

In file included from /builds/linux/include/linux/scatterlist.h:9,
                 from /builds/linux/include/linux/dma-mapping.h:10,
                 from /builds/linux/drivers/cdrom/gdrom.c:16:
/builds/linux/drivers/cdrom/gdrom.c: In function 'gdrom_readdisk_dma':
/builds/linux/drivers/cdrom/gdrom.c:586:61: error: 'rq' undeclared
(first use in this function)
  586 |  __raw_writel(page_to_phys(bio_page(req->bio)) + bio_offset(rq->bio),
      |                                                             ^~

Fixes: 1d2c82001a5f ("gdrom: support highmem")
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Tested-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: fix a regression of code compiling failure in debug.c

The patch "bcache: remove PTR_CACHE" introduces a compiling failure in
debug.c with following error message,
  In file included from drivers/md/bcache/bcache.h:182:0,
                   from drivers/md/bcache/debug.c:9:
  drivers/md/bcache/debug.c: In function 'bch_btree_verify':
  drivers/md/bcache/debug.c:53:19: error: 'c' undeclared (first use in
  this function)
    bio_set_dev(bio, c->cache->bdev);
                     ^
This patch fixes the regression by replacing c->cache->bdev by b->c->
cache->bdev.

Signed-off-by: Coly Li <colyli@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210411134316.80274-8-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: Use 64-bit arithmetic instead of 32-bit

Cast multiple variables to (int64_t) in order to give the compiler
complete information about the proper arithmetic to use. Notice that
these variables are being used in contexts that expect expressions of
type int64_t  (64 bit, signed). And currently, such expressions are
being evaluated using 32-bit arithmetic.

Fixes: d0cf9503e908 ("octeontx2-pf: ethtool fec mode support")
Addresses-Coverity-ID: 1501724 ("Unintentional integer overflow")
Addresses-Coverity-ID: 1501725 ("Unintentional integer overflow")
Addresses-Coverity-ID: 1501726 ("Unintentional integer overflow")
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20210411134316.80274-7-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

md: bcache: Trivial typo fixes in the file journal.c

s/condidate/candidate/
s/folowing/following/

Signed-off-by: Bhaskar Chowdhury <unixbhaskar@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20210411134316.80274-6-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

md: bcache: avoid -Wempty-body warnings

building with 'make W=1' shows a harmless warning for each user of the
EBUG_ON() macro:

drivers/md/bcache/bset.c: In function 'bch_btree_sort_partial':
drivers/md/bcache/util.h:30:55: error: suggest braces around empty body in an 'if' statement [-Werror=empty-body]
   30 | #define EBUG_ON(cond)                   do { if (cond); } while (0)
      |                                                       ^
drivers/md/bcache/bset.c:1312:9: note: in expansion of macro 'EBUG_ON'
 1312 |         EBUG_ON(oldsize >= 0 && bch_count_data(b) != oldsize);
      |         ^~~~~~~

Reword the macro slightly to avoid the warning.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20210411134316.80274-5-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: use NULL instead of using plain integer as pointer

This fixes the following sparse warnings:
drivers/md/bcache/features.c:22:16: warning: Using plain integer as NULL
pointer

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20210411134316.80274-4-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: remove PTR_CACHE

Remove the PTR_CACHE inline and replace it with a direct dereference
of c->cache.

(Coly Li: fix the typo from PTR_BUCKET to PTR_CACHE in commit log)

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20210411134316.80274-3-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: reduce redundant code in bch_cached_dev_run()

In bch_cached_dev_run(), free(env[1])|free(env[2])|free(buf)
show up three times. This patch introduce out tag in
which free(env[1])|free(env[2])|free(buf) are only called
one time. If we need to call free() when errors occur,
we can set error code to ret, and then goto out tag directly.

Signed-off-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>
Signed-off-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20210411134316.80274-2-colyli@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge branch 'md-next' of https://git./linux/kernel/git/song/md into for-5.13/drivers

Pull MD updates from Song:

"These patches fix a race condition with md_release() and md_open()."

* 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md: split mddev_find
  md: factor out a mddev_find_locked helper from mddev_find
  md: md_open returns -EBUSY when entering racing area

md: split mddev_find

Split mddev_find into a simple mddev_find that just finds an existing
mddev by the unit number, and a more complicated mddev_find that deals
with find or allocating a mddev.

This turns out to fix this bug reported by Zhao Heming.

----------------------------- snip ------------------------------
commit d3374825ce57 ("md: make devices disappear when they are no longer
needed.") introduced protection between mddev creating & removing. The
md_open shouldn't create mddev when all_mddevs list doesn't contain
mddev. With currently code logic, there will be very easy to trigger
soft lockup in non-preempt env.

*** env ***
kvm-qemu VM 2C1G with 2 iscsi luns
kernel should be non-preempt

*** script ***

about trigger 1 time with 10 tests

`1  node1="15sp3-mdcluster1"
2  node2="15sp3-mdcluster2"
3
4  mdadm -Ss
5  ssh ${node2} "mdadm -Ss"
6  wipefs -a /dev/sda /dev/sdb
7  mdadm -CR /dev/md0 -b clustered -e 1.2 -n 2 -l mirror /dev/sda \
   /dev/sdb --assume-clean
8
9  for i in {1..100}; do
10    echo ==== $i ====;
11
12    echo "test  ...."
13    ssh ${node2} "mdadm -A /dev/md0 /dev/sda /dev/sdb"
14    sleep 1
15
16    echo "clean  ....."
17    ssh ${node2} "mdadm -Ss"
18 done
`
I use mdcluster env to trigger soft lockup, but it isn't mdcluster
speical bug. To stop md array in mdcluster env will do more jobs than
non-cluster array, which will leave enough time/gap to allow kernel to
run md_open.

*** stack ***

`ID: 2831   TASK: ffff8dd7223b5040  CPU: 0   COMMAND: "mdadm"
 #0 [ffffa15d00a13b90] __schedule at ffffffffb8f1935f
 #1 [ffffa15d00a13ba8] exact_lock at ffffffffb8a4a66d
 #2 [ffffa15d00a13bb0] kobj_lookup at ffffffffb8c62fe3
 #3 [ffffa15d00a13c28] __blkdev_get at ffffffffb89273b9
 #4 [ffffa15d00a13c98] blkdev_get at ffffffffb8927964
 #5 [ffffa15d00a13cb0] do_dentry_open at ffffffffb88dc4b4
 #6 [ffffa15d00a13ce0] path_openat at ffffffffb88f0ccc
 #7 [ffffa15d00a13db8] do_filp_open at ffffffffb88f32bb
 #8 [ffffa15d00a13ee0] do_sys_open at ffffffffb88ddc7d
 #9 [ffffa15d00a13f38] do_syscall_64 at ffffffffb86053cb ffffffffb900008c

or:
[  884.226509]  mddev_put+0x1c/0xe0 [md_mod]
[  884.226515]  md_open+0x3c/0xe0 [md_mod]
[  884.226518]  __blkdev_get+0x30d/0x710
[  884.226520]  ? bd_acquire+0xd0/0xd0
[  884.226522]  blkdev_get+0x14/0x30
[  884.226524]  do_dentry_open+0x204/0x3a0
[  884.226531]  path_openat+0x2fc/0x1520
[  884.226534]  ? seq_printf+0x4e/0x70
[  884.226536]  do_filp_open+0x9b/0x110
[  884.226542]  ? md_release+0x20/0x20 [md_mod]
[  884.226543]  ? seq_read+0x1d8/0x3e0
[  884.226545]  ? kmem_cache_alloc+0x18a/0x270
[  884.226547]  ? do_sys_open+0x1bd/0x260
[  884.226548]  do_sys_open+0x1bd/0x260
[  884.226551]  do_syscall_64+0x5b/0x1e0
[  884.226554]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
`
*** rootcause ***

"mdadm -A" (or other array assemble commands) will start a daemon "mdadm
--monitor" by default. When "mdadm -Ss" is running, the stop action will
wakeup "mdadm --monitor". The "--monitor" daemon will immediately get
info from /proc/mdstat. This time mddev in kernel still exist, so
/proc/mdstat still show md device, which makes "mdadm --monitor" to open
/dev/md0.

The previously "mdadm -Ss" is removing action, the "mdadm --monitor"
open action will trigger md_open which is creating action. Racing is
happening.

`<thread 1>: "mdadm -Ss"
md_release
  mddev_put deletes mddev from all_mddevs
  queue_work for mddev_delayed_delete
  at this time, "/dev/md0" is still available for opening

<thread 2>: "mdadm --monitor ..."
md_open
 + mddev_find can't find mddev of /dev/md0, and create a new mddev and
 |    return.
 + trigger "if (mddev->gendisk != bdev->bd_disk)" and return
      -ERESTARTSYS.
`
In non-preempt kernel, <thread 2> is occupying on current CPU. and
mddev_delayed_delete which was created in <thread 1> also can't be
schedule.

In preempt kernel, it can also trigger above racing. But kernel doesn't
allow one thread running on a CPU all the time. after <thread 2> running
some time, the later "mdadm -A" (refer above script line 13) will call
md_alloc to alloc a new gendisk for mddev. it will break md_open
statement "if (mddev->gendisk != bdev->bd_disk)" and return 0 to caller,
the soft lockup is broken.
------------------------------ snip ------------------------------

Cc: stable@vger.kernel.org
Fixes: d3374825ce57 ("md: make devices disappear when they are no longer needed.")
Reported-by: Heming Zhao <heming.zhao@suse.com>
Reviewed-by: Heming Zhao <heming.zhao@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>

md: factor out a mddev_find_locked helper from mddev_find

Factor out a self-contained helper to just lookup a mddev by the dev_t
"unit".

Cc: stable@vger.kernel.org
Reviewed-by: Heming Zhao <heming.zhao@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Song Liu <song@kernel.org>

md: md_open returns -EBUSY when entering racing area

commit d3374825ce57 ("md: make devices disappear when they are no longer
needed.") introduced protection between mddev creating & removing. The
md_open shouldn't create mddev when all_mddevs list doesn't contain
mddev. With currently code logic, there will be very easy to trigger
soft lockup in non-preempt env.

This patch changes md_open returning from -ERESTARTSYS to -EBUSY, which
will break the infinitely retry when md_open enter racing area.

This patch is partly fix soft lockup issue, full fix needs mddev_find
is split into two functions: mddev_find & mddev_find_or_alloc. And
md_open should call new mddev_find (it only does searching job).

For more detail, please refer with Christoph's "split mddev_find" patch
in later commits.

*** env ***
kvm-qemu VM 2C1G with 2 iscsi luns
kernel should be non-preempt

*** script ***

about trigger every time with below script

```
1  node1="mdcluster1"
2  node2="mdcluster2"
3
4  mdadm -Ss
5  ssh ${node2} "mdadm -Ss"
6  wipefs -a /dev/sda /dev/sdb
7  mdadm -CR /dev/md0 -b clustered -e 1.2 -n 2 -l mirror /dev/sda \
   /dev/sdb --assume-clean
8
9  for i in {1..10}; do
10    echo ==== $i ====;
11
12    echo "test  ...."
13    ssh ${node2} "mdadm -A /dev/md0 /dev/sda /dev/sdb"
14    sleep 1
15
16    echo "clean  ....."
17    ssh ${node2} "mdadm -Ss"
18 done
```

I use mdcluster env to trigger soft lockup, but it isn't mdcluster
speical bug. To stop md array in mdcluster env will do more jobs than
non-cluster array, which will leave enough time/gap to allow kernel to
run md_open.

*** stack ***

```
[  884.226509]  mddev_put+0x1c/0xe0 [md_mod]
[  884.226515]  md_open+0x3c/0xe0 [md_mod]
[  884.226518]  __blkdev_get+0x30d/0x710
[  884.226520]  ? bd_acquire+0xd0/0xd0
[  884.226522]  blkdev_get+0x14/0x30
[  884.226524]  do_dentry_open+0x204/0x3a0
[  884.226531]  path_openat+0x2fc/0x1520
[  884.226534]  ? seq_printf+0x4e/0x70
[  884.226536]  do_filp_open+0x9b/0x110
[  884.226542]  ? md_release+0x20/0x20 [md_mod]
[  884.226543]  ? seq_read+0x1d8/0x3e0
[  884.226545]  ? kmem_cache_alloc+0x18a/0x270
[  884.226547]  ? do_sys_open+0x1bd/0x260
[  884.226548]  do_sys_open+0x1bd/0x260
[  884.226551]  do_syscall_64+0x5b/0x1e0
[  884.226554]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
```

*** rootcause ***

"mdadm -A" (or other array assemble commands) will start a daemon "mdadm
--monitor" by default. When "mdadm -Ss" is running, the stop action will
wakeup "mdadm --monitor". The "--monitor" daemon will immediately get
info from /proc/mdstat. This time mddev in kernel still exist, so
/proc/mdstat still show md device, which makes "mdadm --monitor" to open
/dev/md0.

The previously "mdadm -Ss" is removing action, the "mdadm --monitor"
open action will trigger md_open which is creating action. Racing is
happening.

```
<thread 1>: "mdadm -Ss"
md_release
  mddev_put deletes mddev from all_mddevs
  queue_work for mddev_delayed_delete
  at this time, "/dev/md0" is still available for opening

<thread 2>: "mdadm --monitor ..."
md_open
 + mddev_find can't find mddev of /dev/md0, and create a new mddev and
 |    return.
 + trigger "if (mddev->gendisk != bdev->bd_disk)" and return
      -ERESTARTSYS.
```

In non-preempt kernel, <thread 2> is occupying on current CPU. and
mddev_delayed_delete which was created in <thread 1> also can't be
schedule.

In preempt kernel, it can also trigger above racing. But kernel doesn't
allow one thread running on a CPU all the time. after <thread 2> running
some time, the later "mdadm -A" (refer above script line 13) will call
md_alloc to alloc a new gendisk for mddev. it will break md_open
statement "if (mddev->gendisk != bdev->bd_disk)" and return 0 to caller,
the soft lockup is broken.

Cc: stable@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Zhao Heming <heming.zhao@suse.com>
Signed-off-by: Song Liu <song@kernel.org>

drbd: use DEFINE_SPINLOCK() for spinlock

spinlock can be initialized automatically with DEFINE_SPINLOCK()
rather than explicitly calling spin_lock_init().

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Guobin Huang <huangguobin4@huawei.com>
Link: https://lore.kernel.org/r/1617710988-49205-1-git-send-email-huangguobin4@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

swim3: support highmem

swim3 only uses the virtual address of a bio to stash it into the data
transfer using virt_to_bus.  But the ppc32 virt_to_bus just uses the
physical address with an offset.  Replace virt_to_bus with a local hack
that performs the equivalent transformation and stop asking for block
layer bounce buffering.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210406061839.811588-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: always use the track buffer

Always use the track buffer that is already used for addresses outside
the 16MB address capability of the floppy controller.  This allows to
remove a lot of code that relies on kernel virtual addresses.  With
this gone there is just a single place left that looks at the bio,
which can be converted to memcpy_{from,to}_page, thus removing the need
for the extra block-layer bounce buffering for highmem pages.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210406061755.811522-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

swim: don't call blk_queue_bounce_limit

m68k doesn't support highmem, so don't bother enabling the block layer
bounce buffer code.  Just for safety throw in a depend on !HIGHMEM.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210406061725.811389-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

gdrom: support highmem

The gdrom driver only has a single reference to the virtual address of
the bio data, and uses that only to get the physical address.  Switch
to deriving the physical address from the page directly and thus avoid
bounce buffering highmem data.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210406061648.811275-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_nl: Demote half-complete kernel-doc headers

Fixes the following W=1 kernel build warning(s):

 from drivers/block/drbd/drbd_nl.c:24:
 drivers/block/drbd/drbd_nl.c: In function ‘drbd_adm_attach’:
 drivers/block/drbd/drbd_nl.c:1968:10: warning: implicit conversion from ‘enum drbd_state_rv’ to ‘enum drbd_ret_code’ [-Wenum-conversion]
 drivers/block/drbd/drbd_nl.c:930: warning: Function parameter or member 'flags' not described in 'drbd_determine_dev_size'
 drivers/block/drbd/drbd_nl.c:930: warning: Function parameter or member 'rs' not described in 'drbd_determine_dev_size'
 drivers/block/drbd/drbd_nl.c:1148: warning: Function parameter or member 'dc' not described in 'drbd_check_al_size'

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-12-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: xen-blkfront: Demote kernel-doc abuses

Fixes the following W=1 kernel build warning(s):

 drivers/block/xen-blkfront.c:1960: warning: Function parameter or member 'dev' not described in 'blkfront_probe'
 drivers/block/xen-blkfront.c:1960: warning: Function parameter or member 'id' not described in 'blkfront_probe'
 drivers/block/xen-blkfront.c:1960: warning: expecting prototype for Allocate the basic(). Prototype was for blkfront_probe() instead
 drivers/block/xen-blkfront.c:2085: warning: Function parameter or member 'dev' not described in 'blkfront_resume'
 drivers/block/xen-blkfront.c:2085: warning: expecting prototype for or a backend(). Prototype was for blkfront_resume() instead
 drivers/block/xen-blkfront.c:2444: warning: wrong kernel-doc identifier on line:

Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: "Roger Pau Monné" <roger.pau@citrix.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: xen-devel@lists.xenproject.org
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Link: https://lore.kernel.org/r/20210312105530.2219008-11-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_receiver: Demote less than half complete kernel-doc header

Fixes the following W=1 kernel build warning(s):

 drivers/block/drbd/drbd_receiver.c:1641: warning: Function parameter or member 'op' not described in 'drbd_submit_peer_request'
 drivers/block/drbd/drbd_receiver.c:1641: warning: Function parameter or member 'op_flags' not described in 'drbd_submit_peer_request'
 drivers/block/drbd/drbd_receiver.c:1641: warning: Function parameter or member 'fault_type' not described in 'drbd_submit_peer_request'

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-10-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_main: Fix a bunch of function documentation discrepancies

Fixes the following W=1 kernel build warning(s):

 drivers/block/drbd/drbd_main.c:278: warning: Function parameter or member 'connection' not described in 'tl_clear'
 drivers/block/drbd/drbd_main.c:278: warning: Excess function parameter 'device' description in 'tl_clear'
 drivers/block/drbd/drbd_main.c:489: warning: Function parameter or member 'cpu_mask' not described in 'drbd_calc_cpu_mask'
 drivers/block/drbd/drbd_main.c:528: warning: Excess function parameter 'device' description in 'drbd_thread_current_set_cpu'
 drivers/block/drbd/drbd_main.c:549: warning: Function parameter or member 'connection' not described in 'drbd_header_size'
 drivers/block/drbd/drbd_main.c:1204: warning: Function parameter or member 'device' not described in 'send_bitmap_rle_or_plain'
 drivers/block/drbd/drbd_main.c:1204: warning: Function parameter or member 'c' not described in 'send_bitmap_rle_or_plain'
 drivers/block/drbd/drbd_main.c:1335: warning: Function parameter or member 'peer_device' not described in '_drbd_send_ack'
 drivers/block/drbd/drbd_main.c:1335: warning: Excess function parameter 'device' description in '_drbd_send_ack'
 drivers/block/drbd/drbd_main.c:1379: warning: Function parameter or member 'peer_device' not described in 'drbd_send_ack'
 drivers/block/drbd/drbd_main.c:1379: warning: Excess function parameter 'device' description in 'drbd_send_ack'
 drivers/block/drbd/drbd_main.c:1892: warning: Function parameter or member 'connection' not described in 'drbd_send_all'
 drivers/block/drbd/drbd_main.c:1892: warning: Function parameter or member 'sock' not described in 'drbd_send_all'
 drivers/block/drbd/drbd_main.c:1892: warning: Function parameter or member 'buffer' not described in 'drbd_send_all'
 drivers/block/drbd/drbd_main.c:1892: warning: Function parameter or member 'size' not described in 'drbd_send_all'
 drivers/block/drbd/drbd_main.c:1892: warning: Function parameter or member 'msg_flags' not described in 'drbd_send_all'
 drivers/block/drbd/drbd_main.c:3525: warning: Function parameter or member 'flags' not described in 'drbd_queue_bitmap_io'
 drivers/block/drbd/drbd_main.c:3563: warning: Function parameter or member 'flags' not described in 'drbd_bitmap_io'

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-9-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_nl: Make conversion to 'enum drbd_ret_code' explicit

Fixes the following W=1 kernel build warning(s):

 from drivers/block/drbd/drbd_nl.c:24:
 drivers/block/drbd/drbd_nl.c: In function ‘drbd_adm_set_role’:
 drivers/block/drbd/drbd_nl.c:793:11: warning: implicit conversion from ‘enum drbd_state_rv’ to ‘enum drbd_ret_code’ [-Wenum-conversion]
 drivers/block/drbd/drbd_nl.c:795:11: warning: implicit conversion from ‘enum drbd_state_rv’ to ‘enum drbd_ret_code’ [-Wenum-conversion]
 drivers/block/drbd/drbd_nl.c: In function ‘drbd_adm_attach’:
 drivers/block/drbd/drbd_nl.c:1965:10: warning: implicit conversion from ‘enum drbd_state_rv’ to ‘enum drbd_ret_code’ [-Wenum-conversion]
 drivers/block/drbd/drbd_nl.c: In function ‘drbd_adm_connect’:
 drivers/block/drbd/drbd_nl.c:2690:10: warning: implicit conversion from ‘enum drbd_state_rv’ to ‘enum drbd_ret_code’ [-Wenum-conversion]
 drivers/block/drbd/drbd_nl.c: In function ‘drbd_adm_disconnect’:
 drivers/block/drbd/drbd_nl.c:2803:11: warning: implicit conversion from ‘enum drbd_state_rv’ to ‘enum drbd_ret_code’ [-Wenum-conversion]

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-8-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_main: Remove duplicate field initialisation

[P_RETRY_WRITE] is initialised more than once.

Fixes the following W=1 kernel build warning(s):

 drivers/block/drbd/drbd_main.c: In function ‘cmdname’:
 drivers/block/drbd/drbd_main.c:3660:22: warning: initialized field overwritten [-Woverride-init]
 drivers/block/drbd/drbd_main.c:3660:22: note: (near initialization for ‘cmdnames[44]’)

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-7-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_receiver: Demote non-conformant kernel-doc headers

Fixes the following W=1 kernel build warning(s):

 drivers/block/drbd/drbd_receiver.c:265: warning: Function parameter or member 'peer_device' not described in 'drbd_alloc_pages'
 drivers/block/drbd/drbd_receiver.c:265: warning: Excess function parameter 'device' description in 'drbd_alloc_pages'
 drivers/block/drbd/drbd_receiver.c:1362: warning: Function parameter or member 'connection' not described in 'drbd_may_finish_epoch'
 drivers/block/drbd/drbd_receiver.c:1362: warning: Excess function parameter 'device' description in 'drbd_may_finish_epoch'
 drivers/block/drbd/drbd_receiver.c:1451: warning: Function parameter or member 'resource' not described in 'drbd_bump_write_ordering'
 drivers/block/drbd/drbd_receiver.c:1451: warning: Function parameter or member 'bdev' not described in 'drbd_bump_write_ordering'
 drivers/block/drbd/drbd_receiver.c:1451: warning: Excess function parameter 'connection' description in 'drbd_bump_write_ordering'
 drivers/block/drbd/drbd_receiver.c:1643: warning: Function parameter or member 'op' not described in 'drbd_submit_peer_request'
 drivers/block/drbd/drbd_receiver.c:1643: warning: Function parameter or member 'op_flags' not described in 'drbd_submit_peer_request'
 drivers/block/drbd/drbd_receiver.c:1643: warning: Function parameter or member 'fault_type' not described in 'drbd_submit_peer_request'
 drivers/block/drbd/drbd_receiver.c:1643: warning: Excess function parameter 'rw' description in 'drbd_submit_peer_request'
 drivers/block/drbd/drbd_receiver.c:3055: warning: Function parameter or member 'peer_device' not described in 'drbd_asb_recover_0p'
 drivers/block/drbd/drbd_receiver.c:3138: warning: Function parameter or member 'peer_device' not described in 'drbd_asb_recover_1p'
 drivers/block/drbd/drbd_receiver.c:3195: warning: Function parameter or member 'peer_device' not described in 'drbd_asb_recover_2p'
 drivers/block/drbd/drbd_receiver.c:4684: warning: Function parameter or member 'peer_device' not described in 'receive_bitmap_plain'
 drivers/block/drbd/drbd_receiver.c:4684: warning: Function parameter or member 'size' not described in 'receive_bitmap_plain'
 drivers/block/drbd/drbd_receiver.c:4684: warning: Function parameter or member 'p' not described in 'receive_bitmap_plain'
 drivers/block/drbd/drbd_receiver.c:4684: warning: Function parameter or member 'c' not described in 'receive_bitmap_plain'
 drivers/block/drbd/drbd_receiver.c:4738: warning: Function parameter or member 'peer_device' not described in 'recv_bm_rle_bits'
 drivers/block/drbd/drbd_receiver.c:4738: warning: Function parameter or member 'p' not described in 'recv_bm_rle_bits'
 drivers/block/drbd/drbd_receiver.c:4738: warning: Function parameter or member 'c' not described in 'recv_bm_rle_bits'
 drivers/block/drbd/drbd_receiver.c:4738: warning: Function parameter or member 'len' not described in 'recv_bm_rle_bits'
 drivers/block/drbd/drbd_receiver.c:4807: warning: Function parameter or member 'peer_device' not described in 'decode_bitmap_c'
 drivers/block/drbd/drbd_receiver.c:4807: warning: Function parameter or member 'p' not described in 'decode_bitmap_c'
 drivers/block/drbd/drbd_receiver.c:4807: warning: Function parameter or member 'c' not described in 'decode_bitmap_c'
 drivers/block/drbd/drbd_receiver.c:4807: warning: Function parameter or member 'len' not described in 'decode_bitmap_c'

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-6-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_state: Fix some function documentation issues

Fixes the following W=1 kernel build warning(s):

 drivers/block/drbd/drbd_state.c:913: warning: Function parameter or member 'connection' not described in 'is_valid_soft_transition'
 drivers/block/drbd/drbd_state.c:913: warning: Excess function parameter 'device' description in 'is_valid_soft_transition'
 drivers/block/drbd/drbd_state.c:1054: warning: Function parameter or member 'warn' not described in 'sanitize_state'
 drivers/block/drbd/drbd_state.c:1054: warning: Excess function parameter 'warn_sync_abort' description in 'sanitize_state'
 drivers/block/drbd/drbd_state.c:1703: warning: Function parameter or member 'state_change' not described in 'after_state_ch'

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-5-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: mtip32xx: mtip32xx: Mark debugging variable 'start' as __maybe_unused

Fixes the following W=1 kernel build warning(s):

 drivers/block/mtip32xx/mtip32xx.c: In function ‘mtip_standby_immediate’:
 drivers/block/mtip32xx/mtip32xx.c:1216:16: warning: variable ‘start’ set but not used [-Wunused-but-set-variable]

Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-4-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: drbd: drbd_interval: Demote some kernel-doc abuses and fix another header

Fixes the following W=1 kernel build warning(s):

 drivers/block/drbd/drbd_interval.c:11: warning: Function parameter or member 'node' not described in 'interval_end'
 drivers/block/drbd/drbd_interval.c:26: warning: Function parameter or member 'root' not described in 'drbd_insert_interval'
 drivers/block/drbd/drbd_interval.c:26: warning: Function parameter or member 'this' not described in 'drbd_insert_interval'
 drivers/block/drbd/drbd_interval.c:70: warning: Function parameter or member 'root' not described in 'drbd_contains_interval'
 drivers/block/drbd/drbd_interval.c:96: warning: Function parameter or member 'root' not described in 'drbd_remove_interval'
 drivers/block/drbd/drbd_interval.c:96: warning: Function parameter or member 'this' not described in 'drbd_remove_interval'
 drivers/block/drbd/drbd_interval.c:113: warning: Function parameter or member 'root' not described in 'drbd_find_overlap'

Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: drbd-dev@lists.linbit.com
Cc: linux-block@vger.kernel.org
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Link: https://lore.kernel.org/r/20210312105530.2219008-3-lee.jones@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'nvme-5.13-2021-04-06' of git://git.infradead.org/nvme into for-5.13/drivers

Pull NVMe updates from Christoph:

"nvme updates for Linux 5.13

 - fix handling of very large MDTS values (Bart Van Assche)
 - retrigger ANA log update if group descriptor isn't found
   (Hannes Reinecke)
 - fix locking contexts in nvme-tcp and nvmet-tcp (Sagi Grimberg)
 - return proper error code from discovery ctrl (Hou Pu)
 - verify the SGLS field in nvmet-tcp and nvmet-fc (Max Gurtovoy)
 - disallow passthru cmd from targeting a nsid != nsid of the block dev
   (Niklas Cassel)
 - do not allow model_number exceed 40 bytes in nvmet (Noam Gottlieb)
 - enable optional queue idle period tracking in nvmet-tcp
   (Mark Wunderlich)
 - various cleanups and optimizations (Chaitanya Kulkarni, Kanchan Joshi)
 - expose fast_io_fail_tmo in sysfs (Daniel Wagner)
 - implement non-MDTS command limits (Keith Busch)
 - reduce warnings for unhandled command effects (Keith Busch)
 - allocate storage for the SQE as part of the nvme_request (Keith Busch)"

* tag 'nvme-5.13-2021-04-06' of git://git.infradead.org/nvme: (33 commits)
  nvme: fix handling of large MDTS values
  nvme: implement non-mdts command limits
  nvme: disallow passthru cmd from targeting a nsid != nsid of the block dev
  nvme: retrigger ANA log update if group descriptor isn't found
  nvme: export fast_io_fail_tmo to sysfs
  nvme: remove superfluous else in nvme_ctrl_loss_tmo_store
  nvme: use sysfs_emit instead of sprintf
  nvme-fc: check sgl supported by target
  nvme-tcp: check sgl supported by target
  nvmet-tcp: enable optional queue idle period tracking
  nvmet-tcp: fix incorrect locking in state_change sk callback
  nvme-tcp: block BH in sk state_change sk callback
  nvmet: return proper error code from discovery ctrl
  nvme: warn of unhandled effects only once
  nvme: use driver pdu command for passthrough
  nvme-pci: allocate nvme_command within driver pdu
  nvmet: do not allow model_number exceed 40 bytes
  nvmet: remove unnecessary ctrl parameter
  nvmet-fc: update function documentation
  nvme-fc: fix the function documentation comment
  ...

nvme: fix handling of large MDTS values

Instead of triggering an integer overflow and undefined behavior if MDTS is
large, set max_hw_sectors to UINT_MAX.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Keith Busch <kbusch@kernel.org>
[hch: rebased to account for the new nvme_mps_to_sectors helper]
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: implement non-mdts command limits

Commands that access LBA contents without a data transfer between the
host historically have not had a spec defined upper limit. The driver
set the queue constraints for such commands to the max data transfer
size just to be safe, but this artificial constraint frequently limits
devices below their capabilities.

The NVMe Workgroup ratified TP4040 defines how a controller may
advertise their non-MDTS limits. Use these if provided and default to
the current constraints if not. Since the Dataset Management command
limits are defined in logical blocks, but without a namespace to tell us
the logical block size, the code defaults to the safe 512b size.

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

nvme: disallow passthru cmd from targeting a nsid != nsid of the block dev

When a passthru command targets a specific namespace, the ns parameter to
nvme_user_cmd()/nvme_user_cmd64() is set. However, there is currently no
validation that the nsid specified in the passthru command targets the
namespace/nsid represented by the block device that the ioctl was
performed on.

Add a check that validates that the nsid in the passthru command matches
that of the supplied namespace.

Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: Javier González <javier@javigon.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: retrigger ANA log update if group descriptor isn't found

If ANA is enabled but no ANA group descriptor is found when creating
a new namespace the ANA log is most likely out of date, so trigger
a re-read. The namespace will be tagged with the NS_ANA_PENDING flag
to exclude it from path selection until the ANA log has been re-read.

Fixes: 32acab3181c7 ("nvme: implement multipath access to nvme subsystems")
Reported-by: Martin George <marting@netapp.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: export fast_io_fail_tmo to sysfs

Commit 8c4dfea97f15 ("nvme-fabrics: reject I/O to offline device")
introduced fast_io_fail_tmo but didn't export the value to sysfs. The
value can be set during the 'nvme connect'. Export the timeout value
to user space via sysfs to allow runtime configuration.

Cc: Victor Gladkov <Victor.Gladkov@kioxia.com>
Signed-off-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Himanshu Madhani <himanshu.madhaani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: remove superfluous else in nvme_ctrl_loss_tmo_store

If there is an error we will leave the function early. So there
is no need for an else. Remove it.

Signed-off-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: use sysfs_emit instead of sprintf

sysfs_emit is the recommended API to use for formatting strings to be
returned to user space. It is equivalent to scnprintf and aware of the
PAGE_SIZE buffer size.

Suggested-by: Chaitanya Kulkarni <Chaitanya.Kulkarni@wdc.com>
Signed-off-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-fc: check sgl supported by target

SGLs support is mandatory for NVMe/FC, make sure that the target is
aligned to the specification.

Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-tcp: check sgl supported by target

SGLs support is mandatory for NVMe/tcp, make sure that the target is
aligned to the specification.

Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: enable optional queue idle period tracking

Add 'idle_poll_period_usecs' option used by io_work() to support
network devices enabled with advanced interrupt moderation
supporting a relaxed interrupt model. It was discovered that
such a NIC used on the target was unable to support initiator
connection establishment, caused by the existing io_work()
flow that immediately exits after a loop with no activity and
does not re-queue itself.

With this new option a queue is assigned a period of time
that no activity must occur in order to become 'idle'.  Until
the queue is idle the work item is requeued.

The new module option is defined as changeable making it
flexible for testing purposes.

The pre-existing legacy behavior is preserved when no module option
for idle_poll_period_usecs is specified.

Signed-off-by: Mark Wunderlich <mark.wunderlich@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-tcp: fix incorrect locking in state_change sk callback

We are not changing anything in the TCP connection state so
we should not take a write_lock but rather a read lock.

This caused a deadlock when running nvmet-tcp and nvme-tcp
on the same system, where state_change callbacks on the
host and on the controller side have causal relationship
and made lockdep report on this with blktests:

================================
WARNING: inconsistent lock state
5.12.0-rc3 #1 Tainted: G          I
--------------------------------
inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-R} usage.
nvme/1324 [HC0[0]:SC0[0]:HE1:SE1] takes:
ffff888363151000 (clock-AF_INET){++-?}-{2:2}, at: nvme_tcp_state_change+0x21/0x150 [nvme_tcp]
{IN-SOFTIRQ-W} state was registered at:
  __lock_acquire+0x79b/0x18d0
  lock_acquire+0x1ca/0x480
  _raw_write_lock_bh+0x39/0x80
  nvmet_tcp_state_change+0x21/0x170 [nvmet_tcp]
  tcp_fin+0x2a8/0x780
  tcp_data_queue+0xf94/0x1f20
  tcp_rcv_established+0x6ba/0x1f00
  tcp_v4_do_rcv+0x502/0x760
  tcp_v4_rcv+0x257e/0x3430
  ip_protocol_deliver_rcu+0x69/0x6a0
  ip_local_deliver_finish+0x1e2/0x2f0
  ip_local_deliver+0x1a2/0x420
  ip_rcv+0x4fb/0x6b0
  __netif_receive_skb_one_core+0x162/0x1b0
  process_backlog+0x1ff/0x770
  __napi_poll.constprop.0+0xa9/0x5c0
  net_rx_action+0x7b3/0xb30
  __do_softirq+0x1f0/0x940
  do_softirq+0xa1/0xd0
  __local_bh_enable_ip+0xd8/0x100
  ip_finish_output2+0x6b7/0x18a0
  __ip_queue_xmit+0x706/0x1aa0
  __tcp_transmit_skb+0x2068/0x2e20
  tcp_write_xmit+0xc9e/0x2bb0
  __tcp_push_pending_frames+0x92/0x310
  inet_shutdown+0x158/0x300
  __nvme_tcp_stop_queue+0x36/0x270 [nvme_tcp]
  nvme_tcp_stop_queue+0x87/0xb0 [nvme_tcp]
  nvme_tcp_teardown_admin_queue+0x69/0xe0 [nvme_tcp]
  nvme_do_delete_ctrl+0x100/0x10c [nvme_core]
  nvme_sysfs_delete.cold+0x8/0xd [nvme_core]
  kernfs_fop_write_iter+0x2c7/0x460
  new_sync_write+0x36c/0x610
  vfs_write+0x5c0/0x870
  ksys_write+0xf9/0x1d0
  do_syscall_64+0x33/0x40
  entry_SYSCALL_64_after_hwframe+0x44/0xae
irq event stamp: 10687
hardirqs last  enabled at (10687): [<ffffffff9ec376bd>] _raw_spin_unlock_irqrestore+0x2d/0x40
hardirqs last disabled at (10686): [<ffffffff9ec374d8>] _raw_spin_lock_irqsave+0x68/0x90
softirqs last  enabled at (10684): [<ffffffff9f000608>] __do_softirq+0x608/0x940
softirqs last disabled at (10649): [<ffffffff9cdedd31>] do_softirq+0xa1/0xd0

other info that might help us debug this:
 Possible unsafe locking scenario:

       CPU0
       ----
  lock(clock-AF_INET);
  <Interrupt>
    lock(clock-AF_INET);

 *** DEADLOCK ***

5 locks held by nvme/1324:
 #0: ffff8884a01fe470 (sb_writers#4){.+.+}-{0:0}, at: ksys_write+0xf9/0x1d0
 #1: ffff8886e435c090 (&of->mutex){+.+.}-{3:3}, at: kernfs_fop_write_iter+0x216/0x460
 #2: ffff888104d90c38 (kn->active#255){++++}-{0:0}, at: kernfs_remove_self+0x22d/0x330
 #3: ffff8884634538d0 (&queue->queue_lock){+.+.}-{3:3}, at: nvme_tcp_stop_queue+0x52/0xb0 [nvme_tcp]
 #4: ffff888363150d30 (sk_lock-AF_INET){+.+.}-{0:0}, at: inet_shutdown+0x59/0x300

stack backtrace:
CPU: 26 PID: 1324 Comm: nvme Tainted: G          I       5.12.0-rc3 #1
Hardware name: Dell Inc. PowerEdge R640/06NR82, BIOS 2.10.0 11/12/2020
Call Trace:
 dump_stack+0x93/0xc2
 mark_lock_irq.cold+0x2c/0xb3
 ? verify_lock_unused+0x390/0x390
 ? stack_trace_consume_entry+0x160/0x160
 ? lock_downgrade+0x100/0x100
 ? save_trace+0x88/0x5e0
 ? _raw_spin_unlock_irqrestore+0x2d/0x40
 mark_lock+0x530/0x1470
 ? mark_lock_irq+0x1d10/0x1d10
 ? enqueue_timer+0x660/0x660
 mark_usage+0x215/0x2a0
 __lock_acquire+0x79b/0x18d0
 ? tcp_schedule_loss_probe.part.0+0x38c/0x520
 lock_acquire+0x1ca/0x480
 ? nvme_tcp_state_change+0x21/0x150 [nvme_tcp]
 ? rcu_read_unlock+0x40/0x40
 ? tcp_mtu_probe+0x1ae0/0x1ae0
 ? kmalloc_reserve+0xa0/0xa0
 ? sysfs_file_ops+0x170/0x170
 _raw_read_lock+0x3d/0xa0
 ? nvme_tcp_state_change+0x21/0x150 [nvme_tcp]
 nvme_tcp_state_change+0x21/0x150 [nvme_tcp]
 ? sysfs_file_ops+0x170/0x170
 inet_shutdown+0x189/0x300
 __nvme_tcp_stop_queue+0x36/0x270 [nvme_tcp]
 nvme_tcp_stop_queue+0x87/0xb0 [nvme_tcp]
 nvme_tcp_teardown_admin_queue+0x69/0xe0 [nvme_tcp]
 nvme_do_delete_ctrl+0x100/0x10c [nvme_core]
 nvme_sysfs_delete.cold+0x8/0xd [nvme_core]
 kernfs_fop_write_iter+0x2c7/0x460
 new_sync_write+0x36c/0x610
 ? new_sync_read+0x600/0x600
 ? lock_acquire+0x1ca/0x480
 ? rcu_read_unlock+0x40/0x40
 ? lock_is_held_type+0x9a/0x110
 vfs_write+0x5c0/0x870
 ksys_write+0xf9/0x1d0
 ? __ia32_sys_read+0xa0/0xa0
 ? lockdep_hardirqs_on_prepare.part.0+0x198/0x340
 ? syscall_enter_from_user_mode+0x27/0x70
 do_syscall_64+0x33/0x40
 entry_SYSCALL_64_after_hwframe+0x44/0xae

Fixes: 872d26a391da ("nvmet-tcp: add NVMe over TCP target driver")
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-tcp: block BH in sk state_change sk callback

The TCP stack can run from process context for a long time
so we should disable BH here.

Fixes: 3f2304f8c6d6 ("nvme-tcp: add NVMe over TCP host driver")
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: return proper error code from discovery ctrl

Return NVME_SC_INVALID_FIELD from discovery controller like normal
controller when executing identify or get log page command.

Signed-off-by: Hou Pu <houpu.main@gmail.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: warn of unhandled effects only once

We don't need to repeatedly spam the kernel logs with the same warning
about unhandled passthrough IO effects. Just one warning is sufficient
to observe this condition occurs.

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

nvme: use driver pdu command for passthrough

All nvme transport drivers preallocate an nvme command for each request.
Assume to use that command for nvme_setup_cmd() instead of requiring
drivers pass a pointer to it. All nvme drivers must initialize the
generic nvme_request 'cmd' to point to the transport's preallocated
nvme_command.

The generic nvme_request cmd pointer had previously been used only as a
temporary copy for passthrough commands. Since it now points to the
command that gets dispatched, passthrough commands must directly set it
up prior to executing the request.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-pci: allocate nvme_command within driver pdu

Except for pci, all the nvme transport drivers allocate a command within
the driver's pdu. Align pci with everyone else by allocating the nvme
command within pci's pdu and replace the .queue_rq() stack variable with
this.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: do not allow model_number exceed 40 bytes

According to the NVM specifications, the model number size should be
40 bytes (bytes 63:24 of the Identify Controller data structure).
Therefore, any attempt to store a value into model_number which
exceeds 40 bytes should return an error.

Reviewed-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Signed-off-by: Noam Gottlieb <ngottlieb@nvidia.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: remove unnecessary ctrl parameter

The function nvmet_ctrl_find_get() accepts out pointer to nvmet_ctrl
structure. This function returns the same error value from two places
that is :- NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR.

Move this to the caller so we can change the return type to nvmet_ctrl.

Now that we can changed the return type, instead of taking out pointer
to the nvmet_ctrl structure remove that function parameter and return
the valid nvmet_ctrl pointer on success and NULL on failure.

Also, add and rename the goto labels for more readability with comments.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet-fc: update function documentation

Add minimum description of the hosthandle parameter for
nvmet_fc_rcv_ls_req() so that we can get rid of the following warning.

drivers/nvme//target/fc.c:2009: warning: Function parameter or member 'hosthandle' not described in 'nvmet_fc_rcv_ls_req

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-fc: fix the function documentation comment

The nvme_fc_rcv_ls_req() function has first argument as pointer to
remoteport named portprt, but in the documentation comment that is name
is used as remoteport. Fix that to get rid if the compilation warning.

drivers/nvme//host/fc.c:1724: warning: Function parameter or member 'portptr' not described in 'nvme_fc_rcv_ls_req'
drivers/nvme//host/fc.c:1724: warning: Excess function parameter 'remoteport' description in 'nvme_fc_rcv_ls_req'

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: add new line after variable declatation

Add a new line in functions nvme_pr_preempt(), nvme_pr_clear(), and
nvme_pr_release() after variable declaration which follows the rest of
the code in the nvme/host/core.c.

No functional change(s) in this patch.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: don't check nvme_req flags for new req

nvme_clear_request() has a check for flag REQ_DONTPREP and it is called
from nvme_init_request() and nvme_setuo_cmd().

The function nvme_init_request() is called from nvme_alloc_request()
and nvme_alloc_request_qid(). From these two callers new request is
allocated everytime. For newly allocated request RQF_DONTPREP is never
set. Since after getting a tag, block layer sets the req->rq_flags == 0
and never sets the REQ_DONTPREP when returning the request :-

nvme_alloc_request()
blk_mq_alloc_request()
blk_mq_rq_ctx_init()
rq->rq_flags = 0 <----

nvme_alloc_request_qid()
blk_mq_alloc_request_hctx()
blk_mq_rq_ctx_init()
rq->rq_flags = 0 <----

The block layer does set req->rq_flags but REQ_DONTPREP is not one of
them and that is set by the driver.

That means we can unconditinally set the REQ_DONTPREP value to the
rq->rq_flags when nvme_init_request()->nvme_clear_request() is called
from above two callers.

Move the check for REQ_DONTPREP from nvme_clear_nvme_request() into
nvme_setup_cmd().

This is needed since nvme_alloc_request() now gets called from fast
path when NVMeOF target is configured with passthru backend to avoid
unnecessary checks in the fast path.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: mark nvme_setup_passsthru() inline

Since nvmet_setup_passthru() function falls in fast path when called
from the NVMeOF passthru backend, make it inline.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: split init identify into helper

The function nvme_init_ctrl_finish() (formerly nvme_init_identify()) has
grown over the period of time about ~200 lines given the size of nvme id
ctrl data structure.

Move the nvme_id_ctrl data structure related initilzation into helper
nvme_init_identify() and call it from nvme_init_ctrl_finish().

When we move the code into nvme_init_identify() change the local
variable i from int to unsigned int and remove the duplicate kfree()
after nvme_mpath_init() and jump to the label out_free if
nvme_mpath_ini() fails.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: rename nvme_init_identify()

This is a prep patch so that we can move the identify data structure
related code initialization from nvme_init_identify() into a helper.

Rename the function nvmet_init_identify() to nvmet_init_ctrl_finish().

Next patch will move the nvme_id_ctrl related initialization from newly
renamed function nvme_init_ctrl_finish() into the nvme_init_identify()
helper.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: reduce checks for zero command effects

For passthrough I/O commands, effects are usually to be zero.
nvme_passthrough_end() does three checks in futility for this case.
Bail out of function-call/checks.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: use NVME_CTRL_CMIC_ANA macro

Use the proper macro instead of hard-coded value.

Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: replace white spaces with tabs

Instead of the using the whitespaces use tab spacing in the
nvmet_execute_identify_ns().

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: remove an unnecessary function parameter to nvmet_check_ctrl_status

In nvmet_check_ctrl_status() cmd can be derived from nvmet_req. Remove
the local variable cmd in the nvmet_check_ctrl_status() and function
parameter cmd for nvmet_check_ctrl_status(). Derive the cmd value from
req parameter in the nvmet_check_ctrl_status().

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>