blkcg: annotate implicit fall through

There is a plan to build the kernel with -Wimplicit-fallthrough and
this place in the code produced a warning (W=1).

This commit remove the following warning:

  kernel/trace/blktrace.c:725:9: warning: this statement may fall through [-Wimplicit-fallthrough=]

Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme-tcp: support C2HData with SUCCESS flag

A C2HData PDU with the SUCCESS flag set indicates that the I/O was
completed by the controller successfully and means that a subsequent
completion response capsule PDU will be ommitted.

If we see this flag, fisrt we check that LAST_PDU flag is set as well,
and then we complete the request when the data transfer (and data digest
verification if its on) is done.

While we're at it, reuse a bit of code with nvme_fail_request.

Reported-by: Steve Blightman <steve.blightman@oracle.com>
Suggested-by: Oliver Smith-Denny <osmithde@cisco.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Oliver Smith-Denny <osmithde@cisco.com>
Tested-by: Oliver Smith-Denny <osmithde@cisco.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvmet: ignore EOPNOTSUPP for discard

NVMe DSM is a pure hint, so if the underlying device / file system
does not support discard-like operations we should not fail the
operation but rather return success.

Fixes: 3b031d15995f ("nvmet: add error log support for bdev backend")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Tested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: add proper write zeroes setup for the multipath device

Add a gendisk argument to nvme_config_write_zeroes so that the call to
nvme_update_disk_info for the multipath device node updates the
proper request_queue.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Tested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: add proper discard setup for the multipath device

Add a gendisk argument to nvme_config_discard so that the call to
nvme_update_disk_info for the multipath device node updates the
proper request_queue.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Tested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: remove nvme_ns_config_oncs

Just opencode the two function calls in the caller.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Tested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: disable Write Zeroes for qemu controllers

Qemu started out with a broken implementation of Write Zeroes written
by yours truly.  Disable Write Zeroes on qemu for now, eventually
we need to go back and make all the qemu quirks version specific,
but that is left for another time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Tested-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvmet-fc: bring Disconnect into compliance with FC-NVME spec

The FC-NVME spec, when finally approved, modified the disconnect LS
such that the only scope available is the association.

Rework the Disconnect LS processing to be in accordance with the
change.

Signed-off-by: Nigel Kirkland <nigel.kirkland@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvmet-fc: fix issues with targetport assoc_list list walking

There are two changes:

1) The logic in the __nvmet_fc_free_assoc() routine is bad. It uses
"safe" routines assuming pointers will come back valid.  However, the
intervening next structure being linked can be removed from the list and
the resulting safe pointers are bad, resulting in NULL ptrs being hit.

Correct by scheduling a work element to perform the association delete,
which can be done while under lock.

2) Prior patch that added the work element scheduling left a possible
reference on the object if the work element couldn't be scheduled.

Correct by doing the put on a failing schedule_work() call.

Signed-off-by: Nigel Kirkland <nigel.kirkland@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme-fc: reject reconnect if io queue count is reduced to zero

If:

 - A successful connect has occurred with an io queue count greater than
   zero and namespaces detected and running.
 - An error or something occurs which causes a termination of the prior
   association and then starts a reconnect,
 - The reconnect then creates a new controller, but for whatever reason,
   nvme_set_queue_count() results in io queue count set to zero.  This
   will skip io queue and tag set changes.
 - But... the controller will transition to live, calling
   nvme_start_ctrl, which calls nvme_start_queues(), which then releases
   I/Os into the transport which then sends them to the driver.

As there are no queues, things eventually hit the driver looking for a
handle, which was cleared when the original controller was reset, and it
can't proceed. Worst case, things progress, but everything fails.

In the failing scenario, the nvme_set_features(NVME_FEAT_NUM_QUEUES)
command actually failed with a NVME_SC_INTERNAL error.  For some reason,
although nvme_set_queue_count() saw the error and set io queue count to
zero, it doesn't return a failure status to the transport, which allows
the transport to continue using the controller.

Fix the problem by simply rejecting the new association if at least 1
I/O queue can't be created. The association reject will fail the
reconnect attempt and fall into the reconnect retry policy.

Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme-fc: fix numa_node when dev is null

A recent change added a numa_node field to the nvme controller
and has the transport assign the node using dev_to_node().
However, fcloop registers with a NULL device struct, so the
dev_to_node() call oops.

Revise the assignment to assign no node when device struct is null.

Fixes: 103e515efa89b ("nvme: add a numa_node field to struct nvme_ctrl")
Reported-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
[hch: small coding style fixup]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme-fc: use nr_phys_segments to determine existence of sgl

For some nvme command, when issued by the nvme core layer, there
is an internal buffer which can cause blk_rq_payload_bytes() to
return a non-zero value yet there is no actual/real command payload
and sg list.  An example is the WRITE ZEROES command.

To address this, when making choices on whether to dma map an sgl,
use blk_rq_nr_phys_segments() instead of blk_rq_payload_bytes().
When there is a sgl, blk_rq_payload_bytes() will return the amount
of data to be transferred by the sgl.

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

nvme-loop: init nvmet_ctrl fatal_err_work when allocate

After commit 4d43d395fe (workqueue: Try to catch flush_work() without
INIT_WORK()), it can cause warning when delete nvme-loop device, trace
like:

[   76.601272] Call Trace:
[   76.601646]  ? del_timer+0x72/0xa0
[   76.602156]  __cancel_work_timer+0x1ae/0x270
[   76.602791]  cancel_work_sync+0x14/0x20
[   76.603407]  nvmet_ctrl_free+0x1b7/0x2f0 [nvmet]
[   76.604091]  ? free_percpu+0x168/0x300
[   76.604652]  nvmet_sq_destroy+0x106/0x240 [nvmet]
[   76.605346]  nvme_loop_destroy_admin_queue+0x30/0x60 [nvme_loop]
[   76.606220]  nvme_loop_shutdown_ctrl+0xc3/0xf0 [nvme_loop]
[   76.607026]  nvme_loop_delete_ctrl_host+0x19/0x30 [nvme_loop]
[   76.607871]  nvme_do_delete_ctrl+0x75/0xb0
[   76.608477]  nvme_sysfs_delete+0x7d/0xc0
[   76.609057]  dev_attr_store+0x24/0x40
[   76.609603]  sysfs_kf_write+0x4c/0x60
[   76.610144]  kernfs_fop_write+0x19a/0x260
[   76.610742]  __vfs_write+0x1c/0x60
[   76.611246]  vfs_write+0xfa/0x280
[   76.611739]  ksys_write+0x6e/0x120
[   76.612238]  __x64_sys_write+0x1e/0x30
[   76.612787]  do_syscall_64+0xbf/0x3a0
[   76.613329]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

We fix it by moving fatal_err_work init to nvmet_alloc_ctrl(), which may
more reasonable.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: update comment to make the code easier to read

After commit a686ed75c0fb ("nvme: introduce a helper function for
controller deletion), nvme_delete_ctrl_sync no longer use flush_work.
Update comment, accordingly.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: put ns_head ref if namespace fails allocation

In case nvme_alloc_ns fails after we initialize ns_head but before we
add the ns to the controller namespaces list we need to explicitly put
the ns_head reference because when we teardown the controller we
won't find it, causing us to leak a dangling subsystem eventually.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme-trace: fix cdw10 buffer overrun

The field is defined to be a 24 byte array, we don't need to multiply
the sizeof() that field by the number of dwords it covers.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: don't warn on block content change effects

A write or flush IO passthrough command is expected to change the
logical block content, so don't warn on these as no additional handling
is necessary.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nvme: add get-feature to admin cmds tracer

This will print get-feature cmd in more informative way. For example,
run "nvme get-feature /dev/nvme0 -n 1 -f 0x9 -c 10" will trace:

 nvme-3907  [008] ....  1763.635054: nvme_setup_cmd: nvme0: qid=0, cmdid=6, nsid=1, flags=0x0, meta=0x0, cmd=(nvme_admin_get_features fid=0x9 sel=0x0 cdw11=0xa)
<idle>-0     [001] d.h.  1763.635112: nvme_sq: nvme0: qid=0, head=27, tail=27
<idle>-0     [008] ..s.  1763.635121: nvme_complete_rq: nvme0: qid=0, cmdid=6, res=10, retries=0, flags=0x2, status=0

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge branch 'for-5.1/md-post' of https://github.com/liu-song-6/linux into for-5.1/block-post

Pull MD fixes from Song.

* 'for-5.1/md-post' of https://github.com/liu-song-6/linux:
  md: Fix failed allocation of md_register_thread
  It's wrong to add len to sector_nr in raid10 reshape twice
  raid5: set write hint for PPL

md: Fix failed allocation of md_register_thread

mddev->sync_thread can be set to NULL on kzalloc failure downstream.
The patch checks for such a scenario and frees allocated resources.

Committer node:

Added similar fix to raid5.c, as suggested by Guoqing.

Cc: stable@vger.kernel.org # v3.16+
Acked-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: Aditya Pakki <pakki001@umn.edu>
Signed-off-by: Song Liu <songliubraving@fb.com>

It's wrong to add len to sector_nr in raid10 reshape twice

In reshape_request it already adds len to sector_nr already. It's wrong to add len to
sector_nr again after adding pages to bio. If there is bad block it can't copy one chunk
at a time, it needs to goto read_more. Now the sector_nr is wrong. It can cause data
corruption.

Cc: stable@vger.kernel.org # v3.16+
Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>

raid5: set write hint for PPL

When the Partial Parity Log is enabled, circular buffer is used to store
PPL data. Each write to RAID device causes overwrite of data in this buffer
so some write_hint can be set to those request to help drives handle
garbage collection. This patch adds new sysfs attribute which can be used
to specify which write_hint should be assigned to PPL.

Acked-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: Mariusz Dabrowski <mariusz.dabrowski@intel.com>
Signed-off-by: Song Liu <songliubraving@fb.com>

pblk: fix max_io calculation

When calculating the maximun I/O size allowed into the buffer, consider
the write size (ws_opt) used by the write thread in order to cover the
case in which, due to flushes, the mem and subm pointers are disaligned
by (ws_opt - 1). This case currently translates into a stall when
an I/O of the largest possible size is submitted.

Fixes: f9f9d1ae2c66 ("lightnvm: pblk: prevent stall due to wb threshold")
Signed-off-by: Javier González <javier@javigon.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: fix segment calculation for passthrough IO

blk_recount_segments() can be called in bio_add_pc_page() for
calculating how many segments this bio will has after one page is added
to this bio. If the resulted segment number is beyond the queue limit,
the added page will be removed.

The try-and-fix policy requires blk_recount_segments(__blk_recalc_rq_segments)
to not consider the segment number limit. Unfortunately bvec_split_segs()
does check this limit, and causes small segment number returned to
bio_add_pc_page(), then page still may be added to the bio even though
segment number limit becomes broken.

Fixes this issue by not considering segment number limit when calcualting
bio's segment number.

Fixes: dcebd755926b ("block: use bio_for_each_bvec() to compute multi-page bvec count")
Cc: Christoph Hellwig <hch@lst.de>
Cc: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge branch 'stable/for-jens-5.1' of git://git./linux/kernel/git/konrad/xen into for-5.1/block-post

Pull two xen blkback fixes from Konrad.

* 'stable/for-jens-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
  xen/blkback: rework connect_ring() to avoid inconsistent xenstore 'ring-page-order' set by malicious blkfront
  xen/blkback: add stack variable 'blkif' in connect_ring()

block: fix updating bio's front segment size

When the current bvec can be merged to the 1st segment, the bio's front
segment size has to be updated.

However, dcebd755926b doesn't consider that case, then bio's front
segment size may not be correct.

This patch fixes this issue.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Omar Sandoval <osandov@fb.com>
Fixes: dcebd755926b ("block: use bio_for_each_bvec() to compute multi-page bvec count")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: Replace function name in string with __func__

Replace hard coded function name register_blkdev with __func__, to
improve robustness and to conform to the Linux kernel coding
style. Issue found using checkpatch.

Signed-off-by: Keyur Patel <iamkeyur96@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

nbd: propagate genlmsg_reply return code

genlmsg_reply can fail, so propagate its return code

Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

floppy: remove set but not used variable 'q'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/block/floppy.c: In function 'request_done':
drivers/block/floppy.c:2233:24: warning:
 variable 'q' set but not used [-Wunused-but-set-variable]

It's never used and can be removed.

Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

null_blk: fix checking for REQ_FUA

null_handle_bio() erroneously uses the bio_op macro
which masks respective request flag bits including REQ_FUA
out thus failing the check.

Fix by checking bio->bi_opf directly.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: fix NULL pointer dereference in register_disk

If __device_add_disk-->bdi_register_owner-->bdi_register-->
bdi_register_va-->device_create_vargs fails, bdi->dev is still
NULL, __device_add_disk-->register_disk will visit bdi->dev->kobj.
This patch fixes that.

Signed-off-by: zhengbin <zhengbin13@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs: fix guard_bio_eod to check for real EOD errors

guard_bio_eod() can truncate a segment in bio to allow it to do IO on
odd last sectors of a device.

It already checks if the IO starts past EOD, but it does not consider
the possibility of an IO request starting within device boundaries can
contain more than one segment past EOD.

In such cases, truncated_bytes can be bigger than PAGE_SIZE, and will
underflow bvec->bv_len.

Fix this by checking if truncated_bytes is lower than PAGE_SIZE.

This situation has been found on filesystems such as isofs and vfat,
which doesn't check the device size before mount, if the device is
smaller than the filesystem itself, a readahead on such filesystem,
which spans EOD, can trigger this situation, leading a call to
zero_user() with a wrong size possibly corrupting memory.

I didn't see any crash, or didn't let the system run long enough to
check if memory corruption will be hit somewhere, but adding
instrumentation to guard_bio_end() to check truncated_bytes size, was
enough to see the error.

The following script can trigger the error.

MNT=/mnt
IMG=./DISK.img
DEV=/dev/loop0

mkfs.vfat $IMG
mount $IMG $MNT
cp -R /etc $MNT &> /dev/null
umount $MNT

losetup -D

losetup --find --show --sizelimit 16247280 $IMG
mount $DEV $MNT

find $MNT -type f -exec cat {} + >/dev/null

Kudos to Eric Sandeen for coming up with the reproducer above

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

blk-mq: use HCTX_TYPE_DEFAULT but not 0 to index blk_mq_tag_set->map

Replace set->map[0] with set->map[HCTX_TYPE_DEFAULT] to avoid hardcoding.

Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: optimize bvec iteration in bvec_iter_advance

There is no need to only iterate in chunks of PAGE_SIZE or less in
bvec_iter_advance, given that the callers pass in the chunk length that
they are operating on - either that already is less than PAGE_SIZE
because they do classic page-based iteration, or it is larger because
the caller operates on multi-page bvecs.

This should help shaving off a few cycles of the I/O hot path.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: introduce mp_bvec_for_each_page() for iterating over page

mp_bvec_for_each_segment() is a bit big for the iteration, so introduce
a light-weight helper for iterating over pages, then 32bytes stack
space can be saved.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: optimize blk_bio_segment_split for single-page bvec

Introduce a fast path for single-page bvec IO, then we can avoid
to call bvec_split_segs() unnecessarily.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: optimize __blk_segment_map_sg() for single-page bvec

Introduce a fast path for single-page bvec IO, then blk_bvec_map_sg()
can be avoided.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: introduce bvec_nth_page()

Single-page bvec can often be seen in small BS workloads, so
introduce bvec_nth_page() for avoiding to call nth_page() unnecessarily,
which looks not cheap.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

iomap: wire up the iopoll method

Store the request queue the last bio was submitted to in the iocb
private data in addition to the cookie so that we find the right block
device.  Also refactor the common direct I/O bio submission code into a
nice little helper.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Modified to use bio_set_polled().

Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: add bio_set_polled() helper

For the upcoming async polled IO, we can't sleep allocating requests.
If we do, then we introduce a deadlock where the submitter already
has async polled IO in-flight, but can't wait for them to complete
since polled requests must be active found and reaped.

Utilize the helper in the blockdev DIRECT_IO code.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: wire up block device iopoll method

Just call blk_poll on the iocb cookie, we can derive the block device
from the inode trivially.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs: add an iopoll method to struct file_operations

This new methods is used to explicitly poll for I/O completion for an
iocb.  It must be called for any iocb submitted asynchronously (that
is with a non-null ki_complete) which has the IOCB_HIPRI flag set.

The method is assisted by a new ki_cookie field in struct iocb to store
the polling cookie.

Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

xen/blkback: rework connect_ring() to avoid inconsistent xenstore 'ring-page-order' set by malicious blkfront

The xenstore 'ring-page-order' is used globally for each blkback queue and
therefore should be read from xenstore only once. However, it is obtained
in read_per_ring_refs() which might be called multiple times during the
initialization of each blkback queue.

If the blkfront is malicious and the 'ring-page-order' is set in different
value by blkfront every time before blkback reads it, this may end up at
the "WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));" in
xen_blkif_disconnect() when frontend is destroyed.

This patch reworks connect_ring() to read xenstore 'ring-page-order' only
once.

Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

loop: set GENHD_FL_NO_PART_SCAN after blkdev_reread_part()

Commit 0da03cab87e6
("loop: Fix deadlock when calling blkdev_reread_part()") moves
blkdev_reread_part() out of the loop_ctl_mutex. However,
GENHD_FL_NO_PART_SCAN is set before __blkdev_reread_part(). As a result,
__blkdev_reread_part() will fail the check of GENHD_FL_NO_PART_SCAN and
will not rescan the loop device to delete all partitions.

Below are steps to reproduce the issue:

step1 # dd if=/dev/zero of=tmp.raw bs=1M count=100
step2 # losetup -P /dev/loop0 tmp.raw
step3 # parted /dev/loop0 mklabel gpt
step4 # parted -a none -s /dev/loop0 mkpart primary 64s 1
step5 # losetup -d /dev/loop0

Step5 will not be able to delete /dev/loop0p1 (introduced by step4) and
there is below kernel warning message:

[  464.414043] __loop_clr_fd: partition scan of loop0 failed (rc=-22)

This patch sets GENHD_FL_NO_PART_SCAN after blkdev_reread_part().

Fixes: 0da03cab87e6 ("loop: Fix deadlock when calling blkdev_reread_part()")
Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

loop: do not print warn message if partition scan is successful

Do not print warn message when the partition scan returns 0.

Fixes: d57f3374ba48 ("loop: Move special partition reread handling in loop_clr_fd()")
Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: bounce: make sure that bvec table is updated

Block bounce needs to allocate new page for doing IO, and the
new page has to be updated to bvec table.

Commit 6dc4f100c switches __blk_queue_bounce() to use the new
bio_for_each_segment_all() interface. Unfortunately the new
bio_for_each_segment_all() can't be used to update bvec table.

This patch fixes this issue by retrieving bvec from the table
directly, then the new allocated page can be updated to the bio.
This way is safe because the cloned bio is single page bvec.

Fixes: 6dc4f100c ("block: allow bio_for_each_segment_all() to iterate over multi-page bvec")
Cc: Christoph Hellwig <hch@lst.de>
Cc: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge branch 'nvme-5.1' of git://git.infradead.org/nvme into for-5.1/block

Pull NVMe changes for 5.1 from Christoph

* 'nvme-5.1' of git://git.infradead.org/nvme: (22 commits)
  nvme-rdma: use nr_phys_segments when map rq to sgl
  nvmet: convert to SPDX identifiers
  nvmet-rdma: convert to SPDX identifiers
  nvme-loop: convert to SPDX identifiers
  nvmet-fcloop: convert to SPDX identifiers
  nvmet-fc: convert to SPDX identifiers
  nvme: convert to SPDX identifiers
  nvme-pci: convert to SPDX identifiers
  nvme-lightnvm: convert to SPDX identifiers
  nvme-rdma: convert to SPDX identifiers
  nvme-fc: convert to SPDX identifiers
  nvme-fabrics: convert to SPDX identifiers
  nvme-tcp.h: fix SPDX header
  nvme_ioctl.h: remove duplicate GPL boilerplate
  nvme: return error from nvme_alloc_ns()
  nvme: avoid that deleting a controller triggers a circular locking complaint
  nvme: introduce a helper function for controller deletion
  nvme: unexport nvme_delete_ctrl_sync()
  nvme-pci: check kstrtoint() return value in queue_count_set()
  nvme-fabrics: document the poll function argument
  ...

nvme-rdma: use nr_phys_segments when map rq to sgl

Use blk_rq_nr_phys_segments() instead of blk_rq_payload_bytes() to check
if a command contains data to be mapped.  This fixes the case where
a struct request contains LBAs, but it has no payload, such as
Write Zeroes support.

Fixes: 6e02318eaea5 ("nvme: add support for the Write Zeroes command")
Reported-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Tested-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvmet-rdma: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-loop: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvmet-fcloop: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvmet-fc: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-pci: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-lightnvm: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-rdma: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-fc: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-fabrics: convert to SPDX identifiers

Update license to use SPDX-License-Identifier instead of verbose license
text.

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

nvme-tcp.h: fix SPDX header

For .h files we need to use /* */ style comments.

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

nvme_ioctl.h: remove duplicate GPL boilerplate

We already have a ЅPDX header, so no need to duplicate the information.

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

nvme: return error from nvme_alloc_ns()

nvme_alloc_ns() might fail, so we should be returning an error code.

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

nvme: avoid that deleting a controller triggers a circular locking complaint

Rework nvme_delete_ctrl_sync() such that it does not have to wait for
queued work. This patch avoids that test nvme/008 triggers the following
complaint:

WARNING: possible circular locking dependency detected
5.0.0-rc6-dbg+ #10 Not tainted
------------------------------------------------------
nvme/7918 is trying to acquire lock:
000000009a1a7b69 ((work_completion)(&ctrl->delete_work)){+.+.}, at: __flush_work+0x379/0x410

but task is already holding lock:
00000000ef5a45b4 (kn->count#389){++++}, at: kernfs_remove_self+0x196/0x210

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (kn->count#389){++++}:
       lock_acquire+0xc5/0x1e0
       __kernfs_remove+0x42a/0x4a0
       kernfs_remove_by_name_ns+0x45/0x90
       remove_files.isra.1+0x3a/0x90
       sysfs_remove_group+0x5c/0xc0
       sysfs_remove_groups+0x39/0x60
       device_remove_attrs+0x68/0xb0
       device_del+0x24d/0x570
       cdev_device_del+0x1a/0x50
       nvme_delete_ctrl_work+0xbd/0xe0
       process_one_work+0x4f1/0xa40
       worker_thread+0x67/0x5b0
       kthread+0x1cf/0x1f0
       ret_from_fork+0x24/0x30

-> #0 ((work_completion)(&ctrl->delete_work)){+.+.}:
       __lock_acquire+0x1323/0x17b0
       lock_acquire+0xc5/0x1e0
       __flush_work+0x399/0x410
       flush_work+0x10/0x20
       nvme_delete_ctrl_sync+0x65/0x70
       nvme_sysfs_delete+0x4f/0x60
       dev_attr_store+0x3e/0x50
       sysfs_kf_write+0x87/0xa0
       kernfs_fop_write+0x186/0x240
       __vfs_write+0xd7/0x430
       vfs_write+0xfa/0x260
       ksys_write+0xab/0x130
       __x64_sys_write+0x43/0x50
       do_syscall_64+0x71/0x210
       entry_SYSCALL_64_after_hwframe+0x49/0xbe

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(kn->count#389);
                               lock((work_completion)(&ctrl->delete_work));
                               lock(kn->count#389);
  lock((work_completion)(&ctrl->delete_work));

 *** DEADLOCK ***

3 locks held by nvme/7918:
 #0: 00000000e2223b44 (sb_writers#6){.+.+}, at: vfs_write+0x1eb/0x260
 #1: 000000003404976f (&of->mutex){+.+.}, at: kernfs_fop_write+0x128/0x240
 #2: 00000000ef5a45b4 (kn->count#389){++++}, at: kernfs_remove_self+0x196/0x210

stack backtrace:
CPU: 4 PID: 7918 Comm: nvme Not tainted 5.0.0-rc6-dbg+ #10
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
Call Trace:
 dump_stack+0x86/0xca
 print_circular_bug.isra.36.cold.54+0x173/0x1d5
 check_prev_add.constprop.45+0x996/0x1110
 __lock_acquire+0x1323/0x17b0
 lock_acquire+0xc5/0x1e0
 __flush_work+0x399/0x410
 flush_work+0x10/0x20
 nvme_delete_ctrl_sync+0x65/0x70
 nvme_sysfs_delete+0x4f/0x60
 dev_attr_store+0x3e/0x50
 sysfs_kf_write+0x87/0xa0
 kernfs_fop_write+0x186/0x240
 __vfs_write+0xd7/0x430
 vfs_write+0xfa/0x260
 ksys_write+0xab/0x130
 __x64_sys_write+0x43/0x50
 do_syscall_64+0x71/0x210
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: introduce a helper function for controller deletion

This patch does not change any functionality but makes the next patch
in this series easier to read.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme: unexport nvme_delete_ctrl_sync()

Since nvme_delete_ctrl_sync() is not called from any other kernel module,
unexport it.

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

nvme-pci: check kstrtoint() return value in queue_count_set()

This patch avoids that the compiler complains about 'ret' being set
but not being used when building with W=1.

Fixes: 3b6592f70ad7 ("nvme: utilize two queue maps, one for reads and one for writes") # v5.0-rc1
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-fabrics: document the poll function argument

This patch avoids that the kernel-doc tool reports a warning when
building with W=1.

Fixes: 26c682274e0a ("nvme-fabrics: allow nvmf_connect_io_queue to poll") # v5.0-rc1
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvmet: fix indentation

This patch avoids that smatch complains about inconsistent indentation.

Fixes: a07b4970f464 ("nvmet: add a generic NVMe target") # v4.10
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

nvme-multipath: round-robin I/O policy

Implement a simple round-robin I/O policy for multipathing.  Path
selection is done in two rounds, first iterating across all optimized
paths, and if that doesn't return any valid paths, iterate over all
optimized and non-optimized paths.  If no paths are found, use the
existing algorithm.  Also add a sysfs attribute 'iopolicy' to switch
between the current NUMA-aware I/O policy and the 'round-robin' I/O
policy.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

block: avoid to READ fields of null bio

rq->bio can be NULL sometimes, such as flush request, so don't
read bio->bi_seg_front_size until this 'bio' is checked as valid.

Cc: Bart Van Assche <bvanassche@acm.org>
Reported-by: Bart Van Assche <bvanassche@acm.org>
Fixes: dcebd755926b0f39dd1e ("block: use bio_for_each_bvec() to compute multi-page bvec count")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'v5.0-rc6' into for-5.1/block

Pull in 5.0-rc6 to avoid a dumb merge conflict with fs/iomap.c.
This is needed since io_uring is now based on the block branch,
to avoid a conflict between the multi-page bvecs and the bits
of io_uring that touch the core block parts.

* tag 'v5.0-rc6': (525 commits)
  Linux 5.0-rc6
  x86/mm: Make set_pmd_at() paravirt aware
  MAINTAINERS: Update the ocores i2c bus driver maintainer, etc
  blk-mq: remove duplicated definition of blk_mq_freeze_queue
  Blk-iolatency: warn on negative inflight IO counter
  blk-iolatency: fix IO hang due to negative inflight counter
  MAINTAINERS: unify reference to xen-devel list
  x86/mm/cpa: Fix set_mce_nospec()
  futex: Handle early deadlock return correctly
  futex: Fix barrier comment
  net: dsa: b53: Fix for failure when irq is not defined in dt
  blktrace: Show requests without sector
  mips: cm: reprime error cause
  mips: loongson64: remove unreachable(), fix loongson_poweroff().
  sit: check if IPv6 enabled before calling ip6_err_gen_icmpv6_unreach()
  geneve: should not call rt6_lookup() when ipv6 was disabled
  KVM: nVMX: unconditionally cancel preemption timer in free_nested (CVE-2019-7221)
  KVM: x86: work around leak of uninitialized stack contents (CVE-2019-7222)
  kvm: fix kvm_ioctl_create_device() reference counting (CVE-2019-6974)
  signal: Better detection of synchronous signals
  ...

block: kill BLK_MQ_F_SG_MERGE

QUEUE_FLAG_NO_SG_MERGE has been killed, so kill BLK_MQ_F_SG_MERGE too.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: kill QUEUE_FLAG_NO_SG_MERGE

Since bdced438acd83ad83a6c ("block: setup bi_phys_segments after splitting"),
physical segment number is mainly figured out in blk_queue_split() for
fast path, and the flag of BIO_SEG_VALID is set there too.

Now only blk_recount_segments() and blk_recalc_rq_segments() use this
flag.

Basically blk_recount_segments() is bypassed in fast path given BIO_SEG_VALID
is set in blk_queue_split().

For another user of blk_recalc_rq_segments():

- run in partial completion branch of blk_update_request, which is an unusual case

- run in blk_cloned_rq_check_limits(), still not a big problem if the flag is killed
since dm-rq is the only user.

Multi-page bvec is enabled now, not doing S/G merging is rather pointless with the
current setup of the I/O path, as it isn't going to save you a significant amount
of cycles.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: document usage of bio iterator helpers

Now multi-page bvec is supported, some helpers may return page by
page, meantime some may return segment by segment, this patch
documents the usage.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: always define BIO_MAX_PAGES as 256

Now multi-page bvec can cover CONFIG_THP_SWAP, so we don't need to
increase BIO_MAX_PAGES for it.

CONFIG_THP_SWAP needs to split one THP into normal pages and adds
them all to one bio. With multipage-bvec, it just takes one bvec to
hold them all.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: enable multipage bvecs

This patch pulls the trigger for multi-page bvecs.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: allow bio_for_each_segment_all() to iterate over multi-page bvec

This patch introduces one extra iterator variable to bio_for_each_segment_all(),
then we can allow bio_for_each_segment_all() to iterate over multi-page bvec.

Given it is just one mechannical & simple change on all bio_for_each_segment_all()
users, this patch does tree-wide change in one single patch, so that we can
avoid to use a temporary helper for this conversion.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bcache: avoid to use bio_for_each_segment_all() in bch_bio_alloc_pages()

bch_bio_alloc_pages() is always called on one new bio, so it is safe
to access the bvec table directly. Given it is the only kind of this
case, open code the bvec table access since bio_for_each_segment_all()
will be changed to support for iterating over multipage bvec.

Acked-by: Coly Li <colyli@suse.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: loop: pass multi-page bvec to iov_iter

iov_iter is implemented on bvec itererator helpers, so it is safe to pass
multi-page bvec to it, and this way is much more efficient than passing one
page in each bvec.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

btrfs: use mp_bvec_last_segment to get bio's last page

Preparing for supporting multi-page bvec.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/buffer.c: use bvec iterator to truncate the bio

Once multi-page bvec is enabled, the last bvec may include more than one
page, this patch use mp_bvec_last_segment() to truncate the bio.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: introduce mp_bvec_last_segment()

BTRFS and guard_bio_eod() need to get the last singlepage segment
from one multipage bvec, so introduce this helper to make them happy.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: use bio_for_each_bvec() to map sg

It is more efficient to use bio_for_each_bvec() to map sg, meantime
we have to consider splitting multipage bvec as done in blk_bio_segment_split().

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: use bio_for_each_bvec() to compute multi-page bvec count

First it is more efficient to use bio_for_each_bvec() in both
blk_bio_segment_split() and __blk_recalc_rq_segments() to compute how
many multi-page bvecs there are in the bio.

Secondly once bio_for_each_bvec() is used, the bvec may need to be
splitted because its length can be very longer than max segment size,
so we have to split the big bvec into several segments.

Thirdly when splitting multi-page bvec into segments, the max segment
limit may be reached, so the bio split need to be considered under
this situation too.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: introduce bio_for_each_bvec() and rq_for_each_bvec()

bio_for_each_bvec() is used for iterating over multi-page bvec for bio
split & merge code.

rq_for_each_bvec() can be used for drivers which may handle the
multi-page bvec directly, so far loop is one perfect use case.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: introduce multi-page bvec helpers

This patch introduces helpers of 'mp_bvec_iter_*' for multi-page bvec
support.

The introduced helpers treate one bvec as real multi-page segment,
which may include more than one pages.

The existed helpers of bvec_iter_* are interfaces for supporting current
bvec iterator which is thought as single-page by drivers, fs, dm and
etc. These introduced helpers will build single-page bvec in flight, so
this way won't break current bio/bvec users, which needn't any change.

Follows some multi-page bvec background:

- bvecs stored in bio->bi_io_vec is always multi-page style

- bvec(struct bio_vec) represents one physically contiguous I/O
  buffer, now the buffer may include more than one page after
  multi-page bvec is supported, and all these pages represented
  by one bvec is physically contiguous. Before multi-page bvec
  support, at most one page is included in one bvec, we call it
  single-page bvec.

- .bv_page of the bvec points to the 1st page in the multi-page bvec

- .bv_offset of the bvec is the offset of the buffer in the bvec

The effect on the current drivers/filesystem/dm/bcache/...:

- almost everyone supposes that one bvec only includes one single
  page, so we keep the sp interface not changed, for example,
  bio_for_each_segment() still returns single-page bvec

- bio_for_each_segment_all() will return single-page bvec too

- during iterating, iterator variable(struct bvec_iter) is always
  updated in multi-page bvec style, and bvec_iter_advance() is kept
  not changed

- returned(copied) single-page bvec is built in flight by bvec
  helpers from the stored multi-page bvec

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: remove bvec_iter_rewind()

Commit 7759eb23fd980 ("block: remove bio_rewind_iter()") removes
bio_rewind_iter(), then no one uses bvec_iter_rewind() any more,
so remove it.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: don't use bio->bi_vcnt to figure out segment number

It is wrong to use bio->bi_vcnt to figure out how many segments
there are in the bio even though CLONED flag isn't set on this bio,
because this bio may be splitted or advanced.

So always use bio_segments() in blk_recount_segments(), and it shouldn't
cause any performance loss now because the physical segment number is figured
out in blk_queue_split() and BIO_SEG_VALID is set meantime since
bdced438acd83ad83a6c ("block: setup bi_phys_segments after splitting").

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Fixes: 76d8137a3113 ("blk-merge: recaculate segment if it isn't less than max segments")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

btrfs: look at bi_size for repair decisions

bio_readpage_error currently uses bi_vcnt to decide if it is worth
retrying an I/O.  But the vector count is mostly an implementation
artifact - it really should figure out if there is more than a
single sector worth retrying.  Use bi_size for that and shift by
PAGE_SHIFT.  This really should be blocks/sectors, but given that
btrfs doesn't support a sector size different from the PAGE_SIZE
using the page size keeps the changes to a minimum.

Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: avoid setting none scheduler if it's already none

There's no reason to freeze queue and remove scheduler
if there's no scheduler already.

Signed-off-by: Aleksei Zakharov <zakharov.a.g@yandex.ru>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block: avoid setting wbt_lat_usec to current value

There's no reason to set wbt min lat and freeze request queue
if current value is the same.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Aleksei Zakharov <zakharov.a.g@yandex.ru>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: fix race condition on GC

This patch fixes a race condition where a write is mapped to the last
sectors of a line. The write is synced to the device but the L2P is not
updated yet. When the line is garbage collected before the L2P update
is performed, the sectors are ignored by the GC logic and the line is
freed before all sectors are moved. When the L2P is finally updated, it
contains a mapping to a freed line, subsequent reads of the
corresponding LBAs fail.

This patch introduces a per line counter specifying the number of
sectors that are synced to the device but have not been updated in the
L2P. Lines with a counter of greater than zero will not be selected
for GC.

Signed-off-by: Heiner Litz <hlitz@ucsc.edu>
Reviewed-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: prevent stall due to wb threshold

In order to respect mw_cuinits, pblk's write buffer maintains a
backpointer to protect data not yet persisted; when writing to the write
buffer, this backpointer defines a threshold that pblk's rate-limiter
enforces.

On small PU configurations, the following scenarios might take place: (i)
the threshold is larger than the write buffer and (ii) the threshold is
smaller than the write buffer, but larger than the maximun allowed
split bio - 256KB at this moment (Note that writes are not always
split - we only do this when we the size of the buffer is smaller
than the buffer). In both cases, pblk's rate-limiter prevents the I/O to
be written to the buffer, thus stalling.

This patch fixes the original backpointer implementation by considering
the threshold both on buffer creation and on the rate-limiters path,
when bio_split is triggered (case (ii) above).

Fixes: 766c8ceb16fc ("lightnvm: pblk: guarantee that backpointer is respected on writer stall")
Signed-off-by: Javier González <javier@javigon.com>
Reviewed-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: extend line wp balance check

pblk stripes writes of minimal write size across all non-offline chunks
in a line, which means that the maximum write pointer delta should not
exceed the minimal write size.

Extend the line write pointer balance check to cover this case, and
ignore the offline chunk wps.

This will render us a warning during recovery if something unexpected
has happened to the chunk write pointers (i.e. powerloss,  a spurious
chunk reset, ..).

Reported-by: Zhoujie Wu <zjwu@marvell.com>
Tested-by: Zhoujie Wu <zjwu@marvell.com>
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: fix TRACE_INCLUDE_PATH

As the comment block in include/trace/define_trace.h says,
TRACE_INCLUDE_PATH should be a relative path to the define_trace.h

../../drivers/lightnvm is the correct relative path.

../../../drivers/lightnvm is working by coincidence because the top
Makefile adds -I$(srctree)/arch/$(SRCARCH)/include as a header
search path, but we should not rely on it.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: Switch to use new generic UUID API

There are new types and helpers that are supposed to be used in new code.

As a preparation to get rid of legacy types and API functions do
the conversion here.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: Use u64 instead of __le64 for CPU visible side

Sparse complains about using strict data types:

drivers/lightnvm/pblk-read.c:254:43: warning: incorrect type in assignment (different base types)
drivers/lightnvm/pblk-read.c:254:43:    expected restricted __le64 <noident>
drivers/lightnvm/pblk-read.c:254:43:    got unsigned long long [unsigned] [usertype] <noident>
drivers/lightnvm/pblk-read.c:255:29: warning: cast from restricted __le64
drivers/lightnvm/pblk-read.c:268:29: warning: cast from restricted __le64
drivers/lightnvm/pblk-read.c:328:41: warning: incorrect type in assignment (different base types)
drivers/lightnvm/pblk-read.c:328:41:    expected restricted __le64 <noident>
drivers/lightnvm/pblk-read.c:328:41:    got unsigned long long [unsigned] [usertype] <noident>

In the code it seems explicit that lba_list_mem and lba_list_media members
of struct pblk_pr_ctx are used on CPU side, which means they should not be
of strict types.

Change types of lba_list_mem and lba_list_media members to be u64.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: use vfree to free metadata on error path

As chunk metadata is allocated using vmalloc, we need to free it
using vfree.

Fixes: 090ee26fd512 ("lightnvm: use internal allocation for chunk log page")
Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

lightnvm: pblk: stop taking the free lock in in pblk_lines_free

pblk_line_meta_free might sleep (it can end up calling vfree, depending
on how we allocate lba lists), and this can lead to a BUG()
if we wake up on a different cpu and release the lock.

As there is no point of grabbing the free lock when pblk has shut down,
remove the lock.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Linux 5.0-rc6