linux-2.6-microblaze.git
3 years agoblock: remove unused BLK_QC_T_EAGAIN flag
Jeffle Xu [Fri, 25 Sep 2020 06:00:31 +0000 (14:00 +0800)]
block: remove unused BLK_QC_T_EAGAIN flag

commit 7b6620d7db56 ("block: remove REQ_NOWAIT_INLINE") removed the
REQ_NOWAIT_INLINE related code, but the diff wasn't applied to
blk_types.h somehow.

Then commit 2771cefeac49 ("block: remove the REQ_NOWAIT_INLINE flag")
removed the REQ_NOWAIT_INLINE flag while the BLK_QC_T_EAGAIN flag still
remains.

Fixes: 7b6620d7db56 ("block: remove REQ_NOWAIT_INLINE")
Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge tag 'nvme-5.9-2020-09-24' of git://git.infradead.org/nvme into block-5.9
Jens Axboe [Thu, 24 Sep 2020 19:42:40 +0000 (13:42 -0600)]
Merge tag 'nvme-5.9-2020-09-24' of git://git.infradead.org/nvme into block-5.9

Pull NVMe fixes from Christoph:

"nvme fixes for 5.9

  - fix error during controller probe that cause double free irqs
    (Keith Busch)
  - FC connection establishment fix (James Smart)
  - properly handle completions for invalid tags (Xianting Tian)
  - pass the correct nsid to the command effects and supported log
    (Chaitanya Kulkarni)"

* tag 'nvme-5.9-2020-09-24' of git://git.infradead.org/nvme:
  nvme-core: don't use NVME_NSID_ALL for command effects and supported log
  nvme-fc: fail new connections to a deleted host or remote port
  nvme-pci: fix NULL req in completion handler
  nvme: return errors for hwmon init

3 years agonvme-core: don't use NVME_NSID_ALL for command effects and supported log
Chaitanya Kulkarni [Tue, 22 Sep 2020 19:49:38 +0000 (12:49 -0700)]
nvme-core: don't use NVME_NSID_ALL for command effects and supported log

In the function nvme_get_effects_log() it uses NVME_NSID_ALL which has
namespace scope. The command effect log page is controller specific.

Replace NVME_NSID_ALL with 0x00 which specifies the controller scope
instead of namespace scope.

Fixes: 84fef62d135b ("nvme: check admin passthru command effects")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=209287
Reported-by: Huai-Cheng Kuo <hh81478072@gmail.com>
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-fc: fail new connections to a deleted host or remote port
James Smart [Thu, 17 Sep 2020 20:33:22 +0000 (13:33 -0700)]
nvme-fc: fail new connections to a deleted host or remote port

The lldd may have made calls to delete a remote port or local port and
the delete is in progress when the cli then attempts to create a new
controller. Currently, this proceeds without error although it can't be
very successful.

Fix this by validating that both the host port and remote port are
present when a new controller is to be created.

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-pci: fix NULL req in completion handler
Xianting Tian [Tue, 22 Sep 2020 06:25:17 +0000 (14:25 +0800)]
nvme-pci: fix NULL req in completion handler

Currently, we use nvmeq->q_depth as the upper limit for a valid tag in
nvme_handle_cqe(), it is not correct. Because the available tag number
is recorded in tagset, which is not equal to nvmeq->q_depth.

The nvme driver registers interrupts for queues before initializing the
tagset, because it uses the number of successful request_irq() calls to
configure the tagset parameters. This allows a race condition with the
current tag validity check if the controller happens to produce an
interrupt with a corrupted CQE before the tagset is initialized.

Replace the driver's indirect tag check with the one already provided by
the block layer.

Signed-off-by: Xianting Tian <tian.xianting@h3c.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme: return errors for hwmon init
Keith Busch [Thu, 17 Sep 2020 15:50:25 +0000 (08:50 -0700)]
nvme: return errors for hwmon init

Initializing the nvme hwmon retrieves a log from the controller. If the
controller is broken, we need to return the appropriate error so that
subsequent initialization doesn't attempt to continue.

Reported-by: Tong Zhang <ztong0001@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agoMerge tag 'nvme-5.9-2020-09-17' of git://git.infradead.org/nvme into block-5.9
Jens Axboe [Thu, 17 Sep 2020 17:49:44 +0000 (11:49 -0600)]
Merge tag 'nvme-5.9-2020-09-17' of git://git.infradead.org/nvme into block-5.9

Pull NVMe fixes from Christoph:

"nvme fixes for 5.9

  - another quirk for the controller from hell (David Milburn)
  - fix a Kconfig dependency (Necip Fazil Yildiran)
  - char devices / passthrough refcount fixes (Chaitanya Kulkarni)"

* tag 'nvme-5.9-2020-09-17' of git://git.infradead.org/nvme:
  nvmet: get transport reference for passthru ctrl
  nvme-core: get/put ctrl and transport module in nvme_dev_open/release()
  nvme-tcp: fix kconfig dependency warning when !CRYPTO
  nvme-pci: disable the write zeros command for Intel 600P/P3100

3 years agonvmet: get transport reference for passthru ctrl
Christoph Hellwig [Thu, 17 Sep 2020 01:11:01 +0000 (18:11 -0700)]
nvmet: get transport reference for passthru ctrl

Grab a reference to the transport driver to ensure it can't be unloaded
while a passthrough controller is active.

Fixes: c1fef73f793b ("nvmet: add passthru code to process commands")
Reported-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
3 years agonvme-core: get/put ctrl and transport module in nvme_dev_open/release()
Chaitanya Kulkarni [Wed, 16 Sep 2020 03:53:25 +0000 (20:53 -0700)]
nvme-core: get/put ctrl and transport module in nvme_dev_open/release()

Get and put the reference to the ctrl in the nvme_dev_open() and
nvme_dev_release() before and after module get/put for ctrl in char
device file operations.

Introduce char_dev relase function, get/put the controller and module
which allows us to fix the potential Oops which can be easily reproduced
with a passthru ctrl (although the problem also exists with pure user
access):

Entering kdb (current=0xffff8887f8290000, pid 3128) on processor 30 Oops: (null)
due to oops @ 0xffffffffa01019ad
CPU: 30 PID: 3128 Comm: bash Tainted: G        W  OE     5.8.0-rc4nvme-5.9+ #35
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.4
RIP: 0010:nvme_free_ctrl+0x234/0x285 [nvme_core]
Code: 57 10 a0 e8 73 bf 02 e1 ba 3d 11 00 00 48 c7 c6 98 33 10 a0 48 c7 c7 1d 57 10 a0 e8 5b bf 02 e1 8
RSP: 0018:ffffc90001d63de0 EFLAGS: 00010246
RAX: ffffffffa05c0440 RBX: ffff8888119e45a0 RCX: 0000000000000000
RDX: 0000000000000000 RSI: ffff8888177e9550 RDI: ffff8888119e43b0
RBP: ffff8887d4768000 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: ffffc90001d63c90 R12: ffff8888119e43b0
R13: ffff8888119e5108 R14: dead000000000100 R15: ffff8888119e5108
FS:  00007f1ef27b0740(0000) GS:ffff888817600000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffffffa05c0470 CR3: 00000007f6bee000 CR4: 00000000003406e0
Call Trace:
 device_release+0x27/0x80
 kobject_put+0x98/0x170
 nvmet_passthru_ctrl_disable+0x4a/0x70 [nvmet]
 nvmet_passthru_enable_store+0x4c/0x90 [nvmet]
 configfs_write_file+0xe6/0x150
 vfs_write+0xba/0x1e0
 ksys_write+0x5f/0xe0
 do_syscall_64+0x52/0xb0
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7f1ef1eb2840
Code: Bad RIP value.
RSP: 002b:00007fffdbff0eb8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f1ef1eb2840
RDX: 0000000000000002 RSI: 00007f1ef27d2000 RDI: 0000000000000001
RBP: 00007f1ef27d2000 R08: 000000000000000a R09: 00007f1ef27b0740
R10: 0000000000000001 R11: 0000000000000246 R12: 00007f1ef2186400
R13: 0000000000000002 R14: 0000000000000001 R15: 0000000000000000

With this patch fix we take the module ref count in nvme_dev_open() and
release that ref count in newly introduced nvme_dev_release().

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-tcp: fix kconfig dependency warning when !CRYPTO
Necip Fazil Yildiran [Mon, 14 Sep 2020 15:01:21 +0000 (18:01 +0300)]
nvme-tcp: fix kconfig dependency warning when !CRYPTO

When NVME_TCP is enabled and CRYPTO is disabled, it results in the
following Kbuild warning:

WARNING: unmet direct dependencies detected for CRYPTO_CRC32C
  Depends on [n]: CRYPTO [=n]
  Selected by [y]:
  - NVME_TCP [=y] && INET [=y] && BLK_DEV_NVME [=y]

The reason is that NVME_TCP selects CRYPTO_CRC32C without depending on or
selecting CRYPTO while CRYPTO_CRC32C is subordinate to CRYPTO.

Honor the kconfig menu hierarchy to remove kconfig dependency warnings.

Fixes: 79fd751d61aa ("nvme: tcp: selects CRYPTO_CRC32C for nvme-tcp")
Signed-off-by: Necip Fazil Yildiran <fazilyildiran@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-pci: disable the write zeros command for Intel 600P/P3100
David Milburn [Thu, 10 Sep 2020 21:18:50 +0000 (16:18 -0500)]
nvme-pci: disable the write zeros command for Intel 600P/P3100

The write zeros command does not work with 4k range.

bash-4.4# ./blkdiscard /dev/nvme0n1p2
bash-4.4# strace -efallocate xfs_io -c "fzero 536895488 2048" /dev/nvme0n1p2
fallocate(3, FALLOC_FL_ZERO_RANGE, 536895488, 2048) = 0
+++ exited with 0 +++
bash-4.4# dd bs=1 if=/dev/nvme0n1p2 skip=536895488 count=512 | hexdump -C
00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00000200

bash-4.4# ./blkdiscard /dev/nvme0n1p2
bash-4.4# strace -efallocate xfs_io -c "fzero 536895488 4096" /dev/nvme0n1p2
fallocate(3, FALLOC_FL_ZERO_RANGE, 536895488, 4096) = 0
+++ exited with 0 +++
bash-4.4# dd bs=1 if=/dev/nvme0n1p2 skip=536895488 count=512 | hexdump -C
00000000  5c 61 5c b0 96 21 1b 5e  85 0c 07 32 9c 8c eb 3c  |\a\..!.^...2...<|
00000010  4a a2 06 ca 67 15 2d 8e  29 8d a8 a0 7e 46 8c 62  |J...g.-.)...~F.b|
00000020  bb 4c 6c c1 6b f5 ae a5  e4 a9 bc 93 4f 60 ff 7a  |.Ll.k.......O`.z|

Reported-by: Eric Sandeen <esandeen@redhat.com>
Signed-off-by: David Milburn <dmilburn@redhat.com>
Tested-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agos390/dasd: Fix zero write for FBA devices
Jan Höppner [Mon, 14 Sep 2020 11:56:47 +0000 (13:56 +0200)]
s390/dasd: Fix zero write for FBA devices

A discard request that writes zeros using the global kernel internal
ZERO_PAGE will fail for machines with more than 2GB of memory due to the
location of the ZERO_PAGE.

Fix this by using a driver owned global zero page allocated with GFP_DMA
flag set.

Fixes: 28b841b3a7cb ("s390/dasd: Add discard support for FBA devices")
Signed-off-by: Jan Höppner <hoeppner@linux.ibm.com>
Reviewed-by: Stefan Haberland <sth@linux.ibm.com>
Cc: <stable@vger.kernel.org> # 4.14+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge tag 'nvme-5.9-2020-09-10' of git://git.infradead.org/nvme into block-5.9
Jens Axboe [Thu, 10 Sep 2020 13:12:22 +0000 (07:12 -0600)]
Merge tag 'nvme-5.9-2020-09-10' of git://git.infradead.org/nvme into block-5.9

Pull NVMe fixes from Christoph.

"nvme fixes for 5.9

 - cancel async events before freeing them (David Milburn)
 - revert a broken race fix (James Smart)
 - fix command processing during resets (Sagi Grimberg)"

* tag 'nvme-5.9-2020-09-10' of git://git.infradead.org/nvme:
  nvme-fabrics: allow to queue requests for live queues
  nvme-tcp: cancel async events before freeing event struct
  nvme-rdma: cancel async events before freeing event struct
  nvme-fc: cancel async events before freeing event struct
  nvme: Revert: Fix controller creation races with teardown flow

3 years agoblock: Set same_page to false in __bio_try_merge_page if ret is false
Ritesh Harjani [Wed, 9 Sep 2020 03:14:25 +0000 (08:44 +0530)]
block: Set same_page to false in __bio_try_merge_page if ret is false

If we hit the UINT_MAX limit of bio->bi_iter.bi_size and so we are anyway
not merging this page in this bio, then it make sense to make same_page
also as false before returning.

Without this patch, we hit below WARNING in iomap.
This mostly happens with very large memory system and / or after tweaking
vm dirty threshold params to delay writeback of dirty data.

WARNING: CPU: 18 PID: 5130 at fs/iomap/buffered-io.c:74 iomap_page_release+0x120/0x150
 CPU: 18 PID: 5130 Comm: fio Kdump: loaded Tainted: G        W         5.8.0-rc3 #6
 Call Trace:
  __remove_mapping+0x154/0x320 (unreliable)
  iomap_releasepage+0x80/0x180
  try_to_release_page+0x94/0xe0
  invalidate_inode_page+0xc8/0x110
  invalidate_mapping_pages+0x1dc/0x540
  generic_fadvise+0x3c8/0x450
  xfs_file_fadvise+0x2c/0xe0 [xfs]
  vfs_fadvise+0x3c/0x60
  ksys_fadvise64_64+0x68/0xe0
  sys_fadvise64+0x28/0x40
  system_call_exception+0xf8/0x1c0
  system_call_common+0xf0/0x278

Fixes: cc90bc68422 ("block: fix "check bi_size overflow before merge"")
Reported-by: Shivaprasad G Bhat <sbhat@linux.ibm.com>
Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Signed-off-by: Ritesh Harjani <riteshh@linux.ibm.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme-fabrics: allow to queue requests for live queues
Sagi Grimberg [Tue, 8 Sep 2020 19:56:08 +0000 (12:56 -0700)]
nvme-fabrics: allow to queue requests for live queues

Right now we are failing requests based on the controller state (which
is checked inline in nvmf_check_ready) however we should definitely
accept requests if the queue is live.

When entering controller reset, we transition the controller into
NVME_CTRL_RESETTING, and then return BLK_STS_RESOURCE for non-mpath
requests (have blk_noretry_request set).

This is also the case for NVME_REQ_USER for the wrong reason. There
shouldn't be any reason for us to reject this I/O in a controller reset.
We do want to prevent passthru commands on the admin queue because we
need the controller to fully initialize first before we let user passthru
admin commands to be issued.

In a non-mpath setup, this means that the requests will simply be
requeued over and over forever not allowing the q_usage_counter to drop
its final reference, causing controller reset to hang if running
concurrently with heavy I/O.

Fixes: 35897b920c8a ("nvme-fabrics: fix and refine state checks in __nvmf_check_ready")
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agoblock: only call sched requeue_request() for scheduled requests
Omar Sandoval [Tue, 8 Sep 2020 20:46:37 +0000 (13:46 -0700)]
block: only call sched requeue_request() for scheduled requests

Yang Yang reported the following crash caused by requeueing a flush
request in Kyber:

  [    2.517297] Unable to handle kernel paging request at virtual address ffffffd8071c0b00
  ...
  [    2.517468] pc : clear_bit+0x18/0x2c
  [    2.517502] lr : sbitmap_queue_clear+0x40/0x228
  [    2.517503] sp : ffffff800832bc60 pstate : 00c00145
  ...
  [    2.517599] Process ksoftirqd/5 (pid: 51, stack limit = 0xffffff8008328000)
  [    2.517602] Call trace:
  [    2.517606]  clear_bit+0x18/0x2c
  [    2.517619]  kyber_finish_request+0x74/0x80
  [    2.517627]  blk_mq_requeue_request+0x3c/0xc0
  [    2.517637]  __scsi_queue_insert+0x11c/0x148
  [    2.517640]  scsi_softirq_done+0x114/0x130
  [    2.517643]  blk_done_softirq+0x7c/0xb0
  [    2.517651]  __do_softirq+0x208/0x3bc
  [    2.517657]  run_ksoftirqd+0x34/0x60
  [    2.517663]  smpboot_thread_fn+0x1c4/0x2c0
  [    2.517667]  kthread+0x110/0x120
  [    2.517669]  ret_from_fork+0x10/0x18

This happens because Kyber doesn't track flush requests, so
kyber_finish_request() reads a garbage domain token. Only call the
scheduler's requeue_request() hook if RQF_ELVPRIV is set (like we do for
the finish_request() hook in blk_mq_free_request()). Now that we're
handling it in blk-mq, also remove the check from BFQ.

Reported-by: Yang Yang <yang.yang@vivo.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme-tcp: cancel async events before freeing event struct
David Milburn [Wed, 2 Sep 2020 22:42:53 +0000 (17:42 -0500)]
nvme-tcp: cancel async events before freeing event struct

Cancel async event work in case async event has been queued up, and
nvme_tcp_submit_async_event() runs after event has been freed.

Signed-off-by: David Milburn <dmilburn@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-rdma: cancel async events before freeing event struct
David Milburn [Wed, 2 Sep 2020 22:42:52 +0000 (17:42 -0500)]
nvme-rdma: cancel async events before freeing event struct

Cancel async event work in case async event has been queued up, and
nvme_rdma_submit_async_event() runs after event has been freed.

Signed-off-by: David Milburn <dmilburn@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-fc: cancel async events before freeing event struct
David Milburn [Wed, 2 Sep 2020 22:42:54 +0000 (17:42 -0500)]
nvme-fc: cancel async events before freeing event struct

Cancel async event work in case async event has been queued up, and
nvme_fc_submit_async_event() runs after event has been freed.

Signed-off-by: David Milburn <dmilburn@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme: Revert: Fix controller creation races with teardown flow
James Smart [Fri, 28 Aug 2020 19:01:50 +0000 (12:01 -0700)]
nvme: Revert: Fix controller creation races with teardown flow

The indicated patch introduced a barrier in the sysfs_delete attribute
for the controller that rejects the request if the controller isn't
created. "Created" is defined as at least 1 call to nvme_start_ctrl().

This is problematic in error-injection testing.  If an error occurs on
the initial attempt to create an association and the controller enters
reconnect(s) attempts, the admin cannot delete the controller until
either there is a successful association created or ctrl_loss_tmo
times out.

Where this issue is particularly hurtful is when the "admin" is the
nvme-cli, it is performing a connection to a discovery controller, and
it is initiated via auto-connect scripts.  With the FC transport, if the
first connection attempt fails, the controller enters a normal reconnect
state but returns control to the cli thread that created the controller.
In this scenario, the cli attempts to read the discovery log via ioctl,
which fails, causing the cli to see it as an empty log and then proceeds
to delete the discovery controller. The delete is rejected and the
controller is left live. If the discovery controller reconnect then
succeeds, there is no action to delete it, and it sits live doing nothing.

Cc: <stable@vger.kernel.org> # v5.7+
Fixes: ce1518139e69 ("nvme: Fix controller creation races with teardown flow")
Signed-off-by: James Smart <james.smart@broadcom.com>
CC: Israel Rukshin <israelr@mellanox.com>
CC: Max Gurtovoy <maxg@mellanox.com>
CC: Christoph Hellwig <hch@lst.de>
CC: Keith Busch <kbusch@kernel.org>
CC: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agoblock: restore a specific error code in bdev_del_partition
Christoph Hellwig [Tue, 8 Sep 2020 14:15:06 +0000 (16:15 +0200)]
block: restore a specific error code in bdev_del_partition

mdadm relies on the fact that deleting an invalid partition returns
-ENXIO or -ENOTTY to detect if a block device is a partition or a
whole device.

Fixes: 08fc1ab6d748 ("block: fix locking in bdev_del_partition")
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: allow for_each_bvec to support zero len bvec
Ming Lei [Mon, 17 Aug 2020 10:00:55 +0000 (18:00 +0800)]
block: allow for_each_bvec to support zero len bvec

Block layer usually doesn't support or allow zero-length bvec. Since
commit 1bdc76aea115 ("iov_iter: use bvec iterator to implement
iterate_bvec()"), iterate_bvec() switches to bvec iterator. However,
Al mentioned that 'Zero-length segments are not disallowed' in iov_iter.

Fixes for_each_bvec() so that it can move on after seeing one zero
length bvec.

Fixes: 1bdc76aea115 ("iov_iter: use bvec iterator to implement iterate_bvec()")
Reported-by: syzbot <syzbot+61acc40a49a3e46e25ea@syzkaller.appspotmail.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Link: https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg2262077.html
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblk-stat: make q->stats->lock irqsafe
Tejun Heo [Tue, 1 Sep 2020 18:52:32 +0000 (14:52 -0400)]
blk-stat: make q->stats->lock irqsafe

blk-iocost calls blk_stat_enable_accounting() while holding an irqsafe lock
which triggers a lockdep splat because q->stats->lock isn't irqsafe. Let's
make it irqsafe.

Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: cd006509b0a9 ("blk-iocost: account for IO size when testing latencies")
Cc: stable@vger.kernel.org # v5.8+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblk-iocost: ioc_pd_free() shouldn't assume irq disabled
Tejun Heo [Tue, 1 Sep 2020 18:52:31 +0000 (14:52 -0400)]
blk-iocost: ioc_pd_free() shouldn't assume irq disabled

ioc_pd_free() grabs irq-safe ioc->lock without ensuring that irq is disabled
when it can be called with irq disabled or enabled. This has a small chance
of causing A-A deadlocks and triggers lockdep splats. Use irqsave operations
instead.

Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 7caa47151ab2 ("blkcg: implement blk-iocost")
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: fix locking in bdev_del_partition
Christoph Hellwig [Tue, 1 Sep 2020 09:59:41 +0000 (11:59 +0200)]
block: fix locking in bdev_del_partition

We need to hold the whole device bd_mutex to protect against
other thread concurrently deleting out partition before we get
to it, and thus causing a use after free.

Fixes: cddae808aeb7 ("block: pass a hd_struct to delete_partition")
Reported-by: syzbot+6448f3c229bc52b82f69@syzkaller.appspotmail.com
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: release disk reference in hd_struct_free_work
Ming Lei [Tue, 1 Sep 2020 10:07:38 +0000 (18:07 +0800)]
block: release disk reference in hd_struct_free_work

Commit e8c7d14ac6c3 ("block: revert back to synchronous request_queue removal")
stops to release request queue from wq context because that commit
supposed all blk_put_queue() is called in context which is allowed
to sleep. However, this assumption isn't true because we release disk's
reference in partition's percpu_ref's ->release() which doesn't allow
to sleep, because the ->release() is run via call_rcu().

Fixes this issue by moving put disk reference into hd_struct_free_work()

Fixes: e8c7d14ac6c3 ("block: revert back to synchronous request_queue removal")
Reported-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Tested-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: ensure bdi->io_pages is always initialized
Jens Axboe [Mon, 31 Aug 2020 17:20:02 +0000 (11:20 -0600)]
block: ensure bdi->io_pages is always initialized

If a driver leaves the limit settings as the defaults, then we don't
initialize bdi->io_pages. This means that file systems may need to
work around bdi->io_pages == 0, which is somewhat messy.

Initialize the default value just like we do for ->ra_pages.

Cc: stable@vger.kernel.org
Fixes: 9491ae4aade6 ("mm: don't cap request size based on read-ahead setting")
Reported-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge branch 'nvme-5.9-rc' of git://git.infradead.org/nvme into block-5.9
Jens Axboe [Sat, 29 Aug 2020 16:54:11 +0000 (10:54 -0600)]
Merge branch 'nvme-5.9-rc' of git://git.infradead.org/nvme into block-5.9

Pull NVMe fixes from Sagi:

"- instance leak and io boundary fixes from Keith
 - fc locking fix from Christophe
 - various tcp/rdma reset during traffic fixes from Me
 - pci use-after-free fix from Tong
 - tcp target null deref fix from Ziye"

* 'nvme-5.9-rc' of git://git.infradead.org/nvme:
  nvme-pci: cancel nvme device request before disabling
  nvme: only use power of two io boundaries
  nvme: fix controller instance leak
  nvmet-fc: Fix a missed _irqsave version of spin_lock in 'nvmet_fc_fod_op_done()'
  nvme: Fix NULL dereference for pci nvme controllers
  nvme-rdma: fix reset hang if controller died in the middle of a reset
  nvme-rdma: fix timeout handler
  nvme-rdma: serialize controller teardown sequences
  nvme-tcp: fix reset hang if controller died in the middle of a reset
  nvme-tcp: fix timeout handler
  nvme-tcp: serialize controller teardown sequences
  nvme: have nvme_wait_freeze_timeout return if it timed out
  nvme-fabrics: don't check state NVME_CTRL_NEW for request acceptance
  nvmet-tcp: Fix NULL dereference when a connect data comes in h2cdata pdu

3 years agonvme-pci: cancel nvme device request before disabling
Tong Zhang [Fri, 28 Aug 2020 14:17:08 +0000 (10:17 -0400)]
nvme-pci: cancel nvme device request before disabling

This patch addresses an irq free warning and null pointer dereference
error problem when nvme devices got timeout error during initialization.
This problem happens when nvme_timeout() function is called while
nvme_reset_work() is still in execution. This patch fixed the problem by
setting flag of the problematic request to NVME_REQ_CANCELLED before
calling nvme_dev_disable() to make sure __nvme_submit_sync_cmd() returns
an error code and let nvme_submit_sync_cmd() fail gracefully.
The following is console output.

[   62.472097] nvme nvme0: I/O 13 QID 0 timeout, disable controller
[   62.488796] nvme nvme0: could not set timestamp (881)
[   62.494888] ------------[ cut here ]------------
[   62.495142] Trying to free already-free IRQ 11
[   62.495366] WARNING: CPU: 0 PID: 7 at kernel/irq/manage.c:1751 free_irq+0x1f7/0x370
[   62.495742] Modules linked in:
[   62.495902] CPU: 0 PID: 7 Comm: kworker/u4:0 Not tainted 5.8.0+ #8
[   62.496206] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-48-gd9c812dda519-p4
[   62.496772] Workqueue: nvme-reset-wq nvme_reset_work
[   62.497019] RIP: 0010:free_irq+0x1f7/0x370
[   62.497223] Code: e8 ce 49 11 00 48 83 c4 08 4c 89 e0 5b 5d 41 5c 41 5d 41 5e 41 5f c3 44 89 f6 48 c70
[   62.498133] RSP: 0000:ffffa96800043d40 EFLAGS: 00010086
[   62.498391] RAX: 0000000000000000 RBX: ffff9b87fc458400 RCX: 0000000000000000
[   62.498741] RDX: 0000000000000001 RSI: 0000000000000096 RDI: ffffffff9693d72c
[   62.499091] RBP: ffff9b87fd4c8f60 R08: ffffa96800043bfd R09: 0000000000000163
[   62.499440] R10: ffffa96800043bf8 R11: ffffa96800043bfd R12: ffff9b87fd4c8e00
[   62.499790] R13: ffff9b87fd4c8ea4 R14: 000000000000000b R15: ffff9b87fd76b000
[   62.500140] FS:  0000000000000000(0000) GS:ffff9b87fdc00000(0000) knlGS:0000000000000000
[   62.500534] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   62.500816] CR2: 0000000000000000 CR3: 000000003aa0a000 CR4: 00000000000006f0
[   62.501165] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   62.501515] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   62.501864] Call Trace:
[   62.501993]  pci_free_irq+0x13/0x20
[   62.502167]  nvme_reset_work+0x5d0/0x12a0
[   62.502369]  ? update_load_avg+0x59/0x580
[   62.502569]  ? ttwu_queue_wakelist+0xa8/0xc0
[   62.502780]  ? try_to_wake_up+0x1a2/0x450
[   62.502979]  process_one_work+0x1d2/0x390
[   62.503179]  worker_thread+0x45/0x3b0
[   62.503361]  ? process_one_work+0x390/0x390
[   62.503568]  kthread+0xf9/0x130
[   62.503726]  ? kthread_park+0x80/0x80
[   62.503911]  ret_from_fork+0x22/0x30
[   62.504090] ---[ end trace de9ed4a70f8d71e2 ]---
[  123.912275] nvme nvme0: I/O 12 QID 0 timeout, disable controller
[  123.914670] nvme nvme0: 1/0/0 default/read/poll queues
[  123.916310] BUG: kernel NULL pointer dereference, address: 0000000000000000
[  123.917469] #PF: supervisor write access in kernel mode
[  123.917725] #PF: error_code(0x0002) - not-present page
[  123.917976] PGD 0 P4D 0
[  123.918109] Oops: 0002 [#1] SMP PTI
[  123.918283] CPU: 0 PID: 7 Comm: kworker/u4:0 Tainted: G        W         5.8.0+ #8
[  123.918650] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-48-gd9c812dda519-p4
[  123.919219] Workqueue: nvme-reset-wq nvme_reset_work
[  123.919469] RIP: 0010:__blk_mq_alloc_map_and_request+0x21/0x80
[  123.919757] Code: 66 0f 1f 84 00 00 00 00 00 41 55 41 54 55 48 63 ee 53 48 8b 47 68 89 ee 48 89 fb 8b4
[  123.920657] RSP: 0000:ffffa96800043d40 EFLAGS: 00010286
[  123.920912] RAX: ffff9b87fc4fee40 RBX: ffff9b87fc8cb008 RCX: 0000000000000000
[  123.921258] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff9b87fc618000
[  123.921602] RBP: 0000000000000000 R08: ffff9b87fdc2c4a0 R09: ffff9b87fc616000
[  123.921949] R10: 0000000000000000 R11: ffff9b87fffd1500 R12: 0000000000000000
[  123.922295] R13: 0000000000000000 R14: ffff9b87fc8cb200 R15: ffff9b87fc8cb000
[  123.922641] FS:  0000000000000000(0000) GS:ffff9b87fdc00000(0000) knlGS:0000000000000000
[  123.923032] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  123.923312] CR2: 0000000000000000 CR3: 000000003aa0a000 CR4: 00000000000006f0
[  123.923660] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  123.924007] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  123.924353] Call Trace:
[  123.924479]  blk_mq_alloc_tag_set+0x137/0x2a0
[  123.924694]  nvme_reset_work+0xed6/0x12a0
[  123.924898]  process_one_work+0x1d2/0x390
[  123.925099]  worker_thread+0x45/0x3b0
[  123.925280]  ? process_one_work+0x390/0x390
[  123.925486]  kthread+0xf9/0x130
[  123.925642]  ? kthread_park+0x80/0x80
[  123.925825]  ret_from_fork+0x22/0x30
[  123.926004] Modules linked in:
[  123.926158] CR2: 0000000000000000
[  123.926322] ---[ end trace de9ed4a70f8d71e3 ]---
[  123.926549] RIP: 0010:__blk_mq_alloc_map_and_request+0x21/0x80
[  123.926832] Code: 66 0f 1f 84 00 00 00 00 00 41 55 41 54 55 48 63 ee 53 48 8b 47 68 89 ee 48 89 fb 8b4
[  123.927734] RSP: 0000:ffffa96800043d40 EFLAGS: 00010286
[  123.927989] RAX: ffff9b87fc4fee40 RBX: ffff9b87fc8cb008 RCX: 0000000000000000
[  123.928336] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff9b87fc618000
[  123.928679] RBP: 0000000000000000 R08: ffff9b87fdc2c4a0 R09: ffff9b87fc616000
[  123.929025] R10: 0000000000000000 R11: ffff9b87fffd1500 R12: 0000000000000000
[  123.929370] R13: 0000000000000000 R14: ffff9b87fc8cb200 R15: ffff9b87fc8cb000
[  123.929715] FS:  0000000000000000(0000) GS:ffff9b87fdc00000(0000) knlGS:0000000000000000
[  123.930106] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  123.930384] CR2: 0000000000000000 CR3: 000000003aa0a000 CR4: 00000000000006f0
[  123.930731] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  123.931077] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400

Co-developed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Tong Zhang <ztong0001@gmail.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme: only use power of two io boundaries
Keith Busch [Thu, 27 Aug 2020 17:38:57 +0000 (10:38 -0700)]
nvme: only use power of two io boundaries

The kernel requires a power of two for boundaries because that's the
only way it can efficiently split commands that cross them. A
controller, however, may report a non-power of two boundary.

The driver had been rounding the controller's value to one the kernel
can use, but splitting on the wrong boundary provides no benefit on the
device side, and incurs additional submission overhead from non-optimal
splits.

Don't provide any boundary hint if the controller's value can't be used
and log a warning when first scanning a disk's unreported IO boundary.
Since the chunk sector logic has grown, move it to a separate function.

Cc: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme: fix controller instance leak
Keith Busch [Wed, 26 Aug 2020 17:53:04 +0000 (10:53 -0700)]
nvme: fix controller instance leak

If the driver has to unbind from the controller for an early failure
before the subsystem has been set up, there won't be a subsystem holding
the controller's instance, so the controller needs to free its own
instance in this case.

Fixes: 733e4b69d508d ("nvme: Assign subsys instance from first ctrl")
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvmet-fc: Fix a missed _irqsave version of spin_lock in 'nvmet_fc_fod_op_done()'
Christophe JAILLET [Fri, 21 Aug 2020 07:58:19 +0000 (09:58 +0200)]
nvmet-fc: Fix a missed _irqsave version of spin_lock in 'nvmet_fc_fod_op_done()'

The way 'spin_lock()' and 'spin_lock_irqsave()' are used is not consistent
in this function.

Use 'spin_lock_irqsave()' also here, as there is no guarantee that
interruptions are disabled at that point, according to surrounding code.

Fixes: a97ec51b37ef ("nvmet_fc: Rework target side abort handling")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme: Fix NULL dereference for pci nvme controllers
Sagi Grimberg [Mon, 24 Aug 2020 22:47:25 +0000 (15:47 -0700)]
nvme: Fix NULL dereference for pci nvme controllers

PCIe controllers do not have fabric opts, verify they exist before
showing ctrl_loss_tmo or reconnect_delay attributes.

Fixes: 764075fdcb2f ("nvme: expose reconnect_delay and ctrl_loss_tmo via sysfs")
Reported-by: Tobias Markus <tobias@markus-regensburg.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-rdma: fix reset hang if controller died in the middle of a reset
Sagi Grimberg [Thu, 30 Jul 2020 20:42:42 +0000 (13:42 -0700)]
nvme-rdma: fix reset hang if controller died in the middle of a reset

If the controller becomes unresponsive in the middle of a reset, we
will hang because we are waiting for the freeze to complete, but that
cannot happen since we have commands that are inflight holding the
q_usage_counter, and we can't blindly fail requests that times out.

So give a timeout and if we cannot wait for queue freeze before
unfreezing, fail and have the error handling take care how to
proceed (either schedule a reconnect of remove the controller).

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-rdma: fix timeout handler
Sagi Grimberg [Wed, 29 Jul 2020 09:36:03 +0000 (02:36 -0700)]
nvme-rdma: fix timeout handler

When a request times out in a LIVE state, we simply trigger error
recovery and let the error recovery handle the request cancellation,
however when a request times out in a non LIVE state, we make sure to
complete it immediately as it might block controller setup or teardown
and prevent forward progress.

However tearing down the entire set of I/O and admin queues causes
freeze/unfreeze imbalance (q->mq_freeze_depth) because and is really
an overkill to what we actually need, which is to just fence controller
teardown that may be running, stop the queue, and cancel the request if
it is not already completed.

Now that we have the controller teardown_lock, we can safely serialize
request cancellation. This addresses a hang caused by calling extra
queue freeze on controller namespaces, causing unfreeze to not complete
correctly.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-rdma: serialize controller teardown sequences
Sagi Grimberg [Thu, 6 Aug 2020 01:13:58 +0000 (18:13 -0700)]
nvme-rdma: serialize controller teardown sequences

In the timeout handler we may need to complete a request because the
request that timed out may be an I/O that is a part of a serial sequence
of controller teardown or initialization. In order to complete the
request, we need to fence any other context that may compete with us
and complete the request that is timing out.

In this case, we could have a potential double completion in case
a hard-irq or a different competing context triggered error recovery
and is running inflight request cancellation concurrently with the
timeout handler.

Protect using a ctrl teardown_lock to serialize contexts that may
complete a cancelled request due to error recovery or a reset.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-tcp: fix reset hang if controller died in the middle of a reset
Sagi Grimberg [Thu, 30 Jul 2020 20:25:34 +0000 (13:25 -0700)]
nvme-tcp: fix reset hang if controller died in the middle of a reset

If the controller becomes unresponsive in the middle of a reset, we will
hang because we are waiting for the freeze to complete, but that cannot
happen since we have commands that are inflight holding the
q_usage_counter, and we can't blindly fail requests that times out.

So give a timeout and if we cannot wait for queue freeze before
unfreezing, fail and have the error handling take care how to proceed
(either schedule a reconnect of remove the controller).

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-tcp: fix timeout handler
Sagi Grimberg [Tue, 28 Jul 2020 20:16:36 +0000 (13:16 -0700)]
nvme-tcp: fix timeout handler

When a request times out in a LIVE state, we simply trigger error
recovery and let the error recovery handle the request cancellation,
however when a request times out in a non LIVE state, we make sure to
complete it immediately as it might block controller setup or teardown
and prevent forward progress.

However tearing down the entire set of I/O and admin queues causes
freeze/unfreeze imbalance (q->mq_freeze_depth) because and is really
an overkill to what we actually need, which is to just fence controller
teardown that may be running, stop the queue, and cancel the request if
it is not already completed.

Now that we have the controller teardown_lock, we can safely serialize
request cancellation. This addresses a hang caused by calling extra
queue freeze on controller namespaces, causing unfreeze to not complete
correctly.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-tcp: serialize controller teardown sequences
Sagi Grimberg [Thu, 6 Aug 2020 01:13:48 +0000 (18:13 -0700)]
nvme-tcp: serialize controller teardown sequences

In the timeout handler we may need to complete a request because the
request that timed out may be an I/O that is a part of a serial sequence
of controller teardown or initialization. In order to complete the
request, we need to fence any other context that may compete with us
and complete the request that is timing out.

In this case, we could have a potential double completion in case
a hard-irq or a different competing context triggered error recovery
and is running inflight request cancellation concurrently with the
timeout handler.

Protect using a ctrl teardown_lock to serialize contexts that may
complete a cancelled request due to error recovery or a reset.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme: have nvme_wait_freeze_timeout return if it timed out
Sagi Grimberg [Thu, 30 Jul 2020 20:24:45 +0000 (13:24 -0700)]
nvme: have nvme_wait_freeze_timeout return if it timed out

Users can detect if the wait has completed or not and take appropriate
actions based on this information (e.g. weather to continue
initialization or rather fail and schedule another initialization
attempt).

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme-fabrics: don't check state NVME_CTRL_NEW for request acceptance
Sagi Grimberg [Fri, 14 Aug 2020 18:46:51 +0000 (11:46 -0700)]
nvme-fabrics: don't check state NVME_CTRL_NEW for request acceptance

NVME_CTRL_NEW should never see any I/O, because in order to start
initialization it has to transition to NVME_CTRL_CONNECTING and from
there it will never return to this state.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvmet-tcp: Fix NULL dereference when a connect data comes in h2cdata pdu
Ziye Yang [Fri, 21 Aug 2020 16:48:10 +0000 (00:48 +0800)]
nvmet-tcp: Fix NULL dereference when a connect data comes in h2cdata pdu

When handling commands without in-capsule data, we assign the ttag
assuming we already have the queue commands array allocated (based
on the queue size information in the connect data payload). However
if the connect itself did not send the connect data in-capsule we
have yet to allocate the queue commands,and we will assign a bogus
ttag and suffer a NULL dereference when we receive the corresponding
h2cdata pdu.

Fix this by checking if we already allocated commands before
dereferencing it when handling h2cdata, if we didn't, its for sure a
connect and we should use the preallocated connect command.

Signed-off-by: Ziye Yang <ziye.yang@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
3 years agoMerge branch 'md-fixes' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md...
Jens Axboe [Fri, 28 Aug 2020 13:52:02 +0000 (07:52 -0600)]
Merge branch 'md-fixes' of https://git./linux/kernel/git/song/md into block-5.9

Pull MD fix from Song.

* 'md-fixes' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md/raid5: make sure stripe_size as power of two

3 years agomd/raid5: make sure stripe_size as power of two
Yufen Yu [Thu, 20 Aug 2020 13:22:05 +0000 (09:22 -0400)]
md/raid5: make sure stripe_size as power of two

Commit 3b5408b98e4d ("md/raid5: support config stripe_size by sysfs
entry") make stripe_size as a configurable value. It just requires
stripe_size as multiple of 4KB.

In fact, we should make sure stripe_size as power of two. Otherwise,
stripe_shift which is the result of ilog2 can not represent the real
stripe_size. Then, stripe_hash() and stripe_hash_locks_hash() may
get unexpected value.

Fixes: 3b5408b98e4d ("md/raid5: support config stripe_size by sysfs entry")
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agoloop: Set correct device size when using LOOP_CONFIGURE
Martijn Coenen [Tue, 25 Aug 2020 07:18:29 +0000 (09:18 +0200)]
loop: Set correct device size when using LOOP_CONFIGURE

The device size calculation was done before processing the loop
configuration, which meant that the we set the size on the underlying
block device incorrectly in case lo_offset/lo_sizelimit were set in the
configuration. Delay computing the size until we've setup the device
parameters correctly.

Fixes: 3448914e8cc5("loop: Add LOOP_CONFIGURE ioctl")
Reported-by: Lennart Poettering <mzxreary@0pointer.de>
Tested-by: Yang Xu <xuyang2018.jy@cn.fujitsu.com>
Signed-off-by: Martijn Coenen <maco@android.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonbd: restore default timeout when setting it to zero
Hou Pu [Mon, 10 Aug 2020 12:00:44 +0000 (08:00 -0400)]
nbd: restore default timeout when setting it to zero

If we configured io timeout of nbd0 to 100s. Later after we
finished using it, we configured nbd0 again and set the io
timeout to 0. We expect it would timeout after 30 seconds
and keep retry. But in fact we could not change the timeout
when we set it to 0. the timeout is still the original 100s.

So change the timeout to default 30s when we set it to zero.
It also behaves same as commit 2da22da57348 ("nbd: fix zero
cmd timeout handling v2").

It becomes more important if we were reconfigure a nbd device
and the io timeout it set to zero. Because it could take 30s
to detect the new socket and thus io could be completed more
quickly compared to 100s.

Signed-off-by: Hou Pu <houpu@bytedance.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonull_blk: fix passing of REQ_FUA flag in null_handle_rq
Hou Pu [Fri, 21 Aug 2020 08:34:42 +0000 (04:34 -0400)]
null_blk: fix passing of REQ_FUA flag in null_handle_rq

REQ_FUA should be checked using rq->cmd_flags instead of req_op().

Fixes: deb78b419dfda ("nullb: emulate cache")
Signed-off-by: Hou Pu <houpu@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvmet: Disable keep-alive timer when kato is cleared to 0h
Amit Engel [Wed, 19 Aug 2020 08:31:11 +0000 (11:31 +0300)]
nvmet: Disable keep-alive timer when kato is cleared to 0h

Based on nvme spec, when keep alive timeout is set to zero
the keep-alive timer should be disabled.

Signed-off-by: Amit Engel <amit.engel@dell.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: redirect commands on dying queue
Chao Leng [Tue, 18 Aug 2020 07:11:32 +0000 (09:11 +0200)]
nvme: redirect commands on dying queue

If a command send through nvme-multipath failed on a dying queue, resend it
on another path.

Signed-off-by: Chao Leng <lengchao@huawei.com>
[hch: rebased on top of the completion refactoring]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: just check the status code type in nvme_is_path_error
Christoph Hellwig [Tue, 18 Aug 2020 07:11:31 +0000 (09:11 +0200)]
nvme: just check the status code type in nvme_is_path_error

Check the SCT sub-field for a path related status instead of enumerating
invididual status code.  As of NVMe 1.4 this adds "Internal Path Error"
and "Controller Pathing Error" to the list, but it also future proofs for
additional status codes added to the category.

Suggested-by: Chao Leng <lengchao@huawei.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: refactor command completion
Christoph Hellwig [Tue, 18 Aug 2020 07:11:30 +0000 (09:11 +0200)]
nvme: refactor command completion

Lift all the code to decide the dispostition of a completed command
from nvme_complete_rq and nvme_failover_req into a new helper, which
returns an emum of the potential actions.  nvme_complete_rq then
just switches on those and calls the proper helper for the action.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: rename and document nvme_end_request
Christoph Hellwig [Tue, 18 Aug 2020 07:11:29 +0000 (09:11 +0200)]
nvme: rename and document nvme_end_request

nvme_end_request is a bit misnamed, as it wraps around the
blk_mq_complete_* API.  It's semantics also are non-trivial, so give it
a more descriptive name and add a comment explaining the semantics.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: skip noiob for zoned devices
Keith Busch [Fri, 7 Aug 2020 16:32:35 +0000 (09:32 -0700)]
nvme: skip noiob for zoned devices

Zoned block devices reuse the chunk_sectors queue limit to define zone
boundaries. If a such a device happens to also report an optimal
boundary, do not use that to define the chunk_sectors as that may
intermittently interfere with io splitting and zone size queries.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme-pci: fix PRP pool size
Christoph Hellwig [Tue, 18 Aug 2020 17:51:59 +0000 (19:51 +0200)]
nvme-pci: fix PRP pool size

All operations are based on the controller, not the host page size.
Switch the dma pool to use the controller page size as well to avoid
massive overallocations on large page size systems.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme-pci: Use u32 for nvme_dev.q_depth and nvme_queue.q_depth
John Garry [Fri, 14 Aug 2020 15:34:25 +0000 (23:34 +0800)]
nvme-pci: Use u32 for nvme_dev.q_depth and nvme_queue.q_depth

Recently nvme_dev.q_depth was changed from an int to u16 type.

This falls over for the queue depth calculation in nvme_pci_enable(),
where NVME_CAP_MQES(dev->ctrl.cap) + 1 may overflow as a u16, as
NVME_CAP_MQES() is a 16b number also. That happens for me, and this is the
result:

root@ubuntu:/home/john# [148.272996] Unable to handle kernel NULL pointer
dereference at virtual address 0000000000000010
Mem abort info:
ESR = 0x96000004
EC = 0x25: DABT (current EL), IL = 32 bits
SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS = 0x00000004
CM = 0, WnR = 0
user pgtable: 4k pages, 48-bit VAs, pgdp=00000a27bf3c9000
[0000000000000010] pgd=0000000000000000, p4d=0000000000000000
Internal error: Oops: 96000004 [#1] PREEMPT SMP
Modules linked in: nvme nvme_core
CPU: 56 PID: 256 Comm: kworker/u195:0 Not tainted
5.8.0-next-20200812 #27
Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 -
V1.16.01 03/15/2019
Workqueue: nvme-reset-wq nvme_reset_work [nvme]
pstate: 80c00009 (Nzcv daif +PAN +UAO BTYPE=--)
pc : __sg_alloc_table_from_pages+0xec/0x238
lr : __sg_alloc_table_from_pages+0xc8/0x238
sp : ffff800013ccbad0
x29: ffff800013ccbad0 x28: ffff0a27b3d380a8
x27: 0000000000000000 x26: 0000000000002dc2
x25: 0000000000000dc0 x24: 0000000000000000
x23: 0000000000000000 x22: ffff800013ccbbe8
x21: 0000000000000010 x20: 0000000000000000
x19: 00000000fffff000 x18: ffffffffffffffff
x17: 00000000000000c0 x16: fffffe289eaf6380
x15: ffff800011b59948 x14: ffff002bc8fe98f8
x13: ff00000000000000 x12: ffff8000114ca000
x11: 0000000000000000 x10: ffffffffffffffff
x9 : ffffffffffffffc0 x8 : ffff0a27b5f9b6a0
x7 : 0000000000000000 x6 : 0000000000000001
x5 : ffff0a27b5f9b680 x4 : 0000000000000000
x3 : ffff0a27b5f9b680 x2 : 0000000000000000
 x1 : 0000000000000001 x0 : 0000000000000000
 Call trace:
__sg_alloc_table_from_pages+0xec/0x238
sg_alloc_table_from_pages+0x18/0x28
iommu_dma_alloc+0x474/0x678
dma_alloc_attrs+0xd8/0xf0
nvme_alloc_queue+0x114/0x160 [nvme]
nvme_reset_work+0xb34/0x14b4 [nvme]
process_one_work+0x1e8/0x360
worker_thread+0x44/0x478
kthread+0x150/0x158
ret_from_fork+0x10/0x34
 Code: f94002c3 6b01017f 540007c2 11000486 (f8645aa5)
---[ end trace 89bb2b72d59bf925 ]---

Fix by making onto a u32.

Also use u32 for nvme_dev.q_depth, as we assign this value from
nvme_dev.q_depth, and nvme_dev.q_depth will possibly hold 65536 - this
avoids the same crash as above.

Fixes: 61f3b8963097 ("nvme-pci: use unsigned for io queue depth")
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: Use spin_lock_irq() when taking the ctrl->lock
Logan Gunthorpe [Wed, 12 Aug 2020 23:24:44 +0000 (17:24 -0600)]
nvme: Use spin_lock_irq() when taking the ctrl->lock

When locking the ctrl->lock spinlock IRQs need to be disabled to avoid a
dead lock. The new spin_lock() calls recently added produce the
following lockdep warning when running the blktest nvme/003:

    ================================
    WARNING: inconsistent lock state
    --------------------------------
    inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
    ksoftirqd/2/22 [HC0[0]:SC1[1]:HE0:SE0] takes:
    ffff888276a8c4c0 (&ctrl->lock){+.?.}-{2:2}, at: nvme_keep_alive_end_io+0x50/0xc0
    {SOFTIRQ-ON-W} state was registered at:
      lock_acquire+0x164/0x500
      _raw_spin_lock+0x28/0x40
      nvme_get_effects_log+0x37/0x1c0
      nvme_init_identify+0x9e4/0x14f0
      nvme_reset_work+0xadd/0x2360
      process_one_work+0x66b/0xb70
      worker_thread+0x6e/0x6c0
      kthread+0x1e7/0x210
      ret_from_fork+0x22/0x30
    irq event stamp: 1449221
    hardirqs last  enabled at (1449220): [<ffffffff81c58e69>] ktime_get+0xf9/0x140
    hardirqs last disabled at (1449221): [<ffffffff83129665>] _raw_spin_lock_irqsave+0x25/0x60
    softirqs last  enabled at (1449210): [<ffffffff83400447>] __do_softirq+0x447/0x595
    softirqs last disabled at (1449215): [<ffffffff81b489b5>] run_ksoftirqd+0x35/0x50

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

           CPU0
           ----
      lock(&ctrl->lock);
      <Interrupt>
        lock(&ctrl->lock);

     *** DEADLOCK ***

    no locks held by ksoftirqd/2/22.

    stack backtrace:
    CPU: 2 PID: 22 Comm: ksoftirqd/2 Not tainted 5.8.0-rc4-eid-vmlocalyes-dbg-00157-g7236657c6b3a #1450
    Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.12.0-1 04/01/2014
    Call Trace:
     dump_stack+0xc8/0x11a
     print_usage_bug.cold.63+0x235/0x23e
     mark_lock+0xa9c/0xcf0
     __lock_acquire+0xd9a/0x2b50
     lock_acquire+0x164/0x500
     _raw_spin_lock_irqsave+0x40/0x60
     nvme_keep_alive_end_io+0x50/0xc0
     blk_mq_end_request+0x158/0x210
     nvme_complete_rq+0x146/0x500
     nvme_loop_complete_rq+0x26/0x30 [nvme_loop]
     blk_done_softirq+0x187/0x1e0
     __do_softirq+0x118/0x595
     run_ksoftirqd+0x35/0x50
     smpboot_thread_fn+0x1d3/0x310
     kthread+0x1e7/0x210
     ret_from_fork+0x22/0x30

Fixes: be93e87e7802 ("nvme: support for multiple Command Sets Supported and Effects log pages")
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Tested-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvmet: call blk_mq_free_request() directly
Chaitanya Kulkarni [Thu, 6 Aug 2020 22:56:27 +0000 (15:56 -0700)]
nvmet: call blk_mq_free_request() directly

Instead of calling blk_put_request() which calls blk_mq_free_request(),
call blk_mq_free_request() directly for NVMeOF passthru. This is to
mainly avoid an extra function call in the completion path
nvmet_passthru_req_done().

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvmet: fix oops in pt cmd execution
Chaitanya Kulkarni [Thu, 6 Aug 2020 22:48:58 +0000 (15:48 -0700)]
nvmet: fix oops in pt cmd execution

In the existing NVMeOF Passthru core command handling on failure of
nvme_alloc_request() it errors out with rq value set to NULL. In the
error handling path it calls blk_put_request() without checking if
rq is set to NULL or not which produces following Oops:-

[ 1457.346861] BUG: kernel NULL pointer dereference, address: 0000000000000000
[ 1457.347838] #PF: supervisor read access in kernel mode
[ 1457.348464] #PF: error_code(0x0000) - not-present page
[ 1457.349085] PGD 0 P4D 0
[ 1457.349402] Oops: 0000 [#1] SMP NOPTI
[ 1457.349851] CPU: 18 PID: 10782 Comm: kworker/18:2 Tainted: G           OE     5.8.0-rc4nvme-5.9+ #35
[ 1457.350951] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e3214
[ 1457.352347] Workqueue: events nvme_loop_execute_work [nvme_loop]
[ 1457.353062] RIP: 0010:blk_mq_free_request+0xe/0x110
[ 1457.353651] Code: 3f ff ff ff 83 f8 01 75 0d 4c 89 e7 e8 1b db ff ff e9 2d ff ff ff 0f 0b eb ef 66 8
[ 1457.355975] RSP: 0018:ffffc900035b7de0 EFLAGS: 00010282
[ 1457.356636] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000002
[ 1457.357526] RDX: ffffffffa060bd05 RSI: 0000000000000000 RDI: 0000000000000000
[ 1457.358416] RBP: 0000000000000037 R08: 0000000000000000 R09: 0000000000000000
[ 1457.359317] R10: 0000000000000000 R11: 000000000000006d R12: 0000000000000000
[ 1457.360424] R13: ffff8887ffa68600 R14: 0000000000000000 R15: ffff8888150564c8
[ 1457.361322] FS:  0000000000000000(0000) GS:ffff888814600000(0000) knlGS:0000000000000000
[ 1457.362337] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1457.363058] CR2: 0000000000000000 CR3: 000000081c0ac000 CR4: 00000000003406e0
[ 1457.363973] Call Trace:
[ 1457.364296]  nvmet_passthru_execute_cmd+0x150/0x2c0 [nvmet]
[ 1457.364990]  process_one_work+0x24e/0x5a0
[ 1457.365493]  ? __schedule+0x353/0x840
[ 1457.365957]  worker_thread+0x3c/0x380
[ 1457.366426]  ? process_one_work+0x5a0/0x5a0
[ 1457.366948]  kthread+0x135/0x150
[ 1457.367362]  ? kthread_create_on_node+0x60/0x60
[ 1457.367934]  ret_from_fork+0x22/0x30
[ 1457.368388] Modules linked in: nvme_loop(OE) nvmet(OE) nvme_fabrics(OE) null_blk nvme(OE) nvme_corer
[ 1457.368414]  ata_piix crc32c_intel virtio_pci libata virtio_ring serio_raw t10_pi virtio floppy dm_]
[ 1457.380849] CR2: 0000000000000000
[ 1457.381288] ---[ end trace c6cab61bfd1f68fd ]---
[ 1457.381861] RIP: 0010:blk_mq_free_request+0xe/0x110
[ 1457.382469] Code: 3f ff ff ff 83 f8 01 75 0d 4c 89 e7 e8 1b db ff ff e9 2d ff ff ff 0f 0b eb ef 66 8
[ 1457.384749] RSP: 0018:ffffc900035b7de0 EFLAGS: 00010282
[ 1457.385393] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000002
[ 1457.386264] RDX: ffffffffa060bd05 RSI: 0000000000000000 RDI: 0000000000000000
[ 1457.387142] RBP: 0000000000000037 R08: 0000000000000000 R09: 0000000000000000
[ 1457.388029] R10: 0000000000000000 R11: 000000000000006d R12: 0000000000000000
[ 1457.388914] R13: ffff8887ffa68600 R14: 0000000000000000 R15: ffff8888150564c8
[ 1457.389798] FS:  0000000000000000(0000) GS:ffff888814600000(0000) knlGS:0000000000000000
[ 1457.390796] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1457.391508] CR2: 0000000000000000 CR3: 000000081c0ac000 CR4: 00000000003406e0
[ 1457.392525] Kernel panic - not syncing: Fatal exception
[ 1457.394138] Kernel Offset: disabled
[ 1457.394677] ---[ end Kernel panic - not syncing: Fatal exception ]---

We fix this Oops by adding a new goto label out_put_req and reordering
the blk_put_request call to avoid calling blk_put_request() with rq
value is set to NULL. Here we also update the rest of the code
accordingly.

Fixes: 06b7164dfdc0 ("nvmet: add passthru code to process commands")
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvmet: add ns tear down label for pt-cmd handling
Chaitanya Kulkarni [Thu, 6 Aug 2020 20:02:23 +0000 (13:02 -0700)]
nvmet: add ns tear down label for pt-cmd handling

In the current implementation before submitting the passthru cmd we
may come across error e.g. getting ns from passthru controller,
allocating a request from passthru controller, etc. For all the failure
cases it only uses single goto label fail_out.

In the target code, we follow the pattern to have a separate label for
each error out the case when setting up multiple things before the actual
action.

This patch follows the same pattern and renames generic fail_out label
to out_put_ns and updates the error out cases in the
nvmet_passthru_execute_cmd() where it is needed.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: multipath: round-robin: eliminate "fallback" variable
Martin Wilck [Thu, 6 Aug 2020 13:19:32 +0000 (15:19 +0200)]
nvme: multipath: round-robin: eliminate "fallback" variable

If we find an optimized path, we quit the loop immediately. Thus we can use
just one variable for the next path, slighly simplifying the code.

Signed-off-by: Martin Wilck <mwilck@suse.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme: multipath: round-robin: fix single non-optimized path case
Martin Wilck [Thu, 6 Aug 2020 13:19:31 +0000 (15:19 +0200)]
nvme: multipath: round-robin: fix single non-optimized path case

If there's only one usable, non-optimized path, nvme_round_robin_path()
returns NULL, which is wrong. Fix it by falling back to "old", like in
the single optimized path case. Also, if the active path isn't changed,
there's no need to re-assign the pointer.

Fixes: 3f6e3246db0e ("nvme-multipath: fix logic for non-optimized paths")
Signed-off-by: Martin Wilck <mwilck@suse.com>
Signed-off-by: Martin George <marting@netapp.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvme-fc: Fix wrong return value in __nvme_fc_init_request()
Tianjia Zhang [Sun, 2 Aug 2020 11:15:45 +0000 (19:15 +0800)]
nvme-fc: Fix wrong return value in __nvme_fc_init_request()

On an error exit path, a negative error code should be returned
instead of a positive return value.

Fixes: e399441de9115 ("nvme-fabrics: Add host support for FC transport")
Cc: James Smart <jsmart2021@gmail.com>
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvmet-passthru: Reject commands with non-sgl flags set
Logan Gunthorpe [Wed, 29 Jul 2020 19:10:09 +0000 (13:10 -0600)]
nvmet-passthru: Reject commands with non-sgl flags set

Any command with a non-SGL flag set (like fuse flags) should be
rejected.

Fixes: c1fef73f793b ("nvmet: add passthru code to process commands")
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonvmet: fix a memory leak
Sagi Grimberg [Thu, 30 Jul 2020 18:51:00 +0000 (11:51 -0700)]
nvmet: fix a memory leak

We forgot to free new_model_number

Fixes: 013b7ebe5a0d ("nvmet: make ctrl model configurable")
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblkcg: fix memleak for iolatency
Yufen Yu [Tue, 11 Aug 2020 02:21:16 +0000 (22:21 -0400)]
blkcg: fix memleak for iolatency

Normally, blkcg_iolatency_exit() will free related memory in iolatency
when cleanup queue. But if blk_throtl_init() return error and queue init
fail, blkcg_iolatency_exit() will not do that for us. Then it cause
memory leak.

Fixes: d70675121546 ("block: introduce blk-iolatency io controller")
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMAINTAINERS: Add missing header files to BLOCK LAYER section
Geert Uytterhoeven [Wed, 19 Aug 2020 12:32:43 +0000 (14:32 +0200)]
MAINTAINERS: Add missing header files to BLOCK LAYER section

The various <linux/blk*.h> header files are part of the Block Layer.
Add them to the corresponding section in the MAINTAINERS file, so
scripts/get_maintainer.pl will pick them up.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: fix get_max_io_size()
Keith Busch [Thu, 6 Aug 2020 21:58:37 +0000 (14:58 -0700)]
block: fix get_max_io_size()

A previous commit aligning splits to physical block sizes inadvertently
modified one return case such that that it now returns 0 length splits
when the number of sectors doesn't exceed the physical offset. This
later hits a BUG in bio_split(). Restore the previous working behavior.

Fixes: 9cc5169cd478b ("block: Improve physical block alignment of split bios")
Reported-by: Eric Deal <eric.deal@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblk-mq: insert request not through ->queue_rq into sw/scheduler queue
Ming Lei [Tue, 18 Aug 2020 09:07:28 +0000 (17:07 +0800)]
blk-mq: insert request not through ->queue_rq into sw/scheduler queue

c616cbee97ae ("blk-mq: punt failed direct issue to dispatch list") supposed
to add request which has been through ->queue_rq() to the hw queue dispatch
list, however it adds request running out of budget or driver tag to hw queue
too. This way basically bypasses request merge, and causes too many request
dispatched to LLD, and system% is unnecessary increased.

Fixes this issue by adding request not through ->queue_rq into sw/scheduler
queue, and this way is safe because no ->queue_rq is called on this request
yet.

High %system can be observed on Azure storvsc device, and even soft lock
is observed. This patch reduces %system during heavy sequential IO,
meantime decreases soft lockup risk.

Fixes: c616cbee97ae ("blk-mq: punt failed direct issue to dispatch list")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock/rnbd: Ensure err is always initialized in process_rdma
Nathan Chancellor [Tue, 18 Aug 2020 06:49:25 +0000 (23:49 -0700)]
block/rnbd: Ensure err is always initialized in process_rdma

Clang warns:

drivers/block/rnbd/rnbd-srv.c:150:6: warning: variable 'err' is used
uninitialized whenever 'if' condition is true
[-Wsometimes-uninitialized]
        if (IS_ERR(bio)) {
            ^~~~~~~~~~~
drivers/block/rnbd/rnbd-srv.c:177:9: note: uninitialized use occurs here
        return err;
               ^~~
drivers/block/rnbd/rnbd-srv.c:150:2: note: remove the 'if' if its
condition is always false
        if (IS_ERR(bio)) {
        ^~~~~~~~~~~~~~~~~~
drivers/block/rnbd/rnbd-srv.c:126:9: note: initialize the variable 'err'
to silence this warning
        int err;
               ^
                = 0
1 warning generated.

err is indeed uninitialized when this statement is taken. Ensure that it
is assigned the error value of bio before jumping to the error handling
label.

Fixes: 735d77d4fd28 ("rnbd: remove rnbd_dev_submit_io")
Reported-by: Brooke Basile <brookebasile@gmail.com>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Acked-by: Jack Wang <jinpu.wang@cloud.ionos.com>
Link: https://github.com/ClangBuiltLinux/linux/issues/1134
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobfq: fix blkio cgroup leakage v4
Dmitry Monakhov [Tue, 11 Aug 2020 06:43:40 +0000 (06:43 +0000)]
bfq: fix blkio cgroup leakage v4

Changes from v1:
    - update commit description with proper ref-accounting justification

commit db37a34c563b ("block, bfq: get a ref to a group when adding it to a service tree")
introduce leak forbfq_group and blkcg_gq objects because of get/put
imbalance.
In fact whole idea of original commit is wrong because bfq_group entity
can not dissapear under us because it is referenced by child bfq_queue's
entities from here:
 -> bfq_init_entity()
    ->bfqg_and_blkg_get(bfqg);
    ->entity->parent = bfqg->my_entity

 -> bfq_put_queue(bfqq)
    FINAL_PUT
    ->bfqg_and_blkg_put(bfqq_group(bfqq))
    ->kmem_cache_free(bfq_pool, bfqq);

So parent entity can not disappear while child entity is in tree,
and child entities already has proper protection.
This patch revert commit db37a34c563b ("block, bfq: get a ref to a group when adding it to a service tree")

bfq_group leak trace caused by bad commit:
-> blkg_alloc
   -> bfq_pq_alloc
     -> bfqg_get (+1)
->bfq_activate_bfqq
  ->bfq_activate_requeue_entity
    -> __bfq_activate_entity
       ->bfq_get_entity
         ->bfqg_and_blkg_get (+1)  <==== : Note1
->bfq_del_bfqq_busy
  ->bfq_deactivate_entity+0x53/0xc0 [bfq]
    ->__bfq_deactivate_entity+0x1b8/0x210 [bfq]
      -> bfq_forget_entity(is_in_service = true)
 entity->on_st_or_in_serv = false   <=== :Note2
 if (is_in_service)
     return;  ==> do not touch reference
-> blkcg_css_offline
 -> blkcg_destroy_blkgs
  -> blkg_destroy
   -> bfq_pd_offline
    -> __bfq_deactivate_entity
         if (!entity->on_st_or_in_serv) /* true, because (Note2)
return false;
 -> bfq_pd_free
    -> bfqg_put() (-1, byt bfqg->ref == 2) because of (Note2)
So bfq_group and blkcg_gq  will leak forever, see test-case below.

##TESTCASE_BEGIN:
#!/bin/bash

max_iters=${1:-100}
#prep cgroup mounts
mount -t tmpfs cgroup_root /sys/fs/cgroup
mkdir /sys/fs/cgroup/blkio
mount -t cgroup -o blkio none /sys/fs/cgroup/blkio

# Prepare blkdev
grep blkio /proc/cgroups
truncate -s 1M img
losetup /dev/loop0 img
echo bfq > /sys/block/loop0/queue/scheduler

grep blkio /proc/cgroups
for ((i=0;i<max_iters;i++))
do
    mkdir -p /sys/fs/cgroup/blkio/a
    echo 0 > /sys/fs/cgroup/blkio/a/cgroup.procs
    dd if=/dev/loop0 bs=4k count=1 of=/dev/null iflag=direct 2> /dev/null
    echo 0 > /sys/fs/cgroup/blkio/cgroup.procs
    rmdir /sys/fs/cgroup/blkio/a
    grep blkio /proc/cgroups
done
##TESTCASE_END:

Fixes: db37a34c563b ("block, bfq: get a ref to a group when adding it to a service tree")
Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Signed-off-by: Dmitry Monakhov <dmtrmonakhov@yandex-team.ru>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: Fix page_is_mergeable() for compound pages
Matthew Wilcox (Oracle) [Mon, 17 Aug 2020 19:52:06 +0000 (20:52 +0100)]
block: Fix page_is_mergeable() for compound pages

If we pass in an offset which is larger than PAGE_SIZE, then
page_is_mergeable() thinks it's not mergeable with the previous bio_vec,
leading to a large number of bio_vecs being used.  Use a slightly more
obvious test that the two pages are compatible with each other.

Fixes: 52d52d1c98a9 ("block: only allow contiguous page structs in a bio_vec")
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: virtio_blk: fix handling single range discard request
Ming Lei [Mon, 17 Aug 2020 09:52:40 +0000 (17:52 +0800)]
block: virtio_blk: fix handling single range discard request

1f23816b8eb8 ("virtio_blk: add discard and write zeroes support") starts
to support multi-range discard for virtio-blk. However, the virtio-blk
disk may report max discard segment as 1, at least that is exactly what
qemu is doing.

So far, block layer switches to normal request merge if max discard segment
limit is 1, and multiple bios can be merged to single segment. This way may
cause memory corruption in virtblk_setup_discard_write_zeroes().

Fix the issue by handling single max discard segment in straightforward
way.

Fixes: 1f23816b8eb8 ("virtio_blk: add discard and write zeroes support")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Changpeng Liu <changpeng.liu@intel.com>
Cc: Daniel Verkamp <dverkamp@chromium.org>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: respect queue limit of max discard segment
Ming Lei [Mon, 17 Aug 2020 09:52:39 +0000 (17:52 +0800)]
block: respect queue limit of max discard segment

When queue_max_discard_segments(q) is 1, blk_discard_mergable() will
return false for discard request, then normal request merge is applied.
However, only queue_max_segments() is checked, so max discard segment
limit isn't respected.

Check max discard segment limit in the request merge code for fixing
the issue.

Discard request failure of virtio_blk is fixed.

Fixes: 69840466086d ("block: fix the DISCARD request merge")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: loop: set discard granularity and alignment for block device backed loop
Ming Lei [Mon, 17 Aug 2020 10:01:30 +0000 (18:01 +0800)]
block: loop: set discard granularity and alignment for block device backed loop

In case of block device backend, if the backend supports write zeros, the
loop device will set queue flag of QUEUE_FLAG_DISCARD. However,
limits.discard_granularity isn't setup, and this way is wrong,
see the following description in Documentation/ABI/testing/sysfs-block:

A discard_granularity of 0 means that the device does not support
discard functionality.

Especially 9b15d109a6b2 ("block: improve discard bio alignment in
__blkdev_issue_discard()") starts to take q->limits.discard_granularity
for computing max discard sectors. And zero discard granularity may cause
kernel oops, or fail discard request even though the loop queue claims
discard support via QUEUE_FLAG_DISCARD.

Fix the issue by setup discard granularity and alignment.

Fixes: c52abf563049 ("loop: Better discard support for block devices")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Coly Li <colyli@suse.de>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Xiao Ni <xni@redhat.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Evan Green <evgreen@chromium.org>
Cc: Gwendal Grignou <gwendal@chromium.org>
Cc: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Cc: Andrzej Pietrasiewicz <andrzej.p@collabora.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblk-mq: order adding requests to hctx->dispatch and checking SCHED_RESTART
Ming Lei [Mon, 17 Aug 2020 10:01:15 +0000 (18:01 +0800)]
blk-mq: order adding requests to hctx->dispatch and checking SCHED_RESTART

SCHED_RESTART code path is relied to re-run queue for dispatch requests
in hctx->dispatch. Meantime the SCHED_RSTART flag is checked when adding
requests to hctx->dispatch.

memory barriers have to be used for ordering the following two pair of OPs:

1) adding requests to hctx->dispatch and checking SCHED_RESTART in
blk_mq_dispatch_rq_list()

2) clearing SCHED_RESTART and checking if there is request in hctx->dispatch
in blk_mq_sched_restart().

Without the added memory barrier, either:

1) blk_mq_sched_restart() may miss requests added to hctx->dispatch meantime
blk_mq_dispatch_rq_list() observes SCHED_RESTART, and not run queue in
dispatch side

or

2) blk_mq_dispatch_rq_list still sees SCHED_RESTART, and not run queue
in dispatch side, meantime checking if there is request in
hctx->dispatch from blk_mq_sched_restart() is missed.

IO hang in ltp/fs_fill test is reported by kernel test robot:

https://lkml.org/lkml/2020/7/26/77

Turns out it is caused by the above out-of-order OPs. And the IO hang
can't be observed any more after applying this patch.

Fixes: bd166ef183c2 ("blk-mq-sched: add framework for MQ capable IO schedulers")
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Jeffery <djeffery@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobsg-lib: convert comma to semicolon
Xu Wang [Mon, 17 Aug 2020 02:16:49 +0000 (02:16 +0000)]
bsg-lib: convert comma to semicolon

Replace a comma between expression statements by a semicolon.

Signed-off-by: Xu Wang <vulab@iscas.ac.cn>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: blk-mq.c: fix @at_head kernel-doc warning
Randy Dunlap [Sun, 16 Aug 2020 23:39:34 +0000 (16:39 -0700)]
block: blk-mq.c: fix @at_head kernel-doc warning

Fix a kernel-doc warning in block/blk-mq.c:

../block/blk-mq.c:1844: warning: Function parameter or member 'at_head' not described in 'blk_mq_request_bypass_insert'

Fixes: 01e99aeca397 ("blk-mq: insert passthrough request into hctx->dispatch directly")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: André Almeida <andrealmeid@collabora.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: linux-block@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoLinux 5.9-rc1
Linus Torvalds [Sun, 16 Aug 2020 20:04:57 +0000 (13:04 -0700)]
Linux 5.9-rc1

3 years agoMerge tag 'io_uring-5.9-2020-08-15' of git://git.kernel.dk/linux-block
Linus Torvalds [Sun, 16 Aug 2020 17:55:12 +0000 (10:55 -0700)]
Merge tag 'io_uring-5.9-2020-08-15' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "A few differerent things in here.

  Seems like syzbot got some more io_uring bits wired up, and we got a
  handful of reports and the associated fixes are in here.

  General fixes too, and a lot of them marked for stable.

  Lastly, a bit of fallout from the async buffered reads, where we now
  more easily trigger short reads. Some applications don't really like
  that, so the io_read() code now handles short reads internally, and
  got a cleanup along the way so that it's now easier to read (and
  documented). We're now passing tests that failed before"

* tag 'io_uring-5.9-2020-08-15' of git://git.kernel.dk/linux-block:
  io_uring: short circuit -EAGAIN for blocking read attempt
  io_uring: sanitize double poll handling
  io_uring: internally retry short reads
  io_uring: retain iov_iter state over io_read/io_write calls
  task_work: only grab task signal lock when needed
  io_uring: enable lookup of links holding inflight files
  io_uring: fail poll arm on queue proc failure
  io_uring: hold 'ctx' reference around task_work queue + execute
  fs: RWF_NOWAIT should imply IOCB_NOIO
  io_uring: defer file table grabbing request cleanup for locked requests
  io_uring: add missing REQ_F_COMP_LOCKED for nested requests
  io_uring: fix recursive completion locking on oveflow flush
  io_uring: use TWA_SIGNAL for task_work uncondtionally
  io_uring: account locked memory before potential error case
  io_uring: set ctx sq/cq entry count earlier
  io_uring: Fix NULL pointer dereference in loop_rw_iter()
  io_uring: add comments on how the async buffered read retry works
  io_uring: io_async_buf_func() need not test page bit

3 years agoparisc: fix PMD pages allocation by restoring pmd_alloc_one()
Mike Rapoport [Sun, 16 Aug 2020 14:24:03 +0000 (17:24 +0300)]
parisc: fix PMD pages allocation by restoring pmd_alloc_one()

Commit 1355c31eeb7e ("asm-generic: pgalloc: provide generic pmd_alloc_one()
and pmd_free_one()") converted parisc to use generic version of
pmd_alloc_one() but it missed the fact that parisc uses order-1 pages for
PMD.

Restore the original version of pmd_alloc_one() for parisc, just use
GFP_PGTABLE_KERNEL that implies __GFP_ZERO instead of GFP_KERNEL and
memset.

Fixes: 1355c31eeb7e ("asm-generic: pgalloc: provide generic pmd_alloc_one() and pmd_free_one()")
Reported-by: Meelis Roos <mroos@linux.ee>
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Tested-by: Meelis Roos <mroos@linux.ee>
Reviewed-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Link: https://lkml.kernel.org/r/9f2b5ebd-e4a4-0fa1-6cd3-4b9f6892d1ad@linux.ee
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agoMerge tag 'block-5.9-2020-08-14' of git://git.kernel.dk/linux-block
Linus Torvalds [Sun, 16 Aug 2020 03:36:42 +0000 (20:36 -0700)]
Merge tag 'block-5.9-2020-08-14' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A few fixes on the block side of things:

   - Discard granularity fix (Coly)

   - rnbd cleanups (Guoqing)

   - md error handling fix (Dan)

   - md sysfs fix (Junxiao)

   - Fix flush request accounting, which caused an IO slowdown for some
     configurations (Ming)

   - Properly propagate loop flag for partition scanning (Lennart)"

* tag 'block-5.9-2020-08-14' of git://git.kernel.dk/linux-block:
  block: fix double account of flush request's driver tag
  loop: unset GENHD_FL_NO_PART_SCAN on LOOP_CONFIGURE
  rnbd: no need to set bi_end_io in rnbd_bio_map_kern
  rnbd: remove rnbd_dev_submit_io
  md-cluster: Fix potential error pointer dereference in resize_bitmaps()
  block: check queue's limits.discard_granularity in __blkdev_issue_discard()
  md: get sysfs entry after redundancy attr group create

3 years agoMerge tag 'riscv-for-linus-5.9-mw1' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 16 Aug 2020 01:54:42 +0000 (18:54 -0700)]
Merge tag 'riscv-for-linus-5.9-mw1' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fix from Palmer Dabbelt:
 "I collected a single fix during the merge window: we managed to break
  the early trap setup on !MMU, this fixes it"

* tag 'riscv-for-linus-5.9-mw1' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: Setup exception vector for nommu platform

3 years agoMerge tag 'sh-for-5.9' of git://git.libc.org/linux-sh
Linus Torvalds [Sun, 16 Aug 2020 01:50:32 +0000 (18:50 -0700)]
Merge tag 'sh-for-5.9' of git://git.libc.org/linux-sh

Pull arch/sh updates from Rich Felker:
 "Cleanup, SECCOMP_FILTER support, message printing fixes, and other
  changes to arch/sh"

* tag 'sh-for-5.9' of git://git.libc.org/linux-sh: (34 commits)
  sh: landisk: Add missing initialization of sh_io_port_base
  sh: bring syscall_set_return_value in line with other architectures
  sh: Add SECCOMP_FILTER
  sh: Rearrange blocks in entry-common.S
  sh: switch to copy_thread_tls()
  sh: use the generic dma coherent remap allocator
  sh: don't allow non-coherent DMA for NOMMU
  dma-mapping: consolidate the NO_DMA definition in kernel/dma/Kconfig
  sh: unexport register_trapped_io and match_trapped_io_handler
  sh: don't include <asm/io_trapped.h> in <asm/io.h>
  sh: move the ioremap implementation out of line
  sh: move ioremap_fixed details out of <asm/io.h>
  sh: remove __KERNEL__ ifdefs from non-UAPI headers
  sh: sort the selects for SUPERH alphabetically
  sh: remove -Werror from Makefiles
  sh: Replace HTTP links with HTTPS ones
  arch/sh/configs: remove obsolete CONFIG_SOC_CAMERA*
  sh: stacktrace: Remove stacktrace_ops.stack()
  sh: machvec: Modernize printing of kernel messages
  sh: pci: Modernize printing of kernel messages
  ...

3 years agoio_uring: short circuit -EAGAIN for blocking read attempt
Jens Axboe [Sat, 15 Aug 2020 22:58:42 +0000 (15:58 -0700)]
io_uring: short circuit -EAGAIN for blocking read attempt

One case was missed in the short IO retry handling, and that's hitting
-EAGAIN on a blocking attempt read (eg from io-wq context). This is a
problem on sockets that are marked as non-blocking when created, they
don't carry any REQ_F_NOWAIT information to help us terminate them
instead of perpetually retrying.

Fixes: 227c0c9673d8 ("io_uring: internally retry short reads")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoio_uring: sanitize double poll handling
Jens Axboe [Sat, 15 Aug 2020 18:44:50 +0000 (11:44 -0700)]
io_uring: sanitize double poll handling

There's a bit of confusion on the matching pairs of poll vs double poll,
depending on if the request is a pure poll (IORING_OP_POLL_ADD) or
poll driven retry.

Add io_poll_get_double() that returns the double poll waitqueue, if any,
and io_poll_get_single() that returns the original poll waitqueue. With
that, remove the argument to io_poll_remove_double().

Finally ensure that wait->private is cleared once the double poll handler
has run, so that remove knows it's already been seen.

Cc: stable@vger.kernel.org # v5.8
Reported-by: syzbot+7f617d4a9369028b8a2c@syzkaller.appspotmail.com
Fixes: 18bceab101ad ("io_uring: allow POLL_ADD with double poll_wait() users")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge tag 'perf-tools-2020-08-14' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 15 Aug 2020 18:17:15 +0000 (11:17 -0700)]
Merge tag 'perf-tools-2020-08-14' of git://git./linux/kernel/git/acme/linux

Pull more perf tools updates from Arnaldo Carvalho de Melo:
 "Fixes:
   - Fixes for 'perf bench numa'.

   - Always memset source before memcpy in 'perf bench mem'.

   - Quote CC and CXX for their arguments to fix build in environments
     using those variables to pass more than just the compiler names.

   - Fix module symbol processing, addressing regression detected via
     "perf test".

   - Allow multiple probes in record+script_probe_vfs_getname.sh 'perf
     test' entry.

  Improvements:
   - Add script to autogenerate socket family name id->string table from
     copy of kernel header, used so far in 'perf trace'.

   - 'perf ftrace' improvements to provide similar options for this
     utility so that one can go from 'perf record', 'perf trace', etc to
     'perf ftrace' just by changing the name of the subcommand.

   - Prefer new "sched:sched_waking" trace event when it exists in 'perf
     sched' post processing.

   - Update POWER9 metrics to utilize other metrics.

   - Fall back to querying debuginfod if debuginfo not found locally.

  Miscellaneous:
   - Sync various kvm headers with kernel sources"

* tag 'perf-tools-2020-08-14' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux: (40 commits)
  perf ftrace: Make option description initials all capital letters
  perf build-ids: Fall back to debuginfod query if debuginfo not found
  perf bench numa: Remove dead code in parse_nodes_opt()
  perf stat: Update POWER9 metrics to utilize other metrics
  perf ftrace: Add change log
  perf: ftrace: Add set_tracing_options() to set all trace options
  perf ftrace: Add option --tid to filter by thread id
  perf ftrace: Add option -D/--delay to delay tracing
  perf: ftrace: Allow set graph depth by '--graph-opts'
  perf ftrace: Add support for trace option tracing_thresh
  perf ftrace: Add option 'verbose' to show more info for graph tracer
  perf ftrace: Add support for tracing option 'irq-info'
  perf ftrace: Add support for trace option funcgraph-irqs
  perf ftrace: Add support for trace option sleep-time
  perf ftrace: Add support for tracing option 'func_stack_trace'
  perf tools: Add general function to parse sublevel options
  perf ftrace: Add option '--inherit' to trace children processes
  perf ftrace: Show trace column header
  perf ftrace: Add option '-m/--buffer-size' to set per-cpu buffer size
  perf ftrace: Factor out function write_tracing_file_int()
  ...

3 years agoMerge tag 'x86-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 15 Aug 2020 17:38:03 +0000 (10:38 -0700)]
Merge tag 'x86-urgent-2020-08-15' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "Misc fixes and small updates all around the place:

   - Fix mitigation state sysfs output

   - Fix an FPU xstate/sxave code assumption bug triggered by
     Architectural LBR support

   - Fix Lightning Mountain SoC TSC frequency enumeration bug

   - Fix kexec debug output

   - Fix kexec memory range assumption bug

   - Fix a boundary condition in the crash kernel code

   - Optimize porgatory.ro generation a bit

   - Enable ACRN guests to use X2APIC mode

   - Reduce a __text_poke() IRQs-off critical section for the benefit of
     PREEMPT_RT"

* tag 'x86-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/alternatives: Acquire pte lock with interrupts enabled
  x86/bugs/multihit: Fix mitigation reporting when VMX is not in use
  x86/fpu/xstate: Fix an xstate size check warning with architectural LBRs
  x86/purgatory: Don't generate debug info for purgatory.ro
  x86/tsr: Fix tsc frequency enumeration bug on Lightning Mountain SoC
  kexec_file: Correctly output debugging information for the PT_LOAD ELF header
  kexec: Improve & fix crash_exclude_mem_range() to handle overlapping ranges
  x86/crash: Correct the address boundary of function parameters
  x86/acrn: Remove redundant chars from ACRN signature
  x86/acrn: Allow ACRN guest to use X2APIC mode

3 years agoMerge tag 'sched-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 15 Aug 2020 17:36:40 +0000 (10:36 -0700)]
Merge tag 'sched-urgent-2020-08-15' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Ingo Molnar:
 "Two fixes: fix a new tracepoint's output value, and fix the formatting
  of show-state syslog printouts"

* tag 'sched-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/debug: Fix the alignment of the show-state debug output
  sched: Fix use of count for nr_running tracepoint

3 years agoMerge tag 'perf-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 15 Aug 2020 17:34:24 +0000 (10:34 -0700)]
Merge tag 'perf-urgent-2020-08-15' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "Misc fixes, an expansion of perf syscall access to CAP_PERFMON
  privileged tools, plus a RAPL HW-enablement for Intel SPR platforms"

* tag 'perf-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/x86/rapl: Add support for Intel SPR platform
  perf/x86/rapl: Support multiple RAPL unit quirks
  perf/x86/rapl: Fix missing psys sysfs attributes
  hw_breakpoint: Remove unused __register_perf_hw_breakpoint() declaration
  kprobes: Remove show_registers() function prototype
  perf/core: Take over CAP_SYS_PTRACE creds to CAP_PERFMON capability

3 years agoMerge tag 'locking-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 15 Aug 2020 17:32:18 +0000 (10:32 -0700)]
Merge tag 'locking-urgent-2020-08-15' of git://git./linux/kernel/git/tip/tip

Pull locking fixlets from Ingo Molnar:
 "A documentation fix and a 'fallthrough' macro update"

* tag 'locking-urgent-2020-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  futex: Convert to use the preferred 'fallthrough' macro
  Documentation/locking/locktypes: Fix a typo

3 years agoMerge tag '9p-for-5.9-rc1' of git://github.com/martinetd/linux
Linus Torvalds [Sat, 15 Aug 2020 15:34:36 +0000 (08:34 -0700)]
Merge tag '9p-for-5.9-rc1' of git://github.com/martinetd/linux

Pull 9p updates from Dominique Martinet:

 - some code cleanup

 - a couple of static analysis fixes

 - setattr: try to pick a fid associated with the file rather than the
   dentry, which might sometimes matter

* tag '9p-for-5.9-rc1' of git://github.com/martinetd/linux:
  9p: Remove unneeded cast from memory allocation
  9p: remove unused code in 9p
  net/9p: Fix sparse endian warning in trans_fd.c
  9p: Fix memory leak in v9fs_mount
  9p: retrieve fid from file when file instance exist.

3 years agoMerge tag '5.9-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sat, 15 Aug 2020 15:31:39 +0000 (08:31 -0700)]
Merge tag '5.9-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "Three small cifs/smb3 fixes, one for stable fixing mkdir path with
  the 'idsfromsid' mount option"

* tag '5.9-rc-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6:
  SMB3: Fix mkdir when idsfromsid configured on mount
  cifs: Convert to use the fallthrough macro
  cifs: Fix an error pointer dereference in cifs_mount()

3 years agoMerge tag 'nfs-for-5.9-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Sat, 15 Aug 2020 15:26:55 +0000 (08:26 -0700)]
Merge tag 'nfs-for-5.9-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client updates from Trond Myklebust:
 "Stable fixes:
   - pNFS: Don't return layout segments that are being used for I/O
   - pNFS: Don't move layout segments off the active list when being used for I/O

  Features:
   - NFS: Add support for user xattrs through the NFSv4.2 protocol
   - NFS: Allow applications to speed up readdir+statx() using AT_STATX_DONT_SYNC
   - NFSv4.0 allow nconnect for v4.0

  Bugfixes and cleanups:
   - nfs: ensure correct writeback errors are returned on close()
   - nfs: nfs_file_write() should check for writeback errors
   - nfs: Fix getxattr kernel panic and memory overflow
   - NFS: Fix the pNFS/flexfiles mirrored read failover code
   - SUNRPC: dont update timeout value on connection reset
   - freezer: Add unsafe versions of freezable_schedule_timeout_interruptible for NFS
   - sunrpc: destroy rpc_inode_cachep after unregister_filesystem"

* tag 'nfs-for-5.9-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (32 commits)
  NFS: Fix flexfiles read failover
  fs: nfs: delete repeated words in comments
  rpc_pipefs: convert comma to semicolon
  nfs: Fix getxattr kernel panic and memory overflow
  NFS: Don't return layout segments that are in use
  NFS: Don't move layouts to plh_return_segs list while in use
  NFS: Add layout segment info to pnfs read/write/commit tracepoints
  NFS: Add tracepoints for layouterror and layoutstats.
  NFS: Report the stateid + status in trace_nfs4_layoutreturn_on_close()
  SUNRPC dont update timeout value on connection reset
  nfs: nfs_file_write() should check for writeback errors
  nfs: ensure correct writeback errors are returned on close()
  NFSv4.2: xattr cache: get rid of cache discard work queue
  NFS: remove redundant initialization of variable result
  NFSv4.0 allow nconnect for v4.0
  freezer: Add unsafe versions of freezable_schedule_timeout_interruptible for NFS
  sunrpc: destroy rpc_inode_cachep after unregister_filesystem
  NFSv4.2: add client side xattr caching.
  NFSv4.2: hook in the user extended attribute handlers
  NFSv4.2: add the extended attribute proc functions.
  ...

3 years agoMerge tag 'edac_updates_for_5.9_pt2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 15 Aug 2020 15:25:41 +0000 (08:25 -0700)]
Merge tag 'edac_updates_for_5.9_pt2' of git://git./linux/kernel/git/ras/ras

Pull edac fix from Tony Luck:
 "Fix for the ie31200 driver that missed the first pull"

* tag 'edac_updates_for_5.9_pt2' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras:
  EDAC/ie31200: Fallback if host bridge device is already initialized

3 years agoMerge tag 'devicetree-fixes-for-5.9' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 15 Aug 2020 15:19:58 +0000 (08:19 -0700)]
Merge tag 'devicetree-fixes-for-5.9' of git://git./linux/kernel/git/robh/linux

Pull devicetree fixes from Rob Herring:
 "Another round of 'allOf' removals and whitespace clean-ups of schemas"

* tag 'devicetree-fixes-for-5.9' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  dt-bindings: Remove more cases of 'allOf' containing a '$ref'
  dt-bindings: Whitespace clean-ups in schema files

3 years agoMerge tag 'acpi-5.9-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Sat, 15 Aug 2020 15:18:22 +0000 (08:18 -0700)]
Merge tag 'acpi-5.9-rc1-2' of git://git./linux/kernel/git/rafael/linux-pm

Pull more ACPI updates from Rafael Wysocki:
 "Add new hardware support to the ACPI driver for AMD SoCs, the x86 clk
  driver and the Designware i2c driver (changes from Akshu Agrawal and
  Pu Wen)"

* tag 'acpi-5.9-rc1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  clk: x86: Support RV architecture
  ACPI: APD: Add a fmw property is_raven
  clk: x86: Change name from ST to FCH
  ACPI: APD: Change name from ST to FCH
  i2c: designware: Add device HID for Hygon I2C controller

3 years agoMerge tag 'pm-5.9-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Sat, 15 Aug 2020 15:17:01 +0000 (08:17 -0700)]
Merge tag 'pm-5.9-rc1-3' of git://git./linux/kernel/git/rafael/linux-pm

Pull one more power management update from Rafael Wysocki:
 "Modify the intel_pstate driver to allow it to work in the passive mode
  with hardware-managed P-states (HWP) enabled"

* tag 'pm-5.9-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: intel_pstate: Implement passive mode with HWP enabled

3 years agoMerge tag 'mfd-next-5.9-1' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd
Linus Torvalds [Sat, 15 Aug 2020 15:09:38 +0000 (08:09 -0700)]
Merge tag 'mfd-next-5.9-1' of git://git./linux/kernel/git/lee/mfd

Pull MFD updates from Lee Jones:
 "Core Frameworks
   - Make better attempt at matching device with the correct OF node
   - Allow batch removal of hierarchical sub-devices

  New Drivers
   - Add STM32 Clocksource driver
   - Add support for Khadas System Control Microcontroller

  Driver Removal
   - Remove unused driver for TI's SMSC ECE1099

  New Device Support
   - Add support for Intel Emmitsburg PCH to Intel LPSS PCI
   - Add support for Intel Tiger Lake PCH-H to Intel LPSS PCI
   - Add support for Dialog DA revision to Dialog DA9063

  New Functionality
   - Add support for AXP803 to be probed by I2C

  Fix-ups
   - Numerous W=1 warning fixes
   - Device Tree changes (stm32-lptimer, gateworks-gsc, khadas,mcu, stmfx, cros-ec, j721e-system-controller)
   - Enabled Regmap 'fast I/O' in stm32-lptimer
   - Change BUG_ON to WARN_ON in arizona-core
   - Remove superfluous code/initialisation (madera, max14577)
   - Trivial formatting/spelling issues (madera-core, madera-i2c, da9055, max77693-private)
   - Switch to of_platform_populate() in sprd-sc27xx-spi
   - Expand out set/get brightness/pwm macros in lm3533-ctrlbank
   - Disable IRQs on suspend in motorola-cpcap
   - Clean-up error handling in intel_soc_pmic_mrfld
   - Ensure correct removal order of sub-devices in madera
   - Many s/HTTP/HTTPS/ link changes
   - Ensure name used with Regmap is unique in syscon

  Bug Fixes
   - Properly 'put' clock on unbind and error in arizona-core
   - Fix revision handling in da9063
   - Fix 'assignment of read-only location' error in kempld-core
   - Avoid using the Regmap API when atomic in rn5t618
   - Redefine volatile register description in rn5t618
   - Use locking to protect event handler in dln2"

* tag 'mfd-next-5.9-1' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd: (76 commits)
  mfd: syscon: Use a unique name with regmap_config
  mfd: Replace HTTP links with HTTPS ones
  mfd: dln2: Run event handler loop under spinlock
  mfd: madera: Improve handling of regulator unbinding
  mfd: mfd-core: Add mechanism for removal of a subset of children
  mfd: intel_soc_pmic_mrfld: Simplify the return expression of intel_scu_ipc_dev_iowrite8()
  mfd: max14577: Remove redundant initialization of variable current_bits
  mfd: rn5t618: Fix caching of battery related registers
  mfd: max77693-private: Drop a duplicated word
  mfd: da9055: pdata.h: Drop a duplicated word
  mfd: rn5t618: Make restart handler atomic safe
  mfd: kempld-core: Fix 'assignment of read-only location' error
  mfd: axp20x: Allow the AXP803 to be probed by I2C
  mfd: da9063: Add support for latest DA silicon revision
  mfd: da9063: Fix revision handling to correctly select reg tables
  dt-bindings: mfd: st,stmfx: Remove I2C unit name
  dt-bindings: mfd: ti,j721e-system-controller.yaml: Add J721e system controller
  mfd: motorola-cpcap: Disable interrupt for suspend
  mfd: smsc-ece1099: Remove driver
  mfd: core: Add OF_MFD_CELL_REG() helper
  ...

3 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 15 Aug 2020 15:02:03 +0000 (08:02 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge more updates from Andrew Morton:
 "Subsystems affected by this patch series: mm/hotfixes, lz4, exec,
  mailmap, mm/thp, autofs, sysctl, mm/kmemleak, mm/misc and lib"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (35 commits)
  virtio: pci: constify ioreadX() iomem argument (as in generic implementation)
  ntb: intel: constify ioreadX() iomem argument (as in generic implementation)
  rtl818x: constify ioreadX() iomem argument (as in generic implementation)
  iomap: constify ioreadX() iomem argument (as in generic implementation)
  sh: use generic strncpy()
  sh: clkfwk: remove r8/r16/r32
  include/asm-generic/vmlinux.lds.h: align ro_after_init
  mm: annotate a data race in page_zonenum()
  mm/swap.c: annotate data races for lru_rotate_pvecs
  mm/rmap: annotate a data race at tlb_flush_batched
  mm/mempool: fix a data race in mempool_free()
  mm/list_lru: fix a data race in list_lru_count_one
  mm/memcontrol: fix a data race in scan count
  mm/page_counter: fix various data races at memsw
  mm/swapfile: fix and annotate various data races
  mm/filemap.c: fix a data race in filemap_fault()
  mm/swap_state: mark various intentional data races
  mm/page_io: mark various intentional data races
  mm/frontswap: mark various intentional data races
  mm/kmemleak: silence KCSAN splats in checksum
  ...

3 years agovirtio: pci: constify ioreadX() iomem argument (as in generic implementation)
Krzysztof Kozlowski [Sat, 15 Aug 2020 00:32:20 +0000 (17:32 -0700)]
virtio: pci: constify ioreadX() iomem argument (as in generic implementation)

The ioreadX() helpers have inconsistent interface.  On some architectures
void *__iomem address argument is a pointer to const, on some not.

Implementations of ioreadX() do not modify the memory under the address so
they can be converted to a "const" version for const-safety and
consistency among architectures.

Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Allen Hubbe <allenbh@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Jon Mason <jdmason@kudzu.us>
Cc: Kalle Valo <kvalo@codeaurora.org>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Rich Felker <dalias@libc.org>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Link: http://lkml.kernel.org/r/20200709072837.5869-5-krzk@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>