linux-2.6-microblaze.git
2 months agoMAINTAINERS: Update Coly Li's email address
Coly Li [Sat, 13 Jun 2026 15:04:58 +0000 (23:04 +0800)]
MAINTAINERS: Update Coly Li's email address

I switch to colyli@fygo.io as my current email address.

Signed-off-by: Coly Li <colyli@fygo.io>
Link: https://patch.msgid.link/20260613150458.682707-1-colyli@fygo.io
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: check bio split for unaligned bvec
Keith Busch [Fri, 12 Jun 2026 22:32:04 +0000 (15:32 -0700)]
block: check bio split for unaligned bvec

Offsets and lengths need to be validated against the dma alignment. This
check was skipped for sufficiently a small bio with a single bvec, which
may allow an invalid request dispatched to the driver. Force the
validation for an unaligned bvec by forcing the bio split path that
handles this condition.

Fixes: 7eac33186957 ("iomap: simplify direct io validity check")
Fixes: 5ff3f74e145a ("block: simplify direct io validity check")
Reported-by: Carlos Maiolino <cem@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/20260612223205.465913-1-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonbd: Reclassify sockets to avoid lockdep circular dependency
Eric Dumazet [Sat, 13 Jun 2026 04:26:19 +0000 (04:26 +0000)]
nbd: Reclassify sockets to avoid lockdep circular dependency

syzbot reported a possible circular locking dependency in udp_sendmsg()
where fs_reclaim can be triggered while holding sk_lock, and fs_reclaim
can eventually depend on another sk_lock (e.g., if NBD is used for swap
or writeback and NBD uses TLS/TCP which acquires sk_lock).

Since the UDP socket and the NBD TCP/TLS socket are different, this is a
false positive. Fix this by reclassifying NBD sockets to a separate lock
class when they are added to the NBD device.

This is similar to what nvme-tcp and other network block devices do.

Fixes: ffa1e7ada456 ("block: Make request_queue lockdep splats show up earlier")
Reported-by: syzbot+607cdcf978b3e79da878@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/netdev/6a2cdafe.428ffe26.258b27.0161.GAE@google.com/T/#u
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260613042619.1108126-1-edumazet@google.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add configurable error injection
Christoph Hellwig [Thu, 11 Jun 2026 14:06:47 +0000 (16:06 +0200)]
block: add configurable error injection

Add a new block error injection interface that allows to inject specific
status code for specific ranges.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Md Haris Iqbal <haris.iqbal@linux.dev>
Link: https://patch.msgid.link/20260611140703.2401204-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add a str_to_blk_op helper
Christoph Hellwig [Thu, 11 Jun 2026 14:06:46 +0000 (16:06 +0200)]
block: add a str_to_blk_op helper

Add a helper to find the REQ_OP_XYZ constant from the "XYZ" string.
This will be used for the error injection debugfs interface.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Md Haris Iqbal <haris.iqbal@linux.dev>
Link: https://patch.msgid.link/20260611140703.2401204-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add a "tag" for block status codes
Christoph Hellwig [Thu, 11 Jun 2026 14:06:45 +0000 (16:06 +0200)]
block: add a "tag" for block status codes

The full name of the status codes is not good for user interfaces as it
can contain white spaces.  Add the name of the status code without the
BLK_STS_ prefix as a tag so that it can be used for user interfaces.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Md Haris Iqbal <haris.iqbal@linux.dev>
Link: https://patch.msgid.link/20260611140703.2401204-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add a macro to initialize the status table
Christoph Hellwig [Thu, 11 Jun 2026 14:06:44 +0000 (16:06 +0200)]
block: add a macro to initialize the status table

Prepare for adding a new value to the error table by adding a macro
to fill it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Md Haris Iqbal <haris.iqbal@linux.dev>
Link: https://patch.msgid.link/20260611140703.2401204-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agofloppy: Drop unused pnp driver data
Uwe Kleine-König (The Capable Hub) [Wed, 10 Jun 2026 07:27:55 +0000 (09:27 +0200)]
floppy: Drop unused pnp driver data

The pnp_device_id array is only used for module data to support
auto-loading the floppy module. So the .driver_data member is unused and
this assignment can be dropped.

While touching that array, align the coding style to what is used most
for these.

This patch doesn't modify the compiled array, only its representation
in source form benefits. The former was confirmed with x86 and arm64
builds.

Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Reviewed-by: Denis Efremov (Oracle) <efremov@linux.com>
Link: https://patch.msgid.link/99dbf851ffb99229ea1dcfd8f58e9ee6a1f05349.1781075967.git.u.kleine-koenig@baylibre.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: propagate in_flight to whole disk on partition I/O
Tang Yizhou [Tue, 26 May 2026 02:15:55 +0000 (10:15 +0800)]
block: propagate in_flight to whole disk on partition I/O

Now when I/O is submitted to a partition, the per-CPU in_flight[]
counter is incremented only on the partition's block_device, not on the
underlying whole disk. This leads to a problem which can be shown by a
fio test:

lsblk
  NAME     MAJ:MIN RM  SIZE RO TYPE MOUNTPOINTS
  mydev    252:1    0   20G  0 disk
  └─mydev1 259:0    0   10G  0 part

iostat -xp 1
  Device       r/s        rkB/s      ... aqu-sz   %util
  mydev    128153.00  512612.00      ...  13.22   72.20
  mydev1   128154.00  512616.00      ...  13.22  100.00

%util is different between mydev and mydev1, which is unexpected.

This is the cumulative effect of a series of patches. The root cause is
commit e016b78201a2 ("block: return just one value from part_in_flight"),
which deleted the branch in part_in_flight() that aggregated the whole-disk
in_flight count on top of the partition's. Then the second commit is
commit 10ec5e86f9b8 ("block: merge part_{inc,dev}_in_flight into their
only callers"), which folded the whole-disk in_flight accounting into
generic_start_io_acct() and generic_end_io_acct(). Those two helpers
were then removed by commit e722fff238bb ("block: remove
generic_{start,end}_io_acct"), and from that point on the whole disk's
in_flight is no longer accounted at all.

In update_io_ticks(), if calling bdev_count_inflight() finds that the
inflight value of the whole device is 0, the accumulation of io_ticks will
be skipped, causing the reported util% value to be underestimated.

Fix it by restoring the whole-disk in_flight accounting.

Fixes: e016b78201a2 ("block: return just one value from part_in_flight")
Suggested-by: Leon Hwang <leon.huangfu@shopee.com>
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: Tang Yizhou <yizhou.tang@shopee.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260526021555.359500-1-yizhou.tang@shopee.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agovirtio-blk: clamp zone report to the report buffer capacity
Michael Bommarito [Sun, 7 Jun 2026 12:48:34 +0000 (08:48 -0400)]
virtio-blk: clamp zone report to the report buffer capacity

virtblk_report_zones() trusts the device-reported number of zones when
walking the report buffer:

nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
   nr_zones);
...
for (i = 0; i < nz && zone_idx < nr_zones; i++) {
ret = virtblk_parse_zone(vblk, &report->zones[i], ...);

The buffer is allocated by virtblk_alloc_report_buffer(), whose size is
capped by the queue's max hardware sectors and max segments and can
therefore hold fewer descriptors than nr_zones. nz is bounded only by
the device-supplied report->nr_zones and the requested nr_zones, never
by the buffer's descriptor capacity. At probe time the request count is
unbounded (blk_revalidate_disk_zones() calls report_zones() with
nr_zones == UINT_MAX), so the device-supplied report->nr_zones is the
sole gate: a device that reports more zones than fit in the buffer
drives the loop to read report->zones[i] past the end of the allocation.

A malicious or buggy virtio-blk device that reports an inflated nr_zones
triggers this during zone revalidation at probe. KASAN reports a
vmalloc-out-of-bounds read in virtblk_report_zones() against the report
buffer allocated a few lines earlier.

Clamp nz to the number of descriptors that actually fit in the report
buffer.

Fixes: 95bfec41bd3d ("virtio-blk: add support for zoned block devices")
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Link: https://patch.msgid.link/20260607124834.3059944-1-michael.bommarito@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: optimize I/O merge hot path with unlikely() hints
Steven Feng [Sat, 6 Jun 2026 02:42:18 +0000 (10:42 +0800)]
block: optimize I/O merge hot path with unlikely() hints

Remove redundant '== false' comparisons and add unlikely() branch
prediction hints in block I/O merge path functions.

These functions (ll_new_hw_segment, ll_merge_requests_fn, and
blk_rq_merge_ok) are executed on every I/O request merge attempt,
making them critical hot paths. Data integrity check failures are
rare events, so marking these conditions as unlikely() helps the
CPU optimize the common case by improving branch prediction.

Changes:
- Replace 'func() == false' with 'unlikely(!func())' for better
  code style and branch prediction

This micro-optimization reduces branch misprediction penalties in
high-frequency I/O merge paths.

Signed-off-by: Steven Feng <steven@joint-cloud.com>
Link: https://patch.msgid.link/tencent_79B652BD0CC23E093F27914380F161E7E505@qq.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agodrivers/block/rbd: Use strscpy() to copy strings into arrays
David Laight [Sat, 6 Jun 2026 20:27:44 +0000 (21:27 +0100)]
drivers/block/rbd: Use strscpy() to copy strings into arrays

Replacing strcpy() with strscpy() ensures than overflow of the target
buffer cannot happen.

Signed-off-by: David Laight <david.laight.linux@gmail.com>
Reviewed-by: Alex Elder <elder@riscstar.com>
Link: https://patch.msgid.link/20260606202744.5113-5-david.laight.linux@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agopartitions: aix: bound the pp_count scan to the ppe array
Bryam Vargas [Sun, 7 Jun 2026 06:41:43 +0000 (06:41 +0000)]
partitions: aix: bound the pp_count scan to the ppe array

aix_partition() reads the physical volume descriptor into a fixed-size
struct pvd and then scans its physical-partition-extent array:

int numpps = be16_to_cpu(pvd->pp_count);
...
for (i = 0; i < numpps; i += 1) {
struct ppe *p = pvd->ppe + i;
...
lp_ix = be16_to_cpu(p->lp_ix);

pvd points at a single kmalloc()'d struct pvd whose ppe[] member holds a
fixed ARRAY_SIZE(pvd->ppe) (1016) entries, but the loop runs up to the
on-disk pp_count.  pp_count is an unvalidated __be16 read straight from
the descriptor, so a crafted AIX image with pp_count larger than 1016
drives the loop to read pvd->ppe[i] past the end of the allocation (up
to 65535 entries, ~2 MB out of bounds).

The partition scan runs without mounting anything, when a block device
with a crafted AIX/IBM partition table appears (an attacker-supplied
image attached with losetup -P, or a device auto-scanned by udev), via
msdos_partition() -> aix_partition().

Clamp the scan to the number of entries the ppe[] array can hold.

Fixes: 6ceea22bbbc8 ("partitions: add aix lvm partition support files")
Cc: stable@vger.kernel.org
Signed-off-by: Bryam Vargas <hexlabsecurity@proton.me>
Acked-by: Philippe De Muyter <phdm@macqel.be>
Link: https://patch.msgid.link/20260607064137.302574-1-hexlabsecurity@proton.me
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: Enable lock context analysis
Bart Van Assche [Fri, 5 Jun 2026 18:01:07 +0000 (11:01 -0700)]
block: Enable lock context analysis

Now that all block/*.c files have been annotated, enable lock context
analysis for all these source files.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/e248ca3aeead238bbc489cf3afdafcbff9e41faf.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/mq-deadline: Make the lock context annotations compatible with Clang
Bart Van Assche [Fri, 5 Jun 2026 18:01:06 +0000 (11:01 -0700)]
block/mq-deadline: Make the lock context annotations compatible with Clang

While sparse ignores the __acquires() and __releases() arguments, Clang
verifies these. Make the arguments of __acquires() and __releases()
acceptable for Clang.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/3b6e336ced91e27213608ffce205ccd24f4ba285.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/Kyber: Make the lock context annotations compatible with Clang
Bart Van Assche [Fri, 5 Jun 2026 18:01:05 +0000 (11:01 -0700)]
block/Kyber: Make the lock context annotations compatible with Clang

While sparse ignores the __acquires() and __releases() arguments, Clang
verifies these. Make the arguments of __acquires() and __releases()
acceptable for Clang.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/91cb8c790fc8b26b8aa742569fbf8c2c1d099dac.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/blk-mq-debugfs: Improve lock context annotations
Bart Van Assche [Fri, 5 Jun 2026 18:01:04 +0000 (11:01 -0700)]
block/blk-mq-debugfs: Improve lock context annotations

Make the existing lock context annotations compatible with Clang. Add
the lock context annotations that are missing.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/f58fe220ff98f9dfddfed4573f40005c773b7fb7.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/blk-iocost: Inline iocg_lock() and iocg_unlock()
Bart Van Assche [Fri, 5 Jun 2026 18:01:03 +0000 (11:01 -0700)]
block/blk-iocost: Inline iocg_lock() and iocg_unlock()

Both iocg_lock() and iocg_unlock() use conditional locking. Fold these
functions into their callers such that unlocking becomes unconditional.

Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/f8c9867788957d2e40a32e23c6d9b866e480ad9d.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/blk-iocost: Split ioc_rqos_throttle()
Bart Van Assche [Fri, 5 Jun 2026 18:01:02 +0000 (11:01 -0700)]
block/blk-iocost: Split ioc_rqos_throttle()

Prepare for inlining iocg_lock() and iocg_unlock() by moving the code
between these two calls into a new function. No functionality has been
changed.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/a6d3ed953cef6669d23a80923bf46600733cbdae.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/crypto: Annotate the crypto functions
Bart Van Assche [Fri, 5 Jun 2026 18:01:01 +0000 (11:01 -0700)]
block/crypto: Annotate the crypto functions

Add the lock context annotations required for Clang's thread-safety
analysis.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Cc: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/297b40e43a7f9b7d20e91a6c44b41a69d01f5c63.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/cgroup: Inline blkg_conf_{open,close}_bdev_frozen()
Bart Van Assche [Fri, 5 Jun 2026 18:01:00 +0000 (11:01 -0700)]
block/cgroup: Inline blkg_conf_{open,close}_bdev_frozen()

The blkg_conf_open_bdev_frozen() calling convention is not compatible
with lock context annotations. Fold both blkg_conf_open_bdev_frozen()
and blkg_conf_close_bdev_frozen() into their only caller. This patch
prepares for enabling lock context analysis.

The type of 'memflags' has been changed from unsigned long into unsigned
int to match the type of current->flags. See also <linux/sched.h>.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/05661d1555decc6dd5389174ba448d803b72ed9a.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/blk-iocost: Combine two error paths in ioc_qos_write()
Bart Van Assche [Fri, 5 Jun 2026 18:00:59 +0000 (11:00 -0700)]
block/blk-iocost: Combine two error paths in ioc_qos_write()

Reduce code duplication by combining two error paths. No functionality
has been changed.

Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/80d4fc1ecd5eaf187c0a31c63a1033a7326d4c7e.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/cgroup: Improve lock context annotations
Bart Van Assche [Fri, 5 Jun 2026 18:00:58 +0000 (11:00 -0700)]
block/cgroup: Improve lock context annotations

Add lock context annotations where these are missing. Move the
blkg_conf_prep() annotation into block/blk-cgroup.h to make it visible
to all blkg_conf_prep() callers.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/58ddd6e2b960bdfa03d0007984386bc0ba351391.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/cgroup: Split blkg_conf_exit()
Bart Van Assche [Fri, 5 Jun 2026 18:00:57 +0000 (11:00 -0700)]
block/cgroup: Split blkg_conf_exit()

Split blkg_conf_exit() into blkg_conf_unprep() and blkg_conf_close_bdev()
because blkg_conf_exit() is not compatible with the Clang thread-safety
annotations. Remove blkg_conf_exit(). Rename blkg_conf_exit_frozen() into
blkg_conf_close_bdev_frozen(). Add thread-safety annotations to the new
functions.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/c1ec1f1c4b675bc5f187f77b3e6436234c6b244c.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/cgroup: Split blkg_conf_prep()
Bart Van Assche [Fri, 5 Jun 2026 18:00:56 +0000 (11:00 -0700)]
block/cgroup: Split blkg_conf_prep()

Move the blkg_conf_open_bdev() call out of blkg_conf_prep() to make it
possible to add lock context annotations to blkg_conf_prep(). Change an
if-statement in blkg_conf_open_bdev() into a WARN_ON_ONCE() call. Export
blkg_conf_open_bdev() because it is called by the BFQ I/O scheduler and
the BFQ I/O scheduler may be built as a kernel module.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/e6ea0387f413217c8561a0ca54ce7b846aa5c7c5.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/bdev: Annotate the blk_holder_ops callback functions
Bart Van Assche [Fri, 5 Jun 2026 18:00:55 +0000 (11:00 -0700)]
block/bdev: Annotate the blk_holder_ops callback functions

The four callback functions in blk_holder_ops all release the
bd_holder_lock. Annotate these functions accordingly.

Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/be51cf81110f691ebd5868ac2f15ceb847805bc8.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: Annotate the queue limits functions
Bart Van Assche [Fri, 5 Jun 2026 18:00:54 +0000 (11:00 -0700)]
block: Annotate the queue limits functions

Let the thread-safety checker verify whether every start of a queue
limits update is followed by a call to a function that finishes a queue
limits update.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Link: https://patch.msgid.link/8f71062b6d0fcf2b80bc8cda701c453224755439.1780682325.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: Add WQ_PERCPU to alloc_workqueue users
Marco Crivellari [Thu, 4 Jun 2026 10:53:47 +0000 (12:53 +0200)]
block: Add WQ_PERCPU to alloc_workqueue users

This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:

   commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
   commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")

The refactoring is going to alter the default behavior of
alloc_workqueue() to be unbound by default.

With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU. For more details see the Link tag below.

In order to keep alloc_workqueue() behavior identical, explicitly request
WQ_PERCPU.

Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@linutronix.de/
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://patch.msgid.link/20260604105347.168322-1-marco.crivellari@suse.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoMerge tag 'nvme-7.2-2026-06-04' of git://git.infradead.org/nvme into for-7.2/block
Jens Axboe [Fri, 5 Jun 2026 11:18:58 +0000 (05:18 -0600)]
Merge tag 'nvme-7.2-2026-06-04' of git://git.infradead.org/nvme into for-7.2/block

Pull NVMe updates from Keith:

"- Per-controller timeouts
 - Multipath telemetry
 - Namespace format validation
 - Various other fixes"

* tag 'nvme-7.2-2026-06-04' of git://git.infradead.org/nvme: (34 commits)
  nvme: export controller reconnect event count via sysfs
  nvme: export controller reset event count via sysfs
  nvme: export I/O failure count when no path is available via sysfs
  nvme: export I/O requeue count when no path is usable via sysfs
  nvme: export command error counters via sysfs
  nvme: export multipath failover count via sysfs
  nvme: export command retry count via sysfs
  nvme: add diag attribute group under sysfs
  nvme-tcp: lockdep: use dynamic lockdep keys per socket instance
  nvme-tcp: move nvme_tcp_reclassify_socket()
  nvme: validate FDP configuration descriptor sizes
  nvmet-auth: validate reply message payload bounds against transfer length
  nvme: refresh multipath head zoned limits from path limits
  nvme: fix FDP fdpcidx bounds check
  nvme-tcp: Use WQ_PERCPU explicitly if wq_unbound is false.
  nvmet: fix pre-auth out-of-bounds heap read in Discovery Get Log Page
  nvme-multipath: set BIO_REMAPPED on bios remapped to per-path namespace disks
  nvme-multipath: require exact iopolicy names for module parameter
  nvme-multipath: pass NS head to nvme_mpath_revalidate_paths()
  nvme-pci: fix out-of-bounds access in nvme_setup_descriptor_pools
  ...

2 months agonvme: export controller reconnect event count via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:55 +0000 (00:06 +0530)]
nvme: export controller reconnect event count via sysfs

When an NVMe-oF link goes down, the driver attempts to recover the
connection by repeatedly reconnecting to the remote controller at
configured intervals. A maximum number of reconnect attempts is also
configured, after which recovery stops and the controller is removed
if the connection cannot be re-established.

The driver maintains a counter, nr_reconnects, which is incremented on
each reconnect attempt. However if in case the reconnect is successful
then this counter reset to zero. Moreover, currently, this counter is
only reported via kernel log messages and is not exposed to userspace.
Since dmesg is a circular buffer, this information may be lost over
time.

So introduce a new accumulator which accumulates nr_reconnect attempts
and also expose this accumulator per-fabric ctrl via a new sysfs
attribute reconnect_count, under diag attribute grroup to provide
persistent visibility into the number of reconnect attempts made by the
host. This information can help users diagnose unstable links or
connectivity issues. Furthermore, this sysfs attribute is also writable
so user may reset it to zero, if needed.

The reconnect_count can also be consumed by monitoring tools such as
nvme-top to improve controller-level observability.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: export controller reset event count via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:54 +0000 (00:06 +0530)]
nvme: export controller reset event count via sysfs

The NVMe controller transitions into the RESETTING state during error
recovery, link instability, firmware activation, or when a reset is
explicitly triggered by the user.

Expose a per-ctrl sysfs attribute reset_count, under diag attribute
group to provide visibility into these RESETTING state transitions.
Observing the frequency of reset events can help users identify issues
such as PCIe errors or unstable fabric links. This counter is also
writable thus allowing user to reset its value, if needed.

This counter can also be consumed by monitoring tools such as nvme-top
to improve controller-level observability.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: export I/O failure count when no path is available via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:53 +0000 (00:06 +0530)]
nvme: export I/O failure count when no path is available via sysfs

When I/O is submitted to the NVMe namespace head and no available path
can handle the request, the driver fails the I/O immediately. Currently,
such failures are only reported via kernel log messages, which may be
lost over time since dmesg is a circular buffer.

Add a new ns-head sysfs counter io_fail_no_available_path_count, under
diag attribute group to expose the number of I/Os that failed due to the
absence of an available path. This provides persistent visibility into
path-related I/O failures and can help users diagnose the cause of I/O
errors. This counter is also writable and so user may reset its value,
if needed.

This counter can also be consumed by monitoring tools such as nvme-top.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: export I/O requeue count when no path is usable via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:52 +0000 (00:06 +0530)]
nvme: export I/O requeue count when no path is usable via sysfs

When the NVMe namespace head determines that there is no currently
available path to handle I/O (for example, while a controller is
resetting/connecting or due to a transient link failure), incoming
I/Os are added to the requeue list.

Currently, there is no visibility into how many I/Os have been requeued
in this situation. Add a new ns-head sysfs counter
io_requeue_no_usable_path_count, under diag attribute group to expose
the number of I/Os that were requeued due to the absence of an available
path. This counter is also writable thus allowing user to reset it, if
needed.

This statistic can help users understand I/O slowdowns or stalls caused
by temporary path unavailability, and can be consumed by monitoring
tools such as nvme-top for real-time observability.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: export command error counters via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:51 +0000 (00:06 +0530)]
nvme: export command error counters via sysfs

When an NVMe command completes with an error status, the driver
logs the error to the kernel log. However, these messages may be
lost or overwritten over time since dmesg is a circular buffer.

Expose per-path and ctrl sysfs attribute command_error_count, under
diag attribute group to provide persistent visibility into error
occurrences. This allows users to observe the total number of commands
that have failed on a given path over time, which can be useful for
diagnosing path health and stability.

This attribute is both readable and writable thus allowing user to reset
these counters. These counters can also be consumed by observability
tools such as nvme-top to provide additional insight into NVMe error
behavior.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: export multipath failover count via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:50 +0000 (00:06 +0530)]
nvme: export multipath failover count via sysfs

When an NVMe command completes with a path-specific error, the NVMe
driver may retry the command on an alternate controller or path if one
is available. These failover events indicate that I/O was redirected
away from the original path.

Currently, the number of times requests are failed over to another
available path is not visible to userspace. Exposing this information
can be useful for diagnosing path health and stability.

Export per-path sysfs attribute "multipath_failover_count" under diag
attribute group. This attribute is both readable and writable and thus
allowing user to reset the counter. This counter can be consumed by
monitoring tools such as nvme-top to help identify paths that
consistently trigger failovers under load.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: export command retry count via sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:49 +0000 (00:06 +0530)]
nvme: export command retry count via sysfs

When Advanced Command Retry Enable (ACRE) is configured, a controller
may interrupt command execution and return a completion status
indicating command interrupted with the DNR bit cleared. In this case,
the driver retries the command based on the Command Retry Delay (CRD)
value provided in the completion status.

Currently, these command retries are handled entirely within the NVMe
driver and are not visible to userspace. As a result, there is no
observability into retry behavior, which can be a useful diagnostic
signal.

Expose a per-namespace sysfs attribute command_retries_count, under
diag attribute group to provide visibility into retry activity. This
information can help identify controller-side congestion under load
and enables comparison across paths in multipath setups (for example,
detecting cases where one path experiences significantly more retries
than another under identical workloads).

This exported metric is intended for diagnostics and monitoring tools
such as nvme-top, and does not change command retry behavior. A new
sysfs attribute named "command_retries_count" is added for this purpose.
This attribute is both readable as well as writable. So user could
reset this counter if needed.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: add diag attribute group under sysfs
Nilay Shroff [Sat, 16 May 2026 18:36:48 +0000 (00:06 +0530)]
nvme: add diag attribute group under sysfs

Add a new diag attribute group under:
/sys/class/nvme/<ctrl>/
/sys/block/<nvme-path-dev>/
/sys/block/<ns-head-dev>/

This new sysfs attribute group will be used to organize NVMe diagnostic
and telemetry-related counters under it.

Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-tcp: lockdep: use dynamic lockdep keys per socket instance
Shin'ichiro Kawasaki [Thu, 4 Jun 2026 02:32:08 +0000 (11:32 +0900)]
nvme-tcp: lockdep: use dynamic lockdep keys per socket instance

When NVMe-TCP controller setup and teardown are repeated with lockdep
enabled, lockdep reports false positives WARN for the following locks:

  1) &q->elevator_lock        : IO scheduler change context
  2) &q->q_usage_counter(io)  : SCSI disk probe context
  3) fs_reclaim               : CPU hotplug bring-up context
  4) cpu_hotplug_lock         : socket establishment context
  5) sk_lock-AF_INET-NVME     : MQ sched dispatch context for the socket
  6) set->srcu                : NVMe controller delete context

The lockdep WARN was observed by running blktests test case nvme/005 for
tcp transport on v7.1-rc1 kernel with a patch. Refer to the Link tag for
the details of the WARN.

This is a false positive because lockdep confuses lock 4) (socket
establishment) with lock 5) (socket in use) for different socket
instances. The locks belong to different sockets, but lockdep treats
them as the same due to shared static lockdep keys.

Fix this by using dynamically allocated lockdep keys per socket instance
instead of static keys nvme_tcp_sk_key[] and nvme_tcp_slock_key[]. Add
nvme_tcp_sk_key and nvme_tcp_slock_key fields to struct nvme_tcp_queue
and pass them to sock_lock_init_class_and_name() for proper lockdep
tracking. Change the argument of nvme_tcp_reclassify_socket() from
'struct socket *' to 'struct nvme_tcp_queue *' to pass both the socket
and the keys. Add CONFIG_DEBUG_LOCK_ALLOC guards to nvme_tcp_alloc_queue()
and nvme_tcp_free_queue() to register and unregister the dynamic keys.
Additionally, move nvme_tcp_reclassify_socket() inside these guards since
it's only needed when lockdep is enabled.

Link: https://lore.kernel.org/linux-nvme/afB5syZbUrppgsDQ@shinmob/
Suggested-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-tcp: move nvme_tcp_reclassify_socket()
Shin'ichiro Kawasaki [Sat, 30 May 2026 05:20:44 +0000 (14:20 +0900)]
nvme-tcp: move nvme_tcp_reclassify_socket()

Move nvme_tcp_reclassify_socket() in tcp.c after the struct
nvme_tcp_queue definition. This is preparation for adding a reference
to struct nvme_tcp_queue in the function, which would otherwise cause a
compile failure due to the struct being defined after the function.

Move the entire CONFIG_DEBUG_LOCK_ALLOC block along with the function
to maintain the code organization.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: validate FDP configuration descriptor sizes
liuxixin [Tue, 2 Jun 2026 14:00:01 +0000 (22:00 +0800)]
nvme: validate FDP configuration descriptor sizes

Validate descriptor sizes while walking the FDP configurations log so
dsze == 0 or a descriptor past the log end cannot cause unbounded
iteration or reads past the buffer.

Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: liuxixin <gliuxen@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvmet-auth: validate reply message payload bounds against transfer length
Tianchu Chen [Fri, 29 May 2026 14:18:39 +0000 (14:18 +0000)]
nvmet-auth: validate reply message payload bounds against transfer length

nvmet_auth_reply() accesses the variable-length rval[] array using
attacker-controlled hl (hash length) and dhvlen (DH value length) fields
without verifying they fit within the allocated buffer of tl bytes.

A malicious NVMe-oF initiator can craft a DHCHAP_REPLY message with a
small transfer length but large hl/dhvlen values, causing out-of-bounds
heap reads when the target processes the DH public key (rval + 2*hl) or
performs the host response memcmp.

With DH authentication configured, the OOB pointer is passed directly to
sg_init_one() and read by crypto_kpp_compute_shared_secret(), reaching
up to 526 bytes past the buffer. This is exploitable pre-authentication.

Add bounds validation ensuring sizeof(*data) + 2*hl + dhvlen <= tl before
any access to the variable-length fields.

Discovered by Atuin - Automated Vulnerability Discovery Engine.

Fixes: db1312dd9548 ("nvmet: implement basic In-Band Authentication")
Cc: stable@vger.kernel.org
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Signed-off-by: Tianchu Chen <flynnnchen@tencent.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agon64cart: use strscpy in n64cart_probe
Thorsten Blum [Sun, 17 May 2026 17:26:17 +0000 (19:26 +0200)]
n64cart: use strscpy in n64cart_probe

strcpy() has been deprecated [1] because it performs no bounds checking
on the destination buffer, which can lead to buffer overflows. While the
current code works correctly, replace strcpy() with the safer strscpy()
to follow secure coding best practices.

[1] https://www.kernel.org/doc/html/latest/process/deprecated.html#strcpy

Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
Link: https://patch.msgid.link/20260517172617.3954-2-thorsten.blum@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock/partitions/acorn: use min in {riscix,linux}_partition
Thorsten Blum [Tue, 2 Jun 2026 16:07:57 +0000 (18:07 +0200)]
block/partitions/acorn: use min in {riscix,linux}_partition

Use min() to replace the open-coded implementations and to simplify
riscix_partition() and linux_partition().

Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
Reviewed-by: Kees Cook <kees@kernel.org>
Link: https://patch.msgid.link/20260602160757.973736-3-thorsten.blum@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock, bfq: release cgroup stats with bfq_group
Yu Kuai [Mon, 1 Jun 2026 06:15:02 +0000 (14:15 +0800)]
block, bfq: release cgroup stats with bfq_group

BFQ cgroup stats contain percpu counters embedded in struct bfq_group,
but the old free path destroys them from bfq_pd_free(), which is tied
to blkg policy-data teardown.

That is not the same lifetime as struct bfq_group. BFQ pins bfq_group
while bfq_queue entities refer to it, so bfq_pd_free() can drop the
policy-data reference while other bfq_group references still exist. The
following blkcg change also defers policy-data release through RCU and
leaves BFQ to run the final bfqg_put() from an RCU callback. For that
conversion, stats teardown must belong to the last bfq_group put, not to
policy-data teardown.

Move stats teardown to bfqg_put() so the embedded counters are destroyed
exactly when the last bfq_group reference is released, before kfree(bfqg).

Without this preparatory change, the RCU-delayed policy-data free
conversion reproduced the following KASAN report:

  BUG: KASAN: slab-use-after-free in percpu_counter_destroy_many+0xf1/0x2e0
  Write of size 8 at addr ffff88811d9409e0 by task test_blkcg/535

  CPU: 0 UID: 0 PID: 535 Comm: test_blkcg Not tainted 7.1.0-rc2-g1e14adca0199 #1 PREEMPT  ea13f83d4b74a12510d20db4a7d9a0fe8275f05c
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-5.fc42 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x54/0x70
   print_address_description+0x77/0x200
   ? percpu_counter_destroy_many+0xf1/0x2e0
   print_report+0x64/0x70
   kasan_report+0x118/0x150
   ? percpu_counter_destroy_many+0xf1/0x2e0
   percpu_counter_destroy_many+0xf1/0x2e0
   __mmdrop+0x1d8/0x350
   finish_task_switch+0x3f5/0x570
   __schedule+0xe8e/0x18a0
   schedule+0xfe/0x1c0
   schedule_timeout+0x7f/0x1d0
   __wait_for_common+0x26c/0x3f0
   wait_for_completion_state+0x21/0x40
   call_usermodehelper_exec+0x271/0x2c0
   __request_module+0x296/0x410
   elv_iosched_store+0x1bc/0x2c0
   queue_attr_store+0x152/0x1c0
   kernfs_fop_write_iter+0x1d7/0x280
   vfs_write+0x580/0x630
   ksys_write+0xec/0x190
   do_syscall_64+0x156/0x490
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  Allocated by task 535:
   kasan_save_track+0x3e/0x80
   __kasan_kmalloc+0x72/0x90
   bfq_pd_alloc+0x60/0x100 [bfq]
   blkg_create+0x3bb/0xbe0
   blkg_lookup_create+0x3a2/0x460
   blkg_conf_start+0x24a/0x2d0
   bfq_io_set_weight+0x17f/0x430 [bfq]
   cgroup_file_write+0x1c5/0x4b0
   kernfs_fop_write_iter+0x1d7/0x280
   vfs_write+0x580/0x630
   ksys_write+0xec/0x190
   do_syscall_64+0x156/0x490
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  Freed by task 0:
   kasan_save_track+0x3e/0x80
   kasan_save_free_info+0x46/0x50
   __kasan_slab_free+0x3a/0x60
   kfree+0x14e/0x4f0
   rcu_core+0x6f3/0xcd0
   handle_softirqs+0x1a0/0x550
   __irq_exit_rcu+0x8c/0x150
   irq_exit_rcu+0xe/0x20
   sysvec_apic_timer_interrupt+0x6e/0x80
   asm_sysvec_apic_timer_interrupt+0x1a/0x20

  Last potentially related work creation:
   kasan_save_stack+0x3e/0x60
   kasan_record_aux_stack+0x99/0xb0
   call_rcu+0x55/0x5c0
   blkg_free_workfn+0x130/0x220
   process_scheduled_works+0x655/0xb60
   worker_thread+0x446/0x600
   kthread+0x1f4/0x230
   ret_from_fork+0x259/0x420
   ret_from_fork_asm+0x1a/0x30

Signed-off-by: Yu Kuai <yukuai@fygo.io>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260601061502.899552-1-yukuai@fygo.io
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonvme: refresh multipath head zoned limits from path limits
Yao Sang [Thu, 28 May 2026 07:36:01 +0000 (15:36 +0800)]
nvme: refresh multipath head zoned limits from path limits

queue_limits_stack_bdev() updates the multipath head limits from the
path queue, but it does not propagate max_open_zones or
max_active_zones. As a result, a zoned multipath namespace head can
keep stale 0/0 values even after a ready path reports finite zoned
resource limits.

When refreshing the head limits in nvme_update_ns_info(), stack the
zoned resource limits directly after stacking the path queue limits.
Use min_not_zero() so the block layer's 0 value keeps its "no limit"
meaning while finite limits are combined conservatively.

This avoids advertising "no limit" on the multipath head while keeping
the zoned-limit handling local to the NVMe multipath update path.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Yao Sang <sangyao@kylinos.cn>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: fix FDP fdpcidx bounds check
liuxixin [Thu, 28 May 2026 10:00:01 +0000 (18:00 +0800)]
nvme: fix FDP fdpcidx bounds check

The fdpcidx bounds check sets n = NUMFDPC + 1 but used > instead of >=,
incorrectly accepting fdp_idx when it equals n (i.e. NUMFDPC + 1).

Fixes: 30b5f20bb2dd ("nvme: register fdp parameters with the block layer")
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: liuxixin <gliuxen@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-tcp: Use WQ_PERCPU explicitly if wq_unbound is false.
Kuniyuki Iwashima [Sat, 30 May 2026 08:10:57 +0000 (08:10 +0000)]
nvme-tcp: Use WQ_PERCPU explicitly if wq_unbound is false.

Since commit 21c05ca88a54 ("workqueue: Add warnings and ensure
one among WQ_PERCPU or WQ_UNBOUND is present"), we must explicitly
set WQ_PERCPU or WQ_UNBOUND when creating workqueue.

nvme_tcp_init_module() sets WQ_UNBOUND when the module param
wq_unbound is set, but otherwise, WQ_PERCPU is missing, triggering
the warning below:

  workqueue: nvme_tcp_wq is using neither WQ_PERCPU or WQ_UNBOUND. Setting WQ_PERCPU.
  WARNING: kernel/workqueue.c:5856 at __alloc_workqueue+0x1d02/0x2070 kernel/workqueue.c:5855, CPU#0: swapper/0/1

Let's set WQ_PERCPU if wq_unbound is false.

Reported-by: syzbot+d078cba4418e65f61984@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/6a1a9a86.323e8352.141b09.0001.GAE@google.com/
Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvmet: fix pre-auth out-of-bounds heap read in Discovery Get Log Page
Bryam Vargas [Wed, 27 May 2026 20:00:00 +0000 (15:00 -0500)]
nvmet: fix pre-auth out-of-bounds heap read in Discovery Get Log Page

nvmet_execute_disc_get_log_page() validates only the dword alignment
of the host-supplied Log Page Offset (lpo).  The 64-bit offset is then
added to a small kzalloc'd buffer that holds the discovery log page
and the result is passed straight to nvmet_copy_to_sgl(), which
memcpy()s data_len bytes out to the host with no source-side bound
check:

    u64 offset      = nvmet_get_log_page_offset(req->cmd);  /* 64-bit host */
    size_t data_len = nvmet_get_log_page_len(req->cmd);     /* 32-bit host */
    ...
    if (offset & 0x3) { ... }                               /* only check */
    ...
    alloc_len = sizeof(*hdr) + entry_size * discovery_log_entries(req);
    buffer = kzalloc(alloc_len, GFP_KERNEL);
    ...
    status = nvmet_copy_to_sgl(req, 0, buffer + offset, data_len);

The Discovery controller is unauthenticated -- nvmet_host_allowed()
returns true unconditionally for the discovery subsystem -- so the call
is reachable pre-authentication by any TCP/RDMA/FC peer that can reach
the nvmet target.  With a discovery log page of ~1 KiB, an attacker
requesting up to 4 KiB starting at offset == alloc_len reads the next
slab page out and gets its content returned over the fabric (an
empirical run on a default nvmet-tcp loopback target leaked 81
canonical kernel pointers in one Get Log Page response).  Pointing the
offset at unmapped kernel memory faults the in-kernel memcpy and
crashes (or panics, on panic_on_oops=1) the target host instead.

The attacker-controlled source-side offset pattern
"nvmet_copy_to_sgl(req, 0, buffer + ATTACKER_OFFSET, ...)" is unique
to nvmet_execute_disc_get_log_page in the entire nvmet codebase: every
other Get Log Page handler in admin-cmd.c either ignores lpo (and
silently starts every response at offset 0) or tracks a local
destination offset with a fixed source pointer.

Validate the host-supplied offset against the log page size, cap the
copy length to what is actually available, and zero-fill any remainder
of the host transfer buffer.  The zero-fill matches the existing
short-response pattern in nvmet_execute_get_log_changed_ns()
(admin-cmd.c) and prevents leaking transport SGL contents when the
host asks for more bytes than the log page contains.

Fixes: a07b4970f464 ("nvmet: add a generic NVMe target")
Cc: stable@vger.kernel.org
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bryam Vargas <hexlabsecurity@proton.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-multipath: set BIO_REMAPPED on bios remapped to per-path namespace disks
Achkinazi, Igor [Thu, 28 May 2026 15:24:27 +0000 (15:24 +0000)]
nvme-multipath: set BIO_REMAPPED on bios remapped to per-path namespace disks

When nvme_ns_head_submit_bio() remaps a bio from the multipath head to a
per-path namespace, bio_set_dev() clears BIO_REMAPPED.  The remapped bio
is then resubmitted through submit_bio_noacct() which calls
bio_check_eod() because BIO_REMAPPED is not set.

This races with nvme_ns_remove() which zeroes the per-path capacity
before synchronize_srcu():

  CPU 0 (IO submission)
  ---------------------
  srcu_read_lock()
  nvme_find_path() -> ns
    [NVME_NS_READY is set]

  CPU 1 (namespace removal)
  -------------------------
  clear_bit(NVME_NS_READY)
  set_capacity(ns->disk, 0)
  synchronize_srcu()  <- blocks

  CPU 0 (IO submission)
  ---------------------
  bio_set_dev(bio, ns->disk->part0)
    [clears BIO_REMAPPED]
  submit_bio_noacct(bio)
    -> bio_check_eod() sees capacity=0
    -> bio fails with IO error

The SRCU read lock prevents synchronize_srcu() from completing, but does
not prevent set_capacity(0) from executing.  The bio fails the EOD check
before it reaches the NVMe driver, so nvme_failover_req() never gets a
chance to redirect it to another path of multipath.  IO errors are
reported to the application despite another path being available.

On older kernels (before commit 0b64682e78f7 "block: skip unnecessary
checks for split bio"), the same race was also reachable through split
remainders resubmitted via submit_bio_noacct().

Fix this by setting BIO_REMAPPED after bio_set_dev() in
nvme_ns_head_submit_bio().  This skips bio_check_eod() on the per-path
device; the EOD check already passed on the multipath head.

NVMe per-path namespace devices are always whole disks (bd_partno=0), so
the blk_partition_remap() skip also gated by BIO_REMAPPED is a no-op.
The flag does not persist across failover and cannot go stale if the
namespace geometry changes between attempts: nvme_failover_req() calls
bio_set_dev() to redirect the bio back to the multipath head, which
clears BIO_REMAPPED.  When nvme_requeue_work() resubmits through
submit_bio_noacct(), bio_check_eod() runs normally against the current
capacity.

Same approach as commit 3a905c37c351 ("block: skip bio_check_eod for
partition-remapped bios").

Fixes: a7c7f7b2b641 ("nvme: use bio_set_dev to assign ->bi_bdev")
Cc: stable@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Igor Achkinazi <igor.achkinazi@dell.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-multipath: require exact iopolicy names for module parameter
liyouhong [Fri, 29 May 2026 08:51:43 +0000 (16:51 +0800)]
nvme-multipath: require exact iopolicy names for module parameter

The iopolicy module parameter uses strncmp prefix matching, so values
like "numax" are accepted as "numa".  The per-subsystem sysfs attribute
already requires an exact match via sysfs_streq().  Parse both through
a shared helper so invalid values are rejected consistently.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: liyouhong <liyouhong@kylinos.cn>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-multipath: pass NS head to nvme_mpath_revalidate_paths()
John Garry [Fri, 29 May 2026 09:52:01 +0000 (09:52 +0000)]
nvme-multipath: pass NS head to nvme_mpath_revalidate_paths()

In nvme_mpath_revalidate_paths(), we are passed a NS pointer and use that
to lookup the NS head and then use that same NS pointer as an iter variable.

It makes more sense pass the NS head and use a local variable for the NS
iter.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: John Garry <john.g.garry@oracle.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agoMerge tag 'md-7.2-20260531' of https://git.kernel.org/pub/scm/linux/kernel/git/mdraid...
Jens Axboe [Mon, 1 Jun 2026 18:52:20 +0000 (12:52 -0600)]
Merge tag 'md-7.2-20260531' of https://git./linux/kernel/git/mdraid/linux into for-7.2/block

Pull MD updates and fixes from Yu Kuai:

"Bug Fixes:
 - Only requeue dm-raid bios when dm is suspending. (Benjamin Marzinski)
 - Reset raid10 read_slot when reusing r10bio for discard. (Chen Cheng)
 - Fix raid1/raid10 deadlock in read error recovery path.
   (Abd-Alrhman Masalkhi)
 - Fix raid1/raid10 error-path detection with md_cloned_bio().
   (Abd-Alrhman Masalkhi)
 - Fix raid1/raid10 bio accounting for split md cloned bios.
   (Abd-Alrhman Masalkhi)
 - Fix raid1 nr_pending leak in REQ_ATOMIC bad-block path.
   (Abd-Alrhman Masalkhi)

 Improvements:
 - Skip redundant raid_disks updates when the value is unchanged.
   (Abd-Alrhman Masalkhi)

 Cleanups:
 - Update MAINTAINERS email addresses. (Yu Kuai, Li Nan)
 - Clean up raid1 read error handling. (Christoph Hellwig)
 - Move the exceed_read_errors condition out of fix_read_error().
   (Christoph Hellwig)
 - Use str_plural() in raid0 dump_zones(). (Thorsten Blum)"

* tag 'md-7.2-20260531' of https://git.kernel.org/pub/scm/linux/kernel/git/mdraid/linux:
  md/raid0: use str_plural helper in dump_zones
  raid1: fix nr_pending leak in REQ_ATOMIC bad-block error path
  md/raid1: move the exceed_read_errors condition out of fix_read_error
  md/raid1: cleanup handle_read_error
  md/raid1,raid10: fix bio accounting for split md cloned bios
  md/raid1,raid10: fix error-path detection with md_cloned_bio()
  md/raid1,raid10: fix deadlock in read error recovery path
  md/raid10: reset read_slot when reusing r10bio for discard
  md: skip redundant raid_disks update when value is unchanged
  dm-raid: only requeue bios when dm is suspending
  MAINTAINERS: Update Li Nan's E-mail address
  MAINTAINERS: update Yu Kuai's email address

2 months agoMAINTAINERS: use new drbd-dev mailing list
Christoph Böhmwalder [Wed, 13 May 2026 06:55:57 +0000 (08:55 +0200)]
MAINTAINERS: use new drbd-dev mailing list

We are migrating from our own infrastructure to lists.linux.dev, so
change the drbd-dev address to point to the new domain.

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://patch.msgid.link/20260513065557.36042-1-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agorbd: check snap_count against RBD_MAX_SNAP_COUNT
Rosen Penev [Sat, 30 May 2026 01:12:55 +0000 (18:12 -0700)]
rbd: check snap_count against RBD_MAX_SNAP_COUNT

snap_count is u32 but the comparison is against a SIZE_MAX-derived value
(~2^61 on 64-bit), which clang flags as always false with
-Wtautological-constant-out-of-range-compare.

The proper check here should be that snap_count does not go over
RBD_MAX_SNAP_COUNT.

Assisted-by: Opencode:Big-pickle
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Alex Elder <elder@riscstar.com>
Link: https://patch.msgid.link/20260530011255.52916-1-rosenp@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agorust: block: fix GenDisk cleanup paths
Haoze Xie [Sat, 30 May 2026 06:11:54 +0000 (14:11 +0800)]
rust: block: fix GenDisk cleanup paths

GenDiskBuilder::build() still has fallible work after
__blk_mq_alloc_disk(), but its error path only recovers the
foreign queue data. That leaks the temporary gendisk and
request_queue until later teardown. If the caller moved the last
Arc<TagSet<T>> into build(), the leaked queue can retain blk-mq
state after the tag set is dropped.

Fix the pre-registration failure path by dropping the temporary
gendisk reference with put_disk() before recovering queue_data,
so disk_release() can tear down the owned queue.

Also pair GenDisk::drop() with put_disk() after del_gendisk().
Once a Rust GenDisk has been added with device_add_disk(),
del_gendisk() only unregisters it; the final gendisk reference
still has to be dropped to complete the release path.

Fixes: 3253aba3408a ("rust: block: introduce `kernel::block::mq` module")
Cc: stable@kernel.org
Reported-by: Yuan Tan <yuantan098@gmail.com>
Reported-by: Xin Liu <bird@lzu.edu.cn>
Reviewed-by: Andreas Hindborg <a.hindborg@kernel.org>
Signed-off-by: Haoze Xie <royenheart@gmail.com>
Signed-off-by: Ren Wei <n05ec@lzu.edu.cn>
Link: https://patch.msgid.link/b70aff9a920cc42110fe5cf454c3099561863519.1780063368.git.royenheart@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agomd/raid0: use str_plural helper in dump_zones
Thorsten Blum [Wed, 27 May 2026 14:19:33 +0000 (16:19 +0200)]
md/raid0: use str_plural helper in dump_zones

Replace the manual ternary "s" pluralization with str_plural() to
simplify the code.

Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
Link: https://patch.msgid.link/20260527141932.1243503-2-thorsten.blum@linux.dev
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agoraid1: fix nr_pending leak in REQ_ATOMIC bad-block error path
Abd-Alrhman Masalkhi [Sat, 30 May 2026 15:14:11 +0000 (15:14 +0000)]
raid1: fix nr_pending leak in REQ_ATOMIC bad-block error path

In raid1_write_request(), each per-mirror loop iteration begins by
incrementing rdev->nr_pending. If a REQ_ATOMIC write encounters a
badblock within the requested range, the code jumps to err_handle
without dropping the reference taken for the current mirror.

err_handle's cleanup loop will only decrements for k < i and
r1_bio->bios[k] is non-NULL. The current slot is therefore skipped,
leaving its nr_pending reference leaked permanently. The reference
prevents the rdev from ever being removed, since raid1_remove_conf()
refuses to remove an rdev with nr_pending > 0.

Fix this by calling rdev_dec_pending() before jumping to err_handle.

Fixes: f2a38abf5f1c ("md/raid1: Atomic write support")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Link: https://patch.msgid.link/20260530151411.4119-1-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd/raid1: move the exceed_read_errors condition out of fix_read_error
Christoph Hellwig [Fri, 29 May 2026 05:43:00 +0000 (07:43 +0200)]
md/raid1: move the exceed_read_errors condition out of fix_read_error

This condition much better fits into the only caller, limiting
fix_read_error to actually fix up data devices after a read error.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260529054308.2720300-3-hch@lst.de
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd/raid1: cleanup handle_read_error
Christoph Hellwig [Fri, 29 May 2026 05:42:59 +0000 (07:42 +0200)]
md/raid1: cleanup handle_read_error

Unwind the main conditional with duplicate conditions and initialize
variables at initialization time where possible.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260529054308.2720300-2-hch@lst.de
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd/raid1,raid10: fix bio accounting for split md cloned bios
Abd-Alrhman Masalkhi [Fri, 1 May 2026 11:46:51 +0000 (13:46 +0200)]
md/raid1,raid10: fix bio accounting for split md cloned bios

Use md_cloned_bio() to control bio accounting instead of relying
on r1bio_existed in raid1 or the io_accounting flag in raid10.

The previous logic does not reliably reflect whether a bio is an
md cloned bio. When a failed bio is split and resubmitted via
bio_submit_split_bioset() on the error path, this can lead to either
double accounting for md cloned bios, or missing accounting for bios
returned from bio_submit_split_bioset()

Fix this by using md_cloned_bio() to detect md cloned bios and
skip accounting accordingly.

Fixes: bb2a9acefaf9 ("md/raid1: switch to use md_account_bio() for io accounting")
Fixes: 820455238366 ("md/raid10: switch to use md_account_bio() for io accounting")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Reviewed-by: Xiao Ni <xiao@kernel.org>
Link: https://patch.msgid.link/20260501114652.590037-4-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd/raid1,raid10: fix error-path detection with md_cloned_bio()
Abd-Alrhman Masalkhi [Fri, 1 May 2026 11:46:50 +0000 (13:46 +0200)]
md/raid1,raid10: fix error-path detection with md_cloned_bio()

Detect the error path using md_cloned_bio() instead of relying
on r1_bio in raid1 or r10_bio->read_slot in raid10, which may be
NULL or -1 after splitting and resubmitting a failed bio.

As a result, the error path may not be recognized and memory
allocations can incorrectly use GFP_NOIO instead of
(GFP_NOIO | __GFP_HIGH), which can lead to a deadlock under
memory pressure.

Fixes: 689389a06ce7 ("md/raid1: simplify handle_read_error().")
Fixes: 545250f24809 ("md/raid10: simplify handle_read_error()")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Reviewed-by: Xiao Ni <xiao@kernel.org>
Link: https://patch.msgid.link/20260501114652.590037-3-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd/raid1,raid10: fix deadlock in read error recovery path
Abd-Alrhman Masalkhi [Fri, 1 May 2026 11:46:49 +0000 (13:46 +0200)]
md/raid1,raid10: fix deadlock in read error recovery path

raid1d and raid10d may resubmit a split md cloned bio while handling
a read error. In this case, resubmitting the bio can lead to a deadlock
if the array is suspended before md_handle_request() acquires an
active_io reference via percpu_ref_tryget_live().

Since the cloned bio already holds an active_io reference,
trying to acquire another reference via percpu_ref_tryget_live()
can lead to a deadlock while the array is suspended.

Fix this by using percpu_ref_get() for md cloned bios.

Fixes: bb2a9acefaf9 ("md/raid1: switch to use md_account_bio() for io accounting")
Fixes: 820455238366 ("md/raid10: switch to use md_account_bio() for io accounting")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Reviewed-by: Xiao Ni <xiao@kernel.org>
Reviewed-by: Yu Kuai <yukuai@fygo.io>
Link: https://patch.msgid.link/20260501114652.590037-2-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd/raid10: reset read_slot when reusing r10bio for discard
Chen Cheng [Fri, 15 May 2026 09:30:19 +0000 (17:30 +0800)]
md/raid10: reset read_slot when reusing r10bio for discard

put_all_bios() always drops devs[i].bio, but it only drops
devs[i].repl_bio when r10_bio->read_slot < 0. If discard reuses an
r10bio that was previously used for a read, read_slot can still be
non-negative, and discard cleanup can skip bio_put() on repl_bio.

Reset read_slot to -1 when preparing an r10bio for discard so the
replacement bio is always released correctly.

Fixes: d30588b2731f ("md/raid10: improve raid10 discard request")
Signed-off-by: Chen Cheng <chencheng@fnnas.com>
Reviewed-by: Xiao Ni <xiao@kernel.org>
Link: https://patch.msgid.link/20260515093019.3436882-1-chencheng@fnnas.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agomd: skip redundant raid_disks update when value is unchanged
Abd-Alrhman Masalkhi [Tue, 28 Apr 2026 13:05:24 +0000 (15:05 +0200)]
md: skip redundant raid_disks update when value is unchanged

Calling update_raid_disks() with the same value as the current one
can trigger unnecessary work. For example, RAID1 will reallocate
resources such as the mempool for r1bio.

Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Link: https://patch.msgid.link/20260428130524.448063-1-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agodm-raid: only requeue bios when dm is suspending
Benjamin Marzinski [Tue, 28 Apr 2026 23:20:10 +0000 (19:20 -0400)]
dm-raid: only requeue bios when dm is suspending

Returning DM_MAPIO_REQUEUE from the target map() function only requeues
the bio during noflush suspends. During regular operations or during
flushing suspends, it fails the bio. Failing the bio during flushing
suspends is the correct behavior here. The bio cannot be handled, and
dm-raid cannot suspend while it is outstanding. But during normal
operations, dm-raid should not push the bio back to dm. Instead, wait
for the reshape to be resumed.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Reviewed-by: Xiao Ni <xiao@kernel.org>
Link: https://patch.msgid.link/20260428232010.2785514-1-bmarzins@redhat.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agoMAINTAINERS: Update Li Nan's E-mail address
Li Nan [Fri, 8 May 2026 09:55:13 +0000 (17:55 +0800)]
MAINTAINERS: Update Li Nan's E-mail address

Change to my new email address on didiglobal.com.

Signed-off-by: Li Nan <magiclinan@didiglobal.com>
Link: https://patch.msgid.link/tencent_8F8173BEDF20E98550D5429DF802F34A7108@qq.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agoMAINTAINERS: update Yu Kuai's email address
Yu Kuai [Wed, 20 May 2026 11:21:48 +0000 (19:21 +0800)]
MAINTAINERS: update Yu Kuai's email address

Update Yu Kuai's maintainer entries to use the new fygo.io address.

Link: https://patch.msgid.link/20260520112627.1264368-1-yukuai@fnnas.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
2 months agoblock: Include bvec.h kernel-doc in the htmldocs
Matthew Wilcox (Oracle) [Thu, 28 May 2026 17:59:04 +0000 (18:59 +0100)]
block: Include bvec.h kernel-doc in the htmldocs

People have gone to the trouble of writing this kernel-doc; the
least we can do is publish it.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: William Kucharski <william.kucharski@linux.dev>
Link: https://patch.msgid.link/20260528175905.1102280-3-willy@infradead.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: Add bvec_folio()
Matthew Wilcox (Oracle) [Thu, 28 May 2026 17:59:03 +0000 (18:59 +0100)]
block: Add bvec_folio()

This is a simple helper which replaces page_folio(bvec->bv_page).
Minor improvement in readability, but the real motivation is to reduce
the number of references to bvec->bv_page so that it can be changed
with less work.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Leon Romanovsky <leon@kernel.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: William Kucharski <william.kucharski@linux.dev>
Link: https://patch.msgid.link/20260528175905.1102280-2-willy@infradead.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonvme: add support multipath passthrough iostats
Keith Busch [Thu, 28 May 2026 01:00:41 +0000 (18:00 -0700)]
nvme: add support multipath passthrough iostats

Don't skip the io accounting for passthrough commands if the user
enabled tracking these.

Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260528010041.1533124-3-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: export passthrough stats enabled
Keith Busch [Thu, 28 May 2026 01:00:40 +0000 (18:00 -0700)]
block: export passthrough stats enabled

A user can enable io accounting for passthrough requests, so export the
helper that checks if the request should be tracked. This will enable
stacking drivers to to report iostats for passthrough workloads. Since
the stacking request_queue may not be the one providing the request, the
API has to add a parameter for the caller to specify which one to check.

Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260528010041.1533124-2-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add a bio_endio_status helper
Christoph Hellwig [Thu, 28 May 2026 08:46:13 +0000 (10:46 +0200)]
block: add a bio_endio_status helper

Add a helper that sets bi_status and call bio_endio() as that is a very
common pattern and convert the core block code over to it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Md Haris Iqbal <haris.iqbal@linux.dev>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Link: https://patch.msgid.link/20260528084632.2505277-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agobvec: make the bvec_iter helpers inline functions
Christoph Hellwig [Wed, 27 May 2026 15:10:22 +0000 (17:10 +0200)]
bvec: make the bvec_iter helpers inline functions

The macros are impossible to follow due to the lack of visual type
information and all the braces.  Replace them with inline helpers to
improve on that.  Because the calling conventions are a bit problematic
with a lot of passing structures by value, all the helpers are marked
as __always_inline so that they are force inlined.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/20260527151043.2349900-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonvme-tcp: cleanup nvme_tcp_init_iter
Christoph Hellwig [Wed, 27 May 2026 15:10:21 +0000 (17:10 +0200)]
nvme-tcp: cleanup nvme_tcp_init_iter

Split the two init cases based on code in the zloop driver.  This
simplifies the code and makes it easier to follow.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/20260527151043.2349900-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoloop: cleanup lo_rw_aio
Christoph Hellwig [Wed, 27 May 2026 15:10:20 +0000 (17:10 +0200)]
loop: cleanup lo_rw_aio

Port over the changes from the zloop driver to remove the need for
the local bio, bvec and offset variables and clean up the code by
that.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/20260527151043.2349900-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: mark biovec_init_pool static
Christoph Hellwig [Wed, 27 May 2026 15:06:46 +0000 (17:06 +0200)]
block: mark biovec_init_pool static

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20260527150646.2349405-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonvme-pci: fix out-of-bounds access in nvme_setup_descriptor_pools
Mateusz Nowicki [Sat, 23 May 2026 08:28:16 +0000 (08:28 +0000)]
nvme-pci: fix out-of-bounds access in nvme_setup_descriptor_pools

nvme_setup_descriptor_pools() indexes dev->descriptor_pools[] using the
numa_node forwarded from hctx->numa_node by its single caller,
nvme_init_hctx_common().  On a non-NUMA kernel hctx->numa_node is
NUMA_NO_NODE (-1).  Because the parameter was declared 'unsigned', the
value becomes UINT_MAX and the index walks off the array (sized to
nr_node_ids), faulting during nvme_alloc_ns() and leaving the namespace
without a /dev node.

Reproduces on any NVMe controller probed by a CONFIG_NUMA=n kernel:

  BUG: unable to handle page fault for address: ffff889101603d38
  RIP: 0010:nvme_init_hctx_common+0x5a/0x190 [nvme]
  Call Trace:
   nvme_init_hctx+0x10/0x20 [nvme]
   nvme_alloc_ns+0x9e/0xa10 [nvme_core]
   nvme_scan_ns+0x301/0x3b0 [nvme_core]
   nvme_scan_ns_async+0x23/0x30 [nvme_core]

Switch the parameter to int and fall back to node 0 when it is
NUMA_NO_NODE; node 0 is always present.

Fixes: d977506f8863 ("nvme-pci: make PRP list DMA pools per-NUMA-node")
Link: https://lore.kernel.org/r/20260309062840.2937858-2-iam@sung-woo.kim
Reported-by: Sung-woo Kim <iam@sung-woo.kim>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Mateusz Nowicki <mateusz.nowicki@posteo.net>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme: target: rdma: fix ndev refcount leak on queue connect
Wentao Liang [Wed, 27 May 2026 08:45:44 +0000 (08:45 +0000)]
nvme: target: rdma: fix ndev refcount leak on queue connect

nvmet_rdma_queue_connect() calls nvmet_rdma_find_get_device() which
acquires a reference on the returned ndev via kref_get(). On the path
where the host queue backlog is exceeded and the function returns
NVME_SC_CONNECT_CTRL_BUSY, reference of ndev is not released, leaking
the kref.

Fix this by adding a goto to the existing put_device label before the
early return.

Fixes: 31deaeb11ba7 ("nvmet-rdma: avoid circular locking dependency on install_queue()")
Cc: stable@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Wentao Liang <vulab@iscas.ac.cn>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-multipath: fix flex array size in struct nvme_ns_head
Nilay Shroff [Wed, 27 May 2026 06:20:00 +0000 (11:50 +0530)]
nvme-multipath: fix flex array size in struct nvme_ns_head

struct nvme_ns_head contains a flexible array member, current_path[],
which is indexed using the NUMA node ID:
head->current_path[numa_node_id()]

The structure is currently allocated as:
size = sizeof(struct nvme_ns_head) +
       (num_possible_nodes() * sizeof(struct nvme_ns *));
head = kzalloc(size, GFP_KERNEL);

This allocation assumes that NUMA node IDs are sequential and densely
packed from 0 .. num_possible_nodes() - 1. While this assumption holds
on many systems, it is not always true on some architectures such as
powerpc.

On some powerpc systems, NUMA node IDs can be sparse. For example:
NUMA:
  NUMA node(s):              6
  NUMA node0 CPU(s):         80-159
  NUMA node8 CPU(s):         0-79
  NUMA node252 CPU(s):
  NUMA node253 CPU(s):
  NUMA node254 CPU(s):
  NUMA node255 CPU(s):

That is, the possible/online NUMA node IDs are: 0, 8, 252, 253, 254, 255
In this case: num_possible_nodes() = 6

So memory is allocated for only 6 entries in current_path[]. However,
the array is later indexed using the actual NUMA node ID. As a result,
accesses such as:
head->current_path[8] or
head->current_path[252]
goes out of bounds, leading to the following KASAN splat:

==================================================================
BUG: KASAN: slab-out-of-bounds in nvme_mpath_revalidate_paths+0x22c/0x290 [nvme_core]
Write of size 8 at addr c00020003bda35b8 by task kworker/u641:2/1997

CPU: 1 UID: 0 PID: 1997 Comm: kworker/u641:2 Not tainted 7.1.0-rc5-dirty #14 PREEMPT(lazy)
Hardware name: 8335-GTH POWER9 0x4e1202 opal:skiboot-v6.5.3-35-g1851b2a06 PowerNV
Workqueue: async async_run_entry_fn
Call Trace:
[c000200037fa7510] [c0000000021c23d4] dump_stack_lvl+0x88/0xdc (unreliable)
[c000200037fa7540] [c0000000009fda90] print_report+0x22c/0x67c
[c000200037fa7630] [c0000000009fd508] kasan_report+0x108/0x220
[c000200037fa7740] [c0000000009fff48] __asan_store8+0xe8/0x120
[c000200037fa7760] [c008000018e76474] nvme_mpath_revalidate_paths+0x22c/0x290 [nvme_core]
[c000200037fa7800] [c008000018e6556c] nvme_update_ns_info+0x4a4/0x5e0 [nvme_core]
[c000200037fa7a50] [c008000018e66270] nvme_alloc_ns+0x6d8/0x1a70 [nvme_core]
[c000200037fa7c20] [c008000018e679fc] nvme_scan_ns+0x3f4/0x630 [nvme_core]
[c000200037fa7d10] [c00000000031f22c] async_run_entry_fn+0x9c/0x3a0
[c000200037fa7db0] [c0000000002fa544] process_one_work+0x414/0xa10
[c000200037fa7ec0] [c0000000002fbf00] worker_thread+0x320/0x640
[c000200037fa7f80] [c00000000030d0f8] kthread+0x278/0x290
[c000200037fa7fe0] [c00000000000ded8] start_kernel_thread+0x14/0x18

Allocated by task 1997 on cpu 1 at 35.928317s:

The buggy address belongs to the object at c00020003bda3000
 which belongs to the cache kmalloc-rnd-15-2k of size 2048
The buggy address is located 16 bytes to the right of
 allocated 1448-byte region [c00020003bda3000c00020003bda35a8)

The buggy address belongs to the physical page:

Memory state around the buggy address:
 c00020003bda3480: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
 c00020003bda3500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>c00020003bda3580: 00 00 00 00 00 fc fc fc fc fc fc fc fc fc fc fc
                                        ^
 c00020003bda3600: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 c00020003bda3680: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
==================================================================

Fix this by allocating the flexible array using nr_node_ids instead
of num_possible_nodes(). Since nr_node_ids represents the maximum
possible NUMA node IDs, indexing current_path[] using numa_node_id()
becomes safe even on systems with sparse node IDs.

Fixes: f333444708f8 ("nvme: take node locality into account when selecting a path")
Tested-by: Mukesh Kumar Chaurasiya (IBM) <mkchauras@gmail.com>
Reviewed-by: Mukesh Kumar Chaurasiya (IBM) <mkchauras@gmail.com>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agoublk: set canceling flag even when disk is not allocated
Ming Lei [Wed, 27 May 2026 14:40:42 +0000 (09:40 -0500)]
ublk: set canceling flag even when disk is not allocated

ublk_start_cancel() previously bailed out early when ublk_get_disk()
returned NULL, treating it as "our disk has been dead".  That is correct
for the post-teardown case, but it also wrongly covers the pre-start
case: ublk_ctrl_start_dev() has not assigned ub->ub_disk yet, while
io_uring is already tearing down the daemon's uring_cmds via
ublk_uring_cmd_cancel_fn().

In that window, the cancel path skips ublk_set_canceling(), so
ubq->canceling stays false, even though ublk_cancel_cmd() goes on to
NULL out every io->cmd.  ublk_ctrl_start_dev() then proceeds to set
ub->ub_disk, call add_disk(), and schedule partition_scan_work.  When
ublk_partition_scan_work() runs bdev_disk_changed() and the resulting
read reaches ublk_queue_rq() -> ublk_queue_cmd(), the ubq->canceling
check passes and the code dereferences the NULL io->cmd:

  BUG: kernel NULL pointer dereference, address: 0000000000000018
  RIP: ublk_queue_cmd drivers/block/ublk_drv.c [inline]
  RIP: ublk_queue_rq+0x73/0x100
  Call Trace:
   blk_mq_dispatch_rq_list+0x1c5/0xca0
   ...
   bdev_disk_changed+0x3d4/0x5e0
   ublk_partition_scan_work+0x89/0xe0
   process_one_work+0x344/0x8a0

Fix it by always setting ub->canceling / ubq->canceling under
cancel_mutex.  When the disk is allocated, keep the existing
quiesce/unquiesce dance so the flag is observed across the
ublk_queue_rq() barrier.  When the disk is not yet allocated, there is
no request_queue and ublk_queue_rq() cannot be running concurrently, so
simply flipping the flag is sufficient: any subsequent I/O - including
the partition scan started by ublk_ctrl_start_dev() - will see
canceling set and be aborted via __ublk_queue_rq_common().

Fixes: 7fc4da6a304b ("ublk: scan partition in async way")
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260527144042.2095194-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonvme: use DEFINE_SIMPLE_SYSFS_GROUP_VISIBLE for multipath_sysfs
John Garry [Wed, 13 May 2026 09:50:30 +0000 (09:50 +0000)]
nvme: use DEFINE_SIMPLE_SYSFS_GROUP_VISIBLE for multipath_sysfs

Use DEFINE_SIMPLE_SYSFS_GROUP_VISIBLE instead of
DEFINE_SYSFS_GROUP_VISIBLE, which means that we can drop
multipath_sysfs_attr_visible().

Incidentally, multipath_sysfs_attr_visible() should have returned a
umode_t.

This idea was suggested by Ben Marzinski elsewhere.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: John Garry <john.g.garry@oracle.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvmet-tcp: check return value of nvmet_tcp_set_queue_sock
Geliang Tang [Tue, 26 May 2026 09:28:05 +0000 (17:28 +0800)]
nvmet-tcp: check return value of nvmet_tcp_set_queue_sock

The return value of nvmet_tcp_set_queue_sock() is currently ignored in
nvmet_tcp_tls_handshake_done(). If it fails (e.g., due to the socket
not being in TCP_ESTABLISHED state), the socket callbacks will not be
properly set, leading to queue and socket leakage.

Fix this by capturing the return value and calling
nvmet_tcp_schedule_release_queue() on failure to ensure proper cleanup.

Fixes: 675b453e0241 ("nvmet-tcp: enable TLS handshake upcall")
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agoblk-throttle: schedule parent dispatch in tg_flush_bios()
Tao Cui [Fri, 22 May 2026 09:15:30 +0000 (17:15 +0800)]
blk-throttle: schedule parent dispatch in tg_flush_bios()

tg_flush_bios() schedules pending_timer on the child tg's own
service_queue, which causes throtl_pending_timer_fn() to dispatch from
the child's pending_tree.  For leaf cgroups this tree is empty, so the
timer fires and exits without dispatching the throttled bio.

The throttled bio sits in the parent's pending_tree with disptime set
to jiffies (THROTL_TG_CANCELING zeroes all dispatch times), but the
parent's timer is never explicitly rescheduled.  The bio only gets
dispatched when the parent timer eventually fires at its previously
scheduled expiry.

Fix by calling throtl_schedule_next_dispatch(sq->parent_sq, true)
instead, matching what tg_set_limit() already does.  This forces the
parent's dispatch cycle to run immediately and flush all canceling
bios without waiting for a stale timer.

For the device deletion path (blk_throtl_cancel_bios), directly
complete throttled bios with EIO via bio_io_error() instead of
dispatching them through the timer -> work -> submission chain.
This avoids a race with the SCSI state machine where bios can reach
the SCSI layer while the device is in SDEV_CANCEL state, causing
ENODEV instead of the expected EIO.

Reported-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Link: https://lore.kernel.org/all/ag2owaQQoigp_fSV@shinmob/
Signed-off-by: Tao Cui <cuitao@kylinos.cn>
Link: https://patch.msgid.link/20260522091530.1901437-1-cuitao@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agorust: block: mq: align init_request numa_node arg with C signature
Andreas Hindborg [Wed, 27 May 2026 09:18:09 +0000 (11:18 +0200)]
rust: block: mq: align init_request numa_node arg with C signature

Commit b040a1a4523d ("block: switch numa_node to int in
blk_mq_hw_ctx and init_request") changed the type of the
`numa_node` argument of `blk_mq_ops::init_request` from
`unsigned int` to `int`. Update the Rust callback signature to
match, so that the function item can be coerced to the C fn
pointer type stored in `blk_mq_ops`.

Without this change the Rust block layer fails to build:

  error[E0308]: mismatched types
     --> rust/kernel/block/mq/operations.rs:274:28
      |
  274 |         init_request: Some(Self::init_request_callback),
      |                       ---- ^^^^^^^^^^^^^^^^^^^^^^^^^^^
      |                       expected fn pointer, found fn item
      |
      = note: expected fn pointer
                `unsafe extern "C" fn(_, _, _, i32) -> _`
                    found fn item
                `unsafe extern "C" fn(_, _, _, u32) -> _ {...}`

The argument is unused on the Rust side, so this is a pure
type-signature change with no functional impact.

Fixes: b040a1a4523d ("block: switch numa_node to int in blk_mq_hw_ctx and init_request")
Signed-off-by: Andreas Hindborg <a.hindborg@kernel.org>
Reviewed-by: Gary Guo <gary@garyguo.net>
Link: https://patch.msgid.link/20260527-block-for-next-2026-05-26-2200-failure-v1-1-4865889e282c@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: partitions: replace __get_free_page() with kmalloc()
Mike Rapoport (Microsoft) [Wed, 27 May 2026 14:33:28 +0000 (17:33 +0300)]
block: partitions: replace __get_free_page() with kmalloc()

check_partition() allocates a buffer to use as backing memory for
seq_buf.

This buffer can be allocated with kmalloc() as there's nothing special
about it to go directly to the page allocator.

kmalloc() provides a better API that does not require ugly casts and
kfree() does not need to know the size of the freed object.

For a single allocation on the cold path the performance difference between
kmalloc() and __get_free_pages() is not measurable as both allocators take
an object/page from a per-CPU list for fast path allocations.

For the slow path the performance is anyway determined by the amount of
reclaim involved rather than by what allocator is used.

Replace use of __get_free_page() with kmalloc() and free_page() with
kfree().

Link: https://lore.kernel.org/all/635405e4-9423-4a25-a6e7-e03c8ea0bcbe@redhat.com
Signed-off-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
Link: https://patch.msgid.link/20260527-block-v2-1-8e06f914c484@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agonvmet-tcp: fix page fragment cache leak in error path
Geliang Tang [Tue, 26 May 2026 09:22:22 +0000 (17:22 +0800)]
nvmet-tcp: fix page fragment cache leak in error path

In nvmet_tcp_alloc_queue(), when a connection is closed during the
allocation process (e.g., nvmet_tcp_set_queue_sock() returns -ENOTCONN),
the error handling jumps to out_destroy_sq and then to out_ida_remove
without draining the page fragment cache.

Although nvmet_tcp_free_cmd() is called in some error paths to release
individual page fragments, the underlying page cache reference held by
queue->pf_cache is never released. The first allocation using pf_cache
is the call to nvmet_tcp_alloc_cmd() for queue->connect, which happens
after ida_alloc() returns successfully. This results in a page leak each
time a connection fails during allocation, which could lead to memory
exhaustion over time if connections are repeatedly opened and closed.

Fix this by calling page_frag_cache_drain() before freeing the queue
structure in the out_ida_remove label.

Fixes: 872d26a391da ("nvmet-tcp: add NVMe over TCP target driver")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-core: fix unsigned comparison warning in nvme_wait_freeze_timeout
Maurizio Lombardi [Thu, 21 May 2026 15:37:16 +0000 (17:37 +0200)]
nvme-core: fix unsigned comparison warning in nvme_wait_freeze_timeout

The timeout variable in nvme_wait_freeze_timeout() is an unsigned type.
Checking if it is <= 0 triggers a compiler warning because an unsigned
variable can never be negative.

Fix this warning by changing the type to long.

Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <error27@gmail.com>
Closes: https://lore.kernel.org/r/202605211257.STzj2Ujv-lkp@intel.com/
Fixes: 23b6d2cbf75f ("nvme: remove redundant timeout argument from nvme_wait_freeze_timeout")
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-multipath: enable PCI P2PDMA for multipath devices
Kiran Kumar Modukuri [Wed, 13 May 2026 18:51:53 +0000 (11:51 -0700)]
nvme-multipath: enable PCI P2PDMA for multipath devices

NVMe multipath does not expose BLK_FEAT_PCI_P2PDMA on the head disk
even when all underlying controllers support it.

Set BLK_FEAT_PCI_P2PDMA unconditionally in nvme_mpath_alloc_disk()
alongside the other features.  nvme_update_ns_info_block() already
calls queue_limits_stack_bdev() to stack each path's limits onto the
head disk, which routes through blk_stack_limits().  The core now
clears BLK_FEAT_PCI_P2PDMA automatically if any path (e.g., FC) does
not support it, consistent with how BLK_FEAT_NOWAIT and BLK_FEAT_POLL
are handled.

Tested-by: Pranjal Shrivastava <praan@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Kiran Kumar Modukuri <kmodukuri@nvidia.com>
Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
Tested=by: Pranjal Shrivastava <praan@google.com>
Link: https://patch.msgid.link/20260513185153.95552-4-kch@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agomd: propagate BLK_FEAT_PCI_P2PDMA from member devices to RAID device
Kiran Kumar Modukuri [Wed, 13 May 2026 18:51:52 +0000 (11:51 -0700)]
md: propagate BLK_FEAT_PCI_P2PDMA from member devices to RAID device

MD RAID does not propagate BLK_FEAT_PCI_P2PDMA from member devices to
the RAID device, preventing peer-to-peer DMA through the RAID layer even
when all underlying devices support it.

Enable BLK_FEAT_PCI_P2PDMA unconditionally in raid0, raid1 and raid10
personalities during queue limits setup.  blk_stack_limits() clears it
automatically if any member device lacks support, consistent with how
BLK_FEAT_NOWAIT and BLK_FEAT_POLL are handled in the block core.

Parity RAID personalities (raid4/5/6) are excluded because they require
CPU access to data pages for parity computation, which is incompatible
with P2P mappings.

Tested with RAID0/1/10 arrays containing multiple NVMe devices with
P2PDMA support, confirming that peer-to-peer transfers work correctly
through the RAID layer.

Tested-by: Pranjal Shrivastava <praan@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Kiran Kumar Modukuri <kmodukuri@nvidia.com>
Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Yu Kuai <yukuai@fygo.io>
Tested=by: Pranjal Shrivastava <praan@google.com>
Link: https://patch.msgid.link/20260513185153.95552-3-kch@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: clear BLK_FEAT_PCI_P2PDMA in blk_stack_limits() for non-supporting devices
Chaitanya Kulkarni [Wed, 13 May 2026 18:51:51 +0000 (11:51 -0700)]
block: clear BLK_FEAT_PCI_P2PDMA in blk_stack_limits() for non-supporting devices

BLK_FEAT_NOWAIT and BLK_FEAT_POLL are cleared in blk_stack_limits()
when an underlying device does not support them.  Apply the same
treatment to BLK_FEAT_PCI_P2PDMA: stacking drivers set it
unconditionally and rely on the core to clear it whenever a
non-supporting member device is stacked.

Tested-by: Pranjal Shrivastava <praan@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
Tested=by: Pranjal Shrivastava <praan@google.com>
Link: https://patch.msgid.link/20260513185153.95552-2-kch@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: remove blkdev_write_begin() and blkdev_write_end()
Tal Zussman [Mon, 25 May 2026 18:25:55 +0000 (14:25 -0400)]
block: remove blkdev_write_begin() and blkdev_write_end()

Remove blkdev_write_begin(), blkdev_write_end(), and their entries in
def_blk_aops. These have been unreachable since commit 487c607df790
("block: use iomap for writes to block devices") switched block device
buffered writes from generic_perform_write() to
iomap_file_buffered_write(), which bypasses aops->write_begin/end.

Signed-off-by: Tal Zussman <tz2294@columbia.edu>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260525-blk-write-cleanup-v1-1-391c073e3831@columbia.edu
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agomtip32xx: fix use-after-free on service thread failure
Yuho Choi [Mon, 25 May 2026 16:25:31 +0000 (12:25 -0400)]
mtip32xx: fix use-after-free on service thread failure

If service thread creation fails after device_add_disk() succeeds,
mtip_block_initialize() calls del_gendisk() and then falls through to
put_disk(). Since mtip32xx uses .free_disk to free struct driver_data,
put_disk() can release dd on the added-disk path.

The same unwind then continues to use dd for blk_mq_free_tag_set() and
mtip_hw_exit(), and mtip_pci_probe() can later free dd again. This can
cause a use-after-free and double free.

Track whether the disk was added in the current initialization call.
For the post-add service-thread failure path, remove the disk, release
the local hardware resources, and return without dropping the final disk
reference. The probe error path can then finish its cleanup and call
put_disk() after it is done using dd. Keep the pre-add path using
put_disk() before blk_mq_free_tag_set(), and clear dd->disk so the outer
probe cleanup frees dd directly.

Fixes: e8b58ef09e84 ("mtip32xx: fix device removal")
Signed-off-by: Yuho Choi <dbgh9129@gmail.com>
Link: https://patch.msgid.link/20260525162531.1406677-1-dbgh9129@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: don't set BIO_QUIET for BLK_STS_AGAIN
Christoph Hellwig [Mon, 18 May 2026 06:33:30 +0000 (08:33 +0200)]
block: don't set BIO_QUIET for BLK_STS_AGAIN

Commit abb30460bda2 ("block: mark bio_wouldblock_error() bio with
BIO_QUIET") added this to suppress buffer_head warnings, but neither
when this commit was added nor now any buffer_head using code actually
ever sets REQ_NOWAIT which can lead to BLK_STS_AGAIN.

Remove the special handling for now.  If we ever plan to use REQ_NOWAIT
for buffer_head based I/O we're better off handling BLK_STS_AGAIN in
the completion handler as it actually needs to retry the I/O as well.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20260518063336.507369-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agodirect-io: remove IOCB_NOWAIT support
Christoph Hellwig [Mon, 18 May 2026 06:33:29 +0000 (08:33 +0200)]
direct-io: remove IOCB_NOWAIT support

None of the file systems using the legacy direct I/O code actually sets
FMODE_NOWAIT, and if they did this would not work, as the write locking
could not handle the retry.  Remove this dead code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christian Brauner <brauner@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://patch.msgid.link/20260518063336.507369-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: Avoid mounting the bdev pseudo-filesystem in userspace
Denis Arefev [Thu, 21 May 2026 07:28:56 +0000 (10:28 +0300)]
block: Avoid mounting the bdev pseudo-filesystem in userspace

The bdev pseudo-filesystem is an internal kernel filesystem with which
userspace should not interfere. Unregister it so that userspace cannot
even attempt to mount it.

This fixes a bug [1] that occurs when attempting to access files,
because the system call move_mount() uses pointers declared in the
inode_operations structure, which for the bdev pseudo-filesystem
are always equal to 0. `inode->i_op = &empty_iops;`

[1]

 BUG: kernel NULL pointer dereference, address: 0000000000000000
 #PF: supervisor instruction fetch in kernel mode
 #PF: error_code(0x0010) - not-present page
 PGD 23380067 P4D 23380067 PUD 23381067 PMD 0
 Oops: 0010 [#1] PREEMPT SMP KASAN NOPTI
 CPU: 2 PID: 17125 Comm: syz-executor.0 Not tainted 6.1.155-syzkaller-00350-g84221fde2681 #0
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
 RIP: 0010:0x0

 Call Trace:
 <TASK>
 lookup_open.isra.0+0x700/0x1180 fs/namei.c:3460
 open_last_lookups fs/namei.c:3550 [inline]
 path_openat+0x953/0x2700 fs/namei.c:3780
 do_filp_open+0x1c5/0x410 fs/namei.c:3810
 do_sys_openat2+0x171/0x4d0 fs/open.c:1318
 do_sys_open fs/open.c:1334 [inline]
 __do_sys_openat fs/open.c:1350 [inline]
 __se_sys_openat fs/open.c:1345 [inline]
 __x64_sys_openat+0x13c/0x1f0 fs/open.c:1345
 do_syscall_x64 arch/x86/entry/common.c:51 [inline]
 do_syscall_64+0x35/0x80 arch/x86/entry/common.c:81
 entry_SYSCALL_64_after_hwframe+0x6e/0xd8

Found by Linux Verification Center (linuxtesting.org) with Syzkaller.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Link: https://lore.kernel.org/all/20131010004732.GJ13318@ZenIV.linux.org.uk/T/#
Cc: stable@vger.kernel.org
Signed-off-by: Denis Arefev <arefev@swemel.ru>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260521072857.5078-1-arefev@swemel.ru
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: switch numa_node to int in blk_mq_hw_ctx and init_request
Mateusz Nowicki [Sat, 23 May 2026 12:52:35 +0000 (12:52 +0000)]
block: switch numa_node to int in blk_mq_hw_ctx and init_request

numa_node in blk_mq_hw_ctx and the matching argument of
blk_mq_ops::init_request can be NUMA_NO_NODE (-1).  Declared as
unsigned int, NUMA_NO_NODE becomes UINT_MAX and walks off
nvme_dev::descriptor_pools[] on CONFIG_NUMA=n [1].

Switch the field and the callback prototype to int and update all
in-tree init_request implementations.  No functional change:
cpu_to_node(), kmalloc_node() and blk_alloc_flush_queue() already
take int.

Link: https://lore.kernel.org/linux-nvme/20260522150628.399288-1-mateusz.nowicki@posteo.net/
Link: https://lore.kernel.org/linux-nvme/20260309062840.2937858-2-iam@sung-woo.kim/
Suggested-by: Caleb Sander Mateos <csander@purestorage.com>
Suggested-by: Sung-woo Kim <iam@sung-woo.kim>
Signed-off-by: Mateusz Nowicki <mateusz.nowicki@posteo.net>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260523125210.272274-1-mateusz.nowicki@posteo.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: skip sync_blockdev() on surprise removal in bdev_mark_dead()
Chao Shi [Fri, 22 May 2026 22:00:25 +0000 (18:00 -0400)]
block: skip sync_blockdev() on surprise removal in bdev_mark_dead()

bdev_mark_dead()'s @surprise == true means the device is already gone.
The filesystem callback fs_bdev_mark_dead() honours this and skips
sync_filesystem(), but the bare block device path (no ->mark_dead op)
lost its !surprise guard when the holder ->mark_dead callback was wired
up (see Fixes), and now calls sync_blockdev() unconditionally, which can
hang forever waiting on writeback that can no longer complete.

syzkaller hit this via nvme_reset_work()'s "I/O queues lost" path:
nvme_mark_namespaces_dead() -> blk_mark_disk_dead() ->
bdev_mark_dead(bdev, true) -> sync_blockdev() blocks in
folio_wait_writeback(), wedging the reset worker and every task waiting
on it.

Skip the sync on surprise removal, matching fs_bdev_mark_dead();
invalidate_bdev() still runs. Orderly removal (surprise == false) is
unchanged.

Found by FuzzNvme(Syzkaller with FEMU fuzzing framework).

Fixes: d8530de5a6e8 ("block: call into the file system for bdev_mark_dead")
Acked-by: Sungwoo Kim <iam@sung-woo.kim>
Acked-by: Dave Tian <daveti@purdue.edu>
Acked-by: Weidong Zhu <weizhu@fiu.edu>
Signed-off-by: Chao Shi <coshi036@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260522220025.1770388-1-coshi036@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblk-mq: add tracepoint block_rq_tag_wait
Aaron Tomlin [Mon, 25 May 2026 00:51:23 +0000 (20:51 -0400)]
blk-mq: add tracepoint block_rq_tag_wait

In high-performance storage environments, particularly when utilising
RAID controllers with shared tag sets (BLK_MQ_F_TAG_HCTX_SHARED), severe
latency spikes can occur when fast devices (SSDs) are starved of hardware
tags when sharing the same blk_mq_tag_set.

Currently, diagnosing this specific hardware queue contention is
difficult. When a CPU thread exhausts the tag pool, blk_mq_get_tag()
forces the current thread to block uninterruptible via io_schedule().
While this can be inferred via sched:sched_switch or dynamically
traced by attaching a kprobe to blk_mq_mark_tag_wait(), there is no
dedicated, out-of-the-box observability for this event.

This patch introduces the block_rq_tag_wait tracepoint in the tag
allocation slow-path. It triggers immediately before the task state
is altered to TASK_UNINTERRUPTIBLE (ensuring safety for PREEMPT_RT
locks). It exposes the exact hardware context (hctx) that is starved,
the specific pool experiencing starvation (driver, software scheduler,
or reserved), and the exact pool depth.

This provides storage engineers with a zero-configuration, low-overhead
mechanism to definitively identify shared-tag bottlenecks. For example,
userspace can trivially replicate tag starvation counters using bpftrace:

    # bpftrace -e 'tracepoint:block:block_rq_tag_wait { @tag_waits[cpu] = count(); }'
    Attaching 1 probe...
    ^C
    @tag_waits[4]: 12
    @tag_waits[12]: 87

Signed-off-by: Aaron Tomlin <atomlin@atomlin.com>
Link: https://patch.msgid.link/20260525005123.722277-1-atomlin@atomlin.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: partitions: fix of_node refcount leak in of_partition()
Wentao Liang [Tue, 26 May 2026 10:21:24 +0000 (10:21 +0000)]
block: partitions: fix of_node refcount leak in of_partition()

of_partition() calls of_node_get() on the parent device node at the
beginning of the function, storing the reference in 'partitions_np'.
This reference is leaked in two paths:

1. The compatibility check at the top of the function returns 0
   without releasing partitions_np when the node exists but is not
   "fixed-partitions" compatible.

2. The function returns 1 at the end after successfully processing
   all partitions without releasing partitions_np.

Fix both leaks by adding of_node_put(partitions_np) on each path.

Fixes: 2e3a191e89f9 ("block: add support for partition table defined in OF")
Cc: stable@vger.kernel.org
Signed-off-by: Wentao Liang <vulab@iscas.ac.cn>
Reviewed-by: Md Haris Iqbal <haris.iqbal@linux.dev>
Link: https://patch.msgid.link/20260526102124.2283846-1-vulab@iscas.ac.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoublk: factor out ublk_init_iod() helper
Caleb Sander Mateos [Wed, 20 May 2026 20:36:54 +0000 (14:36 -0600)]
ublk: factor out ublk_init_iod() helper

The code for initializing struct ublksrv_io_desc on I/O dispatch is
largely duplicated in 3 places. Commit 4d4a512a1f87 ("ublk: add PFN-
based buffer matching in I/O path") added support to ublk_setup_iod()
for matching request buffers against registered UBLK_F_SHMEM_ZC buffers,
but missed adding it to ublk_setup_iod_zoned() for zoned requests. Move
the duplicated logic to a new helper ublk_init_iod(). This way, zone
appends can also benefit from avoiding the data copy.

Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260520203654.1413640-3-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>