We cannot add a socket to an already running nbd device, the reconfigure
for netlink can only active an inactive socket. But for ioctl path, we can
call NBD_SET_SOCK after NBD_DO_IT, reject this using nbd->pid which has
been setted when NBD_DO_IT. Besides, it is the root cause for commit
b98e762e3d ("nbd: freeze the queue while we're adding connections").
Reviewed-by: Yu Kuai <yukuai@fygo.io>
Signed-off-by: Yang Erkun <yangerkun@huawei.com>
Link: https://patch.msgid.link/20260805122930.57647-3-yangerkun@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The second conditional checking nsock->fallback_index validity is the
logical inverse of the first, so drop it and let execution fall through
naturally. Consolidate the two identical dev_err_ratelimited() + return
paths into a single no_fallback label to reduce duplication.
Reviewed-by: Yu Kuai <yukuai@fygo.io>
Signed-off-by: Long Li <leo.lilong@huawei.com>
Link: https://patch.msgid.link/20260805122930.57647-2-yangerkun@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The conversion to explicit netlink serialization dropped the
exclude_sensitive parameter from net_conf_to_skb(), so each caller has
to sanitize by hand. Two dump paths were missed:
drbd_nl_get_connections_dumpit() and the volume-less connection branch
of get_one_status(). Neither op carries GENL_ADMIN_PERM, so any
unprivileged local user could read the CRAM-HMAC secret.
Add a net_conf_to_skb_sanitized() wrapper and route all three callers
through it.
Fixes: 8098eeb693 ("drbd: replace genl_magic with explicit netlink serialization")
Reported-by: Vivek Parikh <vivek.parikh@breachx.ai>
Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://patch.msgid.link/20260814151617.73752-1-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 05c3e88488ed ("srcu: Queue sdp->work when the delay timer is
successfully deleted") added a check in cleanup_srcu_struct() if the
call to srcu_barrier() has been made before calling it, which
revealed a missing call to srcu_barrier() before calling
cleanup_srcu_struct(set->srcu). Fix this.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
Link: https://patch.msgid.link/20260812060510.3220294-1-m.szyprowski@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The EOPNOTSUPP stubs for the sub-cap kfuncs live in ext.c under #ifndef
CONFIG_EXT_SUB_SCHED while the real definitions live in sub.c. Move the
stubs into sub.c so all sub kfunc definitions live in one file. Pure code
move, no functional change.
Signed-off-by: Tejun Heo <tj@kernel.org>
If IORING_URING_CMD_MULTISHOT is set, io_uring_cmd() treats any
non-negative return from ->uring_cmd() as the driver having taken
ownership of the request and returns IOU_ISSUE_SKIP_COMPLETE. But
nothing guarantees that the driver did so, and any handler that just
completes the command inline and returns 0 or a positive result then
leaves the request orphaned, leaking the io_kiocb, the async data,
and the file reference.
The special case isn't needed. ublk returns -EIOCBQUEUED for the
multishot fetch command, which is passed through as-is, and the poll
driven socket timestamp command returns -EAGAIN. Kill it, a multishot
handler that wants to hang on to the request must return -EIOCBQUEUED
or -EAGAIN like any other command.
Fixes: 620a50c927 ("io_uring: uring_cmd: add multishot support")
Cc: stable@vger.kernel.org
Reported-by: syzbot+a4ccdd7ebf452e4d4701@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/6a7a0194.b50370da.49fe0.0056.GAE@google.com/
Link: https://lore.kernel.org/all/20260811115125.1831170-1-vasilisalmpanis@gmail.com/
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_mem_alloc_compound() allocates get_order(size) pages, which rounds a
region size that is not a power of two up to the next order. The pages
past the region are part of the same allocation and cannot be used for
anything else, but io_create_region() accounts reg->size >> PAGE_SHIFT,
so they are never charged against RLIMIT_MEMLOCK. For a ring with 4096
SQ entries and the default CQ size the region is 37 pages while the
allocation is 64.
Account the tail pages together with the region, and fall back to the
exact sized bulk allocation when they do not fit the limit, so a user
close to their limit still gets the region rather than an error.
io_free_region() gives the same amount back, the compound case being the
one that set IO_REGION_F_SINGLE_REF.
Counting how many 4096 entry rings an unprivileged user can create under
a given RLIMIT_MEMLOCK, before and after:
limit (pages) before after
256 2 2
300 2 2
350 3 2
400 3 3
512 5 4
Five rings under a 512 page limit really pin 640 pages.
Fixes: dfbbfbf191 ("io_uring: introduce concept of memory regions")
Link: https://lore.kernel.org/all/87ik5ncj8d.fsf@mailhost.krisman.be/
Signed-off-by: Ali Ahmet Memis <ali@iusegentoo.com>
Link: https://patch.msgid.link/20260806180044.275543-1-ali@iusegentoo.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The ioctl handlers only test REMOVE_PENDING before entering
mtip_hw_ioctl(). Removal can set that bit immediately afterwards and free
dd->port in mtip_hw_exit() while an ioctl still dereferences it. An already
open block device can reach the handlers while del_gendisk() is in
progress.
Serialize both native and compat ioctls with removal. Set REMOVE_PENDING
before taking the mutex so new callers fail after an in-flight ioctl has
drained, and hold the mutex until the port has been torn down.
Fixes: 88523a6155 ("block: Add driver for Micron RealSSD pcie flash cards")
Signed-off-by: Hongyan Xu <getshell@seu.edu.cn>
Link: https://patch.msgid.link/20260806060441.676-1-getshell@seu.edu.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
__ublk_shmem_remove_ranges() removes matching maple tree ranges in
batches, but first stores each range into a temporary xarray so that the
pages can be unpinned after dropping the maple tree lock.
That temporary xarray is filled under the maple tree lock with
xa_store(..., GFP_ATOMIC). If the store fails before mas_erase(), the
current range is left in the tree and the helper returns false. The
outer ublk_shmem_remove_ranges() loop then immediately retries the same
range. While the atomic allocation keeps failing, the teardown path has
no forward progress.
The issue can be reproduced with radix_tree_node failslab injection after
a SHMEM_ZC buffer has already been registered:
# Kernel config:
# CONFIG_BLK_DEV_UBLK=y
# CONFIG_DEBUG_FS=y
# CONFIG_FAULT_INJECTION=y
# CONFIG_FAULT_INJECTION_DEBUG_FS=y
# CONFIG_FAILSLAB=y
echo 10 > /proc/sys/vm/nr_hugepages
mkdir -p /tmp/htlb
mount -t hugetlbfs none /tmp/htlb
fallocate -l 4M /tmp/htlb/ublk_buf
dev_id=$(kublk add -t null --shmem_zc \
--htlb /tmp/htlb/ublk_buf |
awk -F '[ :]' '/dev id/ {print $3}')
echo 1 > /sys/kernel/slab/radix_tree_node/failslab
echo Y > /sys/kernel/debug/failslab/cache-filter
echo Y > /sys/kernel/debug/failslab/ignore-gfp-wait
echo 1 > /sys/kernel/debug/failslab/interval
echo -1 > /sys/kernel/debug/failslab/times
echo 100 > /sys/kernel/debug/failslab/probability
kublk del -n "$dev_id"
On the unfixed kernel the delete command was still running after 3
seconds. Disabling failslab made it return. The fault-injection stack
showed:
should_failslab
kmem_cache_alloc_lru_noprof
__xas_nomem
__xa_store
xa_store
__ublk_shmem_remove_ranges
ublk_cdev_rel
ublk_ctrl_del_dev
Remove the allocation from the teardown loop. Keep the existing batch
limit, but collect {base_pfn, nr_pages} pairs in a fixed-size stack array.
Once a matching range is found, the range is erased from the maple tree
before dropping the lock, so each successful scan makes progress without
depending on any GFP_ATOMIC allocation.
With the same failslab settings, the fixed kernel completed
"kublk del -n $dev_id" successfully in about 45 ms.
Fixes: 309e02dccf ("ublk: avoid unpinning pages under maple tree spinlock")
Signed-off-by: Yao Sang <sangyao@kylinos.cn>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260804125736.2011774-1-sangyao@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
null_zone_no() does sect >> ilog2(dev->zone_size_sects). When
zone_size_sects is 0, ilog2(0) returns -1, producing shift exponent -1
which UBSAN reports as shift-out-of-bounds.
UBSAN: shift-out-of-bounds in drivers/block/null_blk/zoned.c:21:14
shift exponent -1 is negative
Call Trace:
null_zone_no drivers/block/null_blk/zoned.c:21 [inline]
null_process_zoned_cmd+0xf76/0xf80 drivers/block/null_blk/zoned.c:728
null_handle_cmd drivers/block/null_blk/main.c:1455 [inline]
null_queue_rq+0x8bc/0xe70 drivers/block/null_blk/main.c:1703
__blk_mq_issue_directly block/blk-mq.c:2694 [inline]
blk_mq_try_issue_directly+0x3f4/0x880 block/blk-mq.c:2754
blk_mq_submit_bio+0x20c0/0x2a40 block/blk-mq.c:3208
submit_bio_noacct_nocheck+0x2f4/0xa40 block/blk-core.c:790
block_read_full_folio+0x7a6/0x810 fs/buffer.c:2463
filemap_read_folio+0x12c/0x3a0 mm/filemap.c:2510
read_part_sector+0xb6/0x2b0 block/partitions/core.c:724
adfspart_check_ICS+0xb1/0x960 block/partitions/acorn.c:357
check_partition block/partitions/core.c:143 [inline]
blk_add_partitions block/partitions/core.c:591 [inline]
bdev_disk_changed+0x851/0x17a0 block/partitions/core.c:695
blkdev_get_whole+0x372/0x510 block/bdev.c:751
add_disk_final block/genhd.c:412 [inline]
add_disk_fwnode+0x24b/0x3a0 block/genhd.c:606
null_add_dev+0x130b/0x1d70 drivers/block/null_blk/main.c:2052
nullb_device_power_store+0x240/0x380 drivers/block/null_blk/main.c:501
configfs_write_iter+0x337/0x430 fs/configfs/file.c:229
Syzkaller triggers this by creating a zoned null_blk device via
configfs. The Call Trace shows configfs_write_iter in configfs/file.c
handling a write to power file, which calls nullb_device_power_store in
main.c, which calls null_add_dev in main.c, which calls add_disk in
genhd.c, which triggers partition scan via bdev_disk_changed in
partitions/core.c.
A zoned null_blk device with zone_size 0 should not be legal. Existing
code tries to reject it via is_power_of_2() check in zoned.c and
!zone_size check in main.c, but syzkaller can still reach
null_zone_no() with zone_size_sects 0 via two paths:
1. Direct 0 via configfs: zone_size attribute store in main.c has
NULLB_DEVICE_ATTR(zone_size, ulong, NULL) with no validation callback,
so echo 0 > zone_size succeeds before power store. If zoned is false
at power store time, the !zone_size check in main.c is skipped, and
later zoned set true leaves zone_size 0.
2. Large value overflow: mb_to_sects() in zoned.c does
(sector_t)mb * SZ_1M >> SECTOR_SHIFT which is mb * 2048. If mb is
1UL << 53 (9PB), mb * 2048 overflows 64-bit to 0. The value is
power-of-two so is_power_of_2() passes, but mb_to_sects() returns 0.
Check for zero zone_size explicitly in null_init_zoned_dev() in
zoned.c, returning -EINVAL with "must be non-zero power-of-two".
Check for zero zone_size_sects after mb_to_sects() conversion,
returning -EINVAL for overflow case. Keep defensive check in
null_zone_no() returning 0 for zero sectors to avoid shift out-of-bounds
even if zero slips through.
This change should be safe because zone_size is set once in
null_init_zoned_dev() under device lock and never changes after, and 0
is never valid for a zoned device. Returning -EINVAL at init time fails
device creation early with clear error, while defensive return 0 in
null_zone_no() makes zoned command fail via offline zone check.
No new locking is introduced.
Reported-by: syzbot+abd6a8dca0f2b7726060@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=abd6a8dca0f2b7726060
Link: https://lore.kernel.org/all/6a75205c.01d0871a.3a0d52.0033.GAE@google.com/
Fixes: 8a3cf049af68 ("null_blk: add zoned block device emulation")
Cc: stable@vger.kernel.org
Assisted-by: Hermes:muse-spark-1.2 syzkaller
Signed-off-by: Rik van Riel <riel@surriel.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://patch.msgid.link/20260808114239.69167f68@fangorn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Not needed, probably since this file was split out.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20260810144603.453283-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
All block driver code except loop_set_dio() calls
queue_limits_start_update() before it freezes the request queue. Make
loop_set_dio() follow this convention. This patch fixes the following
lockdep complaint:
======================================================
WARNING: possible circular locking dependency detected
7.2.0-rc5-dbg #11 Not tainted
------------------------------------------------------
losetup/2924 is trying to acquire lock:
ffff88816c76da68 (&q->limits_lock){+.+.}-{4:4}, at: loop_set_dio+0x318/0x720 [loop]
but task is already holding lock:
ffff88816c76d430 (&q->q_usage_counter(io)#24){++++}-{0:0}, at: blk_mq_freeze_queue_nomemsave+0x1a/0x30
which lock already depends on the new lock.
Cc: Keith Busch <kbusch@kernel.org>
Fixes: 6c8dec275c ("loop: set dma_alignment from the backing file for direct I/O")
Reported-by: syzbot+cc0de396bac84da51919@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/linux-block/6a7d5368.d5f0ebe7.22d851.0013.GAE@google.com/
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/d919f5285d16afbec6c51ecdf201692a484566e5.1786637565.git.bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, we have only a one area per zcrx instance, and struct
io_zcrx_ifq stores a single pointer. To prepare for adding more areas,
replace it with an array of areas.
We'll be creating them at runtime, and the array is protected by 3
locks: ->pp_lock, ->alloc_lock and ->rq.lock. It takes all of them when
switching arrays, and readers should hold either of them.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://patch.msgid.link/d186adfa0ea26f651804cffef77d477f257eb404.1786108672.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Until now unmapping and destroying dmabuf were the same thing. To keep
it consistent with non-dmabuf, split it into two separate helpers. Unmap
destroys mappings and attachements as it should, and release only
putting down the dmabuf fd reference.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://patch.msgid.link/bd8c1fd8dac85ce0cf839b389ecd6050761965ca.1786108672.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We might want to create an area without having an instance of struct
io_uring_zcrx_ifq_reg. Extract a helper that doesn't have the ifq
registration structure as an argument but takes the buf length
explicitly.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://patch.msgid.link/76751ee4747abb4893058fcaaa36895dfe24a297.1786108672.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The RQ tail is updated by the user space. Cache it to reduce cache line
bouncing. Refilling now tries to exhaust the previous batch of rqes, but
since it could be too low, the iterator is allowed to recalculate the
rqes to process once after synching the tail value.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://patch.msgid.link/b42d2ed8b0e697110b646573b5da2a5f8e85f92e.1786108672.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
RQ head and tail are currently put into the same cache line, which can
cause false sharing problems when refill is run on another CPU. Put them
into separate cache lines.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/f769116a3f5267f06cb9a9a819d482ddea2e0852.1786108672.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_zcrx_ring_refill() caps the loop by mixing the max number of
allocated netmems and the number of available RQEs together, which
caps the number of entries to process the pp cache size. As a result,
when niovs are heavily fragmented, the refilling logic allocates only a
small number of niovs per call on average and sometimes even none.
Keep a separate counter for the number of processed RQ entries, which is
capped by a roughly calculated from the page size value to keep the
cache full. And separately break if it allocates enough niovs.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://patch.msgid.link/143cf439299728759eb0a840c866363fe99293f0.1786108672.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
create_worker_cb() reserves an io-wq worker slot only after its
task-work callback runs. If the callback is canceled before then,
io_worker_cancel_cb() still decrements acct->nr_workers. When an
existing worker retires with its creation callback pending, that
worker has already decremented the same account's worker count.
The resulting undercount permits worker creation beyond the account's
configured limit. On an AST2600 OpenBMC system, an unchanged sensor
daemon reached 4,291 threads with the original kernel. With an
equivalent downstream fix, 25 passive samples under its normal
workload showed 6-9 threads.
Decrement nr_workers only when the canceled callback is not
create_worker_cb(). Continuation callbacks still release their reserved
slot, and both callback types retain the existing running-count,
reference-count, and create-state cleanup.
Fixes: 1d5f5ea7cb ("io-wq: remove worker to owner tw dependency")
Cc: stable@vger.kernel.org
Assisted-by: Codex:gpt-5.6-sol
Signed-off-by: Vishnu Razdan <vrazdan@openai.com>
Reviewed-by: Gabriel Krisman Bertazi <krisman@suse.de>
Link: https://patch.msgid.link/20260811-vrazdan-io-wq-b4-submit-v1-1-719ced16c921@openai.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ddir was only ever forwarded to io_compat_msg_copy_hdr, which
never looked at it. Drop it.
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>
Link: https://patch.msgid.link/20260813004022.3514537-4-krisman@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Similar to commit f4eaf8eda8 ("io_uring/rsrc: Drop io_copy_iov in
favor of iovec API"), avoid the custom copy of a single iovec and just
rely on the iovec api. This lets the compat and native paths share the
buffer-select length lookup
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>
Link: https://patch.msgid.link/20260813004022.3514537-3-krisman@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Similar to commit f4eaf8eda8 ("io_uring/rsrc: Drop io_copy_iov in
favor of iovec API"), avoid custom copy and just rely on the iovec api.
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>
Link: https://patch.msgid.link/20260813004022.3514537-2-krisman@suse.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Disks created via blk_mq_alloc_disk_for_queue() (e.g. SCSI SD disks)
do not have GD_OWNS_QUEUE set. Currently __blk_mark_disk_dead() only
sets QUEUE_FLAG_DYING when GD_OWNS_QUEUE is set, so for such disks
blk_queue_enter() and __bio_queue_enter() cannot detect the dying
state via blk_queue_dying() and remain blocked waiting for I/O that
will never complete after surprise removal.
blk_mark_disk_dead() is the explicit "surprise removal" API -- the
caller has already decided the disk is dead. Setting QUEUE_FLAG_DYING
unconditionally here is appropriate: any in-flight I/O from other
threads should get -ENODEV immediately from blk_queue_enter()
regardless of GD_OWNS_QUEUE ownership.
For disks that already have GD_OWNS_QUEUE set, __blk_mark_disk_dead()
will set the flag again which is harmless.
Fixes: 6f8191fdf4 ("block: simplify disk shutdown")
Cc: stable@vger.kernel.org
Signed-off-by: Lianqin Hu <hulianqin@vivo.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/PUZPR06MB62247E82E66A3ED46CC3E6C7D2DC2@PUZPR06MB6224.apcprd06.prod.outlook.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit 6c6c1fc09d ("modpost: require a MODULE_DESCRIPTION()"),
modpost complains that swim3.ko is missing a module description.
WARNING: modpost: drivers/block/swim3.ko: missing MODULE_DESCRIPTION()
Add one to clear up the warning.
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260811-swim3-module-description-v1-1-28398c5a0e32@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add test_params_01.sh for SET_PARAMS. The test checks valid basic
parameters and several invalid parameter cases.
Also cover zoned parameters, including a non-power-of-2 zone size. This
case must fail in SET_PARAMS instead of being accepted and rejected later
when the device is started.
Signed-off-by: Yao Sang <sangyao@kylinos.cn>
Link: https://patch.msgid.link/20260814023226.354288-4-sangyao@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The normal kublk add command goes through device startup. It does not
tell the shell test whether a bad parameter is rejected by SET_PARAMS or
later by START_DEV.
Add a set_params command. It creates a temporary ublk device, sends
SET_PARAMS with the command line parameters, returns the ioctl result,
and deletes the device before START_DEV.
Signed-off-by: Yao Sang <sangyao@kylinos.cn>
Link: https://patch.msgid.link/20260814023226.354288-3-sangyao@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
UBLK_F_ZONED uses params.basic.chunk_sectors as zone size. ublk uses
ilog2(chunk_sectors) to get number of zones, so the value must be power
of 2.
If chunk_sectors is 96 and dev_sectors is 96 * 16, userspace asks for
16 zones. But the shift calculation gets 24 zones.
Block layer rejects such zone size when the disk is started. But
SET_PARAMS has already returned success, which is confusing for
userspace. Reject it in SET_PARAMS with other zoned parameter checks.
Fixes: 29802d7ca3 ("ublk: enable zoned storage support")
Signed-off-by: Yao Sang <sangyao@kylinos.cn>
Link: https://patch.msgid.link/20260814023226.354288-2-sangyao@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_cgroup_congested() walks the current task's blkcg ancestor chain on
every readahead decision and, once swap is in use, on every anonymous and
shmem folio allocation. The answer is almost always "no", but finding that
out costs two loads per level on two cold cache lines, plus an out-of-line
kthread_blkcg() and an RCU read-side pair. On a fleet profile of hosts
running containers with 5-10 level hierarchies it costs about as much as
all of mutex_lock(), 99.4% of it under __folio_throttle_swaprate().
Gate the walk on a global count of blkcgs with a non-zero
congestion_count. The counter only moves on the 0 <-> 1 transitions of
each blkcg's congestion_count, so the extra atomic stays in the throttle
arm/disarm paths and never appears in steady state. When something is
throttled the counter is non-zero and the walk runs as before.
Signed-off-by: Usama Arif <usama.arif@linux.dev>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://patch.msgid.link/20260814165712.510132-4-usama.arif@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
iocg_kick_delay() turns sufficiently large debt into an explicit
block-cgroup delay with blkcg_set_delay(), setting blkg->use_delay to
-1 and incrementing blkcg->congestion_count. Clearing it again depends
on iocg_kick_delay() running from the period timer, the waitq timer or
the issue path.
ioc_pd_free() removes the iocg from active_iocgs and cancels its waitq
timer, and no further bios can arrive, so once it has run nothing is
left which can reduce the debt and clear the delay. The blkcg stays
marked congested for the rest of its life.
blk_cgroup_congested() then returns true for every task in that cgroup
and its descendants: page_cache_sync_ra() cuts readahead to a single
page, page_cache_async_ra() skips it altogether, and
__folio_throttle_swaprate() takes swap_avail_lock and schedules a
throttle on anonymous folio allocation.
Clear it explicitly, after the list removal and the synchronous
hrtimer_cancel() so that neither timer processing nor an I/O path can
re-arm it. The free callback can also see policy data which was never
attached to a blkg, hence the pd->blkg check.
Fixes: 7caa47151a ("blkcg: implement blk-iocost")
Signed-off-by: Usama Arif <usama.arif@linux.dev>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://patch.msgid.link/20260814165712.510132-3-usama.arif@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io.latency can throttle a group which has no latency target of its own.
When a sibling misses its target, check_scale_change() scales down its
peers, and a peer that reaches queue depth one gets blkcg_use_delay()
called on it on every further scale-down, even with min_lat_nsec == 0.
iolatency_pd_offline() resets the target through
iolatency_set_min_lat_nsec(), which clears the delay only on a nonzero
to zero transition, so it never clears such a peer. Freeing the policy
data then leaves blkg->use_delay set and blkcg->congestion_count
elevated with nothing left that can drop it.
blk_cgroup_congested() then returns true for every task in that cgroup
and its descendants for as long as the cgroup lives: page_cache_sync_ra()
cuts readahead to a single page, page_cache_async_ra() skips it
altogether, and __folio_throttle_swaprate() takes swap_avail_lock and
schedules a throttle on anonymous folio allocation.
Clear the delay in iolatency_pd_free(). By then bio-held blkg
references have drained, or the queue is frozen for policy
deactivation, so check_scale_change() cannot re-arm it. The free
callback can also see policy data which was never attached to a blkg,
hence the pd->blkg check.
Fixes: d706751215 ("block: introduce blk-iolatency io controller")
Signed-off-by: Usama Arif <usama.arif@linux.dev>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://patch.msgid.link/20260814165712.510132-2-usama.arif@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit c807ab520f ("block/mq-deadline: Add I/O priority support")
left the dd parameter unused in deadline_move_request().
Commit fde02699c2 ("block: mq-deadline: Remove support for zone
write locking") left dd unused in deadline_fifo_request() and
deadline_next_request().
Remove these unused function parameters.
Signed-off-by: Hongfu Li <lihongfu@kylinos.cn>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Tao Cui <cuitao@kylinos.cn>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260812040729.27551-1-hongfu.li@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The attribute store methods generated with NULLB_DEVICE_ATTR() refuse to
change the configuration of a live device by testing
NULLB_DEV_FL_CONFIGURED, but that flag is only set by
nullb_device_power_store() after null_add_dev() has returned, and the
store methods take no lock at all. configfs only serializes writes to
the same open file (buffer->mutex), so a write to any attribute can run
concurrently with null_add_dev() and change the device configuration
while it is being used.
null_add_dev() reads the configuration several times, e.g. dev->zoned is
read once to set up the queue limits and once to initialize the zone
resources:
CPU0: echo 1 > nullb0/power CPU1: echo 1 > nullb0/zoned
nullb_device_power_store()
mutex_lock(&lock)
null_add_dev()
if (dev->zoned) -> false
/* no BLK_FEAT_ZONED */ nullb_device_zoned_store()
test_bit(FL_CONFIGURED) -> 0
dev->zoned = true
blk_mq_alloc_disk()
/* queue is not zoned */
if (nullb->dev->zoned) -> true
null_register_zoned_dev()
blk_revalidate_disk_zones()
blk_revalidate_disk_zones() is then called for a queue that does not
have BLK_FEAT_ZONED set, which triggers its WARN_ON_ONCE() and fails the
device setup with -EIO:
WARNING: CPU: 2 PID: 322 at block/blk-zoned.c:2357 blk_revalidate_disk_zones+0x4c/0x560
Clearing dev->zoned in the same window is worse: the queue is created
with BLK_FEAT_ZONED but the zone resources are never initialized, so
add_disk() succeeds for a zoned disk that has no zones. And a store that
lands after the last dev->zoned test leaves dev->zoned set while
dev->zones is still NULL, which null_process_zoned_cmd() dereferences on
the first write.
Fix this by taking the global lock, which nullb_device_power_store()
already holds across null_add_dev() and null_del_dev(), around both the
NULLB_DEV_FL_CONFIGURED test and the update of the device configuration.
The submit_queues and poll_queues apply callbacks are now called with
that lock held, so remove the locking they did themselves.
Since the store methods can run as soon as configfs_register_subsystem()
returns, that is, before null_init() gets to mutex_init(&lock), also
initialize the lock statically with DEFINE_MUTEX().
Fixes: 3bf2bd2073 ("nullb: add configfs interface")
Reported-by: syzbot+643a6dd130546afdf1fb@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/linux-block/6a7d0b3f.ac361c09.22ff0a.004c.GAE@google.com/
Signed-off-by: Niklas Cassel <cassel@kernel.org>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://patch.msgid.link/20260813141456.1625857-2-cassel@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation to supporting dma-buf backed iterators and bios,
introduce bio_iov_iter_set() which attempts to set up the bio directly
from the given iterator. For now, it only supports bvec and expects
users to check the result and fall back to other means if fails, but
later we'll add more types.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/4686a0e47fc14f3f888967a80d45a6f66044f1e0.1785596451.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The _show callback in the NULLB_DEVICE_ATTR macro reads dev->NAME and the
_store path writes it. configfs does not serialize accesses across separate
open file descriptions (buffer->mutex is per-fd), and _show takes no lock,
so a concurrent read and write on the same attribute is a data race. The
_show readers also race against writes to these fields that run after the
configfs item becomes visible, e.g. in nullb_update_nr_hw_queues().
All of those writers now run under the file-scope lock: _store takes it
unconditionally, and the setup-side writers run under power_store() which
holds the same lock. The only remaining unsynchronized accesses are the
plain reads in _show. Rather than annotating every field with
READ_ONCE()/WRITE_ONCE() across files, simply take the file-scope lock in
_show (and in power_show) as well. This closes the remaining _show-vs-write
data races with a single lock and keeps the writers as plain assignments.
configfs attribute access is not on the I/O hot path, so taking the mutex
in _show is acceptable from a performance standpoint. The dev fields
written in null_alloc_dev() and dev->power in nullb_group_drop_item() need
no locking: the former runs from .make_group before the item is published,
and the latter is serialized by configfs frag_sem/frag_dead against
attribute show/store.
Suggested-by: Nilay Shroff <nilay@linux.ibm.com>
Suggested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Zizhi Wo <wozizhi@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://patch.msgid.link/20260725022509.714271-11-wozizhi@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The NULLB_DEVICE_ATTR _store takes no lock: apply_fn attributes
(submit_queues, poll_queues) get dev->NAME written again after apply_fn
returns, outside its lock; APPLY=NULL attributes are entirely lockless.
configfs only serializes stores per-open-file, so concurrent stores on
separate fds race.
For apply_fn attributes, once one store's apply_fn has reconfigured the
hardware, a second (losing) store can still overwrite dev->NAME
afterwards. This leaves dev->submit_queues out of sync with the live
queue count, which is later caught by the WARN_ON_ONCE() in
null_map_queues().
For !apply_fn attributes, power_store()'s null_add_dev() validates and
builds the device under "lock" but only sets CONFIGURED afterwards. A store
slipping in during this window can change a field mid-setup -- for example,
zone_nr_conv can be pushed above nr_zones after it has already been
clamped, leading to an out-of-bounds dev->zones[] access.
Take "lock" in the macro around the apply_fn call, the CONFIGURED test and
the field write, and move it out of nullb_apply_submit_queues()/
nullb_apply_poll_queues() so both paths are covered once. This serializes
stores with power_store's setup and with each other.
Fixes: 45919fbfe1 ("null_blk: Enable modifying 'submit_queues' after an instance has been configured")
Suggested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Zizhi Wo <wozizhi@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://patch.msgid.link/20260725022509.714271-10-wozizhi@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Using guard()/scoped_guard() ties lock release to scope exit, removing the
need for manual mutex_unlock() calls and preventing missed unlocks on error
paths.
The per-attribute apply wrappers are left untouched, as those are reworked
separately by the configfs show/store serialization patches.
Signed-off-by: Zizhi Wo <wozizhi@huawei.com>
Link: https://patch.msgid.link/20260725022509.714271-9-wozizhi@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When shared_tags is enabled, null_setup_tagset() makes the device use the
global tag_set, whose driver_data stays NULL. null_map_queues() therefore
falls back to the module-wide g_submit_queues/g_poll_queues instead of any
per-device value.
Resizing submit_queues or poll_queues via configfs on such a device calls
blk_mq_update_nr_hw_queues() on the shared set, shrinking
set->nr_hw_queues. __blk_mq_realloc_hw_ctxs() only grows the
q->queue_hw_ctx[] allocation, so on shrink it merely exits and NULLs the
now-excess hctx slots. null_map_queues(), however, keeps mapping CPUs with
the unchanged g_submit_queues/g_poll_queues, so mq_map[] ends up pointing
at those NULLed hctx slots. blk_mq_map_swqueue() then dereferences the NULL
hctx (hctx->cpumask), crashing the kernel:
[ 460.218374] KASAN: null-ptr-deref in range [0x0000000000000098-0x000000000000009f]
[ 460.219003] CPU: 24 UID: 0 PID: 1492 Comm: sh Not tainted 7.2.0-rc2+ #67 PREEMPT(full)
[ 460.219792] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-4.fc41 04/01/2014
[ 460.220452] RIP: 0010:blk_mq_map_swqueue+0x4db/0x1430
......
[ 460.228977] Call Trace:
[ 460.229175] <TASK>
[ 460.229354] blk_mq_update_nr_hw_queues+0xd49/0x11c0
[ 460.229779] ? __pfx_blk_mq_update_nr_hw_queues+0x10/0x10
[ 460.230200] nullb_update_nr_hw_queues+0x1a9/0x370 [null_blk]
[ 460.230694] nullb_device_submit_queues_store+0xd9/0x170 [null_blk]
[ 460.231190] ? __pfx_nullb_device_submit_queues_store+0x10/0x10 [null_blk]
[ 460.231776] ? configfs_write_iter+0x35c/0x4e0
[ 460.232122] configfs_write_iter+0x286/0x4e0
[ 460.232460] vfs_write+0x52d/0xd00
[ 460.232779] ? __x64_sys_openat+0x108/0x1d0
[ 460.233106] ? __pfx_vfs_write+0x10/0x10
[ 460.233413] ? fdget_pos+0x1cf/0x4c0
[ 460.233745] ? fput_close+0x133/0x190
[ 460.234038] ? __pfx_expand_files+0x10/0x10
[ 460.234368] ksys_write+0xfc/0x1d0
Reproducer:
modprobe null_blk shared_tags=1 submit_queues=64 poll_queues=1
mkdir /sys/kernel/config/nullb/dev
echo 1 > /sys/kernel/config/nullb/dev/power
echo 1 > /sys/kernel/config/nullb/dev/submit_queues
A per-device resize of a shared tag set is meaningless anyway, so reject it
with -EINVAL in nullb_update_nr_hw_queues() when the device is bound to the
global tag_set.
Fixes: 45919fbfe1 ("null_blk: Enable modifying 'submit_queues' after an instance has been configured")
Suggested-by: Nilay Shroff <nilay@linux.ibm.com>
Assisted-by: Claude-Code:GLM-5.2
Signed-off-by: Zizhi Wo <wozizhi@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20260725022509.714271-8-wozizhi@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
