KCSAN reports a data race between raid1_end_read_request() and
raid1_read_request().
The completion path updates conf->mirrors[disk].head_position in
update_head_pos() without a lock, while the read-balance heuristic reads
the same field locklessly in is_sequential() and choose_best_rdev().
KCSAN report:
=========================
BUG: KCSAN: data-race in raid1_end_read_request / raid1_read_request
write to 0xffff8f0306ba7868 of 8 bytes by interrupt on cpu 9:
raid1_end_read_request+0xb5/0x440
bio_endio+0x3c9/0x3e0
blk_update_request+0x257/0x770
scsi_end_request+0x4d/0x520
scsi_io_completion+0x6f/0x990
scsi_finish_command+0x188/0x280
scsi_complete+0xac/0x160
blk_complete_reqs+0x8e/0xb0
blk_done_softirq+0x1d/0x30
[...]
read to 0xffff8f0306ba7868 of 8 bytes by task 667002 on cpu 11:
raid1_read_request+0x497/0x1a10
raid1_make_request+0xdf/0x1950
md_handle_request+0x2c5/0x700
md_submit_bio+0x126/0x320
__submit_bio+0x2ec/0x3a0
submit_bio_noacct_nocheck+0x572/0x890
[...]
value changed: 0x0000000000000078 -> 0x00000000005fe448
Signed-off-by: Chen Cheng <chencheng@fnnas.com>
Link: https://patch.msgid.link/20260619044114.1208456-1-chencheng@fnnas.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
raid1 supports REQ_NOWAIT reads by avoiding waits in the barrier path
through wait_read_barrier(). However, a read can still block on a
WriteMostly device when the array uses a bitmap and there are
outstanding behind writes.
In that case raid1 unconditionally calls wait_behind_writes(), which
may sleep until all behind writes complete. As a result, a REQ_NOWAIT
read can block despite the caller explicitly requesting non-blocking
behavior.
This ensures that raid1 consistently honors REQ_NOWAIT reads across all
paths that may otherwise wait for behind writes.
Fixes: 5aa705039c ("md: raid1 add nowait support")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Link: https://patch.msgid.link/20260611083514.754922-1-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
llbitmap discard is useful even when no underlying member device supports
it. The discard still converts the llbitmap range to unwritten, so later
reads and recovery do not rely on stale parity for that range.
Let llbitmap discard bypass the raid5 lower discard support check. If lower
discard is not safe or not supported, complete the accounted clone after
md_account_bio() so the llbitmap conversion callbacks run without member
discard bios.
Link: https://patch.msgid.link/20260605072639.2434847-4-yukuai@kernel.org
Signed-off-by: Yu Kuai <yukuai@fygo.io>
Raid5 used to disable discard limits when devices_handle_discard_safely
was not set or when stacked member limits could not support a full-stripe
discard. That hides discard from userspace before raid5 can decide whether
a request can be handled safely.
Follow other virtual drivers and advertise a UINT_MAX discard limit for the
md device. Cache lower discard support in r5conf when setting queue limits,
and reject unsupported discard bios before queuing stripe work.
Link: https://patch.msgid.link/20260605072639.2434847-3-yukuai@kernel.org
Signed-off-by: Yu Kuai <yukuai@fygo.io>
Raid5 handles discard bios internally through make_discard_request() and
never passes them through md_account_bio(). As a result, discard IO is
missing the md-device iostat accounting that normal raid5 IO and discard
IO in other raid levels get from md_account_bio().
Before accounting the bio, trim the request to the full data stripes that
raid5 will actually discard. The first full stripe is the ceiling of the
bio start divided by data-stripe sectors, and the last full stripe is the
floor of the bio end divided by data-stripe sectors. Account that exact
MD logical full-stripe range, then restore the original iterator so bio
completion and iostat still cover the original request.
Link: https://patch.msgid.link/20260605072639.2434847-2-yukuai@kernel.org
Signed-off-by: Yu Kuai <yukuai@fygo.io>
raid1_write_request() increments rdev->nr_pending before checking the
badblocks and then immediately decrements it again when a device is
skipped. Move the increment until after the checks succeed so the
reference accounting is easier to follow.
Consolidate the failure paths so that each error label releases exactly
the resources acquired up to that point. err_dec_pending drops pending
references and frees the r1bio, while err_allow_barrier handles the
barrier release before returning.
When a REQ_ATOMIC write cannot be satisfied due to a badblock range,
complete the bio with BLK_STS_NOTSUPP rather than reporting an I/O
error, since the operation is unsupported rather than having failed
during I/O.
Rename max_write_sectors to max_sectors and remove the redundant local
copy.
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Link: https://patch.msgid.link/20260613182810.1317258-5-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
raid10_make_request() acquires a writes_pending reference with
md_write_start() before calling raid10_handle_discard(). Several failure
paths in raid10_handle_discard() complete the bio and return without
releasing the corresponding reference, causing md_write_end() to be
skipped.
Call md_write_end() before returning from these failure paths to keep
writes_pending accounting balanced.
Additionally, discard split allocation failures can occur after
wait_barrier() succeeds. Those paths return without calling
allow_barrier(), leaking the associated barrier reference.
Release the barrier before returning from those paths.
Fixes: c9aa889b03 ("md: raid10 add nowait support")
Fixes: 4cf58d9529 ("md/raid10: Handle bio_split() errors")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Link: https://patch.msgid.link/20260613182810.1317258-4-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
raid10_make_request() acquires a writes_pending reference with
md_write_start() before dispatching write requests. Several failure
paths in raid10_write_request() complete the bio and return without
reaching the normal write completion path, causing the corresponding
md_write_end() to be skipped.
Make raid10_write_request() return a status indicating whether the write
request was successfully queued. This allows raid10_make_request() to
release the writes_pending reference with md_write_end() when a write
request fails.
Fixes: 4cf58d9529 ("md/raid10: Handle bio_split() errors")
Fixes: c9aa889b03 ("md: raid10 add nowait support")
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com>
Link: https://patch.msgid.link/20260613182810.1317258-3-abd.masalkhi@gmail.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
Pull device mapper updates from Mikulas Patocka:
- small cleanups in dm-vdo, dm-raid, dm-cache, dm-zoned-metadata
- rework of dm-ima
- introduce dm-inlinecrypt
- fix wrong return value in dm-ioctl
- fix rcu stall when polling
* tag 'for-7.2/dm-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm-zoned-metadata: Use strscpy() to copy device name
dm cache: make smq background work limit configurable
dm-inlinecrypt: add support for hardware-wrapped keys
dm: limit target bio polling to one shot
dm-ioctl: report an error if a device has no table
dm: add documentation for dm-inlinecrypt target
dm-inlinecrypt: add target for inline block device encryption
block: export blk-crypto symbols required by dm-inlinecrypt
dm-ima: use active table's size if available
dm-ima: Fail more gracefully in dm_ima_measure_on_*
dm-ima: Handle race between rename and table swap
dm-ima: Fix issues with dm_ima_measure_on_device_rename
dm-ima: remove new_map from dm_ima_measure_on_device_clear
dm-ima: Fix UAF errors and measuring incorrect context
dm-ima: don't copy the active table to the inactive table
dm-ima: Remove status_flags from dm_ima_measure_on_table_load()
dm-ima: remove broken last_target_measured logic
dm-ima: remove dm_ima_reset_data()
dm-raid: only requeue bios when dm is suspending
dm vdo: use get_random_u32() where appropriate
Pull block updates from Jens Axboe:
- NVMe pull request via Keith:
- Per-controller admin and IO timeout sysfs attributes, and
letting the block layer set request timeouts (Maurizio,
Maximilian)
- Multipath passthrough iostats, and PCI P2PDMA enablement for
multipath devices (Keith, Kiran)
- A new diag sysfs attribute group exporting per-controller
counters (retries, multipath failover, error counters, requeue
and failure counts, reset and reconnect events) (Nilay)
- FDP configuration validation and bounds check fixes (liuxixin)
- Various nvmet fixes, including a pre-auth out-of-bounds read in
the Discovery Get Log Page handler, auth payload bounds
validation, and tcp error-path leak fixes (Bryam, Tianchu,
Geliang)
- nvme-tcp lockdep and workqueue fixes (Shin'ichiro, Kuniyuki,
Eric)
- Assorted other fixes and cleanups (John, Yao, Chao, Mateusz,
Achkinazi, Wentao)
- MD pull request via Yu Kuai:
- raid1/raid10 fixes for a deadlock in the read error recovery
path, error-path detection and bio accounting with cloned bios,
and an nr_pending leak in the REQ_ATOMIC bad-block error path
(Abd-Alrhman)
- PCI P2PDMA propagation from member devices to the RAID device
(Kiran)
- dm-raid bio requeue fix, and various smaller fixes and cleanups
(Benjamin, Chen, Li, Thorsten)
- Enable Clang lock context analysis for the block layer, with the
accompanying annotations across queue limits, the blk_holder_ops
callbacks, crypto, cgroup, iocost, kyber and mq-deadline (Bart)
- Block status code infrastructure work: a tagged status table, a
str_to_blk_op() helper, a bio_endio_status() helper, and on top of
that a new configurable block-layer error injection facility
(Christoph)
- DRBD netlink rework, replacing the genl_magic machinery with explicit
netlink serialization and moving the DRBD UAPI headers to
include/uapi/linux/ (Christoph Böhmwalder)
- bvec improvements: a bvec_folio() helper and making the bvec_iter
helpers proper inline functions (Willy, Christoph)
- ublk cleanups and a canceling-flag fix for the disk-not-allocated
case (Caleb, Ming)
- Partition handling fixes: bound the AIX pp_count scan, fix an of_node
refcount leak, and replace __get_free_page() with kmalloc() (Bryam,
Wentao, Mike)
- Convert numa_node to int in blk_mq_hw_ctx and ->init_request, and add
WQ_PERCPU to the block workqueue users (Mateusz, Marco)
- Block statistics and tracing: propagate in-flight to the whole disk
on partition IO, export passthrough stats, and a new
block_rq_tag_wait tracepoint (Tang, Keith, Aaron)
- A round of removals, unexports and cleanups across bio, direct-io and
the bvec helpers (Christoph)
- Various driver fixes (mtip32xx use-after-free, rbd snap_count
validation and strscpy conversion, nbd socket lockdep reclassify,
virtio-blk zone report clamp, floppy) and a batch of MAINTAINERS
email/list updates (Coly, Li, Yu, Christoph Böhmwalder)
- Other little fixes and cleanups all over
* tag 'for-7.2/block-20260615' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux: (117 commits)
MAINTAINERS: Update Coly Li's email address
block: check bio split for unaligned bvec
nbd: Reclassify sockets to avoid lockdep circular dependency
block: add configurable error injection
block: add a str_to_blk_op helper
block: add a "tag" for block status codes
block: add a macro to initialize the status table
floppy: Drop unused pnp driver data
block: propagate in_flight to whole disk on partition I/O
virtio-blk: clamp zone report to the report buffer capacity
block: optimize I/O merge hot path with unlikely() hints
drivers/block/rbd: Use strscpy() to copy strings into arrays
partitions: aix: bound the pp_count scan to the ppe array
block: Enable lock context analysis
block/mq-deadline: Make the lock context annotations compatible with Clang
block/Kyber: Make the lock context annotations compatible with Clang
block/blk-mq-debugfs: Improve lock context annotations
block/blk-iocost: Inline iocg_lock() and iocg_unlock()
block/blk-iocost: Split ioc_rqos_throttle()
block/crypto: Annotate the crypto functions
...
Pull buffer_head updates from Christian Brauner:
"This removes b_end_io from struct buffer_head.
Instead of setting bio->bi_end_io to end_bio_bh_io_sync() which then
calls bh->b_end_io(), the new bh_submit() and __bh_submit() interfaces
set bio->bi_end_io to the appropriate completion handler directly,
replacing two indirect function calls in the completion path with one.
It is also one fewer function pointer in the middle of a writable data
structure that can be corrupted, it shrinks struct buffer_head from
104 to 96 bytes allowing roughly 7% more buffer_heads to be cached in
the same amount of memory, and it removes some atomic operations as
the buffer refcount is no longer incremented before calling the end_io
handler.
All in-tree users (fs/buffer.c itself, ext4, jbd2, ocfs2, gfs2,
nilfs2, and md-bitmap) are converted, and submit_bh(),
mark_buffer_async_write(), and end_buffer_write_sync() are removed"
* tag 'vfs-7.2-rc1.bh' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: (34 commits)
buffer: Remove end_buffer_write_sync()
buffer: Change calling convention for end_buffer_read_sync()
buffer: Remove b_end_io
buffer: Remove submit_bh()
md-bitmap: Convert read_file_page and write_file_page to bh_submit()
nilfs2: Convert nilfs_mdt_submit_block to bh_submit()
nilfs2: Convert nilfs_gccache_submit_read_data to bh_submit()
nilfs2: Convert nilfs_btnode_submit_block to bh_submit()
buffer: Remove mark_buffer_async_write()
gfs2: Convert gfs2_aspace_write_folio to bh_submit()
gfs2: Remove use of b_end_io in gfs2_meta_read_endio()
gfs2: Convert gfs2_dir_readahead to bh_submit()
gfs2: Convert gfs2_metapath_ra to bh_submit()
ocfs2: Convert ocfs2_write_super_or_backup to bh_submit()
ocfs2: Convert ocfs2_read_blocks to bh_submit()
ocfs2: Convert ocfs2_read_block to bh_submit()
ocfs2: Convert ocfs2_write_block to bh_submit()
jbd2: Convert jbd2_write_superblock() to bh_submit()
jbd2: Convert journal commit to bh_submit()
ext4: Convert ext4_commit_super() to bh_submit()
...
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
The maximum number of concurrent background work items (promotions,
demotions, writebacks) in the SMQ policy was hardcoded to 4096, with
a FIXME comment noting it should be made configurable.
This value was originally tuned down from 10240 to balance memory
overhead (~128 bytes per entry, ~512KB at 4096 entries) against I/O
parallelism. However, different workloads and cache sizes may benefit
from different limits:
- Write-heavy workloads may need more writeback concurrency
- Very large caches (10+ TB) may need more promotion slots
- Memory-constrained systems may want a lower limit
Make this configurable via the module parameter "smq_max_background_work"
(defaulting to 4096 to preserve existing behaviour). Clamp the value to
at least 1 to prevent setting 0, which would block all background work.
The parameter only affects newly created cache devices; existing caches
retain their value from creation time.
Signed-off-by: Cao Guanghui <caoguanghui@kylinos.cn>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
commit 2d1f7b65f5 ("dm cache policy smq: fix missing locks in
invalidating cache blocks") added mq->lock around the destructive part of
smq_invalidate_mapping(), but left the e->allocated check outside the
critical section.
That leaves a check-then-act race. Two concurrent invalidators can both
observe e->allocated as true before either of them takes mq->lock. The
first invalidator that acquires the lock removes the entry from the
queues and hash table and then calls free_entry(), which clears
e->allocated and puts the entry back on the free list. The second
invalidator can then acquire mq->lock and continue with the stale result
of the unlocked check.
This can corrupt the SMQ queues or hash table by deleting an entry that
is no longer on those structures. It can also hit the allocation check in
free_entry() when the same entry is freed again.
Move the allocation check under mq->lock so the predicate and the
destructive operations are serialized by the same lock.
Fixes: 2d1f7b65f5 ("dm cache policy smq: fix missing locks in invalidating cache blocks")
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
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: f2a38abf5f ("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>
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: bb2a9acefa ("md/raid1: switch to use md_account_bio() for io accounting")
Fixes: 8204552383 ("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>
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: 689389a06c ("md/raid1: simplify handle_read_error().")
Fixes: 545250f248 ("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>
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: bb2a9acefa ("md/raid1: switch to use md_account_bio() for io accounting")
Fixes: 8204552383 ("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>
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: d30588b273 ("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>
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>
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>
Pull device mapper fix from Mikulas Patocka:
- fix crashes in dm-vdo if GFP_NOWAIT allocation fails
* tag 'for-7.1/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm vdo: use GFP_NOIO for blkdev_issue_zeroout on format path
Add support for hardware-wrapped encryption keys to the
dm-inlinecrypt target.
Introduce a new optional argument <key_type> to indicate
whether the provided key is a raw key or a hardware-wrapped
key. Based on this flag, the appropriate blk-crypto key type
is selected when initializing the key.
This allows dm-inlinecrypt to work with hardware that requires
keys to be wrapped and managed by the underlying inline
encryption engine.
Update the target argument parsing accordingly and pass the
key type to blk_crypto_init_key(). Documentation is also
updated to reflect the new parameter and usage.
Signed-off-by: Linlin Zhang <linlin.zhang@oss.qualcomm.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Fixes: e7f57d2c47 ("dm-inlinecrypt: add target for inline block device encryption")
dm_poll_bio() is the ->poll_bio() callback for a stacked dm device.
The caller only knows about the dm queue, so it may decide to do a
spinning poll if it thinks a single queue is being polled. Passing those
flags unchanged to the mapped clone lets blk_mq_poll() spin on a target
queue from inside dm_poll_bio().
With io_uring IOPOLL on a dm-stripe target this can keep a task in
dm_poll_bio() -> bio_poll() -> blk_mq_poll()
long enough to trigger an RCU CPU stall, before io_uring gets back to
io_iopoll_check() and its need_resched() check.
Keep dm's ->poll_bio() bounded by forcing one-shot polling for target
bios. The caller can invoke dm_poll_bio() again if it wants to keep
polling, and it also gets a chance to reap completions or reschedule
between passes.
Fixes: f22ecf9c14 ("blk-mq: delete task running check in blk_hctx_poll()")
Signed-off-by: Fengnan Chang <changfengnan@bytedance.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
When we send a message to a device that has no table, the return code was
not set. The code would return "2", which is not considered a valid return value.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@vger.kernel.org
Reviewed-by: Benjamin Marzinski <bmarzins@redhat.com>
Add a new device-mapper target "dm-inlinecrypt" that is similar to
dm-crypt but uses the blk-crypto API instead of the regular crypto API.
This allows it to take advantage of inline encryption hardware such as
that commonly built into UFS host controllers.
The table syntax matches dm-crypt's, but for now only a stripped-down
set of parameters is supported. For example, for now AES-256-XTS is the
only supported cipher.
dm-inlinecrypt is based on Android's dm-default-key with the
controversial passthrough support removed. Note that due to the removal
of passthrough support, use of dm-inlinecrypt in combination with
fscrypt causes double encryption of file contents (similar to dm-crypt +
fscrypt), with the fscrypt layer not being able to use the inline
encryption hardware. This makes dm-inlinecrypt unusable on systems such
as Android that use fscrypt and where a more optimized approach is
needed. It is however suitable as a replacement for dm-crypt.
dm-inlinecrypt supports both keyring key and hex key, the former avoids
the key to be exposed in dm-table message. Similar to dm-default-key in
Android, it will fallabck to the software block crypto once the inline
crypto hardware cannot support the expected cipher.
Test:
dmsetup create inlinecrypt_logon --table "0 `blockdev --getsz $1` \
inlinecrypt aes-xts-plain64 :64:logon:fde:dminlinecrypt_test_key 0 $1 0"
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Linlin Zhang <linlin.zhang@oss.qualcomm.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
GFP_NOWAIT is inappropriate when blkdev_issue_zeroout may sleep and
bio_alloc can fail under pressure; use GFP_NOIO for clear_partition and
vdo_clear_layout zeroout calls.
Signed-off-by: Bruce Johnston <bjohnsto@redhat.com>
Signed-off-by: Matthew Sakai <msakai@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Fixes: fc1d438267 ("dm vdo: save the formatted metadata to disk")
It is possible that the dm_ima_measure_on_* functions run at the same
time as a table is getting swapped, but before the md->ima.active_table
is updated by dm_ima_measure_on_device_resume(). Instead of using the
current device size, use the size of the active table that is being
measured (assuming there is one), so the information is consistent.
Also, don't allocate a separate string to hold the capactiy. Just
print it directly to the measurement buffer.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
In all the dm_ima_measure_on_* functions besides
dm_ima_measure_on_table_load(), even if measuring the event fails, it's
still possible to update dm->ima, so that it continues to correctly
track the device state. This means that one measurement failure won't
cause future measurements to record the wrong data.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
a device rename could happen after do_resume() removed the inactive
table that it was swapping to out of the hash cell, but before it was
made the active table. In this case, the table metadata would still
have the old name. Update the swapped table's metadata to avoid this.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
dm_ima_measure_on_device_rename() can be called on a device before it
ever loads a table, so it needs to handle the case where there is no
table metadata. Also, it was only updating the table_metadata on the
active table. If there was an inactive table when the device was renamed
and that table was later swapped in as the active table, it would
still have the old name. dm_ima_measure_on_device_rename() was also
needlessly allocating new memory for the updated table metadata, instead
of just reusing the existing memory.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Now that two processes can't modify md->ima in
dm_ima_measure_on_device_clear() at the same time, there's no need to
track if an inactive table was actually removed. We might as well
clean it up unconditionally, on the off chance that a previous
ima measurement failed and left md->ima.inactive_table behind.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
the dm-ima code did not keep the dm_ima_measure_on_* functions from
running at the same time. This could lead to various errors. If two
processes were updating the device state, one could update the state
first, but the other could measure the state first, causing the the
current device state to appear incorrect. If a table load happened while
a device was resuming, the IMA measurement could report the wrong table
being active. And if two dm_ima_measure_on_* functions ran at the same
time, one of them could free data that the other was accessing, causing
a crash.
All the core dm functions that call a dm_ima_measure_on_* function
update the device state they want to measure under the _hash_lock,
except for do_resume(). But holding the _hash_lock is not a good way to
synchronize these functions. It's a global mutex, that is needed in many
dm operations, and the dm_ima_measure_* functions can sleep, blocking
any dm operation on any device that needs the _hash_lock.
To serialize and order the IMA measurement functions, the
dm_ima_measurements now has two counters, update_idx and measure_idx.
update_idx is incremented while holding the _hash_lock and saved, along
with the device name and uuid, in a dm_ima_context struct. Once the
_hash_lock is dropped, the dm_ima_measure_* function is called. It waits
until measure_idx matches the saved value of update_idx, ensuring that
the updates and measurements happen in the same order if there are
multiple processes changing the device at the same time. Then it
measures the device, updates measure_idx, and wakes up any other
process waiting to do a measurement. This makes sure that the
measurements are serialized and done in the order that the _hash_lock
was acquired in. But they only block other measurements for the same
device, which are unlikely to happen at the same time.
do_resume() is trickier, because it removes the inactive table while
holding the _hash_lock, but doesn't hold it while updating md->map. To
make sure it is also ordered, the IMA code grabs the _hash_lock after
md->map is updated. Then it makes sure that the device isn't being
removed and that another do_resume() hasn't already changed the active
table again, and serializes like the other functions do.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
If an inactive table was cleared, dm_ima_measure_on_table_clear() was
copying the ima.active_table to ima.inactive_table. This is not what
device-mapper does, and it makes the IMA measurements show an inactive
table when there isn't one. Also, once this is removed, the code no
longer needs to keep checking if the active and the inactive table point
to the same memory.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
There is no status flag that is is used for STATUSTYPE_IMA type
status() calls, and STATUSTYPE_IMA itself is not a valid
status flag.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
When it ran out of space for adding more targets to the ima_buf,
dm_ima_measure_on_table_load() would measure the dm device early, and
then add the rest of the targets and measure it again.
last_target_measured was intended to flag the last target measured so
that the device wouldn't get remeasured, if no new targets were added
after the early measurement. But the way to code works, the dm device
will never be measured early unless there is another target to add to
the ima_buf. Instead, if there is only one more target to add, that
target was getting added to the ima_buf, but it wasn't getting
remeasured, because last_target_measured was set. Since
dm_ima_measure_on_table_load() only measures a device early when there
are more targets to add, the final measurement must always happen, and
last_target_measured is unneeded.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
There's no point in saving the string length of DM_IMA_VERSION_STR. It's
a constant, so the compiler will precompute it. dm_create() will already
zero out the rest of dm->ima.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
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. We cannot handle the bio, and we
cannot suspends while it is outstanding. But during normal operations,
we should not push the bio back to dm. Instead, wait for the reshape
to be resumed.
Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
