]> git.ipfire.org Git - thirdparty/kernel/linux.git/log
thirdparty/kernel/linux.git
2 months agonvmet-tcp: Don't free SQ on authentication success
Alistair Francis [Tue, 2 Dec 2025 05:17:53 +0000 (15:17 +1000)] 
nvmet-tcp: Don't free SQ on authentication success

Curently after the host sends a REPLACETLSPSK we free the TLS keys as
part of calling nvmet_auth_sq_free() on success. This means when the
host sends a follow up REPLACETLSPSK we return CONCAT_MISMATCH as the
check for !nvmet_queue_tls_keyid(req->sq) fails.

This patch ensures we don't free the TLS key on success as we might need
it again in the future.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Wilfred Mallawa <wilfred.mallawa@wdc.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvmet-tcp: Don't error if TLS is enabed on a reset
Alistair Francis [Tue, 2 Dec 2025 05:17:52 +0000 (15:17 +1000)] 
nvmet-tcp: Don't error if TLS is enabed on a reset

If the host sends a AUTH_Negotiate Message on the admin queue with
REPLACETLSPSK set then we expect and require a TLS connection and
shouldn't report an error if TLS is enabled.

This change only enforces the nvmet_queue_tls_keyid() check if we aren't
resetting the negotiation.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Wilfred Mallawa <wilfred.mallawa@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agocrypto: remove HKDF library
Eric Biggers [Mon, 2 Mar 2026 07:59:59 +0000 (23:59 -0800)] 
crypto: remove HKDF library

Remove crypto/hkdf.c, since it's no longer used.  Originally it had two
users, but now both of them just inline the needed HMAC computations
using the HMAC library APIs.  That ends up being better, since it
eliminates all the complexity and performance issues associated with the
crypto_shash abstraction and multi-step HMAC input formatting.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: remove selections of no-longer used crypto modules
Eric Biggers [Mon, 2 Mar 2026 07:59:58 +0000 (23:59 -0800)] 
nvme-auth: common: remove selections of no-longer used crypto modules

Now that nvme-auth uses the crypto library instead of crypto_shash,
remove obsolete selections from the NVME_AUTH kconfig option.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: remove nvme_auth_digest_name()
Eric Biggers [Mon, 2 Mar 2026 07:59:57 +0000 (23:59 -0800)] 
nvme-auth: common: remove nvme_auth_digest_name()

Since nvme_auth_digest_name() is no longer used, remove it and the
associated data from the hash_map array.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: target: use crypto library in nvmet_auth_ctrl_hash()
Eric Biggers [Mon, 2 Mar 2026 07:59:56 +0000 (23:59 -0800)] 
nvme-auth: target: use crypto library in nvmet_auth_ctrl_hash()

For the HMAC computation in nvmet_auth_ctrl_hash(), use the crypto
library instead of crypto_shash.  This is simpler, faster, and more
reliable.  Notably, this eliminates the crypto transformation object
allocation for every call, which was very slow.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: target: use crypto library in nvmet_auth_host_hash()
Eric Biggers [Mon, 2 Mar 2026 07:59:55 +0000 (23:59 -0800)] 
nvme-auth: target: use crypto library in nvmet_auth_host_hash()

For the HMAC computation in nvmet_auth_host_hash(), use the crypto
library instead of crypto_shash.  This is simpler, faster, and more
reliable.  Notably, this eliminates the crypto transformation object
allocation for every call, which was very slow.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: target: remove obsolete crypto_has_shash() checks
Eric Biggers [Mon, 2 Mar 2026 07:59:54 +0000 (23:59 -0800)] 
nvme-auth: target: remove obsolete crypto_has_shash() checks

Since nvme-auth is now doing its HMAC computations using the crypto
library, it's guaranteed that all the algorithms actually work.
Therefore, remove the crypto_has_shash() checks which are now obsolete.

However, the caller in nvmet_auth_negotiate() seems to have also been
relying on crypto_has_shash(nvme_auth_hmac_name(host_hmac_id)) to
validate the host_hmac_id.  Therefore, make it validate the ID more
directly by checking whether nvme_auth_hmac_hash_len() returns 0 or not.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: host: remove allocation of crypto_shash
Eric Biggers [Mon, 2 Mar 2026 07:59:53 +0000 (23:59 -0800)] 
nvme-auth: host: remove allocation of crypto_shash

Now that the crypto_shash that is being allocated in
nvme_auth_process_dhchap_challenge() and stored in the
struct nvme_dhchap_queue_context is no longer used, remove it.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: host: use crypto library in nvme_auth_dhchap_setup_ctrl_response()
Eric Biggers [Mon, 2 Mar 2026 07:59:52 +0000 (23:59 -0800)] 
nvme-auth: host: use crypto library in nvme_auth_dhchap_setup_ctrl_response()

For the HMAC computation in nvme_auth_dhchap_setup_ctrl_response(), use
the crypto library instead of crypto_shash.  This is simpler, faster,
and more reliable.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: host: use crypto library in nvme_auth_dhchap_setup_host_response()
Eric Biggers [Mon, 2 Mar 2026 07:59:51 +0000 (23:59 -0800)] 
nvme-auth: host: use crypto library in nvme_auth_dhchap_setup_host_response()

For the HMAC computation in nvme_auth_dhchap_setup_host_response(), use
the crypto library instead of crypto_shash.  This is simpler, faster,
and more reliable.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: use crypto library in nvme_auth_derive_tls_psk()
Eric Biggers [Mon, 2 Mar 2026 07:59:50 +0000 (23:59 -0800)] 
nvme-auth: common: use crypto library in nvme_auth_derive_tls_psk()

For the HKDF-Expand-Label computation in nvme_auth_derive_tls_psk(), use
the crypto library instead of crypto_shash and crypto/hkdf.c.

While this means the HKDF "helper" functions are no longer utilized,
they clearly weren't buying us much: it's simpler to just inline the
HMAC computations directly, and this code needs to be tested anyway.  (A
similar result was seen in fs/crypto/.  As a result, this eliminates the
last user of crypto/hkdf.c, which we'll be able to remove as well.)

As usual this is also a lot more efficient, eliminating the allocation
of a transformation object and multiple other dynamic allocations.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: use crypto library in nvme_auth_generate_digest()
Eric Biggers [Mon, 2 Mar 2026 07:59:49 +0000 (23:59 -0800)] 
nvme-auth: common: use crypto library in nvme_auth_generate_digest()

For the HMAC computation in nvme_auth_generate_digest(), use the crypto
library instead of crypto_shash.  This is simpler, faster, and more
reliable.  Notably, this eliminates the crypto transformation object
allocation for every call, which was very slow.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: use crypto library in nvme_auth_generate_psk()
Eric Biggers [Mon, 2 Mar 2026 07:59:48 +0000 (23:59 -0800)] 
nvme-auth: common: use crypto library in nvme_auth_generate_psk()

For the HMAC computation in nvme_auth_generate_psk(), use the crypto
library instead of crypto_shash.  This is simpler, faster, and more
reliable.  Notably, this eliminates the crypto transformation object
allocation for every call, which was very slow.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: use crypto library in nvme_auth_augmented_challenge()
Eric Biggers [Mon, 2 Mar 2026 07:59:47 +0000 (23:59 -0800)] 
nvme-auth: common: use crypto library in nvme_auth_augmented_challenge()

For the hash and HMAC computations in nvme_auth_augmented_challenge(),
use the crypto library instead of crypto_shash.  This is simpler,
faster, and more reliable.  Notably, this eliminates two crypto
transformation object allocations for every call, which was very slow.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: use crypto library in nvme_auth_transform_key()
Eric Biggers [Mon, 2 Mar 2026 07:59:46 +0000 (23:59 -0800)] 
nvme-auth: common: use crypto library in nvme_auth_transform_key()

For the HMAC computation in nvme_auth_transform_key(), use the crypto
library instead of crypto_shash.  This is simpler, faster, and more
reliable.  Notably, this eliminates the transformation object allocation
for every call, which was very slow.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: add HMAC helper functions
Eric Biggers [Mon, 2 Mar 2026 07:59:45 +0000 (23:59 -0800)] 
nvme-auth: common: add HMAC helper functions

Add some helper functions for computing HMAC-SHA256, HMAC-SHA384, or
HMAC-SHA512 values using the crypto library instead of crypto_shash.
These will enable some significant simplifications and performance
improvements in nvme-auth.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: explicitly verify psk_len == hash_len
Eric Biggers [Mon, 2 Mar 2026 07:59:44 +0000 (23:59 -0800)] 
nvme-auth: common: explicitly verify psk_len == hash_len

nvme_auth_derive_tls_psk() is always called with psk_len == hash_len.
And based on the comments above nvme_auth_generate_psk() and
nvme_auth_derive_tls_psk(), this isn't an implementation choice but
rather just the length the spec uses.  Add a check which makes this
explicit, so that when cleaning up nvme_auth_derive_tls_psk() we don't
have to retain support for arbitrary values of psk_len.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: rename nvme_auth_generate_key() to nvme_auth_parse_key()
Eric Biggers [Mon, 2 Mar 2026 07:59:43 +0000 (23:59 -0800)] 
nvme-auth: rename nvme_auth_generate_key() to nvme_auth_parse_key()

This function does not generate a key.  It parses the key from the
string that the caller passes in.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: add KUnit tests for TLS key derivation
Eric Biggers [Mon, 2 Mar 2026 07:59:42 +0000 (23:59 -0800)] 
nvme-auth: common: add KUnit tests for TLS key derivation

Unit-test the sequence of function calls that derive tls_psk, so that we
can be more confident that changes in the implementation don't break it.

Since the NVMe specification doesn't seem to include any test vectors
for this (nor does its description of the algorithm seem to match what
was actually implemented, for that matter), I just set the expected
values to the values that the code currently produces.  In the case
of SHA-512, nvme_auth_generate_digest() currently returns -EINVAL, so
for now the test tests for that too.  If it is later determined that
some other behavior is needed, the test can be updated accordingly.

Tested with:

    tools/testing/kunit/kunit.py run --kunitconfig drivers/nvme/common/

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: use proper argument types
Eric Biggers [Mon, 2 Mar 2026 07:59:41 +0000 (23:59 -0800)] 
nvme-auth: use proper argument types

For input parameters, use pointer to const.  This makes it easier to
understand which parameters are inputs and which are outputs.

In addition, consistently use char for strings and u8 for binary.  This
makes it easier to understand what is a string and what is binary data.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: common: constify static data
Eric Biggers [Mon, 2 Mar 2026 07:59:40 +0000 (23:59 -0800)] 
nvme-auth: common: constify static data

Fully constify the dhgroup_map and hash_map arrays.  Remove 'const' from
individual fields, as it is now redundant.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agonvme-auth: add NVME_AUTH_MAX_DIGEST_SIZE constant
Eric Biggers [Mon, 2 Mar 2026 07:59:39 +0000 (23:59 -0800)] 
nvme-auth: add NVME_AUTH_MAX_DIGEST_SIZE constant

Define a NVME_AUTH_MAX_DIGEST_SIZE constant and use it in the
appropriate places.

Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Keith Busch <kbusch@kernel.org>
2 months agoblock: fix bio_alloc_bioset slowpath GFP handling
Vasily Gorbik [Sun, 22 Mar 2026 02:35:10 +0000 (03:35 +0100)] 
block: fix bio_alloc_bioset slowpath GFP handling

bio_alloc_bioset() first strips __GFP_DIRECT_RECLAIM from the optimistic
fast allocation attempt with try_alloc_gfp(). If that fast path fails,
the slowpath checks saved_gfp to decide whether blocking allocation is
allowed, but then still calls mempool_alloc() with the stripped gfp mask.
That can lead to a NULL bio pointer being passed into bio_init().

Fix the slowpath by using saved_gfp for the bio and bvec mempool
allocations.

Fixes: b520c4eef83d ("block: split bio_alloc_bioset more clearly into a fast and slowpath")
Reported-by: syzbot+09ddb593eea76a158f42@syzkaller.appspotmail.com
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/p01.gc6e9ad5845ad.ttca29g@ub.hpns
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoublk: move cold paths out of __ublk_batch_dispatch() for icache efficiency
Ming Lei [Wed, 18 Mar 2026 01:41:12 +0000 (09:41 +0800)] 
ublk: move cold paths out of __ublk_batch_dispatch() for icache efficiency

Mark ublk_filter_unused_tags() as noinline since it is only called from
the unlikely(needs_filter) branch. Extract the error-handling block from
__ublk_batch_dispatch() into a new noinline ublk_batch_dispatch_fail()
function to keep the hot path compact and icache-friendly. This also
makes __ublk_batch_dispatch() more readable by separating the error
recovery logic from the normal dispatch flow.

Before: __ublk_batch_dispatch is ~1419 bytes
After:  __ublk_batch_dispatch is ~1090 bytes (-329 bytes, -23%)

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://patch.msgid.link/20260318014112.3125432-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoMerge tag 'md-7.1-20260323' of git://git.kernel.org/pub/scm/linux/kernel/git/mdraid...
Jens Axboe [Sun, 22 Mar 2026 19:37:45 +0000 (13:37 -0600)] 
Merge tag 'md-7.1-20260323' of git://git.kernel.org/pub/scm/linux/kernel/git/mdraid/linux into for-7.1/block

Pull MD changes from Yu Kuia:

"Bug Fixes:
 - md: suppress spurious superblock update error message for dm-raid
   (Chen Cheng)
 - md/raid1: fix the comparing region of interval tree (Xiao Ni)
 - md/raid10: fix deadlock with check operation and nowait requests
   (Josh Hunt)
 - md/raid5: skip 2-failure compute when other disk is R5_LOCKED
   (FengWei Shih)
 - md/md-llbitmap: raise barrier before state machine transition
   (Yu Kuai)
 - md/md-llbitmap: skip reading rdevs that are not in_sync (Yu Kuai)

 Improvements:
 - md/raid5: set chunk_sectors to enable full stripe I/O splitting
   (Yu Kuai)

 Cleanups:
 - md: remove unused mddev argument from export_rdev (Chen Cheng)
 - md/raid5: remove stale md_raid5_kick_device() declaration
   (Chen Cheng)
 - md/raid5: move handle_stripe() comment to correct location
   (Chen Cheng)"

* tag 'md-7.1-20260323' of git://git.kernel.org/pub/scm/linux/kernel/git/mdraid/linux:
  md: remove unused mddev argument from export_rdev
  md/raid5: move handle_stripe() comment to correct location
  md/raid5: remove stale md_raid5_kick_device() declaration
  md/raid1: fix the comparing region of interval tree
  md/raid5: skip 2-failure compute when other disk is R5_LOCKED
  md/md-llbitmap: raise barrier before state machine transition
  md/md-llbitmap: skip reading rdevs that are not in_sync
  md/raid5: set chunk_sectors to enable full stripe I/O splitting
  md/raid10: fix deadlock with check operation and nowait requests
  md: suppress spurious superblock update error message for dm-raid

2 months agomd: remove unused mddev argument from export_rdev
Chen Cheng [Wed, 4 Mar 2026 11:14:17 +0000 (19:14 +0800)] 
md: remove unused mddev argument from export_rdev

The mddev argument in export_rdev() is never used. Remove it to
simplify callers.

Signed-off-by: Chen Cheng <chencheng@fnnas.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Link: https://lore.kernel.org/linux-raid/20260304111417.20777-1-chencheng@fnnas.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
2 months agomd/raid5: move handle_stripe() comment to correct location
Chen Cheng [Wed, 4 Mar 2026 11:10:01 +0000 (19:10 +0800)] 
md/raid5: move handle_stripe() comment to correct location

Move the handle_stripe() documentation comment from above
analyse_stripe() to directly above handle_stripe() where it belongs.

Signed-off-by: Chen Cheng <chencheng@fnnas.com>
Reviewed-by: Yu Kuai <yukuai@fnnas.com>
Link: https://lore.kernel.org/linux-raid/20260304111001.15767-1-chencheng@fnnas.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
2 months agomd/raid5: remove stale md_raid5_kick_device() declaration
Chen Cheng [Wed, 4 Mar 2026 11:09:18 +0000 (19:09 +0800)] 
md/raid5: remove stale md_raid5_kick_device() declaration

Remove the unused md_raid5_kick_device() declaration from raid5.h -
no definition exists for this function.

Signed-off-by: Chen Cheng <chencheng@fnnas.com>
Reviewed-by: Yu Kuai <yukuai@fnnas.com>
Link: https://lore.kernel.org/linux-raid/20260304110919.15071-1-chencheng@fnnas.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
2 months agomd/raid1: fix the comparing region of interval tree
Xiao Ni [Thu, 5 Mar 2026 01:18:33 +0000 (09:18 +0800)] 
md/raid1: fix the comparing region of interval tree

Interval tree uses [start, end] as a region which stores in the tree.
In raid1, it uses the wrong end value. For example:
bio(A,B) is too big and needs to be split to bio1(A,C-1), bio2(C,B).
The region of bio1 is [A,C] and the region of bio2 is [C,B]. So bio1 and
bio2 overlap which is not right.

Fix this problem by using right end value of the region.

Fixes: d0d2d8ba0494 ("md/raid1: introduce wait_for_serialization")
Signed-off-by: Xiao Ni <xni@redhat.com>
Link: https://lore.kernel.org/linux-raid/20260305011839.5118-2-xni@redhat.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
2 months agomd/raid5: skip 2-failure compute when other disk is R5_LOCKED
FengWei Shih [Thu, 19 Mar 2026 05:33:51 +0000 (13:33 +0800)] 
md/raid5: skip 2-failure compute when other disk is R5_LOCKED

When skip_copy is enabled on a doubly-degraded RAID6, a device that is
being written to will be in R5_LOCKED state with R5_UPTODATE cleared.
If a new read triggers fetch_block() while the write is still in
flight, the 2-failure compute path may select this locked device as a
compute target because it is not R5_UPTODATE.

Because skip_copy makes the device page point directly to the bio page,
reconstructing data into it might be risky. Also, since the compute
marks the device R5_UPTODATE, it triggers WARN_ON in ops_run_io()
which checks that R5_SkipCopy and R5_UPTODATE are not both set.

This can be reproduced by running small-range concurrent read/write on
a doubly-degraded RAID6 with skip_copy enabled, for example:

  mdadm -C /dev/md0 -l6 -n6 -R -f /dev/loop[0-3] missing missing
  echo 1 > /sys/block/md0/md/skip_copy
  fio --filename=/dev/md0 --rw=randrw --bs=4k --numjobs=8 \
      --iodepth=32 --size=4M --runtime=30 --time_based --direct=1

Fix by checking R5_LOCKED before proceeding with the compute. The
compute will be retried once the lock is cleared on IO completion.

Signed-off-by: FengWei Shih <dannyshih@synology.com>
Reviewed-by: Yu Kuai <yukuai@fnnas.com>
Link: https://lore.kernel.org/linux-raid/20260319053351.3676794-1-dannyshih@synology.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
2 months agoblock: partitions: Replace pp_buf with struct seq_buf
Kees Cook [Sat, 21 Mar 2026 00:48:44 +0000 (17:48 -0700)] 
block: partitions: Replace pp_buf with struct seq_buf

In preparation for removing the strlcat API[1], replace the char *pp_buf
with a struct seq_buf, which tracks the current write position and
remaining space internally. This allows for:

- Direct use of seq_buf_printf() in place of snprintf()+strlcat()
  pairs, eliminating local tmp buffers throughout.
- Adjacent strlcat() calls that build strings piece-by-piece
  (e.g., strlcat("["); strlcat(name); strlcat("]")) to be collapsed
  into single seq_buf_printf() calls.
- Simpler call sites: seq_buf_puts() takes only the buffer and string,
  with no need to pass PAGE_SIZE at every call.

The backing buffer allocation is unchanged (__get_free_page), and the
output path uses seq_buf_str() to NUL-terminate before passing to
printk().

Link: https://github.com/KSPP/linux/issues/370
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Josh Law <objecting@objecting.org>
Signed-off-by: Kees Cook <kees@kernel.org>
Reviewed-by: Josh Law <objecting@objecting.org>
Link: https://patch.msgid.link/20260321004840.work.670-kees@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoscsi: bsg: add io_uring passthrough handler
Yang Xiuwei [Tue, 17 Mar 2026 07:22:26 +0000 (15:22 +0800)] 
scsi: bsg: add io_uring passthrough handler

Implement the SCSI-specific io_uring command handler for BSG using
struct bsg_uring_cmd.

The handler builds a SCSI request from the io_uring command, maps user
buffers (including fixed buffers), and completes asynchronously via a
request end_io callback and task_work. Completion returns a 32-bit
status and packed residual/sense information via CQE res and res2, and
supports IO_URING_F_NONBLOCK.

Signed-off-by: Yang Xiuwei <yangxiuwei@kylinos.cn>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20260317072226.2598233-4-yangxiuwei@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agobsg: add io_uring command support to generic layer
Yang Xiuwei [Tue, 17 Mar 2026 07:22:25 +0000 (15:22 +0800)] 
bsg: add io_uring command support to generic layer

Add an io_uring command handler to the generic BSG layer. The new
.uring_cmd file operation validates io_uring features and delegates
handling to a per-queue bsg_uring_cmd_fn callback.

Extend bsg_register_queue() so transport drivers can register both
sg_io and io_uring command handlers.

Signed-off-by: Yang Xiuwei <yangxiuwei@kylinos.cn>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20260317072226.2598233-3-yangxiuwei@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agobsg: add bsg_uring_cmd uapi structure
Yang Xiuwei [Tue, 17 Mar 2026 07:22:24 +0000 (15:22 +0800)] 
bsg: add bsg_uring_cmd uapi structure

Add the bsg_uring_cmd structure to the BSG UAPI header to support
io_uring-based SCSI passthrough operations via IORING_OP_URING_CMD.

Signed-off-by: Yang Xiuwei <yangxiuwei@kylinos.cn>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://patch.msgid.link/20260317072226.2598233-2-yangxiuwei@kylinos.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: reject zero length in bio_add_page()
Qu Wenruo [Tue, 17 Mar 2026 08:06:34 +0000 (18:36 +1030)] 
block: reject zero length in bio_add_page()

The function bio_add_page() returns the number of bytes added to the
bio, and if that failed it should return 0.

However there is a special quirk, if a caller is passing a page with
length 0, that function will always return 0 but with different results:

- The page is added to the bio
  If there is enough bvec slot or the folio can be merged with the last
  bvec.

  The return value 0 is just the length passed in, which is also 0.

- The page is not added to the bio
  If the page is not mergeable with the last bvec, or there is no bvec
  slot available.

  The return value 0 means page is not added into the bio.

Unfortunately the caller is not able to distinguish the above two cases,
and will treat the 0 return value as page addition failure.

In that case, this can lead to the double releasing of the last page:

- By the bio cleanup
  Which normally goes through every page of the bio, including the last
  page which is added into the bio.

- By the caller
  Which believes the page is not added into the bio, thus would manually
  release the page.

I do not think anyone should call bio_add_folio()/bio_add_page() with zero
length, but idiots like me can still show up.

So add an extra WARN_ON_ONCE() check for zero length and rejects it
early to avoid double freeing.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Link: https://patch.msgid.link/bc2223c080f38d0b63f968f605c918181c840f40.1773734749.git.wqu@suse.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblk-mq: make blk_mq_hw_ctx_sysfs_entry instances const
Thomas Weißschuh [Mon, 16 Mar 2026 22:43:54 +0000 (23:43 +0100)] 
blk-mq: make blk_mq_hw_ctx_sysfs_entry instances const

The blk_mq_hw_ctx_sysfs_entry structures are never modified,
mark them as const.

Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Link: https://patch.msgid.link/20260316-b4-sysfs-const-attr-block-v1-4-a35d73b986b0@weissschuh.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblk-crypto: make blk_crypto_attr instances const
Thomas Weißschuh [Mon, 16 Mar 2026 22:43:53 +0000 (23:43 +0100)] 
blk-crypto: make blk_crypto_attr instances const

The blk_crypto_attrs structures are never modified, mark them as const.

Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: John Garry <john.g.garry@oracle.com>>
Link: https://patch.msgid.link/20260316-b4-sysfs-const-attr-block-v1-3-a35d73b986b0@weissschuh.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: ia-ranges: make blk_ia_range_sysfs_entry instances const
Thomas Weißschuh [Mon, 16 Mar 2026 22:43:52 +0000 (23:43 +0100)] 
block: ia-ranges: make blk_ia_range_sysfs_entry instances const

The blk_ia_range_sysfs_entry structures are never modified,
mark them as const.

Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Link: https://patch.msgid.link/20260316-b4-sysfs-const-attr-block-v1-2-a35d73b986b0@weissschuh.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: make queue_sysfs_entry instances const
Thomas Weißschuh [Mon, 16 Mar 2026 22:43:51 +0000 (23:43 +0100)] 
block: make queue_sysfs_entry instances const

The queue_sysfs_entry structures are never modified, mark them as const.

Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Link: https://patch.msgid.link/20260316-b4-sysfs-const-attr-block-v1-1-a35d73b986b0@weissschuh.net
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: remove bvec_free
Christoph Hellwig [Mon, 16 Mar 2026 16:11:31 +0000 (17:11 +0100)] 
block: remove bvec_free

bvec_free is only called by bio_free, so inline it there.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> -ck
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Link: https://patch.msgid.link/20260316161144.1607877-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: split bio_alloc_bioset more clearly into a fast and slowpath
Christoph Hellwig [Mon, 16 Mar 2026 16:11:30 +0000 (17:11 +0100)] 
block: split bio_alloc_bioset more clearly into a fast and slowpath

bio_alloc_bioset tries non-waiting slab allocations first for the bio and
bvec array, but does so in a somewhat convoluted way.

Restructure the function so that it first open codes these slab
allocations, and then falls back to the mempools with the original
gfp mask.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> -ck
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Link: https://patch.msgid.link/20260316161144.1607877-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: mark bvec_{alloc,free} static
Christoph Hellwig [Mon, 16 Mar 2026 16:11:29 +0000 (17:11 +0100)] 
block: mark bvec_{alloc,free} static

Only used in bio.c these days.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> -ck
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Link: https://patch.msgid.link/20260316161144.1607877-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agomd/md-llbitmap: raise barrier before state machine transition
Yu Kuai [Mon, 23 Feb 2026 02:40:35 +0000 (10:40 +0800)] 
md/md-llbitmap: raise barrier before state machine transition

Move the barrier raise operation before calling llbitmap_state_machine()
in both llbitmap_start_write() and llbitmap_start_discard(). This
ensures the barrier is in place before any state transitions occur,
preventing potential race conditions where the state machine could
complete before the barrier is properly raised.

Cc: stable@vger.kernel.org
Fixes: 5ab829f1971d ("md/md-llbitmap: introduce new lockless bitmap")
Link: https://lore.kernel.org/linux-raid/20260223024038.3084853-3-yukuai@fnnas.com
Signed-off-by: Yu Kuai <yukuai@fnnas.com>
2 months agomd/md-llbitmap: skip reading rdevs that are not in_sync
Yu Kuai [Mon, 23 Feb 2026 02:40:34 +0000 (10:40 +0800)] 
md/md-llbitmap: skip reading rdevs that are not in_sync

When reading bitmap pages from member disks, the code iterates through
all rdevs and attempts to read from the first available one. However,
it only checks for raid_disk assignment and Faulty flag, missing the
In_sync flag check.

This can cause bitmap data to be read from spare disks that are still
being rebuilt and don't have valid bitmap information yet. Reading
stale or uninitialized bitmap data from such disks can lead to
incorrect dirty bit tracking, potentially causing data corruption
during recovery or normal operation.

Add the In_sync flag check to ensure bitmap pages are only read from
fully synchronized member disks that have valid bitmap data.

Cc: stable@vger.kernel.org
Fixes: 5ab829f1971d ("md/md-llbitmap: introduce new lockless bitmap")
Link: https://lore.kernel.org/linux-raid/20260223024038.3084853-2-yukuai@fnnas.com
Signed-off-by: Yu Kuai <yukuai@fnnas.com>
2 months agomd/raid5: set chunk_sectors to enable full stripe I/O splitting
Yu Kuai [Mon, 23 Feb 2026 03:58:34 +0000 (11:58 +0800)] 
md/raid5: set chunk_sectors to enable full stripe I/O splitting

Set chunk_sectors to the full stripe width (io_opt) so that the block
layer splits I/O at full stripe boundaries. This ensures that large
writes are aligned to full stripes, avoiding the read-modify-write
overhead that occurs with partial stripe writes in RAID-5/6.

When chunk_sectors is set, the block layer's bio splitting logic in
get_max_io_size() uses blk_boundary_sectors_left() to limit I/O size
to the boundary. This naturally aligns split bios to full stripe
boundaries, enabling more efficient full stripe writes.

Test results with 24-disk RAID5 (chunk_size=64k):
  dd if=/dev/zero of=/dev/md0 bs=10M oflag=direct
  Before: 461 MB/s
  After:  520 MB/s  (+12.8%)

Link: https://lore.kernel.org/linux-raid/20260223035834.3132498-1-yukuai@fnnas.com
Suggested-by: Christoph Hellwig <hch@infradead.org>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Yu Kuai <yukuai@fnnas.com>
2 months agomd/raid10: fix deadlock with check operation and nowait requests
Josh Hunt [Tue, 3 Mar 2026 00:56:19 +0000 (19:56 -0500)] 
md/raid10: fix deadlock with check operation and nowait requests

When an array check is running it will raise the barrier at which point
normal requests will become blocked and increment the nr_pending value to
signal there is work pending inside of wait_barrier(). NOWAIT requests
do not block and so will return immediately with an error, and additionally
do not increment nr_pending in wait_barrier(). Upstream change commit
43806c3d5b9b ("raid10: cleanup memleak at raid10_make_request") added a
call to raid_end_bio_io() to fix a memory leak when NOWAIT requests hit
this condition. raid_end_bio_io() eventually calls allow_barrier() and
it will unconditionally do an atomic_dec_and_test(&conf->nr_pending) even
though the corresponding increment on nr_pending didn't happen in the
NOWAIT case.

This can be easily seen by starting a check operation while an application
is doing nowait IO on the same array. This results in a deadlocked state
due to nr_pending value underflowing and so the md resync thread gets stuck
waiting for nr_pending to == 0.

Output of r10conf state of the array when we hit this condition:

crash> struct r10conf
barrier = 1,
        nr_pending = {
          counter = -41
        },
        nr_waiting = 15,
        nr_queued = 0,

Example of md_sync thread stuck waiting on raise_barrier() and other
requests stuck in wait_barrier():

md1_resync
[<0>] raise_barrier+0xce/0x1c0
[<0>] raid10_sync_request+0x1ca/0x1ed0
[<0>] md_do_sync+0x779/0x1110
[<0>] md_thread+0x90/0x160
[<0>] kthread+0xbe/0xf0
[<0>] ret_from_fork+0x34/0x50
[<0>] ret_from_fork_asm+0x1a/0x30

kworker/u1040:2+flush-253:4
[<0>] wait_barrier+0x1de/0x220
[<0>] regular_request_wait+0x30/0x180
[<0>] raid10_make_request+0x261/0x1000
[<0>] md_handle_request+0x13b/0x230
[<0>] __submit_bio+0x107/0x1f0
[<0>] submit_bio_noacct_nocheck+0x16f/0x390
[<0>] ext4_io_submit+0x24/0x40
[<0>] ext4_do_writepages+0x254/0xc80
[<0>] ext4_writepages+0x84/0x120
[<0>] do_writepages+0x7a/0x260
[<0>] __writeback_single_inode+0x3d/0x300
[<0>] writeback_sb_inodes+0x1dd/0x470
[<0>] __writeback_inodes_wb+0x4c/0xe0
[<0>] wb_writeback+0x18b/0x2d0
[<0>] wb_workfn+0x2a1/0x400
[<0>] process_one_work+0x149/0x330
[<0>] worker_thread+0x2d2/0x410
[<0>] kthread+0xbe/0xf0
[<0>] ret_from_fork+0x34/0x50
[<0>] ret_from_fork_asm+0x1a/0x30

Fixes: 43806c3d5b9b ("raid10: cleanup memleak at raid10_make_request")
Cc: stable@vger.kernel.org
Signed-off-by: Josh Hunt <johunt@akamai.com>
Link: https://lore.kernel.org/linux-raid/20260303005619.1352958-1-johunt@akamai.com
Signed-off-by: Yu Kuai <yukuai@fnnas.com>
2 months agomd: suppress spurious superblock update error message for dm-raid
Chen Cheng [Tue, 10 Feb 2026 13:38:47 +0000 (21:38 +0800)] 
md: suppress spurious superblock update error message for dm-raid

dm-raid has external metadata management (mddev->external = 1) and
no persistent superblock (mddev->persistent = 0). For these arrays,
there's no superblock to update, so the error message is spurious.

The error appears as:
md_update_sb: can't update sb for read-only array md0

Fixes: 8c9e376b9d1a ("md: warn about updating super block failure")
Reported-by: Tj <tj.iam.tj@proton.me>
Closes: https://lore.kernel.org/all/20260128082430.96788-1-tj.iam.tj@proton.me/
Signed-off-by: Chen Cheng <chencheng@fnnas.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Link: https://lore.kernel.org/linux-raid/20260210133847.269986-1-chencheng@fnnas.com
Signed-off-by: Yu Kuai <yukuai@fnnas.com>
2 months agoublk: report BLK_SPLIT_INTERVAL_CAPABLE
Caleb Sander Mateos [Fri, 13 Mar 2026 14:47:01 +0000 (07:47 -0700)] 
ublk: report BLK_SPLIT_INTERVAL_CAPABLE

The ublk driver doesn't access request integrity buffers directly, it
only copies them to/from the ublk server in ublk_copy_user_integrity().
ublk_copy_user_integrity() uses bio_for_each_integrity_vec() to walk all
the integrity segments. ublk devices are therefore capable of handling
requests with integrity intervals split across segments. Set
BLK_SPLIT_INTERVAL_CAPABLE in the struct blk_integrity flags for ublk
devices to opt out of the integrity-interval dma_alignment limit.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/20260313144701.1221652-3-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblk-integrity: support arbitrary buffer alignment
Keith Busch [Fri, 13 Mar 2026 14:47:00 +0000 (07:47 -0700)] 
blk-integrity: support arbitrary buffer alignment

A bio segment may have partial interval block data with the rest
continuing into the next segments because direct-io data payloads only
need to align in memory to the device's DMA limits.

At the same time, the protection information may also be split in
multiple segments. The most likely way that may happen is if two
requests merge, or if we're directly using the io_uring user metadata.
The generate/verify, however, only ever accessed the first bip_vec.

Further, it may be possible to unalign the protection fields from the
user space buffer, or if there are odd additional opaque bytes in front
or in back of the protection information metadata region.

Change up the iteration to allow spanning multiple segments. This patch
is mostly a re-write of the protection information handling to allow any
arbitrary alignments, so it's probably easier to review the end result
rather than the diff.

Many controllers are not able to handle interval data composed of
multiple segments when PI is used, so this patch introduces a new
integrity limit that a low level driver can set to notify that it is
capable, default to false. The nvme driver is the first one to enable it
in this patch. Everyone else will force DMA alignment to the logical
block size as before to ensure interval data is always aligned within a
single segment.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Link: https://patch.msgid.link/20260313144701.1221652-2-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblk-cgroup: wait for blkcg cleanup before initializing new disk
Ming Lei [Wed, 11 Mar 2026 03:28:37 +0000 (11:28 +0800)] 
blk-cgroup: wait for blkcg cleanup before initializing new disk

When a queue is shared across disk rebind (e.g., SCSI unbind/bind), the
previous disk's blkcg state is cleaned up asynchronously via
disk_release() -> blkcg_exit_disk(). If the new disk's blkcg_init_disk()
runs before that cleanup finishes, we may overwrite q->root_blkg while
the old one is still alive, and radix_tree_insert() in blkg_create()
fails with -EEXIST because the old blkg entries still occupy the same
queue id slot in blkcg->blkg_tree. This causes the sd probe to fail
with -ENOMEM.

Fix it by waiting in blkcg_init_disk() for root_blkg to become NULL,
which indicates the previous disk's blkcg cleanup has completed.

Fixes: 1059699f87eb ("block: move blkcg initialization/destroy into disk allocation/release handler")
Cc: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260311032837.2368714-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: clear BIO_QOS flags in blk_steal_bios()
Chaitanya Kulkarni [Thu, 26 Feb 2026 03:12:43 +0000 (19:12 -0800)] 
block: clear BIO_QOS flags in blk_steal_bios()

When a bio goes through the rq_qos infrastructure on a path's request
queue, it gets BIO_QOS_THROTTLED or BIO_QOS_MERGED flags set. These
flags indicate that rq_qos_done_bio() should be called on completion
to update rq_qos accounting.

During path failover in nvme_failover_req(), the bio's bi_bdev is
redirected from the failed path's disk to the multipath head's disk
via bio_set_dev(). However, the BIO_QOS flags are not cleared.

When the bio eventually completes (either successfully via a new path
or with an error via bio_io_error()), rq_qos_done_bio() checks for
these flags and calls __rq_qos_done_bio(q->rq_qos, bio) where q is
obtained from the bio's current bi_bdev - which is now the multipath
head's queue, not the original path's queue.

The multipath head's queue does not have rq_qos enabled (q->rq_qos is
NULL), but the code assumes that if BIO_QOS_* flags are set, q->rq_qos
must be valid.

This breaks when a bio is moved between queues during NVMe multipath
failover, leading to a NULL pointer dereference.

Execution Context timeline :-

   * =====> dd process context
   [USER] dd process
     [SYSCALL] write() - dd process context
       submit_bio()
       nvme_ns_head_submit_bio() - path selection
       blk_mq_submit_bio()  #### QOS FLAGS SET HERE

        [USER] dd waits or returns

          ==== I/O in flight on NVMe hardware =====

   ===== End of submission path ====
   ------------------------------------------------------

   * dd ====> Interrupt context;
   [IRQ] NVMe completion interrupt
       nvme_irq()
        nvme_complete_rq()
         nvme_failover_req() ### BIO MOVED TO HEAD
            spin_lock_irqsave (atomic section)
            bio_set_dev() changes bi_bdev
            ### BUG: QOS flags NOT cleared
            kblockd_schedule_work()

   * Interrupt context =====> kblockd workqueue
   [WQ] kblockd workqueue - kworker process
       nvme_requeue_work()
        submit_bio_noacct()
         nvme_ns_head_submit_bio()
          nvme_find_path() returns NULL
           bio_io_error()
            bio_endio()
             rq_qos_done_bio()  ### CRASH ###

   KERNEL PANIC / OOPS

Crash from blktests nvme/058 (rapid namespace remapping):

[ 1339.636033] BUG: kernel NULL pointer dereference, address: 0000000000000000
[ 1339.641025] nvme nvme4: rescanning namespaces.
[ 1339.642064] #PF: supervisor read access in kernel mode
[ 1339.642067] #PF: error_code(0x0000) - not-present page
[ 1339.642070] PGD 0 P4D 0
[ 1339.642073] Oops: Oops: 0000 [#1] SMP NOPTI
[ 1339.642078] CPU: 35 UID: 0 PID: 4579 Comm: kworker/35:2H
               Tainted: G   O     N  6.17.0-rc3nvme+ #5 PREEMPT(voluntary)
[ 1339.642084] Tainted: [O]=OOT_MODULE, [N]=TEST
[ 1339.673446] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
           BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
[ 1339.682359] Workqueue: kblockd nvme_requeue_work [nvme_core]
[ 1339.686613] RIP: 0010:__rq_qos_done_bio+0xd/0x40
[ 1339.690161] Code: 75 dd 5b 5d 41 5c c3 cc cc cc cc 66 90 90 90 90 90 90 90
                     90 90 90 90 90 90 90 90 90 90 0f 1f 44 00 00 55 48 89 f5
             53 48 89 fb <48> 8b 03 48 8b 40 30 48 85 c0 74 0b 48 89 ee
             48 89 df ff d0 0f 1f
[ 1339.703691] RSP: 0018:ffffc900066f3c90 EFLAGS: 00010202
[ 1339.706844] RAX: ffff888148b9ef00 RBX: 0000000000000000 RCX: 0000000000000000
[ 1339.711136] RDX: 00000000000001c0 RSI: ffff8882aaab8a80 RDI: 0000000000000000
[ 1339.715691] RBP: ffff8882aaab8a80 R08: 0000000000000000 R09: 0000000000000000
[ 1339.720472] R10: 0000000000000000 R11: fefefefefefefeff R12: ffff8882aa3b6010
[ 1339.724650] R13: 0000000000000000 R14: ffff8882338bcef0 R15: ffff8882aa3b6020
[ 1339.729029] FS:  0000000000000000(0000) GS:ffff88985c0cf000(0000) knlGS:0000000000000000
[ 1339.734525] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1339.738563] CR2: 0000000000000000 CR3: 0000000111045000 CR4: 0000000000350ef0
[ 1339.742750] DR0: ffffffff845ccbec DR1: ffffffff845ccbed DR2: ffffffff845ccbee
[ 1339.745630] DR3: ffffffff845ccbef DR6: 00000000ffff0ff0 DR7: 0000000000000600
[ 1339.748488] Call Trace:
[ 1339.749512]  <TASK>
[ 1339.750449]  bio_endio+0x71/0x2e0
[ 1339.751833]  nvme_ns_head_submit_bio+0x290/0x320 [nvme_core]
[ 1339.754073]  __submit_bio+0x222/0x5e0
[ 1339.755623]  ? rcu_is_watching+0xd/0x40
[ 1339.757201]  ? submit_bio_noacct_nocheck+0x131/0x370
[ 1339.759210]  submit_bio_noacct_nocheck+0x131/0x370
[ 1339.761189]  ? submit_bio_noacct+0x20/0x620
[ 1339.762849]  nvme_requeue_work+0x4b/0x60 [nvme_core]
[ 1339.764828]  process_one_work+0x20e/0x630
[ 1339.766528]  worker_thread+0x184/0x330
[ 1339.768129]  ? __pfx_worker_thread+0x10/0x10
[ 1339.769942]  kthread+0x10a/0x250
[ 1339.771263]  ? __pfx_kthread+0x10/0x10
[ 1339.772776]  ? __pfx_kthread+0x10/0x10
[ 1339.774381]  ret_from_fork+0x273/0x2e0
[ 1339.775948]  ? __pfx_kthread+0x10/0x10
[ 1339.777504]  ret_from_fork_asm+0x1a/0x30
[ 1339.779163]  </TASK>

Fix this by clearing both BIO_QOS_THROTTLED and BIO_QOS_MERGED flags
when bios are redirected to the multipath head in nvme_failover_req().
This is consistent with the existing code that clears REQ_POLLED and
REQ_NOWAIT flags when the bio changes queues.

Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260226031243.87200-3-kch@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: move bio queue-transition flag fixups into blk_steal_bios()
Chaitanya Kulkarni [Thu, 26 Feb 2026 03:12:42 +0000 (19:12 -0800)] 
block: move bio queue-transition flag fixups into blk_steal_bios()

blk_steal_bios() transfers bios from a request to a bio_list when the
request is requeued to a different queue. The NVMe multipath failover
path (nvme_failover_req) currently open-codes clearing of REQ_POLLED,
bi_cookie, and REQ_NOWAIT on each bio before calling blk_steal_bios().

Move these fixups into blk_steal_bios() itself so that any caller
automatically gets correct flag state when bios cross queue boundaries.
Simplify nvme_failover_req() accordingly.

Signed-off-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://patch.msgid.link/20260226031243.87200-2-kch@nvidia.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoMerge branch 'for-7.1/block-integrity' into for-7.1/block
Jens Axboe [Mon, 9 Mar 2026 20:30:14 +0000 (14:30 -0600)] 
Merge branch 'for-7.1/block-integrity' into for-7.1/block

Merge in integrity changes which are also landing in the VFS tree as
dependencies for fs related changes.

* for-7.1/block-integrity:
  block: pass a maxlen argument to bio_iov_iter_bounce
  block: add fs_bio_integrity helpers
  block: make max_integrity_io_size public
  block: prepare generation / verification helpers for fs usage
  block: add a bdev_has_integrity_csum helper
  block: factor out a bio_integrity_setup_default helper
  block: factor out a bio_integrity_action helper

2 months agoblock: remove bdev_nonrot()
Damien Le Moal [Thu, 26 Feb 2026 07:54:48 +0000 (16:54 +0900)] 
block: remove bdev_nonrot()

bdev_nonrot() is simply the negative return value of bdev_rot().
So replace all call sites of bdev_nonrot() with calls to bdev_rot()
and remove bdev_nonrot().

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: Correct comments on bio_alloc_clone() and bio_init_clone()
John Garry [Fri, 27 Feb 2026 10:44:38 +0000 (10:44 +0000)] 
block: Correct comments on bio_alloc_clone() and bio_init_clone()

Correct the comments that the cloned bio must be freed before the memory
pointed to by @bio_src->bi_io_vecs (is freed).

Christoph Hellwig contributed most the of the update wording.

Signed-off-by: John Garry <john.g.garry@oracle.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoDocumentation: ABI: stable: document the zoned_qd1_writes attribute
Damien Le Moal [Fri, 27 Feb 2026 13:19:51 +0000 (22:19 +0900)] 
Documentation: ABI: stable: document the zoned_qd1_writes attribute

Update the documentation file Documentation/ABI/stable/sysfs-block to
describe the zoned_qd1_writes sysfs queue attribute file.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: default to QD=1 writes for blk-mq rotational zoned devices
Damien Le Moal [Fri, 27 Feb 2026 13:19:50 +0000 (22:19 +0900)] 
block: default to QD=1 writes for blk-mq rotational zoned devices

For blk-mq rotational zoned block devices (e.g. SMR HDDs), default to
having zone write plugging limit write operations to a maximum queue
depth of 1 for all zones. This significantly reduce write seek overhead
and improves SMR HDD write throughput.

For remotely connected disks with a very high network latency this
features might not be useful. However, remotely connected zoned devices
are rare at the moment, and we cannot know the round trip latency to
pick a good default for network attached devices. System administrators
can however disable this feature in that case.

For BIO based (non blk-mq) rotational zoned block devices, the device
driver (e.g. a DM target driver) can directly set an appropriate
default.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: allow submitting all zone writes from a single context
Damien Le Moal [Fri, 27 Feb 2026 13:19:49 +0000 (22:19 +0900)] 
block: allow submitting all zone writes from a single context

In order to maintain sequential write patterns per zone with zoned block
devices, zone write plugging issues only a single write BIO per zone at
any time. This works well but has the side effect that when large
sequential write streams are issued by the user and these streams cross
zone boundaries, the device ends up receiving a discontiguous set of
write commands for different zones. The same also happens when a user
writes simultaneously at high queue depth multiple zones: the device
does not see all sequential writes per zone and receives discontiguous
writes to different zones. While this does not affect the performance of
solid state zoned block devices, when using an SMR HDD, this pattern
change from sequential writes to discontiguous writes to different zones
significantly increases head seek which results in degraded write
throughput.

In order to reduce this seek overhead for rotational media devices,
introduce a per disk zone write plugs kernel thread to issue all write
BIOs to zones. This single zone write issuing context is enabled for
any zoned block device that has a request queue flagged with the new
QUEUE_ZONED_QD1_WRITES flag.

The flag QUEUE_ZONED_QD1_WRITES is visible as the sysfs queue attribute
zoned_qd1_writes for zoned devices. For regular block devices, this
attribute is not visible. For zoned block devices, a user can override
the default value set to force the global write maximum queue depth of
1 for a zoned block device, or clear this attribute to fallback to the
default behavior of zone write plugging which limits writes to QD=1 per
sequential zone.

Writing to a zoned block device flagged with QUEUE_ZONED_QD1_WRITES is
implemented using a list of zone write plugs that have a non-empty BIO
list. Listed zone write plugs are processed by the disk zone write plugs
worker kthread in FIFO order, and all BIOs of a zone write plug are all
processed before switching to the next listed zone write plug. A newly
submitted BIO for a non-FULL zone write plug that is not yet listed
causes the addition of the zone write plug at the end of the disk list
of zone write plugs.

Since the write BIOs queued in a zone write plug BIO list are
necessarilly sequential, for rotational media, using the single zone
write plugs kthread to issue all BIOs maintains a sequential write
pattern and thus reduces seek overhead and improves write throughput.
This processing essentially result in always writing to HDDs at QD=1,
which is not an issue for HDDs operating with write caching enabled.
Performance with write cache disabled is also not degraded thanks to
the efficient write handling of modern SMR HDDs.

A disk list of zone write plugs is defined using the new struct gendisk
zone_wplugs_list, and accesses to this list is protected using the
zone_wplugs_list_lock spinlock.  The per disk kthread
(zone_wplugs_worker) code is implemented by the function
disk_zone_wplugs_worker(). A reference on listed zone write plugs is
always held until all BIOs of the zone write plug are processed by the
worker kthread. BIO issuing at QD=1 is driven using a completion
structure (zone_wplugs_worker_bio_done) and calls to blk_io_wait().

With this change, performance when sequentially writing the zones of a
30 TB SMR SATA HDD connected to an AHCI adapter changes as follows
(1MiB direct I/Os, results in MB/s unit):

                    +--------------------+
    |   Write BW (MB/s)  |
 +------------------+----------+---------+
 | Sequential write | Baseline | Patched |
 |  Queue Depth     | 6.19-rc8 |         |
 +------------------+----------+---------+
 | 1                | 244      | 245     |
 | 2                | 244      | 245     |
 | 4                | 245      | 245     |
 | 8                | 242      | 245     |
 | 16               | 222      | 246     |
 | 32               | 211      | 245     |
 | 64               | 193      | 244     |
 | 128              | 112      | 246     |
 +------------------+----------+---------+

With the current code (baseline), as the sequential write stream crosses
a zone boundary, higher queue depth creates a gap between the
last IO to the previous zone and the first IOs to the following zones,
causing head seeks and degrading performance. Using the disk zone
write plugs worker thread, this pattern disappears and the maximum
throughput of the drive is maintained, leading to over 100%
improvements in throughput for high queue depth write.

Using 16 fio jobs all writing to randomly chosen zones at QD=32 with 1
MiB direct IOs, write throughput also increases significantly.

                    +--------------------+
    |   Write BW (MB/s)  |
 +------------------+----------+---------+
 |   Random write   | Baseline | Patched |
 |  Number of zones | 6.19-rc7 |         |
 +------------------+----------+---------+
 | 1                | 191      | 192     |
 | 2                | 101      | 128     |
 | 4                | 115      | 123     |
 | 8                | 90       | 120     |
 | 16               | 64       | 115     |
 | 32               | 58       | 105     |
 | 64               | 56       | 101     |
 | 128              | 55       | 99      |
 +------------------+----------+---------+

Tests using XFS shows that buffered write speed with 8 jobs writing
files increases by 12% to 35% depending on the workload.

                    +--------------------+
    |   Write BW (MB/s)  |
 +------------------+----------+---------+
 |     Workload     | Baseline | Patched |
 |                  | 6.19-rc7 |         |
 +------------------+----------+---------+
 | 256MiB file size | 212      | 238     |
 +------------------+----------+---------+
 | 4MiB .. 128 MiB  | 213      | 243     |
 | random file size |          |         |
 +------------------+----------+---------+
 | 2MiB .. 8 MiB    | 179      | 242     |
 | random file size |          |         |
 +------------------+----------+---------+

Performance gains are even more significant when using an HBA that
limits the maximum size of commands to a small value, e.g. HBAs
controlled with the mpi3mr driver limit commands to a maximum of 1 MiB.
In such case, the write throughput gains are over 40%.

                    +--------------------+
    |   Write BW (MB/s)  |
 +------------------+----------+---------+
 |     Workload     | Baseline | Patched |
 |                  | 6.19-rc7 |         |
 +------------------+----------+---------+
 | 256MiB file size | 175      | 245     |
 +------------------+----------+---------+
 | 4MiB .. 128 MiB  | 174      | 244     |
 | random file size |          |         |
 +------------------+----------+---------+
 | 2MiB .. 8 MiB    | 171      | 243     |
 | random file size |          |         |
 +------------------+----------+---------+

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: rename struct gendisk zone_wplugs_lock field
Damien Le Moal [Fri, 27 Feb 2026 13:19:48 +0000 (22:19 +0900)] 
block: rename struct gendisk zone_wplugs_lock field

Rename struct gendisk zone_wplugs_lock field to zone_wplugs_hash_lock to
clearly indicates that this is the spinlock used for manipulating the
hash table of zone write plugs.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: remove disk_zone_is_full()
Damien Le Moal [Fri, 27 Feb 2026 13:19:47 +0000 (22:19 +0900)] 
block: remove disk_zone_is_full()

The helper function disk_zone_is_full() is only used in
disk_zone_wplug_is_full(). So remove it and open code it directly in
this single caller.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: rename and simplify disk_get_and_lock_zone_wplug()
Damien Le Moal [Fri, 27 Feb 2026 13:19:46 +0000 (22:19 +0900)] 
block: rename and simplify disk_get_and_lock_zone_wplug()

disk_get_and_lock_zone_wplug() always returns a zone write plug with the
plug lock held. This is unnecessary since this function does not look at
the fields of existing plugs, and new plugs need to be locked only after
their insertion in the disk hash table, when they are being used.

Remove the zone write plug locking from disk_get_and_lock_zone_wplug()
and rename this function disk_get_or_alloc_zone_wplug().
blk_zone_wplug_handle_write() is modified to add locking of the zone
write plug after calling disk_get_or_alloc_zone_wplug() and before
starting to use the plug. This change also simplifies
blk_revalidate_seq_zone() as unlocking the plug becomes unnecessary.

Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: fix zone write plugs refcount handling in disk_zone_wplug_schedule_bio_work()
Damien Le Moal [Fri, 27 Feb 2026 13:19:45 +0000 (22:19 +0900)] 
block: fix zone write plugs refcount handling in disk_zone_wplug_schedule_bio_work()

The function disk_zone_wplug_schedule_bio_work() always takes a
reference on the zone write plug of the BIO work being scheduled. This
ensures that the zone write plug cannot be freed while the BIO work is
being scheduled but has not run yet. However, this unconditional
reference taking is fragile since the reference taken is released by the
BIO work blk_zone_wplug_bio_work() function, which implies that there
always must be a 1:1 relation between the work being scheduled and the
work running.

Make sure to drop the reference taken when scheduling the BIO work if
the work is already scheduled, that is, when queue_work() returns false.

Fixes: 9e78c38ab30b ("block: Hold a reference on zone write plugs to schedule submission")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: fix zone write plug removal
Damien Le Moal [Fri, 27 Feb 2026 13:19:44 +0000 (22:19 +0900)] 
block: fix zone write plug removal

Commit 7b295187287e ("block: Do not remove zone write plugs still in
use") modified disk_should_remove_zone_wplug() to add a check on the
reference count of a zone write plug to prevent removing zone write
plugs from a disk hash table when the plugs are still being referenced
by BIOs or requests in-flight. However, this check does not take into
account that a BIO completion may happen right after its submission by
a zone write plug BIO work, and before the zone write plug BIO work
releases the zone write plug reference count. This situation leads to
disk_should_remove_zone_wplug() returning false as in this case the zone
write plug reference count is at least equal to 3. If the BIO that
completes in such manner transitioned the zone to the FULL condition,
the zone write plug for the FULL zone will remain in the disk hash
table.

Furthermore, relying on a particular value of a zone write plug
reference count to set the BLK_ZONE_WPLUG_UNHASHED flag is fragile as
reading the atomic reference count and doing a comparison with some
value is not overall atomic at all.

Address these issues by reworking the reference counting of zone write
plugs so that removing plugs from a disk hash table can be done
directly from disk_put_zone_wplug() when the last reference on a plug
is dropped.

To do so, replace the function disk_remove_zone_wplug() with
disk_mark_zone_wplug_dead(). This new function sets the zone write plug
flag BLK_ZONE_WPLUG_DEAD (which replaces BLK_ZONE_WPLUG_UNHASHED) and
drops the initial reference on the zone write plug taken when the plug
was added to the disk hash table. This function is called either for
zones that are empty or full, or directly in the case of a forced plug
removal (e.g. when the disk hash table is being destroyed on disk
removal). With this change, disk_should_remove_zone_wplug() is also
removed.

disk_put_zone_wplug() is modified to call the function
disk_free_zone_wplug() to remove a zone write plug from a disk hash
table and free the plug structure (with a call_rcu()), when the last
reference on a zone write plug is dropped. disk_free_zone_wplug()
always checks that the BLK_ZONE_WPLUG_DEAD flag is set.

In order to avoid having multiple zone write plugs for the same zone in
the disk hash table, disk_get_and_lock_zone_wplug() checked for the
BLK_ZONE_WPLUG_UNHASHED flag. This check is removed and a check for
the new BLK_ZONE_WPLUG_DEAD flag is added to
blk_zone_wplug_handle_write(). With this change, we continue preventing
adding multiple zone write plugs for the same zone and at the same time
re-inforce checks on the user behavior by failing new incoming write
BIOs targeting a zone that is marked as dead. This case can happen only
if the user erroneously issues write BIOs to zones that are full, or to
zones that are currently being reset or finished.

Fixes: 7b295187287e ("block: Do not remove zone write plugs still in use")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: annotate struct request_queue with __counted_by_ptr
Bill Wendling [Wed, 25 Feb 2026 20:51:05 +0000 (20:51 +0000)] 
block: annotate struct request_queue with __counted_by_ptr

The queue_hw_ctx field in struct request_queue is an array of pointers
to struct blk_mq_hw_ctx. The number of elements in this array is tracked
by the nr_hw_queues field.

The array is allocated in __blk_mq_realloc_hw_ctxs() using
kcalloc_node() with set->nr_hw_queues elements. q->nr_hw_queues is
subsequently updated to set->nr_hw_queues.

When growing the array, the new array is assigned to queue_hw_ctx before
nr_hw_queues is updated. This is safe because nr_hw_queues (the old
smaller count) is used for bounds checking, which is within the new
larger allocation.

When shrinking the array, nr_hw_queues is updated to the smaller value,
while queue_hw_ctx retains the larger allocation. This is also safe as
the count is within the allocation bounds.

Annotating queue_hw_ctx with __counted_by_ptr(nr_hw_queues) allows the
compiler (with kSAN) to verify that accesses to queue_hw_ctx are within
the valid range defined by nr_hw_queues.

This patch was generated by CodeMender and reviewed by Bill Wendling.
Tested by running blktests.

Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Bill Wendling <morbo@google.com>
[axboe: massage commit message]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add IOC_OPAL_GET_SUM_STATUS ioctl.
Ondrej Kozina [Fri, 6 Feb 2026 14:18:03 +0000 (15:18 +0100)] 
sed-opal: add IOC_OPAL_GET_SUM_STATUS ioctl.

This adds a function for retrieving the set of Locking objects enabled
for Single User Mode (SUM) and the value of the
RangeStartRangeLengthPolicy parameter.

It retrieves data from the LockingInfo table, specifically the
columns SingleUserModeRanges and RangeStartLengthPolicy, which
were added according to the TCG Opal Feature Set: Single User Mode,
as described in chapters 4.4.3.1 and 4.4.3.2.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: increase column attribute type size to 64 bits.
Ondrej Kozina [Fri, 6 Feb 2026 14:18:02 +0000 (15:18 +0100)] 
sed-opal: increase column attribute type size to 64 bits.

Change the column parameter in response_get_column() from u8 to u64
to support the full range of column identifiers.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add IOC_OPAL_ENABLE_DISABLE_LR.
Ondrej Kozina [Fri, 6 Feb 2026 14:18:01 +0000 (15:18 +0100)] 
sed-opal: add IOC_OPAL_ENABLE_DISABLE_LR.

This ioctl is used to set up RLE (read lock enabled) and WLE (write
lock enabled) parameters of the Locking object.

In Single User Mode (SUM), if the RangeStartRangeLengthPolicy parameter
is set in the 'Reactivate' method, only Admin authority maintains the
locking range length and start (offset) attributes of Locking objects
set up for SUM. All other attributes from struct opal_user_lr_setup
(RLE - read locking enabled, WLE - write locking enabled) shall
remain in possession of the User authority associated with the Locking
object set for SUM.

With the IOC_OPAL_ENABLE_DISABLE_LR ioctl, the opal_user_lr_setup
members 'range_start' and 'range_length' of the ioctl argument are
ignored.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add IOC_OPAL_LR_SET_START_LEN ioctl.
Ondrej Kozina [Fri, 6 Feb 2026 14:18:00 +0000 (15:18 +0100)] 
sed-opal: add IOC_OPAL_LR_SET_START_LEN ioctl.

This ioctl is used to set up locking range start (offset)
and locking range length attributes only.

In Single User Mode (SUM), if the RangeStartRangeLengthPolicy parameter
is set in the 'Reactivate' method, only Admin authority maintains the
locking range length and start (offset) attributes of Locking objects
set up for SUM. All other attributes from struct opal_user_lr_setup
(RLE - read locking enabled, WLE - write locking enabled) shall
remain in possession of the User authority associated with the Locking
object set for SUM.

Therefore, we need a separate function for setting up locking range
start and locking range length because it may require two different
authorities (and sessions) if the RangeStartRangeLengthPolicy attribute
is set.

With the IOC_OPAL_LR_SET_START_LEN ioctl, the opal_user_lr_setup
members 'RLE' and 'WLE' of the ioctl argument are ignored.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: refactor (split) IOC_OPAL_LR_SETUP internals.
Ondrej Kozina [Fri, 6 Feb 2026 14:17:59 +0000 (15:17 +0100)] 
sed-opal: refactor (split) IOC_OPAL_LR_SETUP internals.

IOC_OPAL_LR_SETUP is used to set up a locking range entirely under a
single authority (usually Admin1), but for Single User Mode (SUM),
the permissions for attributes (RangeStart, RangeLength)
and (ReadLockEnable, WriteLockEnable, ReadLocked, WriteLocked)
may be split between two different authorities. Typically, it is Admin1
for the former and the User associated with the LockingRange in SUM
for the latter.

This commit only splits the internals in preparation for the introduction
of separate ioctls for setting RangeStart, RangeLength and the rest
using new ioctl calls.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add IOC_OPAL_REACTIVATE_LSP.
Ondrej Kozina [Fri, 6 Feb 2026 14:17:58 +0000 (15:17 +0100)] 
sed-opal: add IOC_OPAL_REACTIVATE_LSP.

This adds the 'Reactivate' method as described in the
"TCG Storage Opal SSC Feature Set: Single User Mode"
document (ch. 3.1.1.1).

The method enables switching an already active SED OPAL2 device,
with appropriate firmware support for Single User Mode (SUM),
to or from SUM.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add Admin1PIN parameter.
Ondrej Kozina [Fri, 6 Feb 2026 14:17:57 +0000 (15:17 +0100)] 
sed-opal: add Admin1PIN parameter.

As desribed in ch. 3.1.1.1.1.3 of TCG Storage Opal SSC Feature Set:
Single User Mode document.

To be used later in Reactivate method implementation.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add RangeStartRangeLengthPolicy parameter.
Ondrej Kozina [Fri, 6 Feb 2026 14:17:56 +0000 (15:17 +0100)] 
sed-opal: add RangeStartRangeLengthPolicy parameter.

As desribed in ch. 3.1.1.1.1.2 of TCG Storage Opal SSC Feature Set:
Single User Mode document.

To be used later in Reactivate method implementation and in function
for retrieving SUM device status.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agosed-opal: add UID of Locking Table.
Ondrej Kozina [Fri, 6 Feb 2026 14:17:55 +0000 (15:17 +0100)] 
sed-opal: add UID of Locking Table.

As described in ch. 6.3, Table 240 in TCG Storage
Architecture Core Specification document.

It's also referenced in TCG Storage Opal SSC Feature Set:
Single User Mode document, ch. 3.1.1.1 Reactivate method.

It will be used later in Reactivate method implemetation
for sed-opal interface.

Signed-off-by: Ondrej Kozina <okozina@redhat.com>
Reviewed-and-tested-by: Milan Broz <gmazyland@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: pass a maxlen argument to bio_iov_iter_bounce
Christoph Hellwig [Mon, 23 Feb 2026 13:20:07 +0000 (05:20 -0800)] 
block: pass a maxlen argument to bio_iov_iter_bounce

Allow the file system to limit the size processed in a single
bounce operation.  This is needed when generating integrity data
so that the size of a single integrity segment can't overflow.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add fs_bio_integrity helpers
Christoph Hellwig [Mon, 23 Feb 2026 13:20:06 +0000 (05:20 -0800)] 
block: add fs_bio_integrity helpers

Add a set of helpers for file system initiated integrity information.
These include mempool backed allocations and verifying based on a passed
in sector and size which is often available from file system completion
routines.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: make max_integrity_io_size public
Christoph Hellwig [Mon, 23 Feb 2026 13:20:05 +0000 (05:20 -0800)] 
block: make max_integrity_io_size public

File systems that generate integrity will need this, so move it out
of the block private or blk-mq specific headers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: prepare generation / verification helpers for fs usage
Christoph Hellwig [Mon, 23 Feb 2026 13:20:04 +0000 (05:20 -0800)] 
block: prepare generation / verification helpers for fs usage

Return the status from verify instead of directly stashing it in the bio,
and rename the helpers to use the usual bio_ prefix for things operating
on a bio.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: add a bdev_has_integrity_csum helper
Christoph Hellwig [Mon, 23 Feb 2026 13:20:03 +0000 (05:20 -0800)] 
block: add a bdev_has_integrity_csum helper

Factor out a helper to see if the block device has an integrity checksum
from bdev_stable_writes so that it can be reused for other checks.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: factor out a bio_integrity_setup_default helper
Christoph Hellwig [Mon, 23 Feb 2026 13:20:02 +0000 (05:20 -0800)] 
block: factor out a bio_integrity_setup_default helper

Add a helper to set the seed and check flag based on useful defaults
from the profile.

Note that this includes a small behavior change, as we now only set the
seed if any action is set, which is fine as nothing will look at it
otherwise.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoblock: factor out a bio_integrity_action helper
Christoph Hellwig [Mon, 23 Feb 2026 13:20:01 +0000 (05:20 -0800)] 
block: factor out a bio_integrity_action helper

Split the logic to see if a bio needs integrity metadata from
bio_integrity_prep into a reusable helper than can be called from
file system code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Tested-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 months agoLinux 7.0-rc3 v7.0-rc3
Linus Torvalds [Sun, 8 Mar 2026 23:56:54 +0000 (16:56 -0700)] 
Linux 7.0-rc3

2 months agoMerge tag 'efi-fixes-for-v7.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 8 Mar 2026 19:13:09 +0000 (12:13 -0700)] 
Merge tag 'efi-fixes-for-v7.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git/efi/efi

Pull EFI fix from Ard Biesheuvel:
 "Fix for the x86 EFI workaround keeping boot services code and data
  regions reserved until after SetVirtualAddressMap() completes:
  deferred struct page initialization may result in some of this memory
  being lost permanently"

* tag 'efi-fixes-for-v7.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git/efi/efi:
  x86/efi: defer freeing of boot services memory

2 months agoMerge tag 'i2c-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sun, 8 Mar 2026 17:17:05 +0000 (10:17 -0700)] 
Merge tag 'i2c-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux

Pull i2c fix from Wolfram Sang:
 "A revert for the i801 driver restoring old locking behaviour"

* tag 'i2c-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: i801: Revert "i2c: i801: replace acpi_lock with I2C bus lock"

2 months agoMerge tag 'x86-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 8 Mar 2026 01:12:06 +0000 (17:12 -0800)] 
Merge tag 'x86-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:

 - Fix SEV guest boot failures in certain circumstances, due to
   very early code relying on a BSS-zeroed variable that isn't
   actually zeroed yet an may contain non-zero bootup values

   Move the variable into the .data section go gain even earlier
   zeroing

 - Expose & allow the IBPB-on-Entry feature on SNP guests, which
   was not properly exposed to guests due to initial implementational
   caution

 - Fix O= build failure when CONFIG_EFI_SBAT_FILE is using relative
   file paths

 - Fix the various SNC (Sub-NUMA Clustering) topology enumeration
   bugs/artifacts (sched-domain build errors mostly).

   SNC enumeration data got more complicated with Granite Rapids X
   (GNR) and Clearwater Forest X (CWF), which exposed these bugs
   and made their effects more serious

 - Also use the now sane(r) SNC code to fix resctrl SNC detection bugs

 - Work around a historic libgcc unwinder bug in the vdso32 sigreturn
   code (again), which regressed during an overly aggressive recent
   cleanup of DWARF annotations

* tag 'x86-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/entry/vdso32: Work around libgcc unwinder bug
  x86/resctrl: Fix SNC detection
  x86/topo: Fix SNC topology mess
  x86/topo: Replace x86_has_numa_in_package
  x86/topo: Add topology_num_nodes_per_package()
  x86/numa: Store extra copy of numa_nodes_parsed
  x86/boot: Handle relative CONFIG_EFI_SBAT_FILE file paths
  x86/sev: Allow IBPB-on-Entry feature for SNP guests
  x86/boot/sev: Move SEV decompressor variables into the .data section

2 months agoMerge tag 'timers-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 8 Mar 2026 01:09:15 +0000 (17:09 -0800)] 
Merge tag 'timers-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull timer fix from Ingo Molnar:
 "Make clock_adjtime() syscall timex validation slightly more permissive
  for auxiliary clocks, to not reject syscalls based on the status field
  that do not try to modify the status field.

  This makes the ABI behavior in clock_adjtime() consistent with
  CLOCK_REALTIME"

* tag 'timers-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  timekeeping: Fix timex status validation for auxiliary clocks

2 months agoMerge tag 'sched-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 8 Mar 2026 01:07:13 +0000 (17:07 -0800)] 
Merge tag 'sched-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull scheduler fix from Ingo Molnar:
 "Fix a DL scheduler bug that may corrupt internal metrics during PI and
  setscheduler() syscalls, resulting in kernel warnings and misbehavior.

  Found during stress-testing"

* tag 'sched-urgent-2026-03-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/deadline: Fix missing ENQUEUE_REPLENISH during PI de-boosting

2 months agoeventpoll: Convert epoll_put_uevent() to scoped user access
Eric Dumazet [Sat, 7 Mar 2026 20:07:15 +0000 (20:07 +0000)] 
eventpoll: Convert epoll_put_uevent() to scoped user access

Saves two function calls, and one stac/clac pair.

stac/clac is rather expensive on older cpus like Zen 2.

A synthetic network stress test gives a ~1.5% increase of pps
on AMD Zen 2.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Kuniyuki Iwashima <kuniyu@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 months agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 7 Mar 2026 22:04:50 +0000 (14:04 -0800)] 
Merge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Two core changes and the rest in drivers, one core change to quirk the
  behaviour of the Iomega Zip drive and one to fix a hang caused by tag
  reallocation problems, which has mostly been seen by the iscsi client.

  Note the latter fixes the problem but still has a slight sysfs memory
  leak, so will be amended in the next pull request (once we've run the
  fix for the fix through our testing)"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: target: Fix recursive locking in __configfs_open_file()
  scsi: devinfo: Add BLIST_SKIP_IO_HINTS for Iomega ZIP
  scsi: mpi3mr: Clear reset history on ready and recheck state after timeout
  scsi: core: Fix refcount leak for tagset_refcnt

2 months agoMerge tag 'fbdev-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sat, 7 Mar 2026 21:21:43 +0000 (13:21 -0800)] 
Merge tag 'fbdev-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/linux-fbdev

Pull fbdev fix from Helge Deller:
 "Silence build error in au1100fb driver found by kernel test robot"

* tag 'fbdev-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/linux-fbdev:
  fbdev: au1100fb: Fix build on MIPS64

2 months agoMerge tag 'parisc-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/delle...
Linus Torvalds [Sat, 7 Mar 2026 20:38:16 +0000 (12:38 -0800)] 
Merge tag 'parisc-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:
 "While testing Sasha Levin's 'kallsyms: embed source file:line info in
  kernel stack traces' patch series, which increases the typical kernel
  image size, I found some issues with the parisc initial kernel mapping
  which may prevent the kernel to boot.

  The three small patches here fix this"

* tag 'parisc-for-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix initial page table creation for boot
  parisc: Check kernel mapping earlier at bootup
  parisc: Increase initial mapping to 64 MB with KALLSYMS

2 months agoMerge tag 'bpf-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
Linus Torvalds [Sat, 7 Mar 2026 20:20:37 +0000 (12:20 -0800)] 
Merge tag 'bpf-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf

Pull bpf fixes from Alexei Starovoitov:

 - Fix u32/s32 bounds when ranges cross min/max boundary (Eduard
   Zingerman)

 - Fix precision backtracking with linked registers (Eduard Zingerman)

 - Fix linker flags detection for resolve_btfids (Ihor Solodrai)

 - Fix race in update_ftrace_direct_add/del (Jiri Olsa)

 - Fix UAF in bpf_trampoline_link_cgroup_shim (Lang Xu)

* tag 'bpf-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf:
  resolve_btfids: Fix linker flags detection
  selftests/bpf: add reproducer for spurious precision propagation through calls
  bpf: collect only live registers in linked regs
  Revert "selftests/bpf: Update reg_bound range refinement logic"
  selftests/bpf: test refining u32/s32 bounds when ranges cross min/max boundary
  bpf: Fix u32/s32 bounds when ranges cross min/max boundary
  bpf: Fix a UAF issue in bpf_trampoline_link_cgroup_shim
  ftrace: Add missing ftrace_lock to update_ftrace_direct_add/del

2 months agoMerge tag 'rcu-fixes.v7.0-20260307a' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 7 Mar 2026 19:56:55 +0000 (11:56 -0800)] 
Merge tag 'rcu-fixes.v7.0-20260307a' of git://git.kernel.org/pub/scm/linux/kernel/git/rcu/linux

Pull RCU selftest fixes from Boqun Feng:
 "Fix a regression in RCU torture test pre-defined scenarios caused by
  commit 7dadeaa6e851 ("sched: Further restrict the preemption modes")
  which limits PREEMPT_NONE to architectures that do not support
  preemption at all and PREEMPT_VOLUNTARY to those architectures that do
  not yet have PREEMPT_LAZY support.

  Since major architectures (e.g. x86 and arm64) no longer support
  CONFIG_PREEMPT_NONE and CONFIG_PREEMPT_VOLUNTARY, using them in
  rcutorture, rcuscale, refscale, and scftorture pre-defined scenarios
  causes config checking errors.

  Switch these kconfigs to PREEMPT_LAZY"

* tag 'rcu-fixes.v7.0-20260307a' of git://git.kernel.org/pub/scm/linux/kernel/git/rcu/linux:
  scftorture: Update due to x86 not supporting none/voluntary preemption
  refscale: Update due to x86 not supporting none/voluntary preemption
  rcuscale: Update due to x86 not supporting none/voluntary preemption
  rcutorture: Update due to x86 not supporting none/voluntary preemption

2 months agoMerge tag 'trace-v7.0-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace...
Linus Torvalds [Sat, 7 Mar 2026 17:50:54 +0000 (09:50 -0800)] 
Merge tag 'trace-v7.0-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace

Pull tracing fixes from Steven Rostedt:

 - Fix possible NULL pointer dereference in trace_data_alloc()

   On the trace_data_alloc() error path, it can call trigger_data_free()
   with a NULL pointer. This used to be a kfree() but was changed to
   trigger_data_free() to clean up any partial initialization. The issue
   is that trigger_data_free() does not expect a NULL pointer. Have
   trigger_data_free() return safely on NULL pointer.

 - Fix multiple events on the command line and bootconfig

   If multiple events are enabled on the command line separately and not
   grouped, only the last event gets enabled. That is:

      trace_event=sched_switch trace_event=sched_waking

   will only enable sched_waking whereas:

      trace_event=sched_switch,sched_waking

   will enable both.

   The bootconfig makes it even worse as the second way is the more
   common method.

   The issue is that a temporary buffer is used to store the events to
   enable later in boot. Each time the cmdline callback is called, it
   overwrites what was previously there.

   Have the callback append the next value (delimited by a comma) if the
   temporary buffer already has content.

 - Fix command line trace_buffer_size if >= 2G

   The logic to allocate the trace buffer uses "int" for the size
   parameter in the command line code causing overflow issues if more
   that 2G is specified.

* tag 'trace-v7.0-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  tracing: Fix trace_buf_size= cmdline parameter with sizes >= 2G
  tracing: Fix enabling multiple events on the kernel command line and bootconfig
  tracing: Add NULL pointer check to trigger_data_free()

2 months agoresolve_btfids: Fix linker flags detection
Ihor Solodrai [Thu, 5 Mar 2026 01:47:30 +0000 (17:47 -0800)] 
resolve_btfids: Fix linker flags detection

The "|| echo -lzstd" default makes zstd an unconditional link
dependency of resolve_btfids. On systems where libzstd-dev is not
installed and pkg-config fails, the linker fails:

  ld: cannot find -lzstd: No such file or directory

libzstd is a transitive dependency of libelf, so the -lzstd flag is
strictly necessary only for static builds [1].

Remove ZSTD_LIBS variable, and instead set LIBELF_LIBS depending on
whether the build is static or not. Use $(HOSTPKG_CONFIG) as primary
source of the flags list.

Also add a default value for HOSTPKG_CONFIG in case it's not built via
the toplevel Makefile. Pass it from selftests/bpf too.

[1] https://lore.kernel.org/bpf/4ff82800-2daa-4b9f-95a9-6f512859ee70@linux.dev/

Reported-by: BPF CI Bot (Claude Opus 4.6) <bot+bpf-ci@kernel.org>
Reported-by: Vitaly Chikunov <vt@altlinux.org>
Closes: https://lore.kernel.org/bpf/aaWqMcK-2AQw5dx8@altlinux.org/
Fixes: 4021848a903e ("selftests/bpf: Pass through build flags to bpftool and resolve_btfids")
Signed-off-by: Ihor Solodrai <ihor.solodrai@linux.dev>
Reviewed-by: Paul Chaignon <paul.chaignon@gmail.com>
Link: https://lore.kernel.org/r/20260305014730.3123382-1-ihor.solodrai@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 months agoMerge tag 'hwmon-for-v7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/groec...
Linus Torvalds [Sat, 7 Mar 2026 16:39:59 +0000 (08:39 -0800)] 
Merge tag 'hwmon-for-v7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging

Pull hwmon fixes from Guenter Roeck:

 - Fix initialization commands for AHT20

 - Correct a malformed email address (emc1403)

 - Check the it87_lock() return value

 - Fix inverted polarity (max6639)

 - Fix overflows, underflows, sign extension, and other problems in
   macsmc

 - Fix stack overflow in debugfs read (pmbus/q54sj108a2)

 - Drop support for SMARC-sAM67 (discontinued and never released to
   market)

* tag 'hwmon-for-v7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (pmbus/q54sj108a2) fix stack overflow in debugfs read
  hwmon: (max6639) fix inverted polarity
  dt-bindings: hwmon: sl28cpld: Drop sa67mcu compatible
  hwmon: (it87) Check the it87_lock() return value
  Revert "hwmon: add SMARC-sAM67 support"
  hwmon: (aht10) Fix initialization commands for AHT20
  hwmon: (emc1403) correct a malformed email address
  hwmon: (macsmc) Fix overflows, underflows, and sign extension
  hwmon: (macsmc) Fix regressions in Apple Silicon SMC hwmon driver

2 months agoMerge tag 'driver-core-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 7 Mar 2026 16:16:48 +0000 (08:16 -0800)] 
Merge tag 'driver-core-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/driver-core/driver-core

Pull driver core fix from Danilo Krummrich:

 - Revert "driver core: enforce device_lock for driver_match_device()":

   When a device is already present in the system and a driver is
   registered on the same bus, we iterate over all devices registered on
   this bus to see if one of them matches. If we come across an already
   bound one where the corresponding driver crashed while holding the
   device lock (e.g. in probe()) we can't make any progress anymore.

   Thus, revert and clarify that an implementer of struct bus_type must
   not expect match() to be called with the device lock held.

* tag 'driver-core-7.0-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/driver-core/driver-core:
  Revert "driver core: enforce device_lock for driver_match_device()"

2 months agoMerge tag 'for-linus-7.0-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 7 Mar 2026 15:44:32 +0000 (07:44 -0800)] 
Merge tag 'for-linus-7.0-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:

 - a cleanup of arch/x86/kernel/head_64.S removing the pre-built page
   tables for Xen guests

 - a small comment update

 - another cleanup for Xen PVH guests mode

 - fix an issue with Xen PV-devices backed by driver domains

* tag 'for-linus-7.0-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/xenbus: better handle backend crash
  xenbus: add xenbus_device parameter to xenbus_read_driver_state()
  x86/PVH: Use boot params to pass RSDP address in start_info page
  x86/xen: update outdated comment
  xen/acpi-processor: fix _CST detection using undersized evaluation buffer
  x86/xen: Build identity mapping page tables dynamically for XENPV

2 months agoMerge branch 'bpf-fix-precision-backtracking-bug-with-linked-registers'
Alexei Starovoitov [Sat, 7 Mar 2026 02:24:41 +0000 (18:24 -0800)] 
Merge branch 'bpf-fix-precision-backtracking-bug-with-linked-registers'

Eduard Zingerman says:

====================
bpf: Fix precision backtracking bug with linked registers

Emil Tsalapatis reported a verifier bug hit by the scx_lavd sched_ext
scheduler. The essential part of the verifier log looks as follows:

  436: ...
  // checkpoint hit for 438: (1d) if r7 == r8 goto ...
  frame 3: propagating r2,r7,r8
  frame 2: propagating r6
  mark_precise: frame3: last_idx ...
  mark_precise: frame3: regs=r2,r7,r8 stack= before 436: ...
  mark_precise: frame3: regs=r2,r7 stack= before 435: ...
  mark_precise: frame3: regs=r2,r7 stack= before 434: (85) call bpf_trace_vprintk#177
  verifier bug: backtracking call unexpected regs 84

The log complains that registers r2 and r7 are tracked as precise
while processing the bpf_trace_vprintk() call in precision backtracking.
This can't be right, as r2 is reset by the call and there is nothing
to backtrack it to. The precision propagation is triggered when
a checkpoint is hit at instruction 438, r2 is dead at that instruction.

This happens because of the following sequence of events:
- Instruction 438 is first reached with registers r2 and r7 having
  the same id via a path that does not call bpf_trace_vprintk():
  - Checkpoint is created at 438.
  - The jump at 438 is predicted, hence r7 and registers linked to it
    (r2) are propagated as precise, marking r2 and r7 precise in the
    checkpoint.
- Instruction 438 is reached a second time with r2 undefined and via
  a path that calls bpf_trace_vprintk():
  - Checkpoint is hit.
  - propagate_precision() picks registers r2 and r7 and propagates
    precision marks for those up to the helper call.

The root cause is the fact that states_equal() and
propagate_precision() assume that the precision flag can't be set for a
dead register (as computed by compute_live_registers()).
However, this is not the case when linked registers are at play.
Fix this by accounting for live register flags in
collect_linked_regs().
---
====================

Link: https://patch.msgid.link/20260306-linked-regs-and-propagate-precision-v1-0-18e859be570d@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2 months agoselftests/bpf: add reproducer for spurious precision propagation through calls
Eduard Zingerman [Sat, 7 Mar 2026 00:02:48 +0000 (16:02 -0800)] 
selftests/bpf: add reproducer for spurious precision propagation through calls

Add a test for the scenario described in the previous commit:
an iterator loop with two paths where one ties r2/r7 via
shared scalar id and skips a call, while the other goes
through the call. Precision marks from the linked registers
get spuriously propagated to the call path via
propagate_precision(), hitting "backtracking call unexpected
regs" in backtrack_insn().

Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/r/20260306-linked-regs-and-propagate-precision-v1-2-18e859be570d@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>