This is just apply Kuai's patch in [1] with mirror changes.
blk_mq_realloc_hw_ctxs() will free the 'queue_hw_ctx'(e.g. undate
submit_queues through configfs for null_blk), while it might still be
used from other context(e.g. switch elevator to none):
t1 t2
elevator_switch
blk_mq_unquiesce_queue
blk_mq_run_hw_queues
queue_for_each_hw_ctx
// assembly code for hctx = (q)->queue_hw_ctx[i]
mov 0x48(%rbp),%rdx -> read old queue_hw_ctx
__blk_mq_update_nr_hw_queues
blk_mq_realloc_hw_ctxs
hctxs = q->queue_hw_ctx
q->queue_hw_ctx = new_hctxs
kfree(hctxs)
movslq %ebx,%rax
mov (%rdx,%rax,8),%rdi ->uaf
This problem was found by code review, and I comfirmed that the concurrent
scenario do exist(specifically 'q->queue_hw_ctx' can be changed during
blk_mq_run_hw_queues()), however, the uaf problem hasn't been repoduced yet
without hacking the kernel.
Sicne the queue is freezed in __blk_mq_update_nr_hw_queues(), fix the
problem by protecting 'queue_hw_ctx' through rcu where it can be accessed
without grabbing 'q_usage_counter'.
[1] https://lore.kernel.org/all/20220225072053.2472431-1-yukuai3@huawei.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Fengnan Chang <changfengnan@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After commit 4e5cc99e1e ("blk-mq: manage hctx map via xarray"), we use
an xarray instead of array to store hctx, but in poll mode, each time
in blk_mq_poll, we need use xa_load to find corresponding hctx, this
introduce some costs. In my test, xa_load may cost 3.8% cpu.
This patch revert previous change, eliminates the overhead of xa_load
and can result in a 3% performance improvement.
Signed-off-by: Fengnan Chang <changfengnan@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch introduces a new structure, struct elevator_resources, to
group together all elevator-related resources that share the same
lifetime. As a first step, this change moves the elevator tag pointer
from struct elv_change_ctx into the new struct elevator_resources.
Additionally, rename blk_mq_alloc_sched_tags_batch() and
blk_mq_free_sched_tags_batch() to blk_mq_alloc_sched_res_batch() and
blk_mq_free_sched_res_batch(), respectively. Introduce two new wrapper
helpers, blk_mq_alloc_sched_res() and blk_mq_free_sched_res(), around
blk_mq_alloc_sched_tags() and blk_mq_free_sched_tags().
These changes pave the way for consolidating the allocation and freeing
of elevator-specific resources into common helper functions. This
refactoring improves encapsulation and prepares the code for future
extensions, allowing additional elevator-specific data to be added to
struct elevator_resources without cluttering struct elv_change_ctx.
Subsequent patches will extend struct elevator_resources to include
other elevator-related data.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Yu Kuai <yukuai@fnnas.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, the nr_hw_queues update path manages two disjoint xarrays —
one for elevator tags and another for elevator type — both used during
elevator switching. Maintaining these two parallel structures for the
same purpose adds unnecessary complexity and potential for mismatched
state.
This patch unifies both xarrays into a single structure, struct
elv_change_ctx, which holds all per-queue elevator change context. A
single xarray, named elv_tbl, now maps each queue (q->id) in a tagset
to its corresponding elv_change_ctx entry, encapsulating the elevator
tags, type and name references.
This unification simplifies the code, improves maintainability, and
clarifies ownership of per-queue elevator state.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Yu Kuai <yukuai@fnnas.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The blk-mq dma iterator has an optimization for requests that align to
the device's iommu merge boundary. This boundary may be larger than the
device's virtual boundary, but the code had been depending on that queue
limit to know ahead of time if the request is guaranteed to align to
that optimization.
Rather than rely on that queue limit, which many devices may not report,
save the lowest set bit of any boundary gap between each segment in the
bio while checking the segments. The request stores the value for
merging and quickly checking per io if the request can use iova
optimizations.
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 7f2799c546 ("blk-mq: cleanup shared tags case in
blk_mq_update_nr_requests()") moves blk_mq_tag_update_sched_shared_tags()
before q->nr_requests is updated, however, it's still using the old
q->nr_requests to resize tag depth.
Fix this problem by passing in expected new tag depth.
Fixes: 7f2799c546 ("blk-mq: cleanup shared tags case in blk_mq_update_nr_requests()")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reported-by: Chris Mason <clm@meta.com>
Link: https://lore.kernel.org/linux-block/20251014130507.4187235-2-clm@meta.com/
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_free_tags() can be called after blk_mq_init_tags(), while
tags->page_list is still not initialized, causing null-ptr-deref.
Fix this problem by initializing tags->page_list at blk_mq_init_tags(),
meanwhile, also free tags directly from error path because there is no
srcu barrier.
Fixes: ad0d05dbdd ("blk-mq: Defer freeing of tags page_list to SRCU callback")
Reported-by: syzbot+5c5d41e80248d610221f@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/68d1b079.a70a0220.1b52b.0000.GAE@google.com/
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Allocate and free sched_tags while queue is freezed can deadlock[1],
this is a long term problem, hence allocate memory before freezing
queue and free memory after queue is unfreezed.
[1] https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Fixes: e3a2b3f931 ("blk-mq: allow changing of queue depth through sysfs")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
No functional changes are intended, make code cleaner and prepare to fix
the grow case in following patches.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For shared tags case, all hctx->sched_tags/tags are the same, it doesn't
make sense to call into blk_mq_tag_update_depth() multiple times for the
same tags.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
queue_requests_store() is the only caller of
blk_mq_update_nr_requests(), and blk_mq_update_nr_requests() is the
only caller of blk_mq_tag_update_depth(), however, they all have
checkings for nr_requests input by user.
Make code cleaner by moving all the checkings to the top function:
1) nr_requests > reserved tags;
2) if there is elevator, 4 <= nr_requests <= 2048;
3) if elevator is none, 4 <= nr_requests <= tag_set->queue_depth;
Meanwhile, case 2 is the only case tags can grow and -ENOMEM might be
returned.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
1) queue_requests_store() is the only caller of
blk_mq_update_nr_requests(), where queue is already freezed, no need to
check mq_freeze_depth;
2) q->tag_set must be set for request based device, and queue_is_mq() is
already checked in blk_mq_queue_attr_visible(), no need to check
q->tag_set.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
bio->issue_time_ns is initialized for every bio, however, it's only used
by blk-iolatency. Add a new queue_flag and only set this flag when
blk-iolatency is enabled, so that extra blk_time_get_ns() can be saved
for disks that blk-iolatency is not enabled.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
bio->issue_time_ns is only used by blk-iolatency, which can only be
enabled for rq-based disk, hence it's not necessary to initialize
the time for bio-based disk.
Meanwhile, if bio is split by blk_crypto_fallback_split_bio_if_needed(),
the issue time is not initialized for new split bio, this can be fixed
as well.
Noted the next patch will optimize better that bio issue time will
only be used when blk-iolatency is really enabled by the disk.
Fixes: 488f6682c8 ("block: blk-crypto-fallback for Inline Encryption")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Replace the spinlock in blk_mq_find_and_get_req() with an SRCU read lock
around the tag iterators.
This is done by:
- Holding the SRCU read lock in blk_mq_queue_tag_busy_iter(),
blk_mq_tagset_busy_iter(), and blk_mq_hctx_has_requests().
- Removing the now-redundant tags->lock from blk_mq_find_and_get_req().
This change fixes lockup issue in scsi_host_busy() in case of shost->host_blocked.
Also avoids big tags->lock when reading disk sysfs attribute `inflight`.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The freeing of the flush queue/request in blk_mq_exit_hctx() can race with
tag iterators that may still be accessing it. To prevent a potential
use-after-free, the deallocation should be deferred until after a grace
period. With this way, we can replace the big tags->lock in tags iterator
code path with srcu for solving the issue.
This patch introduces an SRCU-based deferred freeing mechanism for the
flush queue.
The changes include:
- Adding a `rcu_head` to `struct blk_flush_queue`.
- Creating a new callback function, `blk_free_flush_queue_callback`,
to handle the actual freeing.
- Replacing the direct call to `blk_free_flush_queue()` in
`blk_mq_exit_hctx()` with `call_srcu()`, using the `tags_srcu`
instance to ensure synchronization with tag iterators.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Tag iterators can race with the freeing of the request pages(tags->page_list),
potentially leading to use-after-free issues.
Defer the freeing of the page list and the tags structure itself until
after an SRCU grace period has passed. This ensures that any concurrent
tag iterators have completed before the memory is released. With this
way, we can replace the big tags->lock in tags iterator code path with
srcu for solving the issue.
This is achieved by:
- Adding a new `srcu_struct tags_srcu` to `blk_mq_tag_set` to protect
tag map iteration.
- Adding an `rcu_head` to `struct blk_mq_tags` to be used with
`call_srcu`.
- Moving the page list freeing logic and the `kfree(tags)` call into a
new callback function, `blk_mq_free_tags_callback`.
- In `blk_mq_free_tags`, invoking `call_srcu` to schedule the new
callback for deferred execution.
The read-side protection for the tag iterators will be added in a
subsequent patch.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
To prepare for converting the tag->rqs freeing to be SRCU-based, the
tag_set is needed in the freeing helper functions.
This patch adds 'struct blk_mq_tag_set *' as the first parameter to
blk_mq_free_rq_map() and blk_mq_free_tags(), and updates all their call
sites.
This allows access to the tag_set's SRCU structure in the next step,
which will be used to free the tag maps after a grace period.
No functional change is intended in this patch.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move flush queue allocation into blk_mq_init_hctx() and its release into
blk_mq_exit_hctx(), and prepare for replacing tags->lock with SRCU to
draining inflight request walking. blk_mq_exit_hctx() is the last chance
for us to get valid `tag_set` reference, and we need to add one SRCU to
`tag_set` for freeing flush request via call_srcu().
It is safe to move flush queue & request release into blk_mq_exit_hctx(),
because blk_mq_clear_flush_rq_mapping() clears the flush request
reference int driver tags inflight request table, meantime inflight
request walking is drained.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Current depth_updated has some problems:
1) depth_updated() will be called for each hctx, while all elevators
will update async_depth for the disk level, this is not related to hctx;
2) In blk_mq_update_nr_requests(), if previous hctx update succeed and
this hctx update failed, q->nr_requests will not be updated, while
async_depth is already updated with new nr_reqeuests in previous
depth_updated();
3) All elevators are using q->nr_requests to calculate async_depth now,
however, q->nr_requests is still the old value when depth_updated() is
called from blk_mq_update_nr_requests();
Those problems are first from error path, then mq-deadline, and recently
for bfq and kyber, fix those problems by:
- pass in request_queue instead of hctx;
- move depth_updated() after q->nr_requests is updated in
blk_mq_update_nr_requests();
- add depth_updated() call inside init_sched() method to initialize
async_depth;
- remove init_hctx() method for mq-deadline and bfq that is useless now;
Fixes: 77f1e0a52d ("bfq: update internal depth state when queue depth changes")
Fixes: 39823b47bb ("block/mq-deadline: Fix the tag reservation code")
Fixes: 42e6c6ce03 ("lib/sbitmap: convert shallow_depth from one word to the whole sbitmap")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Li Nan <linan122@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250821060612.1729939-2-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 5989bfe6ac ("block: restore two stage elevator switch while
running nr_hw_queue update") reintroduced a lockdep warning by calling
blk_mq_freeze_queue_nomemsave() before switching the I/O scheduler.
The function blk_mq_elv_switch_none() calls elevator_change_done().
Running this while the queue is frozen causes a lockdep warning.
Fix this by reordering the operations: first, switch the I/O scheduler
to 'none', and then freeze the queue. This ensures that elevator_change_done()
is not called on an already frozen queue. And this way is safe because
elevator_set_none() does freeze queue before switching to none.
Also we still have to rely on blk_mq_elv_switch_back() for switching
back, and it has to cover unfrozen queue case.
Cc: Nilay Shroff <nilay@linux.ibm.com>
Cc: Yu Kuai <yukuai3@huawei.com>
Fixes: 5989bfe6ac ("block: restore two stage elevator switch while running nr_hw_queue update")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250815131737.331692-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move scheduler tags (sched_tags) allocation and deallocation outside
both the ->elevator_lock and ->freeze_lock when updating nr_hw_queues.
This change breaks the dependency chain from the percpu allocator lock
to the elevator lock, helping to prevent potential deadlocks, as
observed in the reported lockdep splat[1].
This commit introduces batch allocation and deallocation helpers for
sched_tags, which are now used from within __blk_mq_update_nr_hw_queues
routine while iterating through the tagset.
With this change, all sched_tags memory management is handled entirely
outside the ->elevator_lock and the ->freeze_lock context, thereby
eliminating the lock dependency that could otherwise manifest during
nr_hw_queues updates.
[1] https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Reported-by: Stefan Haberland <sth@linux.ibm.com>
Closes: https://lore.kernel.org/all/0659ea8d-a463-47c8-9180-43c719e106eb@linux.ibm.com/
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250730074614.2537382-4-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The kmemleak reports memory leaks related to elevator resources that
were originally allocated in the ->init_hctx() method. The following
leak traces are observed after running blktests block/040:
unreferenced object 0xffff8881b82f7400 (size 512):
comm "check", pid 68454, jiffies 4310588881
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace (crc 5bac8b34):
__kvmalloc_node_noprof+0x55d/0x7a0
sbitmap_init_node+0x15a/0x6a0
kyber_init_hctx+0x316/0xb90
blk_mq_init_sched+0x419/0x580
elevator_switch+0x18b/0x630
elv_update_nr_hw_queues+0x219/0x2c0
__blk_mq_update_nr_hw_queues+0x36a/0x6f0
blk_mq_update_nr_hw_queues+0x3a/0x60
0xffffffffc09ceb80
0xffffffffc09d7e0b
configfs_write_iter+0x2b1/0x470
vfs_write+0x527/0xe70
ksys_write+0xff/0x200
do_syscall_64+0x98/0x3c0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
unreferenced object 0xffff8881b82f6000 (size 512):
comm "check", pid 68454, jiffies 4310588881
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace (crc 5bac8b34):
__kvmalloc_node_noprof+0x55d/0x7a0
sbitmap_init_node+0x15a/0x6a0
kyber_init_hctx+0x316/0xb90
blk_mq_init_sched+0x419/0x580
elevator_switch+0x18b/0x630
elv_update_nr_hw_queues+0x219/0x2c0
__blk_mq_update_nr_hw_queues+0x36a/0x6f0
blk_mq_update_nr_hw_queues+0x3a/0x60
0xffffffffc09ceb80
0xffffffffc09d7e0b
configfs_write_iter+0x2b1/0x470
vfs_write+0x527/0xe70
ksys_write+0xff/0x200
do_syscall_64+0x98/0x3c0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
unreferenced object 0xffff8881b82f5800 (size 512):
comm "check", pid 68454, jiffies 4310588881
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace (crc 5bac8b34):
__kvmalloc_node_noprof+0x55d/0x7a0
sbitmap_init_node+0x15a/0x6a0
kyber_init_hctx+0x316/0xb90
blk_mq_init_sched+0x419/0x580
elevator_switch+0x18b/0x630
elv_update_nr_hw_queues+0x219/0x2c0
__blk_mq_update_nr_hw_queues+0x36a/0x6f0
blk_mq_update_nr_hw_queues+0x3a/0x60
0xffffffffc09ceb80
0xffffffffc09d7e0b
configfs_write_iter+0x2b1/0x470
vfs_write+0x527/0xe70
ksys_write+0xff/0x200
do_syscall_64+0x98/0x3c0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
The issue arises while we run nr_hw_queue update, Specifically, we first
reallocate hardware contexts (hctx) via __blk_mq_realloc_hw_ctxs(), and
then later invoke elevator_switch() (assuming q->elevator is not NULL).
The elevator switch code would first exit old elevator (elevator_exit)
and then switches to the new elevator. The elevator_exit loops through
each hctx and invokes the elevator’s per-hctx exit method ->exit_hctx(),
which releases resources allocated during ->init_hctx().
This memleak manifests when we reduce the num of h/w queues - for example,
when the initial update sets the number of queues to X, and a later update
reduces it to Y, where Y < X. In this case, we'd loose the access to old
hctxs while we get to elevator exit code because __blk_mq_realloc_hw_ctxs
would have already released the old hctxs. As we don't now have any
reference left to the old hctxs, we don't have any way to free the
scheduler resources (which are allocate in ->init_hctx()) and kmemleak
complains about it.
This issue was caused due to the commit 596dce110b ("block: simplify
elevator reattachment for updating nr_hw_queues"). That change unified
the two-stage elevator teardown and reattachment into a single call that
occurs after __blk_mq_realloc_hw_ctxs() has already freed the hctxs.
This patch restores the previous two-stage elevator switch logic during
nr_hw_queues updates. First, the elevator is switched to 'none', which
ensures all scheduler resources are properly freed. Then, the hardware
contexts (hctxs) are reallocated, and the software-to-hardware queue
mappings are updated. Finally, the original elevator is reattached. This
sequence prevents loss of references to old hctxs and avoids the scheduler
resource leaks reported by kmemleak.
Reported-by : Yi Zhang <yi.zhang@redhat.com>
Fixes: 596dce110b ("block: simplify elevator reattachment for updating nr_hw_queues")
Closes: https://lore.kernel.org/all/CAHj4cs8oJFvz=daCvjHM5dYCNQH4UXwSySPPU4v-WHce_kZXZA@mail.gmail.com/
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20250724102540.1366308-1-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_zone_update_request_bio() does two things. First it checks if the
request to be completed was written via ZONE APPEND and if yes it then
updates the sector to the one that the data was written to.
This is small enough to be an inline function. But upcoming changes adding
a tracepoint don't work if the function is inlined.
Split the function into two, the first is blk_req_bio_is_zone_append()
checking if the sector needs to be updated. This can still be an inline
function. The second is blk_zone_append_update_request_bio() doing the
sector update.
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20250715115324.53308-3-johannes.thumshirn@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation for fixing device mapper zone write handling, introduce
the inline helper function bio_needs_zone_write_plugging() to test if a
BIO requires handling through zone write plugging using the function
blk_zone_plug_bio(). This function returns true for any write
(op_is_write(bio) == true) operation directed at a zoned block device
using zone write plugging, that is, a block device with a disk that has
a zone write plug hash table.
This helper allows simplifying the check on entry to blk_zone_plug_bio()
and used in to protect calls to it for blk-mq devices and DM devices.
Fixes: f211268ed1 ("dm: Use the block layer zone append emulation")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250625093327.548866-3-dlemoal@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull MD changes from Yu Kuai:
- Fix that normal IO can be starved by sync IO, found by mkfs on newly
created large raid5, with some clean up patches for bdev inflight
counters.
* tag 'md-6.16-20250513' of https://git.kernel.org/pub/scm/linux/kernel/git/mdraid/linux:
md: clean up accounting for issued sync IO
md: fix is_mddev_idle()
md: add a new api sync_io_depth
md: record dm-raid gendisk in mddev
block: export API to get the number of bdev inflight IO
block: clean up blk_mq_in_flight_rw()
block: WARN if bdev inflight counter is negative
block: reuse part_in_flight_rw for part_in_flight
blk-mq: remove blk_mq_in_flight()
In __blk_mq_update_nr_hw_queues(), the current sequence involves:
1. unregistering sysfs/debugfs attributes
2. freeze the queue
3. reallocating the tag set
4. updating the queue map
5. reallocating hardware contexts
6. updating the elevator (which unfreeze the queue again)
7. re-register sysfs/debugfs attributes
If tag set reallocation fails at step 3, the function skips steps 4–6
and proceeds directly to step 7, re-registering the sysfs/debugfs
attributes without unfreezing the queue first. This is incorrect and
can lead to a system hang or lockdep splat, as the queue remains frozen
and is never properly unfrozen.
This patch addresses the issue by explicitly unfreezing the queue before
re-registering the sysfs/debugfs attributes in the event of a tag set
reallocation failure.
Fixes: 9dc7a882ce ("block: move hctx debugfs/sysfs registering out of freezing queue")
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250512092952.135887-1-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After commit 7be835694d ("block: fix that util can be greater than
100%"), it's not used and can be removed.
Link: https://lore.kernel.org/linux-raid/20250506124903.2540268-1-yukuai1@huaweicloud.com
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Move hctx cpuhp add/del out of queue freezing for not connecting freeze
lock with cpuhp locks, then lockdep warning can be avoided.
This way is safe because both needn't queue to be frozen and scheduler
switch isn't allowed, with same reason for moving hctx debugfs/sysfs
register out of queue freeze.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250505141805.2751237-25-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Both blk_mq_map_swqueue() and blk_mq_realloc_hw_ctxs() are called before
the request queue is added to tagset list, so the two won't run concurrently
with blk_mq_update_nr_hw_queues().
When the two functions are only called from queue initialization or
blk_mq_update_nr_hw_queues(), elevator switch can't happen.
So remove ->elevator_lock uses from the two functions.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250505141805.2751237-24-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move hctx debugfs/sysfs register out of freezing queue in
__blk_mq_update_nr_hw_queues(), so that the following lockdep dependency
can be killed:
#2 (&q->q_usage_counter(io)#16){++++}-{0:0}:
#1 (fs_reclaim){+.+.}-{0:0}:
#0 (&sb->s_type->i_mutex_key#3){+.+.}-{4:4}: //debugfs
And registering/un-registering hctx debugfs/sysfs does not require queue to
be frozen:
- hctx sysfs attributes show() are drained when removing kobject, and
there isn't store() implementation for hctx sysfs attributes
- debugfs entry read() is drained too when removing debugfs directory,
and there isn't write() implementation for hctx debugfs too
- so it is safe to register/unregister hctx sysfs/debugfs without
freezing queue because the cod paths changes nothing, and we just
need to keep hctx live
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-23-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move elv_register[unregister]_queue out of ->elevator_lock & queue freezing,
so we can kill many lockdep warnings.
elv_register[unregister]_queue() is serialized, and just dealing with sysfs/
debugfs things, no need to be done with queue frozen:
- when it is called from adding disk, elevator switch isn't possible
because ->queue_kobj isn't added yet
- when it is called from deleting disk, disable_elv_switch() is
responsible for preventing new elevator switch and draining old
elevator switch.
- when it is called from blk_mq_update_nr_hw_queues(), adding/removing
disk and elevator switch can't be allowed or in-progress
With this change, elevator's ->exit() is called before calling
elv_unregister_queue, then user may call into ->show()/store() of elevator's
sysfs attributes, and we have covered this issue by adding `ELEVATOR_FLAG_DYNG`.
For blk-mq debugfs, hctx->sched_tags is always checked with ->elevator_lock by
debugfs code, meantime hctx->sched_tags is updated with ->elevator_lock, so
there isn't such issue.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250505141805.2751237-22-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In blk_mq_update_nr_hw_queues(), nr_hw_queues changes and elevator data
depends on it, and elevator has to be reattached, so call elevator_switch()
to force attachment.
Add elv_update_nr_hw_queues() simply for blk_mq_update_nr_hw_queues() to
reattach elevator, since elevator switch isn't likely when running
blk_mq_update_nr_hw_queues(). This way removes the current switch
none and switch back code.
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-14-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
That makes the function nicely self-contained and can be used
to avoid code duplication.
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-11-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Both adding/deleting disk code are reader of `nr_hw_queues`, so we can't
allow them in-progress when updating nr_hw_queues, kernel panic and
kasan has been reported in [1].
Prevent adding/deleting disk during updating nr_hw_queues by adding
rw_semaphore to tagset, write lock is grabbed in blk_mq_update_nr_hw_queues(),
and read lock is acquired when adding/deleting disk.
Also mark GFP_NOIO allocation scope for adding/deleting disk because
blk_mq_update_nr_hw_queues() is part of some driver's error handler.
This way avoids lot of trouble.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Suggested-by: Nilay Shroff <nilay@linux.ibm.com>
Reported-by: Nilay Shroff <nilay@linux.ibm.com>
Closes: https://lore.kernel.org/linux-block/a5896cdb-a59a-4a37-9f99-20522f5d2987@linux.ibm.com/
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-9-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move blk_mq_add_queue_tag_set() after blk_mq_map_swqueue(), and publish
this request queue to tagset after everything is setup.
This way is safe because BLK_MQ_F_TAG_QUEUE_SHARED isn't used by
blk_mq_map_swqueue(), and this flag is mainly checked in fast IO code
path.
Prepare for removing ->elevator_lock from blk_mq_map_swqueue() which
is supposed to be called when elevator switch can't be done.
Reviewed-by: Nilay Shroff <nilay@linux.ibm.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reported-by: Nilay Shroff <nilay@linux.ibm.com>
Closes: https://lore.kernel.org/linux-block/567cb7ab-23d6-4cee-a915-c8cdac903ddd@linux.ibm.com/
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250505141805.2751237-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The block layer bounce buffering support is unused now, remove it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20250505081138.3435992-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_flush_plug_list() has a fast path if all requests in the plug
are destined for the same request_queue. It calls ->queue_rqs() with the
whole batch of requests, falling back on ->queue_rq() for any requests
not handled by ->queue_rqs(). However, if the requests are destined for
multiple queues, blk_mq_flush_plug_list() has a slow path that calls
blk_mq_dispatch_list() repeatedly to filter the requests by ctx/hctx.
Each queue's requests are inserted into the hctx's dispatch list under a
spinlock, then __blk_mq_sched_dispatch_requests() takes them out of the
dispatch list (taking the spinlock again), and finally
blk_mq_dispatch_rq_list() calls ->queue_rq() on each request.
Acquiring the hctx spinlock twice and calling ->queue_rq() instead of
->queue_rqs() makes the slow path significantly more expensive. Thus,
batching more requests into a single plug (e.g. io_uring_enter syscall)
can counterintuitively hurt performance by causing the plug to span
multiple queues. We have observed 2-3% of CPU time spent acquiring the
hctx spinlock alone on workloads issuing requests to multiple NVMe
devices in the same io_uring SQE batches.
Add a medium path in blk_mq_flush_plug_list() for plugs that don't have
elevators or come from a schedule, but do span multiple queues. Filter
the requests by queue and call ->queue_rqs()/->queue_rq() on the list of
requests destined to each request_queue.
With this change, we no longer see any CPU time spent in _raw_spin_lock
from blk_mq_flush_plug_list and throughput increases accordingly.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250426011728.4189119-4-csander@purestorage.com
[axboe: fix whitespace damage]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Factor out the logic from blk_mq_flush_plug_list() that calls
->queue_rqs() with a fallback to ->queue_rq() into a helper function
blk_mq_dispatch_queue_requests(). This is in preparation for using this
code with other lists of requests.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250426011728.4189119-3-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_plug_issue_direct(), __blk_mq_flush_plug_list(), and
blk_mq_dispatch_plug_list() take a struct blk_plug * but only use its
mq_list. Pass the struct rq_list * instead in preparation for calling
them with other lists of requests.
Drop "plug" from the function names as they are no longer plug-specific.
Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250426011728.4189119-2-csander@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
On x86, rep stos will be emitted to clear the the blk_mq_alloc_data
struct, as not all members are being explicitly initialied. Depending on
the type of CPU, this is a noticeable slowdown compared to just ensuring
that the struct is fully initialized when setup.
For the 4 spots that setup a struct blk_mq_alloc_data on the stack,
ensure all members are being initialized.
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The 'nr_budgets' argument of blk_mq_dispatch_rq_list() is either the
number of elements in the 'list' argument or zero. Instead of passing
the number of list elements to blk_mq_dispatch_rq_list(), pass a boolean
argument that indicates whether or not blk_mq_dispatch_rq_list() should
request the block driver for a budget for each request in 'list'.
Remove the code for counting list elements from blk_mq_dispatch_rq_list()
callers where possible. Remove the code that decrements nr_budgets from
blk_mq_dispatch_rq_list() because it is superfluous. Each request that
is processed by blk_mq_dispatch_rq_list() is in one of these two states
if 'get_budget' is false:
* Either the request is on 'list' and the budget for the request has to
be released from the error path.
* Or the request is not on 'list' and q->mq_ops->queue_rq() has already
released the budget (ret != BLK_STS_OK) or q->mq_ops->queue_rq() will
release the budget asynchronously (ret == BLK_STS_OK).
Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: John Garry <john.g.garry@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20250415205134.3650042-1-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
->elevator_lock depends on queue freeze lock, see block/blk-sysfs.c.
queue freeze lock depends on fs_reclaim.
So don't grab elevator lock during queue initialization which needs to
call kmalloc(GFP_KERNEL), and we can cut the dependency between
->elevator_lock and fs_reclaim, then the lockdep warning can be killed.
This way is safe because elevator setting isn't ready to run during
queue initialization.
There isn't such issue in __blk_mq_update_nr_hw_queues() because
memalloc_noio_save() is called before acquiring elevator lock.
Fixes the following lockdep warning:
https://lore.kernel.org/linux-block/67e6b425.050a0220.2f068f.007b.GAE@google.com/
Reported-by: syzbot+4c7e0f9b94ad65811efb@syzkaller.appspotmail.com
Cc: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250403105402.1334206-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make sure ->nr_integrity_segments is cloned in blk_rq_prep_clone(),
otherwise requests cloned by device-mapper multipath will not have the
proper nr_integrity_segments values set, then BUG() is hit from
sg_alloc_table_chained().
Fixes: b0fd271d5f ("block: add request clone interface (v2)")
Cc: stable@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250310115453.2271109-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, hctx attributes (nr_tags, nr_reserved_tags, and cpu_list)
are protected using `q->sysfs_lock`. However, these attributes can be
updated in multiple scenarios:
- During the driver's probe method.
- When updating nr_hw_queues.
- When writing to the sysfs attribute nr_requests,
which can modify nr_tags.
The nr_requests attribute is already protected using q->elevator_lock,
but none of the update paths actually use q->sysfs_lock to protect hctx
attributes. So to ensure proper synchronization, replace q->sysfs_lock
with q->elevator_lock when reading hctx attributes through sysfs.
Additionally, blk_mq_update_nr_hw_queues allocates and updates hctx.
The allocation of hctx is protected using q->elevator_lock, however,
updating hctx params happens without any protection, so safeguard hctx
param update path by also using q->elevator_lock.
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250306093956.2818808-1-nilay@linux.ibm.com
[axboe: wrap comment at 80 chars]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A queue's elevator can be updated either when modifying nr_hw_queues
or through the sysfs scheduler attribute. Currently, elevator switching/
updating is protected using q->sysfs_lock, but this has led to lockdep
splats[1] due to inconsistent lock ordering between q->sysfs_lock and
the freeze-lock in multiple block layer call sites.
As the scope of q->sysfs_lock is not well-defined, its (mis)use has
resulted in numerous lockdep warnings. To address this, introduce a new
q->elevator_lock, dedicated specifically for protecting elevator
switches/updates. And we'd now use this new q->elevator_lock instead of
q->sysfs_lock for protecting elevator switches/updates.
While at it, make elv_iosched_load_module() a static function, as it is
only called from elv_iosched_store(). Also, remove redundant parameters
from elv_iosched_load_module() function signature.
[1] https://lore.kernel.org/all/67637e70.050a0220.3157ee.000c.GAE@google.com/
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Link: https://lore.kernel.org/r/20250304102551.2533767-5-nilay@linux.ibm.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Fengnan Chang
Xue He
Nilay Shroff
Keith Busch
Yu Kuai
Ming Lei
Johannes Thumshirn
Damien Le Moal
Jens Axboe
Christoph Hellwig
Caleb Sander Mateos
Bart Van Assche
Nitesh Shetty
Guixin Liu