2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/block.h>
43 #include "blk-mq-sched.h"
46 #ifdef CONFIG_DEBUG_FS
47 struct dentry
*blk_debugfs_root
;
50 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
51 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
56 DEFINE_IDA(blk_queue_ida
);
59 * For the allocated request tables
61 struct kmem_cache
*request_cachep
;
64 * For queue allocation
66 struct kmem_cache
*blk_requestq_cachep
;
69 * Controlling structure to kblockd
71 static struct workqueue_struct
*kblockd_workqueue
;
73 static void blk_clear_congested(struct request_list
*rl
, int sync
)
75 #ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
79 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
80 * flip its congestion state for events on other blkcgs.
82 if (rl
== &rl
->q
->root_rl
)
83 clear_wb_congested(rl
->q
->backing_dev_info
->wb
.congested
, sync
);
87 static void blk_set_congested(struct request_list
*rl
, int sync
)
89 #ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
92 /* see blk_clear_congested() */
93 if (rl
== &rl
->q
->root_rl
)
94 set_wb_congested(rl
->q
->backing_dev_info
->wb
.congested
, sync
);
98 void blk_queue_congestion_threshold(struct request_queue
*q
)
102 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
103 if (nr
> q
->nr_requests
)
105 q
->nr_congestion_on
= nr
;
107 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
110 q
->nr_congestion_off
= nr
;
113 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
115 memset(rq
, 0, sizeof(*rq
));
117 INIT_LIST_HEAD(&rq
->queuelist
);
118 INIT_LIST_HEAD(&rq
->timeout_list
);
121 rq
->__sector
= (sector_t
) -1;
122 INIT_HLIST_NODE(&rq
->hash
);
123 RB_CLEAR_NODE(&rq
->rb_node
);
125 rq
->internal_tag
= -1;
126 rq
->start_time
= jiffies
;
127 set_start_time_ns(rq
);
130 EXPORT_SYMBOL(blk_rq_init
);
132 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
133 unsigned int nbytes
, int error
)
136 bio
->bi_error
= error
;
138 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
139 bio_set_flag(bio
, BIO_QUIET
);
141 bio_advance(bio
, nbytes
);
143 /* don't actually finish bio if it's part of flush sequence */
144 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
148 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
150 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
151 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
152 (unsigned long long) rq
->cmd_flags
);
154 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
155 (unsigned long long)blk_rq_pos(rq
),
156 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
157 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
158 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
160 EXPORT_SYMBOL(blk_dump_rq_flags
);
162 static void blk_delay_work(struct work_struct
*work
)
164 struct request_queue
*q
;
166 q
= container_of(work
, struct request_queue
, delay_work
.work
);
167 spin_lock_irq(q
->queue_lock
);
169 spin_unlock_irq(q
->queue_lock
);
173 * blk_delay_queue - restart queueing after defined interval
174 * @q: The &struct request_queue in question
175 * @msecs: Delay in msecs
178 * Sometimes queueing needs to be postponed for a little while, to allow
179 * resources to come back. This function will make sure that queueing is
180 * restarted around the specified time. Queue lock must be held.
182 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
184 if (likely(!blk_queue_dead(q
)))
185 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
186 msecs_to_jiffies(msecs
));
188 EXPORT_SYMBOL(blk_delay_queue
);
191 * blk_start_queue_async - asynchronously restart a previously stopped queue
192 * @q: The &struct request_queue in question
195 * blk_start_queue_async() will clear the stop flag on the queue, and
196 * ensure that the request_fn for the queue is run from an async
199 void blk_start_queue_async(struct request_queue
*q
)
201 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
202 blk_run_queue_async(q
);
204 EXPORT_SYMBOL(blk_start_queue_async
);
207 * blk_start_queue - restart a previously stopped queue
208 * @q: The &struct request_queue in question
211 * blk_start_queue() will clear the stop flag on the queue, and call
212 * the request_fn for the queue if it was in a stopped state when
213 * entered. Also see blk_stop_queue(). Queue lock must be held.
215 void blk_start_queue(struct request_queue
*q
)
217 WARN_ON(!irqs_disabled());
219 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
222 EXPORT_SYMBOL(blk_start_queue
);
225 * blk_stop_queue - stop a queue
226 * @q: The &struct request_queue in question
229 * The Linux block layer assumes that a block driver will consume all
230 * entries on the request queue when the request_fn strategy is called.
231 * Often this will not happen, because of hardware limitations (queue
232 * depth settings). If a device driver gets a 'queue full' response,
233 * or if it simply chooses not to queue more I/O at one point, it can
234 * call this function to prevent the request_fn from being called until
235 * the driver has signalled it's ready to go again. This happens by calling
236 * blk_start_queue() to restart queue operations. Queue lock must be held.
238 void blk_stop_queue(struct request_queue
*q
)
240 cancel_delayed_work(&q
->delay_work
);
241 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
243 EXPORT_SYMBOL(blk_stop_queue
);
246 * blk_sync_queue - cancel any pending callbacks on a queue
250 * The block layer may perform asynchronous callback activity
251 * on a queue, such as calling the unplug function after a timeout.
252 * A block device may call blk_sync_queue to ensure that any
253 * such activity is cancelled, thus allowing it to release resources
254 * that the callbacks might use. The caller must already have made sure
255 * that its ->make_request_fn will not re-add plugging prior to calling
258 * This function does not cancel any asynchronous activity arising
259 * out of elevator or throttling code. That would require elevator_exit()
260 * and blkcg_exit_queue() to be called with queue lock initialized.
263 void blk_sync_queue(struct request_queue
*q
)
265 del_timer_sync(&q
->timeout
);
268 struct blk_mq_hw_ctx
*hctx
;
271 queue_for_each_hw_ctx(q
, hctx
, i
) {
272 cancel_work_sync(&hctx
->run_work
);
273 cancel_delayed_work_sync(&hctx
->delay_work
);
276 cancel_delayed_work_sync(&q
->delay_work
);
279 EXPORT_SYMBOL(blk_sync_queue
);
282 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
283 * @q: The queue to run
286 * Invoke request handling on a queue if there are any pending requests.
287 * May be used to restart request handling after a request has completed.
288 * This variant runs the queue whether or not the queue has been
289 * stopped. Must be called with the queue lock held and interrupts
290 * disabled. See also @blk_run_queue.
292 inline void __blk_run_queue_uncond(struct request_queue
*q
)
294 if (unlikely(blk_queue_dead(q
)))
298 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
299 * the queue lock internally. As a result multiple threads may be
300 * running such a request function concurrently. Keep track of the
301 * number of active request_fn invocations such that blk_drain_queue()
302 * can wait until all these request_fn calls have finished.
304 q
->request_fn_active
++;
306 q
->request_fn_active
--;
308 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
311 * __blk_run_queue - run a single device queue
312 * @q: The queue to run
315 * See @blk_run_queue. This variant must be called with the queue lock
316 * held and interrupts disabled.
318 void __blk_run_queue(struct request_queue
*q
)
320 if (unlikely(blk_queue_stopped(q
)))
323 __blk_run_queue_uncond(q
);
325 EXPORT_SYMBOL(__blk_run_queue
);
328 * blk_run_queue_async - run a single device queue in workqueue context
329 * @q: The queue to run
332 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
333 * of us. The caller must hold the queue lock.
335 void blk_run_queue_async(struct request_queue
*q
)
337 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
338 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
340 EXPORT_SYMBOL(blk_run_queue_async
);
343 * blk_run_queue - run a single device queue
344 * @q: The queue to run
347 * Invoke request handling on this queue, if it has pending work to do.
348 * May be used to restart queueing when a request has completed.
350 void blk_run_queue(struct request_queue
*q
)
354 spin_lock_irqsave(q
->queue_lock
, flags
);
356 spin_unlock_irqrestore(q
->queue_lock
, flags
);
358 EXPORT_SYMBOL(blk_run_queue
);
360 void blk_put_queue(struct request_queue
*q
)
362 kobject_put(&q
->kobj
);
364 EXPORT_SYMBOL(blk_put_queue
);
367 * __blk_drain_queue - drain requests from request_queue
369 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
371 * Drain requests from @q. If @drain_all is set, all requests are drained.
372 * If not, only ELVPRIV requests are drained. The caller is responsible
373 * for ensuring that no new requests which need to be drained are queued.
375 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
376 __releases(q
->queue_lock
)
377 __acquires(q
->queue_lock
)
381 lockdep_assert_held(q
->queue_lock
);
387 * The caller might be trying to drain @q before its
388 * elevator is initialized.
391 elv_drain_elevator(q
);
393 blkcg_drain_queue(q
);
396 * This function might be called on a queue which failed
397 * driver init after queue creation or is not yet fully
398 * active yet. Some drivers (e.g. fd and loop) get unhappy
399 * in such cases. Kick queue iff dispatch queue has
400 * something on it and @q has request_fn set.
402 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
405 drain
|= q
->nr_rqs_elvpriv
;
406 drain
|= q
->request_fn_active
;
409 * Unfortunately, requests are queued at and tracked from
410 * multiple places and there's no single counter which can
411 * be drained. Check all the queues and counters.
414 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
415 drain
|= !list_empty(&q
->queue_head
);
416 for (i
= 0; i
< 2; i
++) {
417 drain
|= q
->nr_rqs
[i
];
418 drain
|= q
->in_flight
[i
];
420 drain
|= !list_empty(&fq
->flush_queue
[i
]);
427 spin_unlock_irq(q
->queue_lock
);
431 spin_lock_irq(q
->queue_lock
);
435 * With queue marked dead, any woken up waiter will fail the
436 * allocation path, so the wakeup chaining is lost and we're
437 * left with hung waiters. We need to wake up those waiters.
440 struct request_list
*rl
;
442 blk_queue_for_each_rl(rl
, q
)
443 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
444 wake_up_all(&rl
->wait
[i
]);
449 * blk_queue_bypass_start - enter queue bypass mode
450 * @q: queue of interest
452 * In bypass mode, only the dispatch FIFO queue of @q is used. This
453 * function makes @q enter bypass mode and drains all requests which were
454 * throttled or issued before. On return, it's guaranteed that no request
455 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
456 * inside queue or RCU read lock.
458 void blk_queue_bypass_start(struct request_queue
*q
)
460 spin_lock_irq(q
->queue_lock
);
462 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
463 spin_unlock_irq(q
->queue_lock
);
466 * Queues start drained. Skip actual draining till init is
467 * complete. This avoids lenghty delays during queue init which
468 * can happen many times during boot.
470 if (blk_queue_init_done(q
)) {
471 spin_lock_irq(q
->queue_lock
);
472 __blk_drain_queue(q
, false);
473 spin_unlock_irq(q
->queue_lock
);
475 /* ensure blk_queue_bypass() is %true inside RCU read lock */
479 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
482 * blk_queue_bypass_end - leave queue bypass mode
483 * @q: queue of interest
485 * Leave bypass mode and restore the normal queueing behavior.
487 void blk_queue_bypass_end(struct request_queue
*q
)
489 spin_lock_irq(q
->queue_lock
);
490 if (!--q
->bypass_depth
)
491 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
492 WARN_ON_ONCE(q
->bypass_depth
< 0);
493 spin_unlock_irq(q
->queue_lock
);
495 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
497 void blk_set_queue_dying(struct request_queue
*q
)
499 spin_lock_irq(q
->queue_lock
);
500 queue_flag_set(QUEUE_FLAG_DYING
, q
);
501 spin_unlock_irq(q
->queue_lock
);
504 blk_mq_wake_waiters(q
);
506 struct request_list
*rl
;
508 spin_lock_irq(q
->queue_lock
);
509 blk_queue_for_each_rl(rl
, q
) {
511 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
512 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
515 spin_unlock_irq(q
->queue_lock
);
518 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
521 * blk_cleanup_queue - shutdown a request queue
522 * @q: request queue to shutdown
524 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
525 * put it. All future requests will be failed immediately with -ENODEV.
527 void blk_cleanup_queue(struct request_queue
*q
)
529 spinlock_t
*lock
= q
->queue_lock
;
531 /* mark @q DYING, no new request or merges will be allowed afterwards */
532 mutex_lock(&q
->sysfs_lock
);
533 blk_set_queue_dying(q
);
537 * A dying queue is permanently in bypass mode till released. Note
538 * that, unlike blk_queue_bypass_start(), we aren't performing
539 * synchronize_rcu() after entering bypass mode to avoid the delay
540 * as some drivers create and destroy a lot of queues while
541 * probing. This is still safe because blk_release_queue() will be
542 * called only after the queue refcnt drops to zero and nothing,
543 * RCU or not, would be traversing the queue by then.
546 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
548 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
549 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
550 queue_flag_set(QUEUE_FLAG_DYING
, q
);
551 spin_unlock_irq(lock
);
552 mutex_unlock(&q
->sysfs_lock
);
555 * Drain all requests queued before DYING marking. Set DEAD flag to
556 * prevent that q->request_fn() gets invoked after draining finished.
561 __blk_drain_queue(q
, true);
562 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
563 spin_unlock_irq(lock
);
565 /* for synchronous bio-based driver finish in-flight integrity i/o */
566 blk_flush_integrity();
568 /* @q won't process any more request, flush async actions */
569 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
573 blk_mq_free_queue(q
);
574 percpu_ref_exit(&q
->q_usage_counter
);
577 if (q
->queue_lock
!= &q
->__queue_lock
)
578 q
->queue_lock
= &q
->__queue_lock
;
579 spin_unlock_irq(lock
);
581 put_disk_devt(q
->disk_devt
);
583 /* @q is and will stay empty, shutdown and put */
586 EXPORT_SYMBOL(blk_cleanup_queue
);
588 /* Allocate memory local to the request queue */
589 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
591 struct request_queue
*q
= data
;
593 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
596 static void free_request_simple(void *element
, void *data
)
598 kmem_cache_free(request_cachep
, element
);
601 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
603 struct request_queue
*q
= data
;
606 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
608 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
615 static void free_request_size(void *element
, void *data
)
617 struct request_queue
*q
= data
;
620 q
->exit_rq_fn(q
, element
);
624 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
627 if (unlikely(rl
->rq_pool
))
631 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
632 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
633 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
634 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
637 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
638 alloc_request_size
, free_request_size
,
639 q
, gfp_mask
, q
->node
);
641 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
642 alloc_request_simple
, free_request_simple
,
643 q
, gfp_mask
, q
->node
);
651 void blk_exit_rl(struct request_list
*rl
)
654 mempool_destroy(rl
->rq_pool
);
657 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
659 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
661 EXPORT_SYMBOL(blk_alloc_queue
);
663 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
668 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
674 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
675 !atomic_read(&q
->mq_freeze_depth
) ||
677 if (blk_queue_dying(q
))
684 void blk_queue_exit(struct request_queue
*q
)
686 percpu_ref_put(&q
->q_usage_counter
);
689 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
691 struct request_queue
*q
=
692 container_of(ref
, struct request_queue
, q_usage_counter
);
694 wake_up_all(&q
->mq_freeze_wq
);
697 static void blk_rq_timed_out_timer(unsigned long data
)
699 struct request_queue
*q
= (struct request_queue
*)data
;
701 kblockd_schedule_work(&q
->timeout_work
);
704 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
706 struct request_queue
*q
;
708 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
709 gfp_mask
| __GFP_ZERO
, node_id
);
713 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
717 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
721 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
722 if (!q
->backing_dev_info
)
725 q
->backing_dev_info
->ra_pages
=
726 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
727 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
728 q
->backing_dev_info
->name
= "block";
731 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
732 laptop_mode_timer_fn
, (unsigned long) q
);
733 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
734 INIT_LIST_HEAD(&q
->queue_head
);
735 INIT_LIST_HEAD(&q
->timeout_list
);
736 INIT_LIST_HEAD(&q
->icq_list
);
737 #ifdef CONFIG_BLK_CGROUP
738 INIT_LIST_HEAD(&q
->blkg_list
);
740 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
742 kobject_init(&q
->kobj
, &blk_queue_ktype
);
744 mutex_init(&q
->sysfs_lock
);
745 spin_lock_init(&q
->__queue_lock
);
748 * By default initialize queue_lock to internal lock and driver can
749 * override it later if need be.
751 q
->queue_lock
= &q
->__queue_lock
;
754 * A queue starts its life with bypass turned on to avoid
755 * unnecessary bypass on/off overhead and nasty surprises during
756 * init. The initial bypass will be finished when the queue is
757 * registered by blk_register_queue().
760 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
762 init_waitqueue_head(&q
->mq_freeze_wq
);
765 * Init percpu_ref in atomic mode so that it's faster to shutdown.
766 * See blk_register_queue() for details.
768 if (percpu_ref_init(&q
->q_usage_counter
,
769 blk_queue_usage_counter_release
,
770 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
773 if (blkcg_init_queue(q
))
779 percpu_ref_exit(&q
->q_usage_counter
);
781 bdi_put(q
->backing_dev_info
);
783 bioset_free(q
->bio_split
);
785 ida_simple_remove(&blk_queue_ida
, q
->id
);
787 kmem_cache_free(blk_requestq_cachep
, q
);
790 EXPORT_SYMBOL(blk_alloc_queue_node
);
793 * blk_init_queue - prepare a request queue for use with a block device
794 * @rfn: The function to be called to process requests that have been
795 * placed on the queue.
796 * @lock: Request queue spin lock
799 * If a block device wishes to use the standard request handling procedures,
800 * which sorts requests and coalesces adjacent requests, then it must
801 * call blk_init_queue(). The function @rfn will be called when there
802 * are requests on the queue that need to be processed. If the device
803 * supports plugging, then @rfn may not be called immediately when requests
804 * are available on the queue, but may be called at some time later instead.
805 * Plugged queues are generally unplugged when a buffer belonging to one
806 * of the requests on the queue is needed, or due to memory pressure.
808 * @rfn is not required, or even expected, to remove all requests off the
809 * queue, but only as many as it can handle at a time. If it does leave
810 * requests on the queue, it is responsible for arranging that the requests
811 * get dealt with eventually.
813 * The queue spin lock must be held while manipulating the requests on the
814 * request queue; this lock will be taken also from interrupt context, so irq
815 * disabling is needed for it.
817 * Function returns a pointer to the initialized request queue, or %NULL if
821 * blk_init_queue() must be paired with a blk_cleanup_queue() call
822 * when the block device is deactivated (such as at module unload).
825 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
827 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
829 EXPORT_SYMBOL(blk_init_queue
);
831 struct request_queue
*
832 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
834 struct request_queue
*q
;
836 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
842 q
->queue_lock
= lock
;
843 if (blk_init_allocated_queue(q
) < 0) {
844 blk_cleanup_queue(q
);
850 EXPORT_SYMBOL(blk_init_queue_node
);
852 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
855 int blk_init_allocated_queue(struct request_queue
*q
)
857 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
861 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
862 goto out_free_flush_queue
;
864 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
865 goto out_exit_flush_rq
;
867 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
868 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
871 * This also sets hw/phys segments, boundary and size
873 blk_queue_make_request(q
, blk_queue_bio
);
875 q
->sg_reserved_size
= INT_MAX
;
877 /* Protect q->elevator from elevator_change */
878 mutex_lock(&q
->sysfs_lock
);
881 if (elevator_init(q
, NULL
)) {
882 mutex_unlock(&q
->sysfs_lock
);
883 goto out_exit_flush_rq
;
886 mutex_unlock(&q
->sysfs_lock
);
891 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
892 out_free_flush_queue
:
893 blk_free_flush_queue(q
->fq
);
897 EXPORT_SYMBOL(blk_init_allocated_queue
);
899 bool blk_get_queue(struct request_queue
*q
)
901 if (likely(!blk_queue_dying(q
))) {
908 EXPORT_SYMBOL(blk_get_queue
);
910 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
912 if (rq
->rq_flags
& RQF_ELVPRIV
) {
913 elv_put_request(rl
->q
, rq
);
915 put_io_context(rq
->elv
.icq
->ioc
);
918 mempool_free(rq
, rl
->rq_pool
);
922 * ioc_batching returns true if the ioc is a valid batching request and
923 * should be given priority access to a request.
925 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
931 * Make sure the process is able to allocate at least 1 request
932 * even if the batch times out, otherwise we could theoretically
935 return ioc
->nr_batch_requests
== q
->nr_batching
||
936 (ioc
->nr_batch_requests
> 0
937 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
941 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
942 * will cause the process to be a "batcher" on all queues in the system. This
943 * is the behaviour we want though - once it gets a wakeup it should be given
946 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
948 if (!ioc
|| ioc_batching(q
, ioc
))
951 ioc
->nr_batch_requests
= q
->nr_batching
;
952 ioc
->last_waited
= jiffies
;
955 static void __freed_request(struct request_list
*rl
, int sync
)
957 struct request_queue
*q
= rl
->q
;
959 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
960 blk_clear_congested(rl
, sync
);
962 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
963 if (waitqueue_active(&rl
->wait
[sync
]))
964 wake_up(&rl
->wait
[sync
]);
966 blk_clear_rl_full(rl
, sync
);
971 * A request has just been released. Account for it, update the full and
972 * congestion status, wake up any waiters. Called under q->queue_lock.
974 static void freed_request(struct request_list
*rl
, bool sync
,
975 req_flags_t rq_flags
)
977 struct request_queue
*q
= rl
->q
;
981 if (rq_flags
& RQF_ELVPRIV
)
984 __freed_request(rl
, sync
);
986 if (unlikely(rl
->starved
[sync
^ 1]))
987 __freed_request(rl
, sync
^ 1);
990 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
992 struct request_list
*rl
;
993 int on_thresh
, off_thresh
;
995 spin_lock_irq(q
->queue_lock
);
997 blk_queue_congestion_threshold(q
);
998 on_thresh
= queue_congestion_on_threshold(q
);
999 off_thresh
= queue_congestion_off_threshold(q
);
1001 blk_queue_for_each_rl(rl
, q
) {
1002 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1003 blk_set_congested(rl
, BLK_RW_SYNC
);
1004 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1005 blk_clear_congested(rl
, BLK_RW_SYNC
);
1007 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1008 blk_set_congested(rl
, BLK_RW_ASYNC
);
1009 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1010 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1012 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1013 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1015 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1016 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1019 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1020 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1022 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1023 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1027 spin_unlock_irq(q
->queue_lock
);
1032 * __get_request - get a free request
1033 * @rl: request list to allocate from
1034 * @op: operation and flags
1035 * @bio: bio to allocate request for (can be %NULL)
1036 * @gfp_mask: allocation mask
1038 * Get a free request from @q. This function may fail under memory
1039 * pressure or if @q is dead.
1041 * Must be called with @q->queue_lock held and,
1042 * Returns ERR_PTR on failure, with @q->queue_lock held.
1043 * Returns request pointer on success, with @q->queue_lock *not held*.
1045 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1046 struct bio
*bio
, gfp_t gfp_mask
)
1048 struct request_queue
*q
= rl
->q
;
1050 struct elevator_type
*et
= q
->elevator
->type
;
1051 struct io_context
*ioc
= rq_ioc(bio
);
1052 struct io_cq
*icq
= NULL
;
1053 const bool is_sync
= op_is_sync(op
);
1055 req_flags_t rq_flags
= RQF_ALLOCED
;
1057 if (unlikely(blk_queue_dying(q
)))
1058 return ERR_PTR(-ENODEV
);
1060 may_queue
= elv_may_queue(q
, op
);
1061 if (may_queue
== ELV_MQUEUE_NO
)
1064 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1065 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1067 * The queue will fill after this allocation, so set
1068 * it as full, and mark this process as "batching".
1069 * This process will be allowed to complete a batch of
1070 * requests, others will be blocked.
1072 if (!blk_rl_full(rl
, is_sync
)) {
1073 ioc_set_batching(q
, ioc
);
1074 blk_set_rl_full(rl
, is_sync
);
1076 if (may_queue
!= ELV_MQUEUE_MUST
1077 && !ioc_batching(q
, ioc
)) {
1079 * The queue is full and the allocating
1080 * process is not a "batcher", and not
1081 * exempted by the IO scheduler
1083 return ERR_PTR(-ENOMEM
);
1087 blk_set_congested(rl
, is_sync
);
1091 * Only allow batching queuers to allocate up to 50% over the defined
1092 * limit of requests, otherwise we could have thousands of requests
1093 * allocated with any setting of ->nr_requests
1095 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1096 return ERR_PTR(-ENOMEM
);
1098 q
->nr_rqs
[is_sync
]++;
1099 rl
->count
[is_sync
]++;
1100 rl
->starved
[is_sync
] = 0;
1103 * Decide whether the new request will be managed by elevator. If
1104 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1105 * prevent the current elevator from being destroyed until the new
1106 * request is freed. This guarantees icq's won't be destroyed and
1107 * makes creating new ones safe.
1109 * Flush requests do not use the elevator so skip initialization.
1110 * This allows a request to share the flush and elevator data.
1112 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1113 * it will be created after releasing queue_lock.
1115 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1116 rq_flags
|= RQF_ELVPRIV
;
1117 q
->nr_rqs_elvpriv
++;
1118 if (et
->icq_cache
&& ioc
)
1119 icq
= ioc_lookup_icq(ioc
, q
);
1122 if (blk_queue_io_stat(q
))
1123 rq_flags
|= RQF_IO_STAT
;
1124 spin_unlock_irq(q
->queue_lock
);
1126 /* allocate and init request */
1127 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1132 blk_rq_set_rl(rq
, rl
);
1133 blk_rq_set_prio(rq
, ioc
);
1135 rq
->rq_flags
= rq_flags
;
1138 if (rq_flags
& RQF_ELVPRIV
) {
1139 if (unlikely(et
->icq_cache
&& !icq
)) {
1141 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1147 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1150 /* @rq->elv.icq holds io_context until @rq is freed */
1152 get_io_context(icq
->ioc
);
1156 * ioc may be NULL here, and ioc_batching will be false. That's
1157 * OK, if the queue is under the request limit then requests need
1158 * not count toward the nr_batch_requests limit. There will always
1159 * be some limit enforced by BLK_BATCH_TIME.
1161 if (ioc_batching(q
, ioc
))
1162 ioc
->nr_batch_requests
--;
1164 trace_block_getrq(q
, bio
, op
);
1169 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1170 * and may fail indefinitely under memory pressure and thus
1171 * shouldn't stall IO. Treat this request as !elvpriv. This will
1172 * disturb iosched and blkcg but weird is bettern than dead.
1174 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1175 __func__
, dev_name(q
->backing_dev_info
->dev
));
1177 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1180 spin_lock_irq(q
->queue_lock
);
1181 q
->nr_rqs_elvpriv
--;
1182 spin_unlock_irq(q
->queue_lock
);
1187 * Allocation failed presumably due to memory. Undo anything we
1188 * might have messed up.
1190 * Allocating task should really be put onto the front of the wait
1191 * queue, but this is pretty rare.
1193 spin_lock_irq(q
->queue_lock
);
1194 freed_request(rl
, is_sync
, rq_flags
);
1197 * in the very unlikely event that allocation failed and no
1198 * requests for this direction was pending, mark us starved so that
1199 * freeing of a request in the other direction will notice
1200 * us. another possible fix would be to split the rq mempool into
1204 if (unlikely(rl
->count
[is_sync
] == 0))
1205 rl
->starved
[is_sync
] = 1;
1206 return ERR_PTR(-ENOMEM
);
1210 * get_request - get a free request
1211 * @q: request_queue to allocate request from
1212 * @op: operation and flags
1213 * @bio: bio to allocate request for (can be %NULL)
1214 * @gfp_mask: allocation mask
1216 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1217 * this function keeps retrying under memory pressure and fails iff @q is dead.
1219 * Must be called with @q->queue_lock held and,
1220 * Returns ERR_PTR on failure, with @q->queue_lock held.
1221 * Returns request pointer on success, with @q->queue_lock *not held*.
1223 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1224 struct bio
*bio
, gfp_t gfp_mask
)
1226 const bool is_sync
= op_is_sync(op
);
1228 struct request_list
*rl
;
1231 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1233 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1237 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1242 /* wait on @rl and retry */
1243 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1244 TASK_UNINTERRUPTIBLE
);
1246 trace_block_sleeprq(q
, bio
, op
);
1248 spin_unlock_irq(q
->queue_lock
);
1252 * After sleeping, we become a "batching" process and will be able
1253 * to allocate at least one request, and up to a big batch of them
1254 * for a small period time. See ioc_batching, ioc_set_batching
1256 ioc_set_batching(q
, current
->io_context
);
1258 spin_lock_irq(q
->queue_lock
);
1259 finish_wait(&rl
->wait
[is_sync
], &wait
);
1264 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1269 /* create ioc upfront */
1270 create_io_context(gfp_mask
, q
->node
);
1272 spin_lock_irq(q
->queue_lock
);
1273 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1275 spin_unlock_irq(q
->queue_lock
);
1279 /* q->queue_lock is unlocked at this point */
1281 rq
->__sector
= (sector_t
) -1;
1282 rq
->bio
= rq
->biotail
= NULL
;
1286 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1289 return blk_mq_alloc_request(q
, rw
,
1290 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1291 0 : BLK_MQ_REQ_NOWAIT
);
1293 return blk_old_get_request(q
, rw
, gfp_mask
);
1295 EXPORT_SYMBOL(blk_get_request
);
1298 * blk_requeue_request - put a request back on queue
1299 * @q: request queue where request should be inserted
1300 * @rq: request to be inserted
1303 * Drivers often keep queueing requests until the hardware cannot accept
1304 * more, when that condition happens we need to put the request back
1305 * on the queue. Must be called with queue lock held.
1307 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1309 blk_delete_timer(rq
);
1310 blk_clear_rq_complete(rq
);
1311 trace_block_rq_requeue(q
, rq
);
1312 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1314 if (rq
->rq_flags
& RQF_QUEUED
)
1315 blk_queue_end_tag(q
, rq
);
1317 BUG_ON(blk_queued_rq(rq
));
1319 elv_requeue_request(q
, rq
);
1321 EXPORT_SYMBOL(blk_requeue_request
);
1323 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1326 blk_account_io_start(rq
, true);
1327 __elv_add_request(q
, rq
, where
);
1330 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1335 if (now
== part
->stamp
)
1338 inflight
= part_in_flight(part
);
1340 __part_stat_add(cpu
, part
, time_in_queue
,
1341 inflight
* (now
- part
->stamp
));
1342 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1348 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1349 * @cpu: cpu number for stats access
1350 * @part: target partition
1352 * The average IO queue length and utilisation statistics are maintained
1353 * by observing the current state of the queue length and the amount of
1354 * time it has been in this state for.
1356 * Normally, that accounting is done on IO completion, but that can result
1357 * in more than a second's worth of IO being accounted for within any one
1358 * second, leading to >100% utilisation. To deal with that, we call this
1359 * function to do a round-off before returning the results when reading
1360 * /proc/diskstats. This accounts immediately for all queue usage up to
1361 * the current jiffies and restarts the counters again.
1363 void part_round_stats(int cpu
, struct hd_struct
*part
)
1365 unsigned long now
= jiffies
;
1368 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1369 part_round_stats_single(cpu
, part
, now
);
1371 EXPORT_SYMBOL_GPL(part_round_stats
);
1374 static void blk_pm_put_request(struct request
*rq
)
1376 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1377 pm_runtime_mark_last_busy(rq
->q
->dev
);
1380 static inline void blk_pm_put_request(struct request
*rq
) {}
1384 * queue lock must be held
1386 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1388 req_flags_t rq_flags
= req
->rq_flags
;
1394 blk_mq_free_request(req
);
1398 blk_pm_put_request(req
);
1400 elv_completed_request(q
, req
);
1402 /* this is a bio leak */
1403 WARN_ON(req
->bio
!= NULL
);
1405 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1408 * Request may not have originated from ll_rw_blk. if not,
1409 * it didn't come out of our reserved rq pools
1411 if (rq_flags
& RQF_ALLOCED
) {
1412 struct request_list
*rl
= blk_rq_rl(req
);
1413 bool sync
= op_is_sync(req
->cmd_flags
);
1415 BUG_ON(!list_empty(&req
->queuelist
));
1416 BUG_ON(ELV_ON_HASH(req
));
1418 blk_free_request(rl
, req
);
1419 freed_request(rl
, sync
, rq_flags
);
1423 EXPORT_SYMBOL_GPL(__blk_put_request
);
1425 void blk_put_request(struct request
*req
)
1427 struct request_queue
*q
= req
->q
;
1430 blk_mq_free_request(req
);
1432 unsigned long flags
;
1434 spin_lock_irqsave(q
->queue_lock
, flags
);
1435 __blk_put_request(q
, req
);
1436 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1439 EXPORT_SYMBOL(blk_put_request
);
1441 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1444 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1446 if (!ll_back_merge_fn(q
, req
, bio
))
1449 trace_block_bio_backmerge(q
, req
, bio
);
1451 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1452 blk_rq_set_mixed_merge(req
);
1454 req
->biotail
->bi_next
= bio
;
1456 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1457 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1459 blk_account_io_start(req
, false);
1463 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1466 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1468 if (!ll_front_merge_fn(q
, req
, bio
))
1471 trace_block_bio_frontmerge(q
, req
, bio
);
1473 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1474 blk_rq_set_mixed_merge(req
);
1476 bio
->bi_next
= req
->bio
;
1479 req
->__sector
= bio
->bi_iter
.bi_sector
;
1480 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1481 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1483 blk_account_io_start(req
, false);
1487 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1490 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1492 if (segments
>= queue_max_discard_segments(q
))
1494 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1495 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1498 req
->biotail
->bi_next
= bio
;
1500 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1501 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1502 req
->nr_phys_segments
= segments
+ 1;
1504 blk_account_io_start(req
, false);
1507 req_set_nomerge(q
, req
);
1512 * blk_attempt_plug_merge - try to merge with %current's plugged list
1513 * @q: request_queue new bio is being queued at
1514 * @bio: new bio being queued
1515 * @request_count: out parameter for number of traversed plugged requests
1516 * @same_queue_rq: pointer to &struct request that gets filled in when
1517 * another request associated with @q is found on the plug list
1518 * (optional, may be %NULL)
1520 * Determine whether @bio being queued on @q can be merged with a request
1521 * on %current's plugged list. Returns %true if merge was successful,
1524 * Plugging coalesces IOs from the same issuer for the same purpose without
1525 * going through @q->queue_lock. As such it's more of an issuing mechanism
1526 * than scheduling, and the request, while may have elvpriv data, is not
1527 * added on the elevator at this point. In addition, we don't have
1528 * reliable access to the elevator outside queue lock. Only check basic
1529 * merging parameters without querying the elevator.
1531 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1533 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1534 unsigned int *request_count
,
1535 struct request
**same_queue_rq
)
1537 struct blk_plug
*plug
;
1539 struct list_head
*plug_list
;
1541 plug
= current
->plug
;
1547 plug_list
= &plug
->mq_list
;
1549 plug_list
= &plug
->list
;
1551 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1552 bool merged
= false;
1557 * Only blk-mq multiple hardware queues case checks the
1558 * rq in the same queue, there should be only one such
1562 *same_queue_rq
= rq
;
1565 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1568 switch (blk_try_merge(rq
, bio
)) {
1569 case ELEVATOR_BACK_MERGE
:
1570 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1572 case ELEVATOR_FRONT_MERGE
:
1573 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1575 case ELEVATOR_DISCARD_MERGE
:
1576 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1589 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1591 struct blk_plug
*plug
;
1593 struct list_head
*plug_list
;
1594 unsigned int ret
= 0;
1596 plug
= current
->plug
;
1601 plug_list
= &plug
->mq_list
;
1603 plug_list
= &plug
->list
;
1605 list_for_each_entry(rq
, plug_list
, queuelist
) {
1613 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1615 if (bio
->bi_opf
& REQ_RAHEAD
)
1616 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1619 req
->__sector
= bio
->bi_iter
.bi_sector
;
1620 if (ioprio_valid(bio_prio(bio
)))
1621 req
->ioprio
= bio_prio(bio
);
1622 blk_rq_bio_prep(req
->q
, req
, bio
);
1625 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1627 struct blk_plug
*plug
;
1628 int where
= ELEVATOR_INSERT_SORT
;
1629 struct request
*req
, *free
;
1630 unsigned int request_count
= 0;
1631 unsigned int wb_acct
;
1634 * low level driver can indicate that it wants pages above a
1635 * certain limit bounced to low memory (ie for highmem, or even
1636 * ISA dma in theory)
1638 blk_queue_bounce(q
, &bio
);
1640 blk_queue_split(q
, &bio
, q
->bio_split
);
1642 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1643 bio
->bi_error
= -EIO
;
1645 return BLK_QC_T_NONE
;
1648 if (op_is_flush(bio
->bi_opf
)) {
1649 spin_lock_irq(q
->queue_lock
);
1650 where
= ELEVATOR_INSERT_FLUSH
;
1655 * Check if we can merge with the plugged list before grabbing
1658 if (!blk_queue_nomerges(q
)) {
1659 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1660 return BLK_QC_T_NONE
;
1662 request_count
= blk_plug_queued_count(q
);
1664 spin_lock_irq(q
->queue_lock
);
1666 switch (elv_merge(q
, &req
, bio
)) {
1667 case ELEVATOR_BACK_MERGE
:
1668 if (!bio_attempt_back_merge(q
, req
, bio
))
1670 elv_bio_merged(q
, req
, bio
);
1671 free
= attempt_back_merge(q
, req
);
1673 __blk_put_request(q
, free
);
1675 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1677 case ELEVATOR_FRONT_MERGE
:
1678 if (!bio_attempt_front_merge(q
, req
, bio
))
1680 elv_bio_merged(q
, req
, bio
);
1681 free
= attempt_front_merge(q
, req
);
1683 __blk_put_request(q
, free
);
1685 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1692 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1695 * Grab a free request. This is might sleep but can not fail.
1696 * Returns with the queue unlocked.
1698 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1700 __wbt_done(q
->rq_wb
, wb_acct
);
1701 bio
->bi_error
= PTR_ERR(req
);
1706 wbt_track(&req
->issue_stat
, wb_acct
);
1709 * After dropping the lock and possibly sleeping here, our request
1710 * may now be mergeable after it had proven unmergeable (above).
1711 * We don't worry about that case for efficiency. It won't happen
1712 * often, and the elevators are able to handle it.
1714 init_request_from_bio(req
, bio
);
1716 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1717 req
->cpu
= raw_smp_processor_id();
1719 plug
= current
->plug
;
1722 * If this is the first request added after a plug, fire
1725 * @request_count may become stale because of schedule
1726 * out, so check plug list again.
1728 if (!request_count
|| list_empty(&plug
->list
))
1729 trace_block_plug(q
);
1731 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1732 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1733 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1734 blk_flush_plug_list(plug
, false);
1735 trace_block_plug(q
);
1738 list_add_tail(&req
->queuelist
, &plug
->list
);
1739 blk_account_io_start(req
, true);
1741 spin_lock_irq(q
->queue_lock
);
1742 add_acct_request(q
, req
, where
);
1745 spin_unlock_irq(q
->queue_lock
);
1748 return BLK_QC_T_NONE
;
1752 * If bio->bi_dev is a partition, remap the location
1754 static inline void blk_partition_remap(struct bio
*bio
)
1756 struct block_device
*bdev
= bio
->bi_bdev
;
1759 * Zone reset does not include bi_size so bio_sectors() is always 0.
1760 * Include a test for the reset op code and perform the remap if needed.
1762 if (bdev
!= bdev
->bd_contains
&&
1763 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1764 struct hd_struct
*p
= bdev
->bd_part
;
1766 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1767 bio
->bi_bdev
= bdev
->bd_contains
;
1769 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1771 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1775 static void handle_bad_sector(struct bio
*bio
)
1777 char b
[BDEVNAME_SIZE
];
1779 printk(KERN_INFO
"attempt to access beyond end of device\n");
1780 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1781 bdevname(bio
->bi_bdev
, b
),
1783 (unsigned long long)bio_end_sector(bio
),
1784 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1787 #ifdef CONFIG_FAIL_MAKE_REQUEST
1789 static DECLARE_FAULT_ATTR(fail_make_request
);
1791 static int __init
setup_fail_make_request(char *str
)
1793 return setup_fault_attr(&fail_make_request
, str
);
1795 __setup("fail_make_request=", setup_fail_make_request
);
1797 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1799 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1802 static int __init
fail_make_request_debugfs(void)
1804 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1805 NULL
, &fail_make_request
);
1807 return PTR_ERR_OR_ZERO(dir
);
1810 late_initcall(fail_make_request_debugfs
);
1812 #else /* CONFIG_FAIL_MAKE_REQUEST */
1814 static inline bool should_fail_request(struct hd_struct
*part
,
1820 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1823 * Check whether this bio extends beyond the end of the device.
1825 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1832 /* Test device or partition size, when known. */
1833 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1835 sector_t sector
= bio
->bi_iter
.bi_sector
;
1837 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1839 * This may well happen - the kernel calls bread()
1840 * without checking the size of the device, e.g., when
1841 * mounting a device.
1843 handle_bad_sector(bio
);
1851 static noinline_for_stack
bool
1852 generic_make_request_checks(struct bio
*bio
)
1854 struct request_queue
*q
;
1855 int nr_sectors
= bio_sectors(bio
);
1857 char b
[BDEVNAME_SIZE
];
1858 struct hd_struct
*part
;
1862 if (bio_check_eod(bio
, nr_sectors
))
1865 q
= bdev_get_queue(bio
->bi_bdev
);
1868 "generic_make_request: Trying to access "
1869 "nonexistent block-device %s (%Lu)\n",
1870 bdevname(bio
->bi_bdev
, b
),
1871 (long long) bio
->bi_iter
.bi_sector
);
1875 part
= bio
->bi_bdev
->bd_part
;
1876 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1877 should_fail_request(&part_to_disk(part
)->part0
,
1878 bio
->bi_iter
.bi_size
))
1882 * If this device has partitions, remap block n
1883 * of partition p to block n+start(p) of the disk.
1885 blk_partition_remap(bio
);
1887 if (bio_check_eod(bio
, nr_sectors
))
1891 * Filter flush bio's early so that make_request based
1892 * drivers without flush support don't have to worry
1895 if (op_is_flush(bio
->bi_opf
) &&
1896 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1897 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1904 switch (bio_op(bio
)) {
1905 case REQ_OP_DISCARD
:
1906 if (!blk_queue_discard(q
))
1909 case REQ_OP_SECURE_ERASE
:
1910 if (!blk_queue_secure_erase(q
))
1913 case REQ_OP_WRITE_SAME
:
1914 if (!bdev_write_same(bio
->bi_bdev
))
1917 case REQ_OP_ZONE_REPORT
:
1918 case REQ_OP_ZONE_RESET
:
1919 if (!bdev_is_zoned(bio
->bi_bdev
))
1922 case REQ_OP_WRITE_ZEROES
:
1923 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1931 * Various block parts want %current->io_context and lazy ioc
1932 * allocation ends up trading a lot of pain for a small amount of
1933 * memory. Just allocate it upfront. This may fail and block
1934 * layer knows how to live with it.
1936 create_io_context(GFP_ATOMIC
, q
->node
);
1938 if (!blkcg_bio_issue_check(q
, bio
))
1941 trace_block_bio_queue(q
, bio
);
1947 bio
->bi_error
= err
;
1953 * generic_make_request - hand a buffer to its device driver for I/O
1954 * @bio: The bio describing the location in memory and on the device.
1956 * generic_make_request() is used to make I/O requests of block
1957 * devices. It is passed a &struct bio, which describes the I/O that needs
1960 * generic_make_request() does not return any status. The
1961 * success/failure status of the request, along with notification of
1962 * completion, is delivered asynchronously through the bio->bi_end_io
1963 * function described (one day) else where.
1965 * The caller of generic_make_request must make sure that bi_io_vec
1966 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1967 * set to describe the device address, and the
1968 * bi_end_io and optionally bi_private are set to describe how
1969 * completion notification should be signaled.
1971 * generic_make_request and the drivers it calls may use bi_next if this
1972 * bio happens to be merged with someone else, and may resubmit the bio to
1973 * a lower device by calling into generic_make_request recursively, which
1974 * means the bio should NOT be touched after the call to ->make_request_fn.
1976 blk_qc_t
generic_make_request(struct bio
*bio
)
1978 struct bio_list bio_list_on_stack
;
1979 blk_qc_t ret
= BLK_QC_T_NONE
;
1981 if (!generic_make_request_checks(bio
))
1985 * We only want one ->make_request_fn to be active at a time, else
1986 * stack usage with stacked devices could be a problem. So use
1987 * current->bio_list to keep a list of requests submited by a
1988 * make_request_fn function. current->bio_list is also used as a
1989 * flag to say if generic_make_request is currently active in this
1990 * task or not. If it is NULL, then no make_request is active. If
1991 * it is non-NULL, then a make_request is active, and new requests
1992 * should be added at the tail
1994 if (current
->bio_list
) {
1995 bio_list_add(current
->bio_list
, bio
);
1999 /* following loop may be a bit non-obvious, and so deserves some
2001 * Before entering the loop, bio->bi_next is NULL (as all callers
2002 * ensure that) so we have a list with a single bio.
2003 * We pretend that we have just taken it off a longer list, so
2004 * we assign bio_list to a pointer to the bio_list_on_stack,
2005 * thus initialising the bio_list of new bios to be
2006 * added. ->make_request() may indeed add some more bios
2007 * through a recursive call to generic_make_request. If it
2008 * did, we find a non-NULL value in bio_list and re-enter the loop
2009 * from the top. In this case we really did just take the bio
2010 * of the top of the list (no pretending) and so remove it from
2011 * bio_list, and call into ->make_request() again.
2013 BUG_ON(bio
->bi_next
);
2014 bio_list_init(&bio_list_on_stack
);
2015 current
->bio_list
= &bio_list_on_stack
;
2017 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2019 if (likely(blk_queue_enter(q
, false) == 0)) {
2020 ret
= q
->make_request_fn(q
, bio
);
2024 bio
= bio_list_pop(current
->bio_list
);
2026 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2032 current
->bio_list
= NULL
; /* deactivate */
2037 EXPORT_SYMBOL(generic_make_request
);
2040 * submit_bio - submit a bio to the block device layer for I/O
2041 * @bio: The &struct bio which describes the I/O
2043 * submit_bio() is very similar in purpose to generic_make_request(), and
2044 * uses that function to do most of the work. Both are fairly rough
2045 * interfaces; @bio must be presetup and ready for I/O.
2048 blk_qc_t
submit_bio(struct bio
*bio
)
2051 * If it's a regular read/write or a barrier with data attached,
2052 * go through the normal accounting stuff before submission.
2054 if (bio_has_data(bio
)) {
2057 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2058 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2060 count
= bio_sectors(bio
);
2062 if (op_is_write(bio_op(bio
))) {
2063 count_vm_events(PGPGOUT
, count
);
2065 task_io_account_read(bio
->bi_iter
.bi_size
);
2066 count_vm_events(PGPGIN
, count
);
2069 if (unlikely(block_dump
)) {
2070 char b
[BDEVNAME_SIZE
];
2071 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2072 current
->comm
, task_pid_nr(current
),
2073 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2074 (unsigned long long)bio
->bi_iter
.bi_sector
,
2075 bdevname(bio
->bi_bdev
, b
),
2080 return generic_make_request(bio
);
2082 EXPORT_SYMBOL(submit_bio
);
2085 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2086 * for new the queue limits
2088 * @rq: the request being checked
2091 * @rq may have been made based on weaker limitations of upper-level queues
2092 * in request stacking drivers, and it may violate the limitation of @q.
2093 * Since the block layer and the underlying device driver trust @rq
2094 * after it is inserted to @q, it should be checked against @q before
2095 * the insertion using this generic function.
2097 * Request stacking drivers like request-based dm may change the queue
2098 * limits when retrying requests on other queues. Those requests need
2099 * to be checked against the new queue limits again during dispatch.
2101 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2104 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2105 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2110 * queue's settings related to segment counting like q->bounce_pfn
2111 * may differ from that of other stacking queues.
2112 * Recalculate it to check the request correctly on this queue's
2115 blk_recalc_rq_segments(rq
);
2116 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2117 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2125 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2126 * @q: the queue to submit the request
2127 * @rq: the request being queued
2129 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2131 unsigned long flags
;
2132 int where
= ELEVATOR_INSERT_BACK
;
2134 if (blk_cloned_rq_check_limits(q
, rq
))
2138 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2142 if (blk_queue_io_stat(q
))
2143 blk_account_io_start(rq
, true);
2144 blk_mq_sched_insert_request(rq
, false, true, false, false);
2148 spin_lock_irqsave(q
->queue_lock
, flags
);
2149 if (unlikely(blk_queue_dying(q
))) {
2150 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2155 * Submitting request must be dequeued before calling this function
2156 * because it will be linked to another request_queue
2158 BUG_ON(blk_queued_rq(rq
));
2160 if (op_is_flush(rq
->cmd_flags
))
2161 where
= ELEVATOR_INSERT_FLUSH
;
2163 add_acct_request(q
, rq
, where
);
2164 if (where
== ELEVATOR_INSERT_FLUSH
)
2166 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2170 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2173 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2174 * @rq: request to examine
2177 * A request could be merge of IOs which require different failure
2178 * handling. This function determines the number of bytes which
2179 * can be failed from the beginning of the request without
2180 * crossing into area which need to be retried further.
2183 * The number of bytes to fail.
2186 * queue_lock must be held.
2188 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2190 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2191 unsigned int bytes
= 0;
2194 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2195 return blk_rq_bytes(rq
);
2198 * Currently the only 'mixing' which can happen is between
2199 * different fastfail types. We can safely fail portions
2200 * which have all the failfast bits that the first one has -
2201 * the ones which are at least as eager to fail as the first
2204 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2205 if ((bio
->bi_opf
& ff
) != ff
)
2207 bytes
+= bio
->bi_iter
.bi_size
;
2210 /* this could lead to infinite loop */
2211 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2214 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2216 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2218 if (blk_do_io_stat(req
)) {
2219 const int rw
= rq_data_dir(req
);
2220 struct hd_struct
*part
;
2223 cpu
= part_stat_lock();
2225 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2230 void blk_account_io_done(struct request
*req
)
2233 * Account IO completion. flush_rq isn't accounted as a
2234 * normal IO on queueing nor completion. Accounting the
2235 * containing request is enough.
2237 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2238 unsigned long duration
= jiffies
- req
->start_time
;
2239 const int rw
= rq_data_dir(req
);
2240 struct hd_struct
*part
;
2243 cpu
= part_stat_lock();
2246 part_stat_inc(cpu
, part
, ios
[rw
]);
2247 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2248 part_round_stats(cpu
, part
);
2249 part_dec_in_flight(part
, rw
);
2251 hd_struct_put(part
);
2258 * Don't process normal requests when queue is suspended
2259 * or in the process of suspending/resuming
2261 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2264 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2265 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2271 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2278 void blk_account_io_start(struct request
*rq
, bool new_io
)
2280 struct hd_struct
*part
;
2281 int rw
= rq_data_dir(rq
);
2284 if (!blk_do_io_stat(rq
))
2287 cpu
= part_stat_lock();
2291 part_stat_inc(cpu
, part
, merges
[rw
]);
2293 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2294 if (!hd_struct_try_get(part
)) {
2296 * The partition is already being removed,
2297 * the request will be accounted on the disk only
2299 * We take a reference on disk->part0 although that
2300 * partition will never be deleted, so we can treat
2301 * it as any other partition.
2303 part
= &rq
->rq_disk
->part0
;
2304 hd_struct_get(part
);
2306 part_round_stats(cpu
, part
);
2307 part_inc_in_flight(part
, rw
);
2315 * blk_peek_request - peek at the top of a request queue
2316 * @q: request queue to peek at
2319 * Return the request at the top of @q. The returned request
2320 * should be started using blk_start_request() before LLD starts
2324 * Pointer to the request at the top of @q if available. Null
2328 * queue_lock must be held.
2330 struct request
*blk_peek_request(struct request_queue
*q
)
2335 while ((rq
= __elv_next_request(q
)) != NULL
) {
2337 rq
= blk_pm_peek_request(q
, rq
);
2341 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2343 * This is the first time the device driver
2344 * sees this request (possibly after
2345 * requeueing). Notify IO scheduler.
2347 if (rq
->rq_flags
& RQF_SORTED
)
2348 elv_activate_rq(q
, rq
);
2351 * just mark as started even if we don't start
2352 * it, a request that has been delayed should
2353 * not be passed by new incoming requests
2355 rq
->rq_flags
|= RQF_STARTED
;
2356 trace_block_rq_issue(q
, rq
);
2359 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2360 q
->end_sector
= rq_end_sector(rq
);
2361 q
->boundary_rq
= NULL
;
2364 if (rq
->rq_flags
& RQF_DONTPREP
)
2367 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2369 * make sure space for the drain appears we
2370 * know we can do this because max_hw_segments
2371 * has been adjusted to be one fewer than the
2374 rq
->nr_phys_segments
++;
2380 ret
= q
->prep_rq_fn(q
, rq
);
2381 if (ret
== BLKPREP_OK
) {
2383 } else if (ret
== BLKPREP_DEFER
) {
2385 * the request may have been (partially) prepped.
2386 * we need to keep this request in the front to
2387 * avoid resource deadlock. RQF_STARTED will
2388 * prevent other fs requests from passing this one.
2390 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2391 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2393 * remove the space for the drain we added
2394 * so that we don't add it again
2396 --rq
->nr_phys_segments
;
2401 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2402 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2404 rq
->rq_flags
|= RQF_QUIET
;
2406 * Mark this request as started so we don't trigger
2407 * any debug logic in the end I/O path.
2409 blk_start_request(rq
);
2410 __blk_end_request_all(rq
, err
);
2412 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2419 EXPORT_SYMBOL(blk_peek_request
);
2421 void blk_dequeue_request(struct request
*rq
)
2423 struct request_queue
*q
= rq
->q
;
2425 BUG_ON(list_empty(&rq
->queuelist
));
2426 BUG_ON(ELV_ON_HASH(rq
));
2428 list_del_init(&rq
->queuelist
);
2431 * the time frame between a request being removed from the lists
2432 * and to it is freed is accounted as io that is in progress at
2435 if (blk_account_rq(rq
)) {
2436 q
->in_flight
[rq_is_sync(rq
)]++;
2437 set_io_start_time_ns(rq
);
2442 * blk_start_request - start request processing on the driver
2443 * @req: request to dequeue
2446 * Dequeue @req and start timeout timer on it. This hands off the
2447 * request to the driver.
2449 * Block internal functions which don't want to start timer should
2450 * call blk_dequeue_request().
2453 * queue_lock must be held.
2455 void blk_start_request(struct request
*req
)
2457 blk_dequeue_request(req
);
2459 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2460 blk_stat_set_issue_time(&req
->issue_stat
);
2461 req
->rq_flags
|= RQF_STATS
;
2462 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2465 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2468 EXPORT_SYMBOL(blk_start_request
);
2471 * blk_fetch_request - fetch a request from a request queue
2472 * @q: request queue to fetch a request from
2475 * Return the request at the top of @q. The request is started on
2476 * return and LLD can start processing it immediately.
2479 * Pointer to the request at the top of @q if available. Null
2483 * queue_lock must be held.
2485 struct request
*blk_fetch_request(struct request_queue
*q
)
2489 rq
= blk_peek_request(q
);
2491 blk_start_request(rq
);
2494 EXPORT_SYMBOL(blk_fetch_request
);
2497 * blk_update_request - Special helper function for request stacking drivers
2498 * @req: the request being processed
2499 * @error: %0 for success, < %0 for error
2500 * @nr_bytes: number of bytes to complete @req
2503 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2504 * the request structure even if @req doesn't have leftover.
2505 * If @req has leftover, sets it up for the next range of segments.
2507 * This special helper function is only for request stacking drivers
2508 * (e.g. request-based dm) so that they can handle partial completion.
2509 * Actual device drivers should use blk_end_request instead.
2511 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2512 * %false return from this function.
2515 * %false - this request doesn't have any more data
2516 * %true - this request has more data
2518 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2522 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2528 * For fs requests, rq is just carrier of independent bio's
2529 * and each partial completion should be handled separately.
2530 * Reset per-request error on each partial completion.
2532 * TODO: tj: This is too subtle. It would be better to let
2533 * low level drivers do what they see fit.
2535 if (!blk_rq_is_passthrough(req
))
2538 if (error
&& !blk_rq_is_passthrough(req
) &&
2539 !(req
->rq_flags
& RQF_QUIET
)) {
2544 error_type
= "recoverable transport";
2547 error_type
= "critical target";
2550 error_type
= "critical nexus";
2553 error_type
= "timeout";
2556 error_type
= "critical space allocation";
2559 error_type
= "critical medium";
2566 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2567 __func__
, error_type
, req
->rq_disk
?
2568 req
->rq_disk
->disk_name
: "?",
2569 (unsigned long long)blk_rq_pos(req
));
2573 blk_account_io_completion(req
, nr_bytes
);
2577 struct bio
*bio
= req
->bio
;
2578 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2580 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2581 req
->bio
= bio
->bi_next
;
2583 req_bio_endio(req
, bio
, bio_bytes
, error
);
2585 total_bytes
+= bio_bytes
;
2586 nr_bytes
-= bio_bytes
;
2597 * Reset counters so that the request stacking driver
2598 * can find how many bytes remain in the request
2601 req
->__data_len
= 0;
2605 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2607 req
->__data_len
-= total_bytes
;
2609 /* update sector only for requests with clear definition of sector */
2610 if (!blk_rq_is_passthrough(req
))
2611 req
->__sector
+= total_bytes
>> 9;
2613 /* mixed attributes always follow the first bio */
2614 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2615 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2616 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2620 * If total number of sectors is less than the first segment
2621 * size, something has gone terribly wrong.
2623 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2624 blk_dump_rq_flags(req
, "request botched");
2625 req
->__data_len
= blk_rq_cur_bytes(req
);
2628 /* recalculate the number of segments */
2629 blk_recalc_rq_segments(req
);
2633 EXPORT_SYMBOL_GPL(blk_update_request
);
2635 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2636 unsigned int nr_bytes
,
2637 unsigned int bidi_bytes
)
2639 if (blk_update_request(rq
, error
, nr_bytes
))
2642 /* Bidi request must be completed as a whole */
2643 if (unlikely(blk_bidi_rq(rq
)) &&
2644 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2647 if (blk_queue_add_random(rq
->q
))
2648 add_disk_randomness(rq
->rq_disk
);
2654 * blk_unprep_request - unprepare a request
2657 * This function makes a request ready for complete resubmission (or
2658 * completion). It happens only after all error handling is complete,
2659 * so represents the appropriate moment to deallocate any resources
2660 * that were allocated to the request in the prep_rq_fn. The queue
2661 * lock is held when calling this.
2663 void blk_unprep_request(struct request
*req
)
2665 struct request_queue
*q
= req
->q
;
2667 req
->rq_flags
&= ~RQF_DONTPREP
;
2668 if (q
->unprep_rq_fn
)
2669 q
->unprep_rq_fn(q
, req
);
2671 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2674 * queue lock must be held
2676 void blk_finish_request(struct request
*req
, int error
)
2678 struct request_queue
*q
= req
->q
;
2680 if (req
->rq_flags
& RQF_STATS
)
2681 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2683 if (req
->rq_flags
& RQF_QUEUED
)
2684 blk_queue_end_tag(q
, req
);
2686 BUG_ON(blk_queued_rq(req
));
2688 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2689 laptop_io_completion(req
->q
->backing_dev_info
);
2691 blk_delete_timer(req
);
2693 if (req
->rq_flags
& RQF_DONTPREP
)
2694 blk_unprep_request(req
);
2696 blk_account_io_done(req
);
2699 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2700 req
->end_io(req
, error
);
2702 if (blk_bidi_rq(req
))
2703 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2705 __blk_put_request(q
, req
);
2708 EXPORT_SYMBOL(blk_finish_request
);
2711 * blk_end_bidi_request - Complete a bidi request
2712 * @rq: the request to complete
2713 * @error: %0 for success, < %0 for error
2714 * @nr_bytes: number of bytes to complete @rq
2715 * @bidi_bytes: number of bytes to complete @rq->next_rq
2718 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2719 * Drivers that supports bidi can safely call this member for any
2720 * type of request, bidi or uni. In the later case @bidi_bytes is
2724 * %false - we are done with this request
2725 * %true - still buffers pending for this request
2727 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2728 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2730 struct request_queue
*q
= rq
->q
;
2731 unsigned long flags
;
2733 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2736 spin_lock_irqsave(q
->queue_lock
, flags
);
2737 blk_finish_request(rq
, error
);
2738 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2744 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2745 * @rq: the request to complete
2746 * @error: %0 for success, < %0 for error
2747 * @nr_bytes: number of bytes to complete @rq
2748 * @bidi_bytes: number of bytes to complete @rq->next_rq
2751 * Identical to blk_end_bidi_request() except that queue lock is
2752 * assumed to be locked on entry and remains so on return.
2755 * %false - we are done with this request
2756 * %true - still buffers pending for this request
2758 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2759 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2761 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2764 blk_finish_request(rq
, error
);
2770 * blk_end_request - Helper function for drivers to complete the request.
2771 * @rq: the request being processed
2772 * @error: %0 for success, < %0 for error
2773 * @nr_bytes: number of bytes to complete
2776 * Ends I/O on a number of bytes attached to @rq.
2777 * If @rq has leftover, sets it up for the next range of segments.
2780 * %false - we are done with this request
2781 * %true - still buffers pending for this request
2783 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2785 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2787 EXPORT_SYMBOL(blk_end_request
);
2790 * blk_end_request_all - Helper function for drives to finish the request.
2791 * @rq: the request to finish
2792 * @error: %0 for success, < %0 for error
2795 * Completely finish @rq.
2797 void blk_end_request_all(struct request
*rq
, int error
)
2800 unsigned int bidi_bytes
= 0;
2802 if (unlikely(blk_bidi_rq(rq
)))
2803 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2805 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2808 EXPORT_SYMBOL(blk_end_request_all
);
2811 * blk_end_request_cur - Helper function to finish the current request chunk.
2812 * @rq: the request to finish the current chunk for
2813 * @error: %0 for success, < %0 for error
2816 * Complete the current consecutively mapped chunk from @rq.
2819 * %false - we are done with this request
2820 * %true - still buffers pending for this request
2822 bool blk_end_request_cur(struct request
*rq
, int error
)
2824 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2826 EXPORT_SYMBOL(blk_end_request_cur
);
2829 * blk_end_request_err - Finish a request till the next failure boundary.
2830 * @rq: the request to finish till the next failure boundary for
2831 * @error: must be negative errno
2834 * Complete @rq till the next failure boundary.
2837 * %false - we are done with this request
2838 * %true - still buffers pending for this request
2840 bool blk_end_request_err(struct request
*rq
, int error
)
2842 WARN_ON(error
>= 0);
2843 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2845 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2848 * __blk_end_request - Helper function for drivers to complete the request.
2849 * @rq: the request being processed
2850 * @error: %0 for success, < %0 for error
2851 * @nr_bytes: number of bytes to complete
2854 * Must be called with queue lock held unlike blk_end_request().
2857 * %false - we are done with this request
2858 * %true - still buffers pending for this request
2860 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2862 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2864 EXPORT_SYMBOL(__blk_end_request
);
2867 * __blk_end_request_all - Helper function for drives to finish the request.
2868 * @rq: the request to finish
2869 * @error: %0 for success, < %0 for error
2872 * Completely finish @rq. Must be called with queue lock held.
2874 void __blk_end_request_all(struct request
*rq
, int error
)
2877 unsigned int bidi_bytes
= 0;
2879 if (unlikely(blk_bidi_rq(rq
)))
2880 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2882 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2885 EXPORT_SYMBOL(__blk_end_request_all
);
2888 * __blk_end_request_cur - Helper function to finish the current request chunk.
2889 * @rq: the request to finish the current chunk for
2890 * @error: %0 for success, < %0 for error
2893 * Complete the current consecutively mapped chunk from @rq. Must
2894 * be called with queue lock held.
2897 * %false - we are done with this request
2898 * %true - still buffers pending for this request
2900 bool __blk_end_request_cur(struct request
*rq
, int error
)
2902 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2904 EXPORT_SYMBOL(__blk_end_request_cur
);
2907 * __blk_end_request_err - Finish a request till the next failure boundary.
2908 * @rq: the request to finish till the next failure boundary for
2909 * @error: must be negative errno
2912 * Complete @rq till the next failure boundary. Must be called
2913 * with queue lock held.
2916 * %false - we are done with this request
2917 * %true - still buffers pending for this request
2919 bool __blk_end_request_err(struct request
*rq
, int error
)
2921 WARN_ON(error
>= 0);
2922 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2924 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2926 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2929 if (bio_has_data(bio
))
2930 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2932 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2933 rq
->bio
= rq
->biotail
= bio
;
2936 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2939 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2941 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2942 * @rq: the request to be flushed
2945 * Flush all pages in @rq.
2947 void rq_flush_dcache_pages(struct request
*rq
)
2949 struct req_iterator iter
;
2950 struct bio_vec bvec
;
2952 rq_for_each_segment(bvec
, rq
, iter
)
2953 flush_dcache_page(bvec
.bv_page
);
2955 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2959 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2960 * @q : the queue of the device being checked
2963 * Check if underlying low-level drivers of a device are busy.
2964 * If the drivers want to export their busy state, they must set own
2965 * exporting function using blk_queue_lld_busy() first.
2967 * Basically, this function is used only by request stacking drivers
2968 * to stop dispatching requests to underlying devices when underlying
2969 * devices are busy. This behavior helps more I/O merging on the queue
2970 * of the request stacking driver and prevents I/O throughput regression
2971 * on burst I/O load.
2974 * 0 - Not busy (The request stacking driver should dispatch request)
2975 * 1 - Busy (The request stacking driver should stop dispatching request)
2977 int blk_lld_busy(struct request_queue
*q
)
2980 return q
->lld_busy_fn(q
);
2984 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2987 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2988 * @rq: the clone request to be cleaned up
2991 * Free all bios in @rq for a cloned request.
2993 void blk_rq_unprep_clone(struct request
*rq
)
2997 while ((bio
= rq
->bio
) != NULL
) {
2998 rq
->bio
= bio
->bi_next
;
3003 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3006 * Copy attributes of the original request to the clone request.
3007 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3009 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3011 dst
->cpu
= src
->cpu
;
3012 dst
->__sector
= blk_rq_pos(src
);
3013 dst
->__data_len
= blk_rq_bytes(src
);
3014 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3015 dst
->ioprio
= src
->ioprio
;
3016 dst
->extra_len
= src
->extra_len
;
3020 * blk_rq_prep_clone - Helper function to setup clone request
3021 * @rq: the request to be setup
3022 * @rq_src: original request to be cloned
3023 * @bs: bio_set that bios for clone are allocated from
3024 * @gfp_mask: memory allocation mask for bio
3025 * @bio_ctr: setup function to be called for each clone bio.
3026 * Returns %0 for success, non %0 for failure.
3027 * @data: private data to be passed to @bio_ctr
3030 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3031 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3032 * are not copied, and copying such parts is the caller's responsibility.
3033 * Also, pages which the original bios are pointing to are not copied
3034 * and the cloned bios just point same pages.
3035 * So cloned bios must be completed before original bios, which means
3036 * the caller must complete @rq before @rq_src.
3038 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3039 struct bio_set
*bs
, gfp_t gfp_mask
,
3040 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3043 struct bio
*bio
, *bio_src
;
3048 __rq_for_each_bio(bio_src
, rq_src
) {
3049 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3053 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3057 rq
->biotail
->bi_next
= bio
;
3060 rq
->bio
= rq
->biotail
= bio
;
3063 __blk_rq_prep_clone(rq
, rq_src
);
3070 blk_rq_unprep_clone(rq
);
3074 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3076 int kblockd_schedule_work(struct work_struct
*work
)
3078 return queue_work(kblockd_workqueue
, work
);
3080 EXPORT_SYMBOL(kblockd_schedule_work
);
3082 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3084 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3086 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3088 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3089 unsigned long delay
)
3091 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3093 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3095 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3096 unsigned long delay
)
3098 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3100 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3103 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3104 * @plug: The &struct blk_plug that needs to be initialized
3107 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3108 * pending I/O should the task end up blocking between blk_start_plug() and
3109 * blk_finish_plug(). This is important from a performance perspective, but
3110 * also ensures that we don't deadlock. For instance, if the task is blocking
3111 * for a memory allocation, memory reclaim could end up wanting to free a
3112 * page belonging to that request that is currently residing in our private
3113 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3114 * this kind of deadlock.
3116 void blk_start_plug(struct blk_plug
*plug
)
3118 struct task_struct
*tsk
= current
;
3121 * If this is a nested plug, don't actually assign it.
3126 INIT_LIST_HEAD(&plug
->list
);
3127 INIT_LIST_HEAD(&plug
->mq_list
);
3128 INIT_LIST_HEAD(&plug
->cb_list
);
3130 * Store ordering should not be needed here, since a potential
3131 * preempt will imply a full memory barrier
3135 EXPORT_SYMBOL(blk_start_plug
);
3137 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3139 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3140 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3142 return !(rqa
->q
< rqb
->q
||
3143 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3147 * If 'from_schedule' is true, then postpone the dispatch of requests
3148 * until a safe kblockd context. We due this to avoid accidental big
3149 * additional stack usage in driver dispatch, in places where the originally
3150 * plugger did not intend it.
3152 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3154 __releases(q
->queue_lock
)
3156 trace_block_unplug(q
, depth
, !from_schedule
);
3159 blk_run_queue_async(q
);
3162 spin_unlock(q
->queue_lock
);
3165 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3167 LIST_HEAD(callbacks
);
3169 while (!list_empty(&plug
->cb_list
)) {
3170 list_splice_init(&plug
->cb_list
, &callbacks
);
3172 while (!list_empty(&callbacks
)) {
3173 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3176 list_del(&cb
->list
);
3177 cb
->callback(cb
, from_schedule
);
3182 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3185 struct blk_plug
*plug
= current
->plug
;
3186 struct blk_plug_cb
*cb
;
3191 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3192 if (cb
->callback
== unplug
&& cb
->data
== data
)
3195 /* Not currently on the callback list */
3196 BUG_ON(size
< sizeof(*cb
));
3197 cb
= kzalloc(size
, GFP_ATOMIC
);
3200 cb
->callback
= unplug
;
3201 list_add(&cb
->list
, &plug
->cb_list
);
3205 EXPORT_SYMBOL(blk_check_plugged
);
3207 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3209 struct request_queue
*q
;
3210 unsigned long flags
;
3215 flush_plug_callbacks(plug
, from_schedule
);
3217 if (!list_empty(&plug
->mq_list
))
3218 blk_mq_flush_plug_list(plug
, from_schedule
);
3220 if (list_empty(&plug
->list
))
3223 list_splice_init(&plug
->list
, &list
);
3225 list_sort(NULL
, &list
, plug_rq_cmp
);
3231 * Save and disable interrupts here, to avoid doing it for every
3232 * queue lock we have to take.
3234 local_irq_save(flags
);
3235 while (!list_empty(&list
)) {
3236 rq
= list_entry_rq(list
.next
);
3237 list_del_init(&rq
->queuelist
);
3241 * This drops the queue lock
3244 queue_unplugged(q
, depth
, from_schedule
);
3247 spin_lock(q
->queue_lock
);
3251 * Short-circuit if @q is dead
3253 if (unlikely(blk_queue_dying(q
))) {
3254 __blk_end_request_all(rq
, -ENODEV
);
3259 * rq is already accounted, so use raw insert
3261 if (op_is_flush(rq
->cmd_flags
))
3262 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3264 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3270 * This drops the queue lock
3273 queue_unplugged(q
, depth
, from_schedule
);
3275 local_irq_restore(flags
);
3278 void blk_finish_plug(struct blk_plug
*plug
)
3280 if (plug
!= current
->plug
)
3282 blk_flush_plug_list(plug
, false);
3284 current
->plug
= NULL
;
3286 EXPORT_SYMBOL(blk_finish_plug
);
3290 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3291 * @q: the queue of the device
3292 * @dev: the device the queue belongs to
3295 * Initialize runtime-PM-related fields for @q and start auto suspend for
3296 * @dev. Drivers that want to take advantage of request-based runtime PM
3297 * should call this function after @dev has been initialized, and its
3298 * request queue @q has been allocated, and runtime PM for it can not happen
3299 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3300 * cases, driver should call this function before any I/O has taken place.
3302 * This function takes care of setting up using auto suspend for the device,
3303 * the autosuspend delay is set to -1 to make runtime suspend impossible
3304 * until an updated value is either set by user or by driver. Drivers do
3305 * not need to touch other autosuspend settings.
3307 * The block layer runtime PM is request based, so only works for drivers
3308 * that use request as their IO unit instead of those directly use bio's.
3310 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3313 q
->rpm_status
= RPM_ACTIVE
;
3314 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3315 pm_runtime_use_autosuspend(q
->dev
);
3317 EXPORT_SYMBOL(blk_pm_runtime_init
);
3320 * blk_pre_runtime_suspend - Pre runtime suspend check
3321 * @q: the queue of the device
3324 * This function will check if runtime suspend is allowed for the device
3325 * by examining if there are any requests pending in the queue. If there
3326 * are requests pending, the device can not be runtime suspended; otherwise,
3327 * the queue's status will be updated to SUSPENDING and the driver can
3328 * proceed to suspend the device.
3330 * For the not allowed case, we mark last busy for the device so that
3331 * runtime PM core will try to autosuspend it some time later.
3333 * This function should be called near the start of the device's
3334 * runtime_suspend callback.
3337 * 0 - OK to runtime suspend the device
3338 * -EBUSY - Device should not be runtime suspended
3340 int blk_pre_runtime_suspend(struct request_queue
*q
)
3347 spin_lock_irq(q
->queue_lock
);
3348 if (q
->nr_pending
) {
3350 pm_runtime_mark_last_busy(q
->dev
);
3352 q
->rpm_status
= RPM_SUSPENDING
;
3354 spin_unlock_irq(q
->queue_lock
);
3357 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3360 * blk_post_runtime_suspend - Post runtime suspend processing
3361 * @q: the queue of the device
3362 * @err: return value of the device's runtime_suspend function
3365 * Update the queue's runtime status according to the return value of the
3366 * device's runtime suspend function and mark last busy for the device so
3367 * that PM core will try to auto suspend the device at a later time.
3369 * This function should be called near the end of the device's
3370 * runtime_suspend callback.
3372 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3377 spin_lock_irq(q
->queue_lock
);
3379 q
->rpm_status
= RPM_SUSPENDED
;
3381 q
->rpm_status
= RPM_ACTIVE
;
3382 pm_runtime_mark_last_busy(q
->dev
);
3384 spin_unlock_irq(q
->queue_lock
);
3386 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3389 * blk_pre_runtime_resume - Pre runtime resume processing
3390 * @q: the queue of the device
3393 * Update the queue's runtime status to RESUMING in preparation for the
3394 * runtime resume of the device.
3396 * This function should be called near the start of the device's
3397 * runtime_resume callback.
3399 void blk_pre_runtime_resume(struct request_queue
*q
)
3404 spin_lock_irq(q
->queue_lock
);
3405 q
->rpm_status
= RPM_RESUMING
;
3406 spin_unlock_irq(q
->queue_lock
);
3408 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3411 * blk_post_runtime_resume - Post runtime resume processing
3412 * @q: the queue of the device
3413 * @err: return value of the device's runtime_resume function
3416 * Update the queue's runtime status according to the return value of the
3417 * device's runtime_resume function. If it is successfully resumed, process
3418 * the requests that are queued into the device's queue when it is resuming
3419 * and then mark last busy and initiate autosuspend for it.
3421 * This function should be called near the end of the device's
3422 * runtime_resume callback.
3424 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3429 spin_lock_irq(q
->queue_lock
);
3431 q
->rpm_status
= RPM_ACTIVE
;
3433 pm_runtime_mark_last_busy(q
->dev
);
3434 pm_request_autosuspend(q
->dev
);
3436 q
->rpm_status
= RPM_SUSPENDED
;
3438 spin_unlock_irq(q
->queue_lock
);
3440 EXPORT_SYMBOL(blk_post_runtime_resume
);
3443 * blk_set_runtime_active - Force runtime status of the queue to be active
3444 * @q: the queue of the device
3446 * If the device is left runtime suspended during system suspend the resume
3447 * hook typically resumes the device and corrects runtime status
3448 * accordingly. However, that does not affect the queue runtime PM status
3449 * which is still "suspended". This prevents processing requests from the
3452 * This function can be used in driver's resume hook to correct queue
3453 * runtime PM status and re-enable peeking requests from the queue. It
3454 * should be called before first request is added to the queue.
3456 void blk_set_runtime_active(struct request_queue
*q
)
3458 spin_lock_irq(q
->queue_lock
);
3459 q
->rpm_status
= RPM_ACTIVE
;
3460 pm_runtime_mark_last_busy(q
->dev
);
3461 pm_request_autosuspend(q
->dev
);
3462 spin_unlock_irq(q
->queue_lock
);
3464 EXPORT_SYMBOL(blk_set_runtime_active
);
3467 int __init
blk_dev_init(void)
3469 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3470 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3471 FIELD_SIZEOF(struct request
, cmd_flags
));
3472 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3473 FIELD_SIZEOF(struct bio
, bi_opf
));
3475 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3476 kblockd_workqueue
= alloc_workqueue("kblockd",
3477 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3478 if (!kblockd_workqueue
)
3479 panic("Failed to create kblockd\n");
3481 request_cachep
= kmem_cache_create("blkdev_requests",
3482 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3484 blk_requestq_cachep
= kmem_cache_create("request_queue",
3485 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3487 #ifdef CONFIG_DEBUG_FS
3488 blk_debugfs_root
= debugfs_create_dir("block", NULL
);