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_delayed_work_sync(&hctx
->run_work
);
274 cancel_delayed_work_sync(&q
->delay_work
);
277 EXPORT_SYMBOL(blk_sync_queue
);
280 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
281 * @q: The queue to run
284 * Invoke request handling on a queue if there are any pending requests.
285 * May be used to restart request handling after a request has completed.
286 * This variant runs the queue whether or not the queue has been
287 * stopped. Must be called with the queue lock held and interrupts
288 * disabled. See also @blk_run_queue.
290 inline void __blk_run_queue_uncond(struct request_queue
*q
)
292 if (unlikely(blk_queue_dead(q
)))
296 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
297 * the queue lock internally. As a result multiple threads may be
298 * running such a request function concurrently. Keep track of the
299 * number of active request_fn invocations such that blk_drain_queue()
300 * can wait until all these request_fn calls have finished.
302 q
->request_fn_active
++;
304 q
->request_fn_active
--;
306 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
309 * __blk_run_queue - run a single device queue
310 * @q: The queue to run
313 * See @blk_run_queue. This variant must be called with the queue lock
314 * held and interrupts disabled.
316 void __blk_run_queue(struct request_queue
*q
)
318 if (unlikely(blk_queue_stopped(q
)))
321 __blk_run_queue_uncond(q
);
323 EXPORT_SYMBOL(__blk_run_queue
);
326 * blk_run_queue_async - run a single device queue in workqueue context
327 * @q: The queue to run
330 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
331 * of us. The caller must hold the queue lock.
333 void blk_run_queue_async(struct request_queue
*q
)
335 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
336 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
338 EXPORT_SYMBOL(blk_run_queue_async
);
341 * blk_run_queue - run a single device queue
342 * @q: The queue to run
345 * Invoke request handling on this queue, if it has pending work to do.
346 * May be used to restart queueing when a request has completed.
348 void blk_run_queue(struct request_queue
*q
)
352 spin_lock_irqsave(q
->queue_lock
, flags
);
354 spin_unlock_irqrestore(q
->queue_lock
, flags
);
356 EXPORT_SYMBOL(blk_run_queue
);
358 void blk_put_queue(struct request_queue
*q
)
360 kobject_put(&q
->kobj
);
362 EXPORT_SYMBOL(blk_put_queue
);
365 * __blk_drain_queue - drain requests from request_queue
367 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
369 * Drain requests from @q. If @drain_all is set, all requests are drained.
370 * If not, only ELVPRIV requests are drained. The caller is responsible
371 * for ensuring that no new requests which need to be drained are queued.
373 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
374 __releases(q
->queue_lock
)
375 __acquires(q
->queue_lock
)
379 lockdep_assert_held(q
->queue_lock
);
385 * The caller might be trying to drain @q before its
386 * elevator is initialized.
389 elv_drain_elevator(q
);
391 blkcg_drain_queue(q
);
394 * This function might be called on a queue which failed
395 * driver init after queue creation or is not yet fully
396 * active yet. Some drivers (e.g. fd and loop) get unhappy
397 * in such cases. Kick queue iff dispatch queue has
398 * something on it and @q has request_fn set.
400 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
403 drain
|= q
->nr_rqs_elvpriv
;
404 drain
|= q
->request_fn_active
;
407 * Unfortunately, requests are queued at and tracked from
408 * multiple places and there's no single counter which can
409 * be drained. Check all the queues and counters.
412 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
413 drain
|= !list_empty(&q
->queue_head
);
414 for (i
= 0; i
< 2; i
++) {
415 drain
|= q
->nr_rqs
[i
];
416 drain
|= q
->in_flight
[i
];
418 drain
|= !list_empty(&fq
->flush_queue
[i
]);
425 spin_unlock_irq(q
->queue_lock
);
429 spin_lock_irq(q
->queue_lock
);
433 * With queue marked dead, any woken up waiter will fail the
434 * allocation path, so the wakeup chaining is lost and we're
435 * left with hung waiters. We need to wake up those waiters.
438 struct request_list
*rl
;
440 blk_queue_for_each_rl(rl
, q
)
441 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
442 wake_up_all(&rl
->wait
[i
]);
447 * blk_queue_bypass_start - enter queue bypass mode
448 * @q: queue of interest
450 * In bypass mode, only the dispatch FIFO queue of @q is used. This
451 * function makes @q enter bypass mode and drains all requests which were
452 * throttled or issued before. On return, it's guaranteed that no request
453 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
454 * inside queue or RCU read lock.
456 void blk_queue_bypass_start(struct request_queue
*q
)
458 spin_lock_irq(q
->queue_lock
);
460 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
461 spin_unlock_irq(q
->queue_lock
);
464 * Queues start drained. Skip actual draining till init is
465 * complete. This avoids lenghty delays during queue init which
466 * can happen many times during boot.
468 if (blk_queue_init_done(q
)) {
469 spin_lock_irq(q
->queue_lock
);
470 __blk_drain_queue(q
, false);
471 spin_unlock_irq(q
->queue_lock
);
473 /* ensure blk_queue_bypass() is %true inside RCU read lock */
477 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
480 * blk_queue_bypass_end - leave queue bypass mode
481 * @q: queue of interest
483 * Leave bypass mode and restore the normal queueing behavior.
485 void blk_queue_bypass_end(struct request_queue
*q
)
487 spin_lock_irq(q
->queue_lock
);
488 if (!--q
->bypass_depth
)
489 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
490 WARN_ON_ONCE(q
->bypass_depth
< 0);
491 spin_unlock_irq(q
->queue_lock
);
493 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
495 void blk_set_queue_dying(struct request_queue
*q
)
497 spin_lock_irq(q
->queue_lock
);
498 queue_flag_set(QUEUE_FLAG_DYING
, q
);
499 spin_unlock_irq(q
->queue_lock
);
502 * When queue DYING flag is set, we need to block new req
503 * entering queue, so we call blk_freeze_queue_start() to
504 * prevent I/O from crossing blk_queue_enter().
506 blk_freeze_queue_start(q
);
509 blk_mq_wake_waiters(q
);
511 struct request_list
*rl
;
513 spin_lock_irq(q
->queue_lock
);
514 blk_queue_for_each_rl(rl
, q
) {
516 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
517 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
520 spin_unlock_irq(q
->queue_lock
);
523 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
526 * blk_cleanup_queue - shutdown a request queue
527 * @q: request queue to shutdown
529 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
530 * put it. All future requests will be failed immediately with -ENODEV.
532 void blk_cleanup_queue(struct request_queue
*q
)
534 spinlock_t
*lock
= q
->queue_lock
;
536 /* mark @q DYING, no new request or merges will be allowed afterwards */
537 mutex_lock(&q
->sysfs_lock
);
538 blk_set_queue_dying(q
);
542 * A dying queue is permanently in bypass mode till released. Note
543 * that, unlike blk_queue_bypass_start(), we aren't performing
544 * synchronize_rcu() after entering bypass mode to avoid the delay
545 * as some drivers create and destroy a lot of queues while
546 * probing. This is still safe because blk_release_queue() will be
547 * called only after the queue refcnt drops to zero and nothing,
548 * RCU or not, would be traversing the queue by then.
551 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
553 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
554 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
555 queue_flag_set(QUEUE_FLAG_DYING
, q
);
556 spin_unlock_irq(lock
);
557 mutex_unlock(&q
->sysfs_lock
);
560 * Drain all requests queued before DYING marking. Set DEAD flag to
561 * prevent that q->request_fn() gets invoked after draining finished.
566 __blk_drain_queue(q
, true);
567 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
568 spin_unlock_irq(lock
);
570 /* for synchronous bio-based driver finish in-flight integrity i/o */
571 blk_flush_integrity();
573 /* @q won't process any more request, flush async actions */
574 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
578 blk_mq_free_queue(q
);
579 percpu_ref_exit(&q
->q_usage_counter
);
582 if (q
->queue_lock
!= &q
->__queue_lock
)
583 q
->queue_lock
= &q
->__queue_lock
;
584 spin_unlock_irq(lock
);
586 /* @q is and will stay empty, shutdown and put */
589 EXPORT_SYMBOL(blk_cleanup_queue
);
591 /* Allocate memory local to the request queue */
592 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
594 struct request_queue
*q
= data
;
596 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
599 static void free_request_simple(void *element
, void *data
)
601 kmem_cache_free(request_cachep
, element
);
604 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
606 struct request_queue
*q
= data
;
609 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
611 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
618 static void free_request_size(void *element
, void *data
)
620 struct request_queue
*q
= data
;
623 q
->exit_rq_fn(q
, element
);
627 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
630 if (unlikely(rl
->rq_pool
))
634 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
635 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
636 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
637 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
640 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
641 alloc_request_size
, free_request_size
,
642 q
, gfp_mask
, q
->node
);
644 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
645 alloc_request_simple
, free_request_simple
,
646 q
, gfp_mask
, q
->node
);
654 void blk_exit_rl(struct request_list
*rl
)
657 mempool_destroy(rl
->rq_pool
);
660 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
662 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
664 EXPORT_SYMBOL(blk_alloc_queue
);
666 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
671 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
678 * read pair of barrier in blk_freeze_queue_start(),
679 * we need to order reading __PERCPU_REF_DEAD flag of
680 * .q_usage_counter and reading .mq_freeze_depth or
681 * queue dying flag, otherwise the following wait may
682 * never return if the two reads are reordered.
686 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
687 !atomic_read(&q
->mq_freeze_depth
) ||
689 if (blk_queue_dying(q
))
696 void blk_queue_exit(struct request_queue
*q
)
698 percpu_ref_put(&q
->q_usage_counter
);
701 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
703 struct request_queue
*q
=
704 container_of(ref
, struct request_queue
, q_usage_counter
);
706 wake_up_all(&q
->mq_freeze_wq
);
709 static void blk_rq_timed_out_timer(unsigned long data
)
711 struct request_queue
*q
= (struct request_queue
*)data
;
713 kblockd_schedule_work(&q
->timeout_work
);
716 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
718 struct request_queue
*q
;
720 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
721 gfp_mask
| __GFP_ZERO
, node_id
);
725 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
729 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
733 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
734 if (!q
->backing_dev_info
)
737 q
->stats
= blk_alloc_queue_stats();
741 q
->backing_dev_info
->ra_pages
=
742 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
743 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
744 q
->backing_dev_info
->name
= "block";
747 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
748 laptop_mode_timer_fn
, (unsigned long) q
);
749 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
750 INIT_LIST_HEAD(&q
->queue_head
);
751 INIT_LIST_HEAD(&q
->timeout_list
);
752 INIT_LIST_HEAD(&q
->icq_list
);
753 #ifdef CONFIG_BLK_CGROUP
754 INIT_LIST_HEAD(&q
->blkg_list
);
756 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
758 kobject_init(&q
->kobj
, &blk_queue_ktype
);
760 mutex_init(&q
->sysfs_lock
);
761 spin_lock_init(&q
->__queue_lock
);
764 * By default initialize queue_lock to internal lock and driver can
765 * override it later if need be.
767 q
->queue_lock
= &q
->__queue_lock
;
770 * A queue starts its life with bypass turned on to avoid
771 * unnecessary bypass on/off overhead and nasty surprises during
772 * init. The initial bypass will be finished when the queue is
773 * registered by blk_register_queue().
776 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
778 init_waitqueue_head(&q
->mq_freeze_wq
);
781 * Init percpu_ref in atomic mode so that it's faster to shutdown.
782 * See blk_register_queue() for details.
784 if (percpu_ref_init(&q
->q_usage_counter
,
785 blk_queue_usage_counter_release
,
786 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
789 if (blkcg_init_queue(q
))
795 percpu_ref_exit(&q
->q_usage_counter
);
797 blk_free_queue_stats(q
->stats
);
799 bdi_put(q
->backing_dev_info
);
801 bioset_free(q
->bio_split
);
803 ida_simple_remove(&blk_queue_ida
, q
->id
);
805 kmem_cache_free(blk_requestq_cachep
, q
);
808 EXPORT_SYMBOL(blk_alloc_queue_node
);
811 * blk_init_queue - prepare a request queue for use with a block device
812 * @rfn: The function to be called to process requests that have been
813 * placed on the queue.
814 * @lock: Request queue spin lock
817 * If a block device wishes to use the standard request handling procedures,
818 * which sorts requests and coalesces adjacent requests, then it must
819 * call blk_init_queue(). The function @rfn will be called when there
820 * are requests on the queue that need to be processed. If the device
821 * supports plugging, then @rfn may not be called immediately when requests
822 * are available on the queue, but may be called at some time later instead.
823 * Plugged queues are generally unplugged when a buffer belonging to one
824 * of the requests on the queue is needed, or due to memory pressure.
826 * @rfn is not required, or even expected, to remove all requests off the
827 * queue, but only as many as it can handle at a time. If it does leave
828 * requests on the queue, it is responsible for arranging that the requests
829 * get dealt with eventually.
831 * The queue spin lock must be held while manipulating the requests on the
832 * request queue; this lock will be taken also from interrupt context, so irq
833 * disabling is needed for it.
835 * Function returns a pointer to the initialized request queue, or %NULL if
839 * blk_init_queue() must be paired with a blk_cleanup_queue() call
840 * when the block device is deactivated (such as at module unload).
843 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
845 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
847 EXPORT_SYMBOL(blk_init_queue
);
849 struct request_queue
*
850 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
852 struct request_queue
*q
;
854 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
860 q
->queue_lock
= lock
;
861 if (blk_init_allocated_queue(q
) < 0) {
862 blk_cleanup_queue(q
);
868 EXPORT_SYMBOL(blk_init_queue_node
);
870 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
873 int blk_init_allocated_queue(struct request_queue
*q
)
875 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
879 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
880 goto out_free_flush_queue
;
882 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
883 goto out_exit_flush_rq
;
885 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
886 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
889 * This also sets hw/phys segments, boundary and size
891 blk_queue_make_request(q
, blk_queue_bio
);
893 q
->sg_reserved_size
= INT_MAX
;
895 /* Protect q->elevator from elevator_change */
896 mutex_lock(&q
->sysfs_lock
);
899 if (elevator_init(q
, NULL
)) {
900 mutex_unlock(&q
->sysfs_lock
);
901 goto out_exit_flush_rq
;
904 mutex_unlock(&q
->sysfs_lock
);
909 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
910 out_free_flush_queue
:
911 blk_free_flush_queue(q
->fq
);
914 EXPORT_SYMBOL(blk_init_allocated_queue
);
916 bool blk_get_queue(struct request_queue
*q
)
918 if (likely(!blk_queue_dying(q
))) {
925 EXPORT_SYMBOL(blk_get_queue
);
927 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
929 if (rq
->rq_flags
& RQF_ELVPRIV
) {
930 elv_put_request(rl
->q
, rq
);
932 put_io_context(rq
->elv
.icq
->ioc
);
935 mempool_free(rq
, rl
->rq_pool
);
939 * ioc_batching returns true if the ioc is a valid batching request and
940 * should be given priority access to a request.
942 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
948 * Make sure the process is able to allocate at least 1 request
949 * even if the batch times out, otherwise we could theoretically
952 return ioc
->nr_batch_requests
== q
->nr_batching
||
953 (ioc
->nr_batch_requests
> 0
954 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
958 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
959 * will cause the process to be a "batcher" on all queues in the system. This
960 * is the behaviour we want though - once it gets a wakeup it should be given
963 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
965 if (!ioc
|| ioc_batching(q
, ioc
))
968 ioc
->nr_batch_requests
= q
->nr_batching
;
969 ioc
->last_waited
= jiffies
;
972 static void __freed_request(struct request_list
*rl
, int sync
)
974 struct request_queue
*q
= rl
->q
;
976 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
977 blk_clear_congested(rl
, sync
);
979 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
980 if (waitqueue_active(&rl
->wait
[sync
]))
981 wake_up(&rl
->wait
[sync
]);
983 blk_clear_rl_full(rl
, sync
);
988 * A request has just been released. Account for it, update the full and
989 * congestion status, wake up any waiters. Called under q->queue_lock.
991 static void freed_request(struct request_list
*rl
, bool sync
,
992 req_flags_t rq_flags
)
994 struct request_queue
*q
= rl
->q
;
998 if (rq_flags
& RQF_ELVPRIV
)
1001 __freed_request(rl
, sync
);
1003 if (unlikely(rl
->starved
[sync
^ 1]))
1004 __freed_request(rl
, sync
^ 1);
1007 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1009 struct request_list
*rl
;
1010 int on_thresh
, off_thresh
;
1012 spin_lock_irq(q
->queue_lock
);
1013 q
->nr_requests
= nr
;
1014 blk_queue_congestion_threshold(q
);
1015 on_thresh
= queue_congestion_on_threshold(q
);
1016 off_thresh
= queue_congestion_off_threshold(q
);
1018 blk_queue_for_each_rl(rl
, q
) {
1019 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1020 blk_set_congested(rl
, BLK_RW_SYNC
);
1021 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1022 blk_clear_congested(rl
, BLK_RW_SYNC
);
1024 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1025 blk_set_congested(rl
, BLK_RW_ASYNC
);
1026 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1027 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1029 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1030 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1032 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1033 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1036 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1037 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1039 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1040 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1044 spin_unlock_irq(q
->queue_lock
);
1049 * __get_request - get a free request
1050 * @rl: request list to allocate from
1051 * @op: operation and flags
1052 * @bio: bio to allocate request for (can be %NULL)
1053 * @gfp_mask: allocation mask
1055 * Get a free request from @q. This function may fail under memory
1056 * pressure or if @q is dead.
1058 * Must be called with @q->queue_lock held and,
1059 * Returns ERR_PTR on failure, with @q->queue_lock held.
1060 * Returns request pointer on success, with @q->queue_lock *not held*.
1062 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1063 struct bio
*bio
, gfp_t gfp_mask
)
1065 struct request_queue
*q
= rl
->q
;
1067 struct elevator_type
*et
= q
->elevator
->type
;
1068 struct io_context
*ioc
= rq_ioc(bio
);
1069 struct io_cq
*icq
= NULL
;
1070 const bool is_sync
= op_is_sync(op
);
1072 req_flags_t rq_flags
= RQF_ALLOCED
;
1074 if (unlikely(blk_queue_dying(q
)))
1075 return ERR_PTR(-ENODEV
);
1077 may_queue
= elv_may_queue(q
, op
);
1078 if (may_queue
== ELV_MQUEUE_NO
)
1081 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1082 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1084 * The queue will fill after this allocation, so set
1085 * it as full, and mark this process as "batching".
1086 * This process will be allowed to complete a batch of
1087 * requests, others will be blocked.
1089 if (!blk_rl_full(rl
, is_sync
)) {
1090 ioc_set_batching(q
, ioc
);
1091 blk_set_rl_full(rl
, is_sync
);
1093 if (may_queue
!= ELV_MQUEUE_MUST
1094 && !ioc_batching(q
, ioc
)) {
1096 * The queue is full and the allocating
1097 * process is not a "batcher", and not
1098 * exempted by the IO scheduler
1100 return ERR_PTR(-ENOMEM
);
1104 blk_set_congested(rl
, is_sync
);
1108 * Only allow batching queuers to allocate up to 50% over the defined
1109 * limit of requests, otherwise we could have thousands of requests
1110 * allocated with any setting of ->nr_requests
1112 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1113 return ERR_PTR(-ENOMEM
);
1115 q
->nr_rqs
[is_sync
]++;
1116 rl
->count
[is_sync
]++;
1117 rl
->starved
[is_sync
] = 0;
1120 * Decide whether the new request will be managed by elevator. If
1121 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1122 * prevent the current elevator from being destroyed until the new
1123 * request is freed. This guarantees icq's won't be destroyed and
1124 * makes creating new ones safe.
1126 * Flush requests do not use the elevator so skip initialization.
1127 * This allows a request to share the flush and elevator data.
1129 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1130 * it will be created after releasing queue_lock.
1132 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1133 rq_flags
|= RQF_ELVPRIV
;
1134 q
->nr_rqs_elvpriv
++;
1135 if (et
->icq_cache
&& ioc
)
1136 icq
= ioc_lookup_icq(ioc
, q
);
1139 if (blk_queue_io_stat(q
))
1140 rq_flags
|= RQF_IO_STAT
;
1141 spin_unlock_irq(q
->queue_lock
);
1143 /* allocate and init request */
1144 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1149 blk_rq_set_rl(rq
, rl
);
1151 rq
->rq_flags
= rq_flags
;
1154 if (rq_flags
& RQF_ELVPRIV
) {
1155 if (unlikely(et
->icq_cache
&& !icq
)) {
1157 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1163 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1166 /* @rq->elv.icq holds io_context until @rq is freed */
1168 get_io_context(icq
->ioc
);
1172 * ioc may be NULL here, and ioc_batching will be false. That's
1173 * OK, if the queue is under the request limit then requests need
1174 * not count toward the nr_batch_requests limit. There will always
1175 * be some limit enforced by BLK_BATCH_TIME.
1177 if (ioc_batching(q
, ioc
))
1178 ioc
->nr_batch_requests
--;
1180 trace_block_getrq(q
, bio
, op
);
1185 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1186 * and may fail indefinitely under memory pressure and thus
1187 * shouldn't stall IO. Treat this request as !elvpriv. This will
1188 * disturb iosched and blkcg but weird is bettern than dead.
1190 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1191 __func__
, dev_name(q
->backing_dev_info
->dev
));
1193 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1196 spin_lock_irq(q
->queue_lock
);
1197 q
->nr_rqs_elvpriv
--;
1198 spin_unlock_irq(q
->queue_lock
);
1203 * Allocation failed presumably due to memory. Undo anything we
1204 * might have messed up.
1206 * Allocating task should really be put onto the front of the wait
1207 * queue, but this is pretty rare.
1209 spin_lock_irq(q
->queue_lock
);
1210 freed_request(rl
, is_sync
, rq_flags
);
1213 * in the very unlikely event that allocation failed and no
1214 * requests for this direction was pending, mark us starved so that
1215 * freeing of a request in the other direction will notice
1216 * us. another possible fix would be to split the rq mempool into
1220 if (unlikely(rl
->count
[is_sync
] == 0))
1221 rl
->starved
[is_sync
] = 1;
1222 return ERR_PTR(-ENOMEM
);
1226 * get_request - get a free request
1227 * @q: request_queue to allocate request from
1228 * @op: operation and flags
1229 * @bio: bio to allocate request for (can be %NULL)
1230 * @gfp_mask: allocation mask
1232 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1233 * this function keeps retrying under memory pressure and fails iff @q is dead.
1235 * Must be called with @q->queue_lock held and,
1236 * Returns ERR_PTR on failure, with @q->queue_lock held.
1237 * Returns request pointer on success, with @q->queue_lock *not held*.
1239 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1240 struct bio
*bio
, gfp_t gfp_mask
)
1242 const bool is_sync
= op_is_sync(op
);
1244 struct request_list
*rl
;
1247 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1249 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1253 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1258 /* wait on @rl and retry */
1259 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1260 TASK_UNINTERRUPTIBLE
);
1262 trace_block_sleeprq(q
, bio
, op
);
1264 spin_unlock_irq(q
->queue_lock
);
1268 * After sleeping, we become a "batching" process and will be able
1269 * to allocate at least one request, and up to a big batch of them
1270 * for a small period time. See ioc_batching, ioc_set_batching
1272 ioc_set_batching(q
, current
->io_context
);
1274 spin_lock_irq(q
->queue_lock
);
1275 finish_wait(&rl
->wait
[is_sync
], &wait
);
1280 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1285 /* create ioc upfront */
1286 create_io_context(gfp_mask
, q
->node
);
1288 spin_lock_irq(q
->queue_lock
);
1289 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1291 spin_unlock_irq(q
->queue_lock
);
1295 /* q->queue_lock is unlocked at this point */
1297 rq
->__sector
= (sector_t
) -1;
1298 rq
->bio
= rq
->biotail
= NULL
;
1302 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1305 return blk_mq_alloc_request(q
, rw
,
1306 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1307 0 : BLK_MQ_REQ_NOWAIT
);
1309 return blk_old_get_request(q
, rw
, gfp_mask
);
1311 EXPORT_SYMBOL(blk_get_request
);
1314 * blk_requeue_request - put a request back on queue
1315 * @q: request queue where request should be inserted
1316 * @rq: request to be inserted
1319 * Drivers often keep queueing requests until the hardware cannot accept
1320 * more, when that condition happens we need to put the request back
1321 * on the queue. Must be called with queue lock held.
1323 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1325 blk_delete_timer(rq
);
1326 blk_clear_rq_complete(rq
);
1327 trace_block_rq_requeue(q
, rq
);
1328 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1330 if (rq
->rq_flags
& RQF_QUEUED
)
1331 blk_queue_end_tag(q
, rq
);
1333 BUG_ON(blk_queued_rq(rq
));
1335 elv_requeue_request(q
, rq
);
1337 EXPORT_SYMBOL(blk_requeue_request
);
1339 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1342 blk_account_io_start(rq
, true);
1343 __elv_add_request(q
, rq
, where
);
1346 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1351 if (now
== part
->stamp
)
1354 inflight
= part_in_flight(part
);
1356 __part_stat_add(cpu
, part
, time_in_queue
,
1357 inflight
* (now
- part
->stamp
));
1358 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1364 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1365 * @cpu: cpu number for stats access
1366 * @part: target partition
1368 * The average IO queue length and utilisation statistics are maintained
1369 * by observing the current state of the queue length and the amount of
1370 * time it has been in this state for.
1372 * Normally, that accounting is done on IO completion, but that can result
1373 * in more than a second's worth of IO being accounted for within any one
1374 * second, leading to >100% utilisation. To deal with that, we call this
1375 * function to do a round-off before returning the results when reading
1376 * /proc/diskstats. This accounts immediately for all queue usage up to
1377 * the current jiffies and restarts the counters again.
1379 void part_round_stats(int cpu
, struct hd_struct
*part
)
1381 unsigned long now
= jiffies
;
1384 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1385 part_round_stats_single(cpu
, part
, now
);
1387 EXPORT_SYMBOL_GPL(part_round_stats
);
1390 static void blk_pm_put_request(struct request
*rq
)
1392 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1393 pm_runtime_mark_last_busy(rq
->q
->dev
);
1396 static inline void blk_pm_put_request(struct request
*rq
) {}
1400 * queue lock must be held
1402 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1404 req_flags_t rq_flags
= req
->rq_flags
;
1410 blk_mq_free_request(req
);
1414 blk_pm_put_request(req
);
1416 elv_completed_request(q
, req
);
1418 /* this is a bio leak */
1419 WARN_ON(req
->bio
!= NULL
);
1421 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1424 * Request may not have originated from ll_rw_blk. if not,
1425 * it didn't come out of our reserved rq pools
1427 if (rq_flags
& RQF_ALLOCED
) {
1428 struct request_list
*rl
= blk_rq_rl(req
);
1429 bool sync
= op_is_sync(req
->cmd_flags
);
1431 BUG_ON(!list_empty(&req
->queuelist
));
1432 BUG_ON(ELV_ON_HASH(req
));
1434 blk_free_request(rl
, req
);
1435 freed_request(rl
, sync
, rq_flags
);
1439 EXPORT_SYMBOL_GPL(__blk_put_request
);
1441 void blk_put_request(struct request
*req
)
1443 struct request_queue
*q
= req
->q
;
1446 blk_mq_free_request(req
);
1448 unsigned long flags
;
1450 spin_lock_irqsave(q
->queue_lock
, flags
);
1451 __blk_put_request(q
, req
);
1452 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1455 EXPORT_SYMBOL(blk_put_request
);
1457 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1460 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1462 if (!ll_back_merge_fn(q
, req
, bio
))
1465 trace_block_bio_backmerge(q
, req
, bio
);
1467 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1468 blk_rq_set_mixed_merge(req
);
1470 req
->biotail
->bi_next
= bio
;
1472 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1473 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1475 blk_account_io_start(req
, false);
1479 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1482 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1484 if (!ll_front_merge_fn(q
, req
, bio
))
1487 trace_block_bio_frontmerge(q
, req
, bio
);
1489 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1490 blk_rq_set_mixed_merge(req
);
1492 bio
->bi_next
= req
->bio
;
1495 req
->__sector
= bio
->bi_iter
.bi_sector
;
1496 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1497 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1499 blk_account_io_start(req
, false);
1503 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1506 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1508 if (segments
>= queue_max_discard_segments(q
))
1510 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1511 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1514 req
->biotail
->bi_next
= bio
;
1516 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1517 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1518 req
->nr_phys_segments
= segments
+ 1;
1520 blk_account_io_start(req
, false);
1523 req_set_nomerge(q
, req
);
1528 * blk_attempt_plug_merge - try to merge with %current's plugged list
1529 * @q: request_queue new bio is being queued at
1530 * @bio: new bio being queued
1531 * @request_count: out parameter for number of traversed plugged requests
1532 * @same_queue_rq: pointer to &struct request that gets filled in when
1533 * another request associated with @q is found on the plug list
1534 * (optional, may be %NULL)
1536 * Determine whether @bio being queued on @q can be merged with a request
1537 * on %current's plugged list. Returns %true if merge was successful,
1540 * Plugging coalesces IOs from the same issuer for the same purpose without
1541 * going through @q->queue_lock. As such it's more of an issuing mechanism
1542 * than scheduling, and the request, while may have elvpriv data, is not
1543 * added on the elevator at this point. In addition, we don't have
1544 * reliable access to the elevator outside queue lock. Only check basic
1545 * merging parameters without querying the elevator.
1547 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1549 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1550 unsigned int *request_count
,
1551 struct request
**same_queue_rq
)
1553 struct blk_plug
*plug
;
1555 struct list_head
*plug_list
;
1557 plug
= current
->plug
;
1563 plug_list
= &plug
->mq_list
;
1565 plug_list
= &plug
->list
;
1567 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1568 bool merged
= false;
1573 * Only blk-mq multiple hardware queues case checks the
1574 * rq in the same queue, there should be only one such
1578 *same_queue_rq
= rq
;
1581 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1584 switch (blk_try_merge(rq
, bio
)) {
1585 case ELEVATOR_BACK_MERGE
:
1586 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1588 case ELEVATOR_FRONT_MERGE
:
1589 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1591 case ELEVATOR_DISCARD_MERGE
:
1592 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1605 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1607 struct blk_plug
*plug
;
1609 struct list_head
*plug_list
;
1610 unsigned int ret
= 0;
1612 plug
= current
->plug
;
1617 plug_list
= &plug
->mq_list
;
1619 plug_list
= &plug
->list
;
1621 list_for_each_entry(rq
, plug_list
, queuelist
) {
1629 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1631 struct io_context
*ioc
= rq_ioc(bio
);
1633 if (bio
->bi_opf
& REQ_RAHEAD
)
1634 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1636 req
->__sector
= bio
->bi_iter
.bi_sector
;
1637 if (ioprio_valid(bio_prio(bio
)))
1638 req
->ioprio
= bio_prio(bio
);
1640 req
->ioprio
= ioc
->ioprio
;
1642 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1643 blk_rq_bio_prep(req
->q
, req
, bio
);
1645 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1647 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1649 struct blk_plug
*plug
;
1650 int where
= ELEVATOR_INSERT_SORT
;
1651 struct request
*req
, *free
;
1652 unsigned int request_count
= 0;
1653 unsigned int wb_acct
;
1656 * low level driver can indicate that it wants pages above a
1657 * certain limit bounced to low memory (ie for highmem, or even
1658 * ISA dma in theory)
1660 blk_queue_bounce(q
, &bio
);
1662 blk_queue_split(q
, &bio
, q
->bio_split
);
1664 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1665 bio
->bi_error
= -EIO
;
1667 return BLK_QC_T_NONE
;
1670 if (op_is_flush(bio
->bi_opf
)) {
1671 spin_lock_irq(q
->queue_lock
);
1672 where
= ELEVATOR_INSERT_FLUSH
;
1677 * Check if we can merge with the plugged list before grabbing
1680 if (!blk_queue_nomerges(q
)) {
1681 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1682 return BLK_QC_T_NONE
;
1684 request_count
= blk_plug_queued_count(q
);
1686 spin_lock_irq(q
->queue_lock
);
1688 switch (elv_merge(q
, &req
, bio
)) {
1689 case ELEVATOR_BACK_MERGE
:
1690 if (!bio_attempt_back_merge(q
, req
, bio
))
1692 elv_bio_merged(q
, req
, bio
);
1693 free
= attempt_back_merge(q
, req
);
1695 __blk_put_request(q
, free
);
1697 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1699 case ELEVATOR_FRONT_MERGE
:
1700 if (!bio_attempt_front_merge(q
, req
, bio
))
1702 elv_bio_merged(q
, req
, bio
);
1703 free
= attempt_front_merge(q
, req
);
1705 __blk_put_request(q
, free
);
1707 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1714 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1717 * Grab a free request. This is might sleep but can not fail.
1718 * Returns with the queue unlocked.
1720 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1722 __wbt_done(q
->rq_wb
, wb_acct
);
1723 bio
->bi_error
= PTR_ERR(req
);
1728 wbt_track(&req
->issue_stat
, wb_acct
);
1731 * After dropping the lock and possibly sleeping here, our request
1732 * may now be mergeable after it had proven unmergeable (above).
1733 * We don't worry about that case for efficiency. It won't happen
1734 * often, and the elevators are able to handle it.
1736 blk_init_request_from_bio(req
, bio
);
1738 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1739 req
->cpu
= raw_smp_processor_id();
1741 plug
= current
->plug
;
1744 * If this is the first request added after a plug, fire
1747 * @request_count may become stale because of schedule
1748 * out, so check plug list again.
1750 if (!request_count
|| list_empty(&plug
->list
))
1751 trace_block_plug(q
);
1753 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1754 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1755 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1756 blk_flush_plug_list(plug
, false);
1757 trace_block_plug(q
);
1760 list_add_tail(&req
->queuelist
, &plug
->list
);
1761 blk_account_io_start(req
, true);
1763 spin_lock_irq(q
->queue_lock
);
1764 add_acct_request(q
, req
, where
);
1767 spin_unlock_irq(q
->queue_lock
);
1770 return BLK_QC_T_NONE
;
1774 * If bio->bi_dev is a partition, remap the location
1776 static inline void blk_partition_remap(struct bio
*bio
)
1778 struct block_device
*bdev
= bio
->bi_bdev
;
1781 * Zone reset does not include bi_size so bio_sectors() is always 0.
1782 * Include a test for the reset op code and perform the remap if needed.
1784 if (bdev
!= bdev
->bd_contains
&&
1785 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1786 struct hd_struct
*p
= bdev
->bd_part
;
1788 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1789 bio
->bi_bdev
= bdev
->bd_contains
;
1791 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1793 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1797 static void handle_bad_sector(struct bio
*bio
)
1799 char b
[BDEVNAME_SIZE
];
1801 printk(KERN_INFO
"attempt to access beyond end of device\n");
1802 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1803 bdevname(bio
->bi_bdev
, b
),
1805 (unsigned long long)bio_end_sector(bio
),
1806 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1809 #ifdef CONFIG_FAIL_MAKE_REQUEST
1811 static DECLARE_FAULT_ATTR(fail_make_request
);
1813 static int __init
setup_fail_make_request(char *str
)
1815 return setup_fault_attr(&fail_make_request
, str
);
1817 __setup("fail_make_request=", setup_fail_make_request
);
1819 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1821 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1824 static int __init
fail_make_request_debugfs(void)
1826 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1827 NULL
, &fail_make_request
);
1829 return PTR_ERR_OR_ZERO(dir
);
1832 late_initcall(fail_make_request_debugfs
);
1834 #else /* CONFIG_FAIL_MAKE_REQUEST */
1836 static inline bool should_fail_request(struct hd_struct
*part
,
1842 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1845 * Check whether this bio extends beyond the end of the device.
1847 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1854 /* Test device or partition size, when known. */
1855 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1857 sector_t sector
= bio
->bi_iter
.bi_sector
;
1859 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1861 * This may well happen - the kernel calls bread()
1862 * without checking the size of the device, e.g., when
1863 * mounting a device.
1865 handle_bad_sector(bio
);
1873 static noinline_for_stack
bool
1874 generic_make_request_checks(struct bio
*bio
)
1876 struct request_queue
*q
;
1877 int nr_sectors
= bio_sectors(bio
);
1879 char b
[BDEVNAME_SIZE
];
1880 struct hd_struct
*part
;
1884 if (bio_check_eod(bio
, nr_sectors
))
1887 q
= bdev_get_queue(bio
->bi_bdev
);
1890 "generic_make_request: Trying to access "
1891 "nonexistent block-device %s (%Lu)\n",
1892 bdevname(bio
->bi_bdev
, b
),
1893 (long long) bio
->bi_iter
.bi_sector
);
1897 part
= bio
->bi_bdev
->bd_part
;
1898 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1899 should_fail_request(&part_to_disk(part
)->part0
,
1900 bio
->bi_iter
.bi_size
))
1904 * If this device has partitions, remap block n
1905 * of partition p to block n+start(p) of the disk.
1907 blk_partition_remap(bio
);
1909 if (bio_check_eod(bio
, nr_sectors
))
1913 * Filter flush bio's early so that make_request based
1914 * drivers without flush support don't have to worry
1917 if (op_is_flush(bio
->bi_opf
) &&
1918 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1919 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1926 switch (bio_op(bio
)) {
1927 case REQ_OP_DISCARD
:
1928 if (!blk_queue_discard(q
))
1931 case REQ_OP_SECURE_ERASE
:
1932 if (!blk_queue_secure_erase(q
))
1935 case REQ_OP_WRITE_SAME
:
1936 if (!bdev_write_same(bio
->bi_bdev
))
1939 case REQ_OP_ZONE_REPORT
:
1940 case REQ_OP_ZONE_RESET
:
1941 if (!bdev_is_zoned(bio
->bi_bdev
))
1944 case REQ_OP_WRITE_ZEROES
:
1945 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1953 * Various block parts want %current->io_context and lazy ioc
1954 * allocation ends up trading a lot of pain for a small amount of
1955 * memory. Just allocate it upfront. This may fail and block
1956 * layer knows how to live with it.
1958 create_io_context(GFP_ATOMIC
, q
->node
);
1960 if (!blkcg_bio_issue_check(q
, bio
))
1963 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1964 trace_block_bio_queue(q
, bio
);
1965 /* Now that enqueuing has been traced, we need to trace
1966 * completion as well.
1968 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
1975 bio
->bi_error
= err
;
1981 * generic_make_request - hand a buffer to its device driver for I/O
1982 * @bio: The bio describing the location in memory and on the device.
1984 * generic_make_request() is used to make I/O requests of block
1985 * devices. It is passed a &struct bio, which describes the I/O that needs
1988 * generic_make_request() does not return any status. The
1989 * success/failure status of the request, along with notification of
1990 * completion, is delivered asynchronously through the bio->bi_end_io
1991 * function described (one day) else where.
1993 * The caller of generic_make_request must make sure that bi_io_vec
1994 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1995 * set to describe the device address, and the
1996 * bi_end_io and optionally bi_private are set to describe how
1997 * completion notification should be signaled.
1999 * generic_make_request and the drivers it calls may use bi_next if this
2000 * bio happens to be merged with someone else, and may resubmit the bio to
2001 * a lower device by calling into generic_make_request recursively, which
2002 * means the bio should NOT be touched after the call to ->make_request_fn.
2004 blk_qc_t
generic_make_request(struct bio
*bio
)
2007 * bio_list_on_stack[0] contains bios submitted by the current
2009 * bio_list_on_stack[1] contains bios that were submitted before
2010 * the current make_request_fn, but that haven't been processed
2013 struct bio_list bio_list_on_stack
[2];
2014 blk_qc_t ret
= BLK_QC_T_NONE
;
2016 if (!generic_make_request_checks(bio
))
2020 * We only want one ->make_request_fn to be active at a time, else
2021 * stack usage with stacked devices could be a problem. So use
2022 * current->bio_list to keep a list of requests submited by a
2023 * make_request_fn function. current->bio_list is also used as a
2024 * flag to say if generic_make_request is currently active in this
2025 * task or not. If it is NULL, then no make_request is active. If
2026 * it is non-NULL, then a make_request is active, and new requests
2027 * should be added at the tail
2029 if (current
->bio_list
) {
2030 bio_list_add(¤t
->bio_list
[0], bio
);
2034 /* following loop may be a bit non-obvious, and so deserves some
2036 * Before entering the loop, bio->bi_next is NULL (as all callers
2037 * ensure that) so we have a list with a single bio.
2038 * We pretend that we have just taken it off a longer list, so
2039 * we assign bio_list to a pointer to the bio_list_on_stack,
2040 * thus initialising the bio_list of new bios to be
2041 * added. ->make_request() may indeed add some more bios
2042 * through a recursive call to generic_make_request. If it
2043 * did, we find a non-NULL value in bio_list and re-enter the loop
2044 * from the top. In this case we really did just take the bio
2045 * of the top of the list (no pretending) and so remove it from
2046 * bio_list, and call into ->make_request() again.
2048 BUG_ON(bio
->bi_next
);
2049 bio_list_init(&bio_list_on_stack
[0]);
2050 current
->bio_list
= bio_list_on_stack
;
2052 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2054 if (likely(blk_queue_enter(q
, false) == 0)) {
2055 struct bio_list lower
, same
;
2057 /* Create a fresh bio_list for all subordinate requests */
2058 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2059 bio_list_init(&bio_list_on_stack
[0]);
2060 ret
= q
->make_request_fn(q
, bio
);
2064 /* sort new bios into those for a lower level
2065 * and those for the same level
2067 bio_list_init(&lower
);
2068 bio_list_init(&same
);
2069 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2070 if (q
== bdev_get_queue(bio
->bi_bdev
))
2071 bio_list_add(&same
, bio
);
2073 bio_list_add(&lower
, bio
);
2074 /* now assemble so we handle the lowest level first */
2075 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2076 bio_list_merge(&bio_list_on_stack
[0], &same
);
2077 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2081 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2083 current
->bio_list
= NULL
; /* deactivate */
2088 EXPORT_SYMBOL(generic_make_request
);
2091 * submit_bio - submit a bio to the block device layer for I/O
2092 * @bio: The &struct bio which describes the I/O
2094 * submit_bio() is very similar in purpose to generic_make_request(), and
2095 * uses that function to do most of the work. Both are fairly rough
2096 * interfaces; @bio must be presetup and ready for I/O.
2099 blk_qc_t
submit_bio(struct bio
*bio
)
2102 * If it's a regular read/write or a barrier with data attached,
2103 * go through the normal accounting stuff before submission.
2105 if (bio_has_data(bio
)) {
2108 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2109 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2111 count
= bio_sectors(bio
);
2113 if (op_is_write(bio_op(bio
))) {
2114 count_vm_events(PGPGOUT
, count
);
2116 task_io_account_read(bio
->bi_iter
.bi_size
);
2117 count_vm_events(PGPGIN
, count
);
2120 if (unlikely(block_dump
)) {
2121 char b
[BDEVNAME_SIZE
];
2122 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2123 current
->comm
, task_pid_nr(current
),
2124 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2125 (unsigned long long)bio
->bi_iter
.bi_sector
,
2126 bdevname(bio
->bi_bdev
, b
),
2131 return generic_make_request(bio
);
2133 EXPORT_SYMBOL(submit_bio
);
2136 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2137 * for new the queue limits
2139 * @rq: the request being checked
2142 * @rq may have been made based on weaker limitations of upper-level queues
2143 * in request stacking drivers, and it may violate the limitation of @q.
2144 * Since the block layer and the underlying device driver trust @rq
2145 * after it is inserted to @q, it should be checked against @q before
2146 * the insertion using this generic function.
2148 * Request stacking drivers like request-based dm may change the queue
2149 * limits when retrying requests on other queues. Those requests need
2150 * to be checked against the new queue limits again during dispatch.
2152 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2155 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2156 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2161 * queue's settings related to segment counting like q->bounce_pfn
2162 * may differ from that of other stacking queues.
2163 * Recalculate it to check the request correctly on this queue's
2166 blk_recalc_rq_segments(rq
);
2167 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2168 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2176 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2177 * @q: the queue to submit the request
2178 * @rq: the request being queued
2180 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2182 unsigned long flags
;
2183 int where
= ELEVATOR_INSERT_BACK
;
2185 if (blk_cloned_rq_check_limits(q
, rq
))
2189 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2193 if (blk_queue_io_stat(q
))
2194 blk_account_io_start(rq
, true);
2195 blk_mq_sched_insert_request(rq
, false, true, false, false);
2199 spin_lock_irqsave(q
->queue_lock
, flags
);
2200 if (unlikely(blk_queue_dying(q
))) {
2201 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2206 * Submitting request must be dequeued before calling this function
2207 * because it will be linked to another request_queue
2209 BUG_ON(blk_queued_rq(rq
));
2211 if (op_is_flush(rq
->cmd_flags
))
2212 where
= ELEVATOR_INSERT_FLUSH
;
2214 add_acct_request(q
, rq
, where
);
2215 if (where
== ELEVATOR_INSERT_FLUSH
)
2217 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2221 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2224 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2225 * @rq: request to examine
2228 * A request could be merge of IOs which require different failure
2229 * handling. This function determines the number of bytes which
2230 * can be failed from the beginning of the request without
2231 * crossing into area which need to be retried further.
2234 * The number of bytes to fail.
2237 * queue_lock must be held.
2239 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2241 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2242 unsigned int bytes
= 0;
2245 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2246 return blk_rq_bytes(rq
);
2249 * Currently the only 'mixing' which can happen is between
2250 * different fastfail types. We can safely fail portions
2251 * which have all the failfast bits that the first one has -
2252 * the ones which are at least as eager to fail as the first
2255 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2256 if ((bio
->bi_opf
& ff
) != ff
)
2258 bytes
+= bio
->bi_iter
.bi_size
;
2261 /* this could lead to infinite loop */
2262 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2265 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2267 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2269 if (blk_do_io_stat(req
)) {
2270 const int rw
= rq_data_dir(req
);
2271 struct hd_struct
*part
;
2274 cpu
= part_stat_lock();
2276 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2281 void blk_account_io_done(struct request
*req
)
2284 * Account IO completion. flush_rq isn't accounted as a
2285 * normal IO on queueing nor completion. Accounting the
2286 * containing request is enough.
2288 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2289 unsigned long duration
= jiffies
- req
->start_time
;
2290 const int rw
= rq_data_dir(req
);
2291 struct hd_struct
*part
;
2294 cpu
= part_stat_lock();
2297 part_stat_inc(cpu
, part
, ios
[rw
]);
2298 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2299 part_round_stats(cpu
, part
);
2300 part_dec_in_flight(part
, rw
);
2302 hd_struct_put(part
);
2309 * Don't process normal requests when queue is suspended
2310 * or in the process of suspending/resuming
2312 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2315 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2316 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2322 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2329 void blk_account_io_start(struct request
*rq
, bool new_io
)
2331 struct hd_struct
*part
;
2332 int rw
= rq_data_dir(rq
);
2335 if (!blk_do_io_stat(rq
))
2338 cpu
= part_stat_lock();
2342 part_stat_inc(cpu
, part
, merges
[rw
]);
2344 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2345 if (!hd_struct_try_get(part
)) {
2347 * The partition is already being removed,
2348 * the request will be accounted on the disk only
2350 * We take a reference on disk->part0 although that
2351 * partition will never be deleted, so we can treat
2352 * it as any other partition.
2354 part
= &rq
->rq_disk
->part0
;
2355 hd_struct_get(part
);
2357 part_round_stats(cpu
, part
);
2358 part_inc_in_flight(part
, rw
);
2366 * blk_peek_request - peek at the top of a request queue
2367 * @q: request queue to peek at
2370 * Return the request at the top of @q. The returned request
2371 * should be started using blk_start_request() before LLD starts
2375 * Pointer to the request at the top of @q if available. Null
2379 * queue_lock must be held.
2381 struct request
*blk_peek_request(struct request_queue
*q
)
2386 while ((rq
= __elv_next_request(q
)) != NULL
) {
2388 rq
= blk_pm_peek_request(q
, rq
);
2392 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2394 * This is the first time the device driver
2395 * sees this request (possibly after
2396 * requeueing). Notify IO scheduler.
2398 if (rq
->rq_flags
& RQF_SORTED
)
2399 elv_activate_rq(q
, rq
);
2402 * just mark as started even if we don't start
2403 * it, a request that has been delayed should
2404 * not be passed by new incoming requests
2406 rq
->rq_flags
|= RQF_STARTED
;
2407 trace_block_rq_issue(q
, rq
);
2410 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2411 q
->end_sector
= rq_end_sector(rq
);
2412 q
->boundary_rq
= NULL
;
2415 if (rq
->rq_flags
& RQF_DONTPREP
)
2418 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2420 * make sure space for the drain appears we
2421 * know we can do this because max_hw_segments
2422 * has been adjusted to be one fewer than the
2425 rq
->nr_phys_segments
++;
2431 ret
= q
->prep_rq_fn(q
, rq
);
2432 if (ret
== BLKPREP_OK
) {
2434 } else if (ret
== BLKPREP_DEFER
) {
2436 * the request may have been (partially) prepped.
2437 * we need to keep this request in the front to
2438 * avoid resource deadlock. RQF_STARTED will
2439 * prevent other fs requests from passing this one.
2441 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2442 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2444 * remove the space for the drain we added
2445 * so that we don't add it again
2447 --rq
->nr_phys_segments
;
2452 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2453 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2455 rq
->rq_flags
|= RQF_QUIET
;
2457 * Mark this request as started so we don't trigger
2458 * any debug logic in the end I/O path.
2460 blk_start_request(rq
);
2461 __blk_end_request_all(rq
, err
);
2463 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2470 EXPORT_SYMBOL(blk_peek_request
);
2472 void blk_dequeue_request(struct request
*rq
)
2474 struct request_queue
*q
= rq
->q
;
2476 BUG_ON(list_empty(&rq
->queuelist
));
2477 BUG_ON(ELV_ON_HASH(rq
));
2479 list_del_init(&rq
->queuelist
);
2482 * the time frame between a request being removed from the lists
2483 * and to it is freed is accounted as io that is in progress at
2486 if (blk_account_rq(rq
)) {
2487 q
->in_flight
[rq_is_sync(rq
)]++;
2488 set_io_start_time_ns(rq
);
2493 * blk_start_request - start request processing on the driver
2494 * @req: request to dequeue
2497 * Dequeue @req and start timeout timer on it. This hands off the
2498 * request to the driver.
2500 * Block internal functions which don't want to start timer should
2501 * call blk_dequeue_request().
2504 * queue_lock must be held.
2506 void blk_start_request(struct request
*req
)
2508 blk_dequeue_request(req
);
2510 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2511 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2512 req
->rq_flags
|= RQF_STATS
;
2513 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2516 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2519 EXPORT_SYMBOL(blk_start_request
);
2522 * blk_fetch_request - fetch a request from a request queue
2523 * @q: request queue to fetch a request from
2526 * Return the request at the top of @q. The request is started on
2527 * return and LLD can start processing it immediately.
2530 * Pointer to the request at the top of @q if available. Null
2534 * queue_lock must be held.
2536 struct request
*blk_fetch_request(struct request_queue
*q
)
2540 rq
= blk_peek_request(q
);
2542 blk_start_request(rq
);
2545 EXPORT_SYMBOL(blk_fetch_request
);
2548 * blk_update_request - Special helper function for request stacking drivers
2549 * @req: the request being processed
2550 * @error: %0 for success, < %0 for error
2551 * @nr_bytes: number of bytes to complete @req
2554 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2555 * the request structure even if @req doesn't have leftover.
2556 * If @req has leftover, sets it up for the next range of segments.
2558 * This special helper function is only for request stacking drivers
2559 * (e.g. request-based dm) so that they can handle partial completion.
2560 * Actual device drivers should use blk_end_request instead.
2562 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2563 * %false return from this function.
2566 * %false - this request doesn't have any more data
2567 * %true - this request has more data
2569 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2573 trace_block_rq_complete(req
, error
, nr_bytes
);
2578 if (error
&& !blk_rq_is_passthrough(req
) &&
2579 !(req
->rq_flags
& RQF_QUIET
)) {
2584 error_type
= "recoverable transport";
2587 error_type
= "critical target";
2590 error_type
= "critical nexus";
2593 error_type
= "timeout";
2596 error_type
= "critical space allocation";
2599 error_type
= "critical medium";
2606 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2607 __func__
, error_type
, req
->rq_disk
?
2608 req
->rq_disk
->disk_name
: "?",
2609 (unsigned long long)blk_rq_pos(req
));
2613 blk_account_io_completion(req
, nr_bytes
);
2617 struct bio
*bio
= req
->bio
;
2618 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2620 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2621 req
->bio
= bio
->bi_next
;
2623 /* Completion has already been traced */
2624 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2625 req_bio_endio(req
, bio
, bio_bytes
, error
);
2627 total_bytes
+= bio_bytes
;
2628 nr_bytes
-= bio_bytes
;
2639 * Reset counters so that the request stacking driver
2640 * can find how many bytes remain in the request
2643 req
->__data_len
= 0;
2647 req
->__data_len
-= total_bytes
;
2649 /* update sector only for requests with clear definition of sector */
2650 if (!blk_rq_is_passthrough(req
))
2651 req
->__sector
+= total_bytes
>> 9;
2653 /* mixed attributes always follow the first bio */
2654 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2655 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2656 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2659 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2661 * If total number of sectors is less than the first segment
2662 * size, something has gone terribly wrong.
2664 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2665 blk_dump_rq_flags(req
, "request botched");
2666 req
->__data_len
= blk_rq_cur_bytes(req
);
2669 /* recalculate the number of segments */
2670 blk_recalc_rq_segments(req
);
2675 EXPORT_SYMBOL_GPL(blk_update_request
);
2677 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2678 unsigned int nr_bytes
,
2679 unsigned int bidi_bytes
)
2681 if (blk_update_request(rq
, error
, nr_bytes
))
2684 /* Bidi request must be completed as a whole */
2685 if (unlikely(blk_bidi_rq(rq
)) &&
2686 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2689 if (blk_queue_add_random(rq
->q
))
2690 add_disk_randomness(rq
->rq_disk
);
2696 * blk_unprep_request - unprepare a request
2699 * This function makes a request ready for complete resubmission (or
2700 * completion). It happens only after all error handling is complete,
2701 * so represents the appropriate moment to deallocate any resources
2702 * that were allocated to the request in the prep_rq_fn. The queue
2703 * lock is held when calling this.
2705 void blk_unprep_request(struct request
*req
)
2707 struct request_queue
*q
= req
->q
;
2709 req
->rq_flags
&= ~RQF_DONTPREP
;
2710 if (q
->unprep_rq_fn
)
2711 q
->unprep_rq_fn(q
, req
);
2713 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2716 * queue lock must be held
2718 void blk_finish_request(struct request
*req
, int error
)
2720 struct request_queue
*q
= req
->q
;
2722 if (req
->rq_flags
& RQF_STATS
)
2725 if (req
->rq_flags
& RQF_QUEUED
)
2726 blk_queue_end_tag(q
, req
);
2728 BUG_ON(blk_queued_rq(req
));
2730 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2731 laptop_io_completion(req
->q
->backing_dev_info
);
2733 blk_delete_timer(req
);
2735 if (req
->rq_flags
& RQF_DONTPREP
)
2736 blk_unprep_request(req
);
2738 blk_account_io_done(req
);
2741 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2742 req
->end_io(req
, error
);
2744 if (blk_bidi_rq(req
))
2745 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2747 __blk_put_request(q
, req
);
2750 EXPORT_SYMBOL(blk_finish_request
);
2753 * blk_end_bidi_request - Complete a bidi request
2754 * @rq: the request to complete
2755 * @error: %0 for success, < %0 for error
2756 * @nr_bytes: number of bytes to complete @rq
2757 * @bidi_bytes: number of bytes to complete @rq->next_rq
2760 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2761 * Drivers that supports bidi can safely call this member for any
2762 * type of request, bidi or uni. In the later case @bidi_bytes is
2766 * %false - we are done with this request
2767 * %true - still buffers pending for this request
2769 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2770 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2772 struct request_queue
*q
= rq
->q
;
2773 unsigned long flags
;
2775 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2778 spin_lock_irqsave(q
->queue_lock
, flags
);
2779 blk_finish_request(rq
, error
);
2780 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2786 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2787 * @rq: the request to complete
2788 * @error: %0 for success, < %0 for error
2789 * @nr_bytes: number of bytes to complete @rq
2790 * @bidi_bytes: number of bytes to complete @rq->next_rq
2793 * Identical to blk_end_bidi_request() except that queue lock is
2794 * assumed to be locked on entry and remains so on return.
2797 * %false - we are done with this request
2798 * %true - still buffers pending for this request
2800 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2801 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2803 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2806 blk_finish_request(rq
, error
);
2812 * blk_end_request - Helper function for drivers to complete the request.
2813 * @rq: the request being processed
2814 * @error: %0 for success, < %0 for error
2815 * @nr_bytes: number of bytes to complete
2818 * Ends I/O on a number of bytes attached to @rq.
2819 * If @rq has leftover, sets it up for the next range of segments.
2822 * %false - we are done with this request
2823 * %true - still buffers pending for this request
2825 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2827 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2829 EXPORT_SYMBOL(blk_end_request
);
2832 * blk_end_request_all - Helper function for drives to finish the request.
2833 * @rq: the request to finish
2834 * @error: %0 for success, < %0 for error
2837 * Completely finish @rq.
2839 void blk_end_request_all(struct request
*rq
, int error
)
2842 unsigned int bidi_bytes
= 0;
2844 if (unlikely(blk_bidi_rq(rq
)))
2845 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2847 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2850 EXPORT_SYMBOL(blk_end_request_all
);
2853 * __blk_end_request - Helper function for drivers to complete the request.
2854 * @rq: the request being processed
2855 * @error: %0 for success, < %0 for error
2856 * @nr_bytes: number of bytes to complete
2859 * Must be called with queue lock held unlike blk_end_request().
2862 * %false - we are done with this request
2863 * %true - still buffers pending for this request
2865 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2867 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2869 EXPORT_SYMBOL(__blk_end_request
);
2872 * __blk_end_request_all - Helper function for drives to finish the request.
2873 * @rq: the request to finish
2874 * @error: %0 for success, < %0 for error
2877 * Completely finish @rq. Must be called with queue lock held.
2879 void __blk_end_request_all(struct request
*rq
, int error
)
2882 unsigned int bidi_bytes
= 0;
2884 if (unlikely(blk_bidi_rq(rq
)))
2885 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2887 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2890 EXPORT_SYMBOL(__blk_end_request_all
);
2893 * __blk_end_request_cur - Helper function to finish the current request chunk.
2894 * @rq: the request to finish the current chunk for
2895 * @error: %0 for success, < %0 for error
2898 * Complete the current consecutively mapped chunk from @rq. Must
2899 * be called with queue lock held.
2902 * %false - we are done with this request
2903 * %true - still buffers pending for this request
2905 bool __blk_end_request_cur(struct request
*rq
, int error
)
2907 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2909 EXPORT_SYMBOL(__blk_end_request_cur
);
2911 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2914 if (bio_has_data(bio
))
2915 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2917 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2918 rq
->bio
= rq
->biotail
= bio
;
2921 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2924 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2926 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2927 * @rq: the request to be flushed
2930 * Flush all pages in @rq.
2932 void rq_flush_dcache_pages(struct request
*rq
)
2934 struct req_iterator iter
;
2935 struct bio_vec bvec
;
2937 rq_for_each_segment(bvec
, rq
, iter
)
2938 flush_dcache_page(bvec
.bv_page
);
2940 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2944 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2945 * @q : the queue of the device being checked
2948 * Check if underlying low-level drivers of a device are busy.
2949 * If the drivers want to export their busy state, they must set own
2950 * exporting function using blk_queue_lld_busy() first.
2952 * Basically, this function is used only by request stacking drivers
2953 * to stop dispatching requests to underlying devices when underlying
2954 * devices are busy. This behavior helps more I/O merging on the queue
2955 * of the request stacking driver and prevents I/O throughput regression
2956 * on burst I/O load.
2959 * 0 - Not busy (The request stacking driver should dispatch request)
2960 * 1 - Busy (The request stacking driver should stop dispatching request)
2962 int blk_lld_busy(struct request_queue
*q
)
2965 return q
->lld_busy_fn(q
);
2969 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2972 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2973 * @rq: the clone request to be cleaned up
2976 * Free all bios in @rq for a cloned request.
2978 void blk_rq_unprep_clone(struct request
*rq
)
2982 while ((bio
= rq
->bio
) != NULL
) {
2983 rq
->bio
= bio
->bi_next
;
2988 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2991 * Copy attributes of the original request to the clone request.
2992 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2994 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2996 dst
->cpu
= src
->cpu
;
2997 dst
->__sector
= blk_rq_pos(src
);
2998 dst
->__data_len
= blk_rq_bytes(src
);
2999 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3000 dst
->ioprio
= src
->ioprio
;
3001 dst
->extra_len
= src
->extra_len
;
3005 * blk_rq_prep_clone - Helper function to setup clone request
3006 * @rq: the request to be setup
3007 * @rq_src: original request to be cloned
3008 * @bs: bio_set that bios for clone are allocated from
3009 * @gfp_mask: memory allocation mask for bio
3010 * @bio_ctr: setup function to be called for each clone bio.
3011 * Returns %0 for success, non %0 for failure.
3012 * @data: private data to be passed to @bio_ctr
3015 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3016 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3017 * are not copied, and copying such parts is the caller's responsibility.
3018 * Also, pages which the original bios are pointing to are not copied
3019 * and the cloned bios just point same pages.
3020 * So cloned bios must be completed before original bios, which means
3021 * the caller must complete @rq before @rq_src.
3023 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3024 struct bio_set
*bs
, gfp_t gfp_mask
,
3025 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3028 struct bio
*bio
, *bio_src
;
3033 __rq_for_each_bio(bio_src
, rq_src
) {
3034 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3038 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3042 rq
->biotail
->bi_next
= bio
;
3045 rq
->bio
= rq
->biotail
= bio
;
3048 __blk_rq_prep_clone(rq
, rq_src
);
3055 blk_rq_unprep_clone(rq
);
3059 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3061 int kblockd_schedule_work(struct work_struct
*work
)
3063 return queue_work(kblockd_workqueue
, work
);
3065 EXPORT_SYMBOL(kblockd_schedule_work
);
3067 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3069 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3071 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3073 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3074 unsigned long delay
)
3076 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3078 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3080 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3081 unsigned long delay
)
3083 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3085 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3087 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3088 unsigned long delay
)
3090 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3092 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3095 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3096 * @plug: The &struct blk_plug that needs to be initialized
3099 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3100 * pending I/O should the task end up blocking between blk_start_plug() and
3101 * blk_finish_plug(). This is important from a performance perspective, but
3102 * also ensures that we don't deadlock. For instance, if the task is blocking
3103 * for a memory allocation, memory reclaim could end up wanting to free a
3104 * page belonging to that request that is currently residing in our private
3105 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3106 * this kind of deadlock.
3108 void blk_start_plug(struct blk_plug
*plug
)
3110 struct task_struct
*tsk
= current
;
3113 * If this is a nested plug, don't actually assign it.
3118 INIT_LIST_HEAD(&plug
->list
);
3119 INIT_LIST_HEAD(&plug
->mq_list
);
3120 INIT_LIST_HEAD(&plug
->cb_list
);
3122 * Store ordering should not be needed here, since a potential
3123 * preempt will imply a full memory barrier
3127 EXPORT_SYMBOL(blk_start_plug
);
3129 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3131 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3132 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3134 return !(rqa
->q
< rqb
->q
||
3135 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3139 * If 'from_schedule' is true, then postpone the dispatch of requests
3140 * until a safe kblockd context. We due this to avoid accidental big
3141 * additional stack usage in driver dispatch, in places where the originally
3142 * plugger did not intend it.
3144 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3146 __releases(q
->queue_lock
)
3148 trace_block_unplug(q
, depth
, !from_schedule
);
3151 blk_run_queue_async(q
);
3154 spin_unlock(q
->queue_lock
);
3157 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3159 LIST_HEAD(callbacks
);
3161 while (!list_empty(&plug
->cb_list
)) {
3162 list_splice_init(&plug
->cb_list
, &callbacks
);
3164 while (!list_empty(&callbacks
)) {
3165 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3168 list_del(&cb
->list
);
3169 cb
->callback(cb
, from_schedule
);
3174 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3177 struct blk_plug
*plug
= current
->plug
;
3178 struct blk_plug_cb
*cb
;
3183 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3184 if (cb
->callback
== unplug
&& cb
->data
== data
)
3187 /* Not currently on the callback list */
3188 BUG_ON(size
< sizeof(*cb
));
3189 cb
= kzalloc(size
, GFP_ATOMIC
);
3192 cb
->callback
= unplug
;
3193 list_add(&cb
->list
, &plug
->cb_list
);
3197 EXPORT_SYMBOL(blk_check_plugged
);
3199 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3201 struct request_queue
*q
;
3202 unsigned long flags
;
3207 flush_plug_callbacks(plug
, from_schedule
);
3209 if (!list_empty(&plug
->mq_list
))
3210 blk_mq_flush_plug_list(plug
, from_schedule
);
3212 if (list_empty(&plug
->list
))
3215 list_splice_init(&plug
->list
, &list
);
3217 list_sort(NULL
, &list
, plug_rq_cmp
);
3223 * Save and disable interrupts here, to avoid doing it for every
3224 * queue lock we have to take.
3226 local_irq_save(flags
);
3227 while (!list_empty(&list
)) {
3228 rq
= list_entry_rq(list
.next
);
3229 list_del_init(&rq
->queuelist
);
3233 * This drops the queue lock
3236 queue_unplugged(q
, depth
, from_schedule
);
3239 spin_lock(q
->queue_lock
);
3243 * Short-circuit if @q is dead
3245 if (unlikely(blk_queue_dying(q
))) {
3246 __blk_end_request_all(rq
, -ENODEV
);
3251 * rq is already accounted, so use raw insert
3253 if (op_is_flush(rq
->cmd_flags
))
3254 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3256 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3262 * This drops the queue lock
3265 queue_unplugged(q
, depth
, from_schedule
);
3267 local_irq_restore(flags
);
3270 void blk_finish_plug(struct blk_plug
*plug
)
3272 if (plug
!= current
->plug
)
3274 blk_flush_plug_list(plug
, false);
3276 current
->plug
= NULL
;
3278 EXPORT_SYMBOL(blk_finish_plug
);
3282 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3283 * @q: the queue of the device
3284 * @dev: the device the queue belongs to
3287 * Initialize runtime-PM-related fields for @q and start auto suspend for
3288 * @dev. Drivers that want to take advantage of request-based runtime PM
3289 * should call this function after @dev has been initialized, and its
3290 * request queue @q has been allocated, and runtime PM for it can not happen
3291 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3292 * cases, driver should call this function before any I/O has taken place.
3294 * This function takes care of setting up using auto suspend for the device,
3295 * the autosuspend delay is set to -1 to make runtime suspend impossible
3296 * until an updated value is either set by user or by driver. Drivers do
3297 * not need to touch other autosuspend settings.
3299 * The block layer runtime PM is request based, so only works for drivers
3300 * that use request as their IO unit instead of those directly use bio's.
3302 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3305 q
->rpm_status
= RPM_ACTIVE
;
3306 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3307 pm_runtime_use_autosuspend(q
->dev
);
3309 EXPORT_SYMBOL(blk_pm_runtime_init
);
3312 * blk_pre_runtime_suspend - Pre runtime suspend check
3313 * @q: the queue of the device
3316 * This function will check if runtime suspend is allowed for the device
3317 * by examining if there are any requests pending in the queue. If there
3318 * are requests pending, the device can not be runtime suspended; otherwise,
3319 * the queue's status will be updated to SUSPENDING and the driver can
3320 * proceed to suspend the device.
3322 * For the not allowed case, we mark last busy for the device so that
3323 * runtime PM core will try to autosuspend it some time later.
3325 * This function should be called near the start of the device's
3326 * runtime_suspend callback.
3329 * 0 - OK to runtime suspend the device
3330 * -EBUSY - Device should not be runtime suspended
3332 int blk_pre_runtime_suspend(struct request_queue
*q
)
3339 spin_lock_irq(q
->queue_lock
);
3340 if (q
->nr_pending
) {
3342 pm_runtime_mark_last_busy(q
->dev
);
3344 q
->rpm_status
= RPM_SUSPENDING
;
3346 spin_unlock_irq(q
->queue_lock
);
3349 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3352 * blk_post_runtime_suspend - Post runtime suspend processing
3353 * @q: the queue of the device
3354 * @err: return value of the device's runtime_suspend function
3357 * Update the queue's runtime status according to the return value of the
3358 * device's runtime suspend function and mark last busy for the device so
3359 * that PM core will try to auto suspend the device at a later time.
3361 * This function should be called near the end of the device's
3362 * runtime_suspend callback.
3364 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3369 spin_lock_irq(q
->queue_lock
);
3371 q
->rpm_status
= RPM_SUSPENDED
;
3373 q
->rpm_status
= RPM_ACTIVE
;
3374 pm_runtime_mark_last_busy(q
->dev
);
3376 spin_unlock_irq(q
->queue_lock
);
3378 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3381 * blk_pre_runtime_resume - Pre runtime resume processing
3382 * @q: the queue of the device
3385 * Update the queue's runtime status to RESUMING in preparation for the
3386 * runtime resume of the device.
3388 * This function should be called near the start of the device's
3389 * runtime_resume callback.
3391 void blk_pre_runtime_resume(struct request_queue
*q
)
3396 spin_lock_irq(q
->queue_lock
);
3397 q
->rpm_status
= RPM_RESUMING
;
3398 spin_unlock_irq(q
->queue_lock
);
3400 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3403 * blk_post_runtime_resume - Post runtime resume processing
3404 * @q: the queue of the device
3405 * @err: return value of the device's runtime_resume function
3408 * Update the queue's runtime status according to the return value of the
3409 * device's runtime_resume function. If it is successfully resumed, process
3410 * the requests that are queued into the device's queue when it is resuming
3411 * and then mark last busy and initiate autosuspend for it.
3413 * This function should be called near the end of the device's
3414 * runtime_resume callback.
3416 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3421 spin_lock_irq(q
->queue_lock
);
3423 q
->rpm_status
= RPM_ACTIVE
;
3425 pm_runtime_mark_last_busy(q
->dev
);
3426 pm_request_autosuspend(q
->dev
);
3428 q
->rpm_status
= RPM_SUSPENDED
;
3430 spin_unlock_irq(q
->queue_lock
);
3432 EXPORT_SYMBOL(blk_post_runtime_resume
);
3435 * blk_set_runtime_active - Force runtime status of the queue to be active
3436 * @q: the queue of the device
3438 * If the device is left runtime suspended during system suspend the resume
3439 * hook typically resumes the device and corrects runtime status
3440 * accordingly. However, that does not affect the queue runtime PM status
3441 * which is still "suspended". This prevents processing requests from the
3444 * This function can be used in driver's resume hook to correct queue
3445 * runtime PM status and re-enable peeking requests from the queue. It
3446 * should be called before first request is added to the queue.
3448 void blk_set_runtime_active(struct request_queue
*q
)
3450 spin_lock_irq(q
->queue_lock
);
3451 q
->rpm_status
= RPM_ACTIVE
;
3452 pm_runtime_mark_last_busy(q
->dev
);
3453 pm_request_autosuspend(q
->dev
);
3454 spin_unlock_irq(q
->queue_lock
);
3456 EXPORT_SYMBOL(blk_set_runtime_active
);
3459 int __init
blk_dev_init(void)
3461 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3462 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3463 FIELD_SIZEOF(struct request
, cmd_flags
));
3464 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3465 FIELD_SIZEOF(struct bio
, bi_opf
));
3467 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3468 kblockd_workqueue
= alloc_workqueue("kblockd",
3469 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3470 if (!kblockd_workqueue
)
3471 panic("Failed to create kblockd\n");
3473 request_cachep
= kmem_cache_create("blkdev_requests",
3474 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3476 blk_requestq_cachep
= kmem_cache_create("request_queue",
3477 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3479 #ifdef CONFIG_DEBUG_FS
3480 blk_debugfs_root
= debugfs_create_dir("block", NULL
);