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);
129 seqcount_init(&rq->gstate_seq);
130 u64_stats_init(&rq->aborted_gstate_sync);
132 EXPORT_SYMBOL(blk_rq_init);
134 static const struct {
138 [BLK_STS_OK] = { 0, "" },
139 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
140 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
141 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
142 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
143 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
144 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
145 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
146 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
147 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
148 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
150 /* device mapper special case, should not leak out: */
151 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
153 /* everything else not covered above: */
154 [BLK_STS_IOERR] = { -EIO, "I/O" },
157 blk_status_t errno_to_blk_status(int errno)
161 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
162 if (blk_errors[i].errno == errno)
163 return (__force blk_status_t)i;
166 return BLK_STS_IOERR;
168 EXPORT_SYMBOL_GPL(errno_to_blk_status);
170 int blk_status_to_errno(blk_status_t status)
172 int idx = (__force int)status;
174 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
176 return blk_errors[idx].errno;
178 EXPORT_SYMBOL_GPL(blk_status_to_errno);
180 static void print_req_error(struct request *req, blk_status_t status)
182 int idx = (__force int)status;
184 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
187 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
188 __func__, blk_errors[idx].name, req->rq_disk ?
189 req->rq_disk->disk_name : "?",
190 (unsigned long long)blk_rq_pos(req));
193 static void req_bio_endio(struct request *rq, struct bio *bio,
194 unsigned int nbytes, blk_status_t error)
197 bio->bi_status = error;
199 if (unlikely(rq->rq_flags & RQF_QUIET))
200 bio_set_flag(bio, BIO_QUIET);
202 bio_advance(bio, nbytes);
204 /* don't actually finish bio if it's part of flush sequence */
205 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
209 void blk_dump_rq_flags(struct request *rq, char *msg)
211 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
212 rq->rq_disk ? rq->rq_disk->disk_name : "?",
213 (unsigned long long) rq->cmd_flags);
215 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
216 (unsigned long long)blk_rq_pos(rq),
217 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
218 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
219 rq->bio, rq->biotail, blk_rq_bytes(rq));
221 EXPORT_SYMBOL(blk_dump_rq_flags);
223 static void blk_delay_work(struct work_struct *work)
225 struct request_queue *q;
227 q = container_of(work, struct request_queue, delay_work.work);
228 spin_lock_irq(q->queue_lock);
230 spin_unlock_irq(q->queue_lock);
234 * blk_delay_queue - restart queueing after defined interval
235 * @q: The &struct request_queue in question
236 * @msecs: Delay in msecs
239 * Sometimes queueing needs to be postponed for a little while, to allow
240 * resources to come back. This function will make sure that queueing is
241 * restarted around the specified time.
243 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
245 lockdep_assert_held(q->queue_lock);
246 WARN_ON_ONCE(q->mq_ops);
248 if (likely(!blk_queue_dead(q)))
249 queue_delayed_work(kblockd_workqueue, &q->delay_work,
250 msecs_to_jiffies(msecs));
252 EXPORT_SYMBOL(blk_delay_queue);
255 * blk_start_queue_async - asynchronously restart a previously stopped queue
256 * @q: The &struct request_queue in question
259 * blk_start_queue_async() will clear the stop flag on the queue, and
260 * ensure that the request_fn for the queue is run from an async
263 void blk_start_queue_async(struct request_queue *q)
265 lockdep_assert_held(q->queue_lock);
266 WARN_ON_ONCE(q->mq_ops);
268 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
269 blk_run_queue_async(q);
271 EXPORT_SYMBOL(blk_start_queue_async);
274 * blk_start_queue - restart a previously stopped queue
275 * @q: The &struct request_queue in question
278 * blk_start_queue() will clear the stop flag on the queue, and call
279 * the request_fn for the queue if it was in a stopped state when
280 * entered. Also see blk_stop_queue().
282 void blk_start_queue(struct request_queue *q)
284 lockdep_assert_held(q->queue_lock);
285 WARN_ON(!in_interrupt() && !irqs_disabled());
286 WARN_ON_ONCE(q->mq_ops);
288 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
291 EXPORT_SYMBOL(blk_start_queue);
294 * blk_stop_queue - stop a queue
295 * @q: The &struct request_queue in question
298 * The Linux block layer assumes that a block driver will consume all
299 * entries on the request queue when the request_fn strategy is called.
300 * Often this will not happen, because of hardware limitations (queue
301 * depth settings). If a device driver gets a 'queue full' response,
302 * or if it simply chooses not to queue more I/O at one point, it can
303 * call this function to prevent the request_fn from being called until
304 * the driver has signalled it's ready to go again. This happens by calling
305 * blk_start_queue() to restart queue operations.
307 void blk_stop_queue(struct request_queue *q)
309 lockdep_assert_held(q->queue_lock);
310 WARN_ON_ONCE(q->mq_ops);
312 cancel_delayed_work(&q->delay_work);
313 queue_flag_set(QUEUE_FLAG_STOPPED, q);
315 EXPORT_SYMBOL(blk_stop_queue);
318 * blk_sync_queue - cancel any pending callbacks on a queue
322 * The block layer may perform asynchronous callback activity
323 * on a queue, such as calling the unplug function after a timeout.
324 * A block device may call blk_sync_queue to ensure that any
325 * such activity is cancelled, thus allowing it to release resources
326 * that the callbacks might use. The caller must already have made sure
327 * that its ->make_request_fn will not re-add plugging prior to calling
330 * This function does not cancel any asynchronous activity arising
331 * out of elevator or throttling code. That would require elevator_exit()
332 * and blkcg_exit_queue() to be called with queue lock initialized.
335 void blk_sync_queue(struct request_queue *q)
337 del_timer_sync(&q->timeout);
338 cancel_work_sync(&q->timeout_work);
341 struct blk_mq_hw_ctx *hctx;
344 cancel_delayed_work_sync(&q->requeue_work);
345 queue_for_each_hw_ctx(q, hctx, i)
346 cancel_delayed_work_sync(&hctx->run_work);
348 cancel_delayed_work_sync(&q->delay_work);
351 EXPORT_SYMBOL(blk_sync_queue);
354 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
355 * @q: request queue pointer
357 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
358 * set and 1 if the flag was already set.
360 int blk_set_preempt_only(struct request_queue *q)
365 spin_lock_irqsave(q->queue_lock, flags);
366 res = queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
367 spin_unlock_irqrestore(q->queue_lock, flags);
371 EXPORT_SYMBOL_GPL(blk_set_preempt_only);
373 void blk_clear_preempt_only(struct request_queue *q)
377 spin_lock_irqsave(q->queue_lock, flags);
378 queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
379 wake_up_all(&q->mq_freeze_wq);
380 spin_unlock_irqrestore(q->queue_lock, flags);
382 EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
385 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
386 * @q: The queue to run
389 * Invoke request handling on a queue if there are any pending requests.
390 * May be used to restart request handling after a request has completed.
391 * This variant runs the queue whether or not the queue has been
392 * stopped. Must be called with the queue lock held and interrupts
393 * disabled. See also @blk_run_queue.
395 inline void __blk_run_queue_uncond(struct request_queue *q)
397 lockdep_assert_held(q->queue_lock);
398 WARN_ON_ONCE(q->mq_ops);
400 if (unlikely(blk_queue_dead(q)))
404 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
405 * the queue lock internally. As a result multiple threads may be
406 * running such a request function concurrently. Keep track of the
407 * number of active request_fn invocations such that blk_drain_queue()
408 * can wait until all these request_fn calls have finished.
410 q->request_fn_active++;
412 q->request_fn_active--;
414 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
417 * __blk_run_queue - run a single device queue
418 * @q: The queue to run
421 * See @blk_run_queue.
423 void __blk_run_queue(struct request_queue *q)
425 lockdep_assert_held(q->queue_lock);
426 WARN_ON_ONCE(q->mq_ops);
428 if (unlikely(blk_queue_stopped(q)))
431 __blk_run_queue_uncond(q);
433 EXPORT_SYMBOL(__blk_run_queue);
436 * blk_run_queue_async - run a single device queue in workqueue context
437 * @q: The queue to run
440 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
444 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
445 * has canceled q->delay_work, callers must hold the queue lock to avoid
446 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
448 void blk_run_queue_async(struct request_queue *q)
450 lockdep_assert_held(q->queue_lock);
451 WARN_ON_ONCE(q->mq_ops);
453 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
454 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
456 EXPORT_SYMBOL(blk_run_queue_async);
459 * blk_run_queue - run a single device queue
460 * @q: The queue to run
463 * Invoke request handling on this queue, if it has pending work to do.
464 * May be used to restart queueing when a request has completed.
466 void blk_run_queue(struct request_queue *q)
470 WARN_ON_ONCE(q->mq_ops);
472 spin_lock_irqsave(q->queue_lock, flags);
474 spin_unlock_irqrestore(q->queue_lock, flags);
476 EXPORT_SYMBOL(blk_run_queue);
478 void blk_put_queue(struct request_queue *q)
480 kobject_put(&q->kobj);
482 EXPORT_SYMBOL(blk_put_queue);
485 * __blk_drain_queue - drain requests from request_queue
487 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
489 * Drain requests from @q. If @drain_all is set, all requests are drained.
490 * If not, only ELVPRIV requests are drained. The caller is responsible
491 * for ensuring that no new requests which need to be drained are queued.
493 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
494 __releases(q->queue_lock)
495 __acquires(q->queue_lock)
499 lockdep_assert_held(q->queue_lock);
500 WARN_ON_ONCE(q->mq_ops);
506 * The caller might be trying to drain @q before its
507 * elevator is initialized.
510 elv_drain_elevator(q);
512 blkcg_drain_queue(q);
515 * This function might be called on a queue which failed
516 * driver init after queue creation or is not yet fully
517 * active yet. Some drivers (e.g. fd and loop) get unhappy
518 * in such cases. Kick queue iff dispatch queue has
519 * something on it and @q has request_fn set.
521 if (!list_empty(&q->queue_head) && q->request_fn)
524 drain |= q->nr_rqs_elvpriv;
525 drain |= q->request_fn_active;
528 * Unfortunately, requests are queued at and tracked from
529 * multiple places and there's no single counter which can
530 * be drained. Check all the queues and counters.
533 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
534 drain |= !list_empty(&q->queue_head);
535 for (i = 0; i < 2; i++) {
536 drain |= q->nr_rqs[i];
537 drain |= q->in_flight[i];
539 drain |= !list_empty(&fq->flush_queue[i]);
546 spin_unlock_irq(q->queue_lock);
550 spin_lock_irq(q->queue_lock);
554 * With queue marked dead, any woken up waiter will fail the
555 * allocation path, so the wakeup chaining is lost and we're
556 * left with hung waiters. We need to wake up those waiters.
559 struct request_list *rl;
561 blk_queue_for_each_rl(rl, q)
562 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
563 wake_up_all(&rl->wait[i]);
568 * blk_queue_bypass_start - enter queue bypass mode
569 * @q: queue of interest
571 * In bypass mode, only the dispatch FIFO queue of @q is used. This
572 * function makes @q enter bypass mode and drains all requests which were
573 * throttled or issued before. On return, it's guaranteed that no request
574 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
575 * inside queue or RCU read lock.
577 void blk_queue_bypass_start(struct request_queue *q)
579 WARN_ON_ONCE(q->mq_ops);
581 spin_lock_irq(q->queue_lock);
583 queue_flag_set(QUEUE_FLAG_BYPASS, q);
584 spin_unlock_irq(q->queue_lock);
587 * Queues start drained. Skip actual draining till init is
588 * complete. This avoids lenghty delays during queue init which
589 * can happen many times during boot.
591 if (blk_queue_init_done(q)) {
592 spin_lock_irq(q->queue_lock);
593 __blk_drain_queue(q, false);
594 spin_unlock_irq(q->queue_lock);
596 /* ensure blk_queue_bypass() is %true inside RCU read lock */
600 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
603 * blk_queue_bypass_end - leave queue bypass mode
604 * @q: queue of interest
606 * Leave bypass mode and restore the normal queueing behavior.
608 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
609 * this function is called for both blk-sq and blk-mq queues.
611 void blk_queue_bypass_end(struct request_queue *q)
613 spin_lock_irq(q->queue_lock);
614 if (!--q->bypass_depth)
615 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
616 WARN_ON_ONCE(q->bypass_depth < 0);
617 spin_unlock_irq(q->queue_lock);
619 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
621 void blk_set_queue_dying(struct request_queue *q)
623 spin_lock_irq(q->queue_lock);
624 queue_flag_set(QUEUE_FLAG_DYING, q);
625 spin_unlock_irq(q->queue_lock);
628 * When queue DYING flag is set, we need to block new req
629 * entering queue, so we call blk_freeze_queue_start() to
630 * prevent I/O from crossing blk_queue_enter().
632 blk_freeze_queue_start(q);
635 blk_mq_wake_waiters(q);
637 struct request_list *rl;
639 spin_lock_irq(q->queue_lock);
640 blk_queue_for_each_rl(rl, q) {
642 wake_up_all(&rl->wait[BLK_RW_SYNC]);
643 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
646 spin_unlock_irq(q->queue_lock);
649 /* Make blk_queue_enter() reexamine the DYING flag. */
650 wake_up_all(&q->mq_freeze_wq);
652 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
655 * blk_cleanup_queue - shutdown a request queue
656 * @q: request queue to shutdown
658 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
659 * put it. All future requests will be failed immediately with -ENODEV.
661 void blk_cleanup_queue(struct request_queue *q)
663 spinlock_t *lock = q->queue_lock;
665 /* mark @q DYING, no new request or merges will be allowed afterwards */
666 mutex_lock(&q->sysfs_lock);
667 blk_set_queue_dying(q);
671 * A dying queue is permanently in bypass mode till released. Note
672 * that, unlike blk_queue_bypass_start(), we aren't performing
673 * synchronize_rcu() after entering bypass mode to avoid the delay
674 * as some drivers create and destroy a lot of queues while
675 * probing. This is still safe because blk_release_queue() will be
676 * called only after the queue refcnt drops to zero and nothing,
677 * RCU or not, would be traversing the queue by then.
680 queue_flag_set(QUEUE_FLAG_BYPASS, q);
682 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
683 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
684 queue_flag_set(QUEUE_FLAG_DYING, q);
685 spin_unlock_irq(lock);
686 mutex_unlock(&q->sysfs_lock);
689 * Drain all requests queued before DYING marking. Set DEAD flag to
690 * prevent that q->request_fn() gets invoked after draining finished.
695 __blk_drain_queue(q, true);
696 queue_flag_set(QUEUE_FLAG_DEAD, q);
697 spin_unlock_irq(lock);
700 * make sure all in-progress dispatch are completed because
701 * blk_freeze_queue() can only complete all requests, and
702 * dispatch may still be in-progress since we dispatch requests
703 * from more than one contexts
706 blk_mq_quiesce_queue(q);
708 /* for synchronous bio-based driver finish in-flight integrity i/o */
709 blk_flush_integrity();
711 /* @q won't process any more request, flush async actions */
712 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
716 blk_mq_free_queue(q);
717 percpu_ref_exit(&q->q_usage_counter);
720 if (q->queue_lock != &q->__queue_lock)
721 q->queue_lock = &q->__queue_lock;
722 spin_unlock_irq(lock);
724 /* @q is and will stay empty, shutdown and put */
727 EXPORT_SYMBOL(blk_cleanup_queue);
729 /* Allocate memory local to the request queue */
730 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
732 struct request_queue *q = data;
734 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
737 static void free_request_simple(void *element, void *data)
739 kmem_cache_free(request_cachep, element);
742 static void *alloc_request_size(gfp_t gfp_mask, void *data)
744 struct request_queue *q = data;
747 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
749 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
756 static void free_request_size(void *element, void *data)
758 struct request_queue *q = data;
761 q->exit_rq_fn(q, element);
765 int blk_init_rl(struct request_list *rl, struct request_queue *q,
768 if (unlikely(rl->rq_pool) || q->mq_ops)
772 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
773 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
774 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
775 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
778 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
779 alloc_request_size, free_request_size,
780 q, gfp_mask, q->node);
782 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
783 alloc_request_simple, free_request_simple,
784 q, gfp_mask, q->node);
789 if (rl != &q->root_rl)
790 WARN_ON_ONCE(!blk_get_queue(q));
795 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
798 mempool_destroy(rl->rq_pool);
799 if (rl != &q->root_rl)
804 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
806 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
808 EXPORT_SYMBOL(blk_alloc_queue);
811 * blk_queue_enter() - try to increase q->q_usage_counter
812 * @q: request queue pointer
813 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
815 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
817 const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
820 bool success = false;
823 rcu_read_lock_sched();
824 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
826 * The code that sets the PREEMPT_ONLY flag is
827 * responsible for ensuring that that flag is globally
828 * visible before the queue is unfrozen.
830 if (preempt || !blk_queue_preempt_only(q)) {
833 percpu_ref_put(&q->q_usage_counter);
836 rcu_read_unlock_sched();
841 if (flags & BLK_MQ_REQ_NOWAIT)
845 * read pair of barrier in blk_freeze_queue_start(),
846 * we need to order reading __PERCPU_REF_DEAD flag of
847 * .q_usage_counter and reading .mq_freeze_depth or
848 * queue dying flag, otherwise the following wait may
849 * never return if the two reads are reordered.
853 ret = wait_event_interruptible(q->mq_freeze_wq,
854 (atomic_read(&q->mq_freeze_depth) == 0 &&
855 (preempt || !blk_queue_preempt_only(q))) ||
857 if (blk_queue_dying(q))
864 void blk_queue_exit(struct request_queue *q)
866 percpu_ref_put(&q->q_usage_counter);
869 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
871 struct request_queue *q =
872 container_of(ref, struct request_queue, q_usage_counter);
874 wake_up_all(&q->mq_freeze_wq);
877 static void blk_rq_timed_out_timer(struct timer_list *t)
879 struct request_queue *q = from_timer(q, t, timeout);
881 kblockd_schedule_work(&q->timeout_work);
884 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
886 struct request_queue *q;
888 q = kmem_cache_alloc_node(blk_requestq_cachep,
889 gfp_mask | __GFP_ZERO, node_id);
893 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
897 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
901 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
902 if (!q->backing_dev_info)
905 q->stats = blk_alloc_queue_stats();
909 q->backing_dev_info->ra_pages =
910 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
911 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
912 q->backing_dev_info->name = "block";
915 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
916 laptop_mode_timer_fn, 0);
917 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
918 INIT_WORK(&q->timeout_work, NULL);
919 INIT_LIST_HEAD(&q->queue_head);
920 INIT_LIST_HEAD(&q->timeout_list);
921 INIT_LIST_HEAD(&q->icq_list);
922 #ifdef CONFIG_BLK_CGROUP
923 INIT_LIST_HEAD(&q->blkg_list);
925 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
927 kobject_init(&q->kobj, &blk_queue_ktype);
929 #ifdef CONFIG_BLK_DEV_IO_TRACE
930 mutex_init(&q->blk_trace_mutex);
932 mutex_init(&q->sysfs_lock);
933 spin_lock_init(&q->__queue_lock);
936 * By default initialize queue_lock to internal lock and driver can
937 * override it later if need be.
939 q->queue_lock = &q->__queue_lock;
942 * A queue starts its life with bypass turned on to avoid
943 * unnecessary bypass on/off overhead and nasty surprises during
944 * init. The initial bypass will be finished when the queue is
945 * registered by blk_register_queue().
948 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
950 init_waitqueue_head(&q->mq_freeze_wq);
953 * Init percpu_ref in atomic mode so that it's faster to shutdown.
954 * See blk_register_queue() for details.
956 if (percpu_ref_init(&q->q_usage_counter,
957 blk_queue_usage_counter_release,
958 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
961 if (blkcg_init_queue(q))
967 percpu_ref_exit(&q->q_usage_counter);
969 blk_free_queue_stats(q->stats);
971 bdi_put(q->backing_dev_info);
973 bioset_free(q->bio_split);
975 ida_simple_remove(&blk_queue_ida, q->id);
977 kmem_cache_free(blk_requestq_cachep, q);
980 EXPORT_SYMBOL(blk_alloc_queue_node);
983 * blk_init_queue - prepare a request queue for use with a block device
984 * @rfn: The function to be called to process requests that have been
985 * placed on the queue.
986 * @lock: Request queue spin lock
989 * If a block device wishes to use the standard request handling procedures,
990 * which sorts requests and coalesces adjacent requests, then it must
991 * call blk_init_queue(). The function @rfn will be called when there
992 * are requests on the queue that need to be processed. If the device
993 * supports plugging, then @rfn may not be called immediately when requests
994 * are available on the queue, but may be called at some time later instead.
995 * Plugged queues are generally unplugged when a buffer belonging to one
996 * of the requests on the queue is needed, or due to memory pressure.
998 * @rfn is not required, or even expected, to remove all requests off the
999 * queue, but only as many as it can handle at a time. If it does leave
1000 * requests on the queue, it is responsible for arranging that the requests
1001 * get dealt with eventually.
1003 * The queue spin lock must be held while manipulating the requests on the
1004 * request queue; this lock will be taken also from interrupt context, so irq
1005 * disabling is needed for it.
1007 * Function returns a pointer to the initialized request queue, or %NULL if
1008 * it didn't succeed.
1011 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1012 * when the block device is deactivated (such as at module unload).
1015 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1017 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1019 EXPORT_SYMBOL(blk_init_queue);
1021 struct request_queue *
1022 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1024 struct request_queue *q;
1026 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1030 q->request_fn = rfn;
1032 q->queue_lock = lock;
1033 if (blk_init_allocated_queue(q) < 0) {
1034 blk_cleanup_queue(q);
1040 EXPORT_SYMBOL(blk_init_queue_node);
1042 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1045 int blk_init_allocated_queue(struct request_queue *q)
1047 WARN_ON_ONCE(q->mq_ops);
1049 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
1053 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1054 goto out_free_flush_queue;
1056 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1057 goto out_exit_flush_rq;
1059 INIT_WORK(&q->timeout_work, blk_timeout_work);
1060 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1063 * This also sets hw/phys segments, boundary and size
1065 blk_queue_make_request(q, blk_queue_bio);
1067 q->sg_reserved_size = INT_MAX;
1069 /* Protect q->elevator from elevator_change */
1070 mutex_lock(&q->sysfs_lock);
1073 if (elevator_init(q, NULL)) {
1074 mutex_unlock(&q->sysfs_lock);
1075 goto out_exit_flush_rq;
1078 mutex_unlock(&q->sysfs_lock);
1083 q->exit_rq_fn(q, q->fq->flush_rq);
1084 out_free_flush_queue:
1085 blk_free_flush_queue(q->fq);
1088 EXPORT_SYMBOL(blk_init_allocated_queue);
1090 bool blk_get_queue(struct request_queue *q)
1092 if (likely(!blk_queue_dying(q))) {
1099 EXPORT_SYMBOL(blk_get_queue);
1101 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1103 if (rq->rq_flags & RQF_ELVPRIV) {
1104 elv_put_request(rl->q, rq);
1106 put_io_context(rq->elv.icq->ioc);
1109 mempool_free(rq, rl->rq_pool);
1113 * ioc_batching returns true if the ioc is a valid batching request and
1114 * should be given priority access to a request.
1116 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1122 * Make sure the process is able to allocate at least 1 request
1123 * even if the batch times out, otherwise we could theoretically
1126 return ioc->nr_batch_requests == q->nr_batching ||
1127 (ioc->nr_batch_requests > 0
1128 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1132 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1133 * will cause the process to be a "batcher" on all queues in the system. This
1134 * is the behaviour we want though - once it gets a wakeup it should be given
1137 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1139 if (!ioc || ioc_batching(q, ioc))
1142 ioc->nr_batch_requests = q->nr_batching;
1143 ioc->last_waited = jiffies;
1146 static void __freed_request(struct request_list *rl, int sync)
1148 struct request_queue *q = rl->q;
1150 if (rl->count[sync] < queue_congestion_off_threshold(q))
1151 blk_clear_congested(rl, sync);
1153 if (rl->count[sync] + 1 <= q->nr_requests) {
1154 if (waitqueue_active(&rl->wait[sync]))
1155 wake_up(&rl->wait[sync]);
1157 blk_clear_rl_full(rl, sync);
1162 * A request has just been released. Account for it, update the full and
1163 * congestion status, wake up any waiters. Called under q->queue_lock.
1165 static void freed_request(struct request_list *rl, bool sync,
1166 req_flags_t rq_flags)
1168 struct request_queue *q = rl->q;
1172 if (rq_flags & RQF_ELVPRIV)
1173 q->nr_rqs_elvpriv--;
1175 __freed_request(rl, sync);
1177 if (unlikely(rl->starved[sync ^ 1]))
1178 __freed_request(rl, sync ^ 1);
1181 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1183 struct request_list *rl;
1184 int on_thresh, off_thresh;
1186 WARN_ON_ONCE(q->mq_ops);
1188 spin_lock_irq(q->queue_lock);
1189 q->nr_requests = nr;
1190 blk_queue_congestion_threshold(q);
1191 on_thresh = queue_congestion_on_threshold(q);
1192 off_thresh = queue_congestion_off_threshold(q);
1194 blk_queue_for_each_rl(rl, q) {
1195 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1196 blk_set_congested(rl, BLK_RW_SYNC);
1197 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1198 blk_clear_congested(rl, BLK_RW_SYNC);
1200 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1201 blk_set_congested(rl, BLK_RW_ASYNC);
1202 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1203 blk_clear_congested(rl, BLK_RW_ASYNC);
1205 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1206 blk_set_rl_full(rl, BLK_RW_SYNC);
1208 blk_clear_rl_full(rl, BLK_RW_SYNC);
1209 wake_up(&rl->wait[BLK_RW_SYNC]);
1212 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1213 blk_set_rl_full(rl, BLK_RW_ASYNC);
1215 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1216 wake_up(&rl->wait[BLK_RW_ASYNC]);
1220 spin_unlock_irq(q->queue_lock);
1225 * __get_request - get a free request
1226 * @rl: request list to allocate from
1227 * @op: operation and flags
1228 * @bio: bio to allocate request for (can be %NULL)
1229 * @flags: BLQ_MQ_REQ_* flags
1231 * Get a free request from @q. This function may fail under memory
1232 * pressure or if @q is dead.
1234 * Must be called with @q->queue_lock held and,
1235 * Returns ERR_PTR on failure, with @q->queue_lock held.
1236 * Returns request pointer on success, with @q->queue_lock *not held*.
1238 static struct request *__get_request(struct request_list *rl, unsigned int op,
1239 struct bio *bio, blk_mq_req_flags_t flags)
1241 struct request_queue *q = rl->q;
1243 struct elevator_type *et = q->elevator->type;
1244 struct io_context *ioc = rq_ioc(bio);
1245 struct io_cq *icq = NULL;
1246 const bool is_sync = op_is_sync(op);
1248 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1249 __GFP_DIRECT_RECLAIM;
1250 req_flags_t rq_flags = RQF_ALLOCED;
1252 lockdep_assert_held(q->queue_lock);
1254 if (unlikely(blk_queue_dying(q)))
1255 return ERR_PTR(-ENODEV);
1257 may_queue = elv_may_queue(q, op);
1258 if (may_queue == ELV_MQUEUE_NO)
1261 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1262 if (rl->count[is_sync]+1 >= q->nr_requests) {
1264 * The queue will fill after this allocation, so set
1265 * it as full, and mark this process as "batching".
1266 * This process will be allowed to complete a batch of
1267 * requests, others will be blocked.
1269 if (!blk_rl_full(rl, is_sync)) {
1270 ioc_set_batching(q, ioc);
1271 blk_set_rl_full(rl, is_sync);
1273 if (may_queue != ELV_MQUEUE_MUST
1274 && !ioc_batching(q, ioc)) {
1276 * The queue is full and the allocating
1277 * process is not a "batcher", and not
1278 * exempted by the IO scheduler
1280 return ERR_PTR(-ENOMEM);
1284 blk_set_congested(rl, is_sync);
1288 * Only allow batching queuers to allocate up to 50% over the defined
1289 * limit of requests, otherwise we could have thousands of requests
1290 * allocated with any setting of ->nr_requests
1292 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1293 return ERR_PTR(-ENOMEM);
1295 q->nr_rqs[is_sync]++;
1296 rl->count[is_sync]++;
1297 rl->starved[is_sync] = 0;
1300 * Decide whether the new request will be managed by elevator. If
1301 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1302 * prevent the current elevator from being destroyed until the new
1303 * request is freed. This guarantees icq's won't be destroyed and
1304 * makes creating new ones safe.
1306 * Flush requests do not use the elevator so skip initialization.
1307 * This allows a request to share the flush and elevator data.
1309 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1310 * it will be created after releasing queue_lock.
1312 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1313 rq_flags |= RQF_ELVPRIV;
1314 q->nr_rqs_elvpriv++;
1315 if (et->icq_cache && ioc)
1316 icq = ioc_lookup_icq(ioc, q);
1319 if (blk_queue_io_stat(q))
1320 rq_flags |= RQF_IO_STAT;
1321 spin_unlock_irq(q->queue_lock);
1323 /* allocate and init request */
1324 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1329 blk_rq_set_rl(rq, rl);
1331 rq->rq_flags = rq_flags;
1332 if (flags & BLK_MQ_REQ_PREEMPT)
1333 rq->rq_flags |= RQF_PREEMPT;
1336 if (rq_flags & RQF_ELVPRIV) {
1337 if (unlikely(et->icq_cache && !icq)) {
1339 icq = ioc_create_icq(ioc, q, gfp_mask);
1345 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1348 /* @rq->elv.icq holds io_context until @rq is freed */
1350 get_io_context(icq->ioc);
1354 * ioc may be NULL here, and ioc_batching will be false. That's
1355 * OK, if the queue is under the request limit then requests need
1356 * not count toward the nr_batch_requests limit. There will always
1357 * be some limit enforced by BLK_BATCH_TIME.
1359 if (ioc_batching(q, ioc))
1360 ioc->nr_batch_requests--;
1362 trace_block_getrq(q, bio, op);
1367 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1368 * and may fail indefinitely under memory pressure and thus
1369 * shouldn't stall IO. Treat this request as !elvpriv. This will
1370 * disturb iosched and blkcg but weird is bettern than dead.
1372 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1373 __func__, dev_name(q->backing_dev_info->dev));
1375 rq->rq_flags &= ~RQF_ELVPRIV;
1378 spin_lock_irq(q->queue_lock);
1379 q->nr_rqs_elvpriv--;
1380 spin_unlock_irq(q->queue_lock);
1385 * Allocation failed presumably due to memory. Undo anything we
1386 * might have messed up.
1388 * Allocating task should really be put onto the front of the wait
1389 * queue, but this is pretty rare.
1391 spin_lock_irq(q->queue_lock);
1392 freed_request(rl, is_sync, rq_flags);
1395 * in the very unlikely event that allocation failed and no
1396 * requests for this direction was pending, mark us starved so that
1397 * freeing of a request in the other direction will notice
1398 * us. another possible fix would be to split the rq mempool into
1402 if (unlikely(rl->count[is_sync] == 0))
1403 rl->starved[is_sync] = 1;
1404 return ERR_PTR(-ENOMEM);
1408 * get_request - get a free request
1409 * @q: request_queue to allocate request from
1410 * @op: operation and flags
1411 * @bio: bio to allocate request for (can be %NULL)
1412 * @flags: BLK_MQ_REQ_* flags.
1414 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1415 * this function keeps retrying under memory pressure and fails iff @q is dead.
1417 * Must be called with @q->queue_lock held and,
1418 * Returns ERR_PTR on failure, with @q->queue_lock held.
1419 * Returns request pointer on success, with @q->queue_lock *not held*.
1421 static struct request *get_request(struct request_queue *q, unsigned int op,
1422 struct bio *bio, blk_mq_req_flags_t flags)
1424 const bool is_sync = op_is_sync(op);
1426 struct request_list *rl;
1429 lockdep_assert_held(q->queue_lock);
1430 WARN_ON_ONCE(q->mq_ops);
1432 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1434 rq = __get_request(rl, op, bio, flags);
1438 if (op & REQ_NOWAIT) {
1440 return ERR_PTR(-EAGAIN);
1443 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1448 /* wait on @rl and retry */
1449 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1450 TASK_UNINTERRUPTIBLE);
1452 trace_block_sleeprq(q, bio, op);
1454 spin_unlock_irq(q->queue_lock);
1458 * After sleeping, we become a "batching" process and will be able
1459 * to allocate at least one request, and up to a big batch of them
1460 * for a small period time. See ioc_batching, ioc_set_batching
1462 ioc_set_batching(q, current->io_context);
1464 spin_lock_irq(q->queue_lock);
1465 finish_wait(&rl->wait[is_sync], &wait);
1470 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1471 static struct request *blk_old_get_request(struct request_queue *q,
1472 unsigned int op, blk_mq_req_flags_t flags)
1475 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1476 __GFP_DIRECT_RECLAIM;
1479 WARN_ON_ONCE(q->mq_ops);
1481 /* create ioc upfront */
1482 create_io_context(gfp_mask, q->node);
1484 ret = blk_queue_enter(q, flags);
1486 return ERR_PTR(ret);
1487 spin_lock_irq(q->queue_lock);
1488 rq = get_request(q, op, NULL, flags);
1490 spin_unlock_irq(q->queue_lock);
1495 /* q->queue_lock is unlocked at this point */
1497 rq->__sector = (sector_t) -1;
1498 rq->bio = rq->biotail = NULL;
1503 * blk_get_request_flags - allocate a request
1504 * @q: request queue to allocate a request for
1505 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1506 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1508 struct request *blk_get_request_flags(struct request_queue *q, unsigned int op,
1509 blk_mq_req_flags_t flags)
1511 struct request *req;
1513 WARN_ON_ONCE(op & REQ_NOWAIT);
1514 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1517 req = blk_mq_alloc_request(q, op, flags);
1518 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1519 q->mq_ops->initialize_rq_fn(req);
1521 req = blk_old_get_request(q, op, flags);
1522 if (!IS_ERR(req) && q->initialize_rq_fn)
1523 q->initialize_rq_fn(req);
1528 EXPORT_SYMBOL(blk_get_request_flags);
1530 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1533 return blk_get_request_flags(q, op, gfp_mask & __GFP_DIRECT_RECLAIM ?
1534 0 : BLK_MQ_REQ_NOWAIT);
1536 EXPORT_SYMBOL(blk_get_request);
1539 * blk_requeue_request - put a request back on queue
1540 * @q: request queue where request should be inserted
1541 * @rq: request to be inserted
1544 * Drivers often keep queueing requests until the hardware cannot accept
1545 * more, when that condition happens we need to put the request back
1546 * on the queue. Must be called with queue lock held.
1548 void blk_requeue_request(struct request_queue *q, struct request *rq)
1550 lockdep_assert_held(q->queue_lock);
1551 WARN_ON_ONCE(q->mq_ops);
1553 blk_delete_timer(rq);
1554 blk_clear_rq_complete(rq);
1555 trace_block_rq_requeue(q, rq);
1556 wbt_requeue(q->rq_wb, &rq->issue_stat);
1558 if (rq->rq_flags & RQF_QUEUED)
1559 blk_queue_end_tag(q, rq);
1561 BUG_ON(blk_queued_rq(rq));
1563 elv_requeue_request(q, rq);
1565 EXPORT_SYMBOL(blk_requeue_request);
1567 static void add_acct_request(struct request_queue *q, struct request *rq,
1570 blk_account_io_start(rq, true);
1571 __elv_add_request(q, rq, where);
1574 static void part_round_stats_single(struct request_queue *q, int cpu,
1575 struct hd_struct *part, unsigned long now,
1576 unsigned int inflight)
1579 __part_stat_add(cpu, part, time_in_queue,
1580 inflight * (now - part->stamp));
1581 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1587 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1588 * @q: target block queue
1589 * @cpu: cpu number for stats access
1590 * @part: target partition
1592 * The average IO queue length and utilisation statistics are maintained
1593 * by observing the current state of the queue length and the amount of
1594 * time it has been in this state for.
1596 * Normally, that accounting is done on IO completion, but that can result
1597 * in more than a second's worth of IO being accounted for within any one
1598 * second, leading to >100% utilisation. To deal with that, we call this
1599 * function to do a round-off before returning the results when reading
1600 * /proc/diskstats. This accounts immediately for all queue usage up to
1601 * the current jiffies and restarts the counters again.
1603 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1605 struct hd_struct *part2 = NULL;
1606 unsigned long now = jiffies;
1607 unsigned int inflight[2];
1610 if (part->stamp != now)
1614 part2 = &part_to_disk(part)->part0;
1615 if (part2->stamp != now)
1622 part_in_flight(q, part, inflight);
1625 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1627 part_round_stats_single(q, cpu, part, now, inflight[0]);
1629 EXPORT_SYMBOL_GPL(part_round_stats);
1632 static void blk_pm_put_request(struct request *rq)
1634 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1635 pm_runtime_mark_last_busy(rq->q->dev);
1638 static inline void blk_pm_put_request(struct request *rq) {}
1641 void __blk_put_request(struct request_queue *q, struct request *req)
1643 req_flags_t rq_flags = req->rq_flags;
1649 blk_mq_free_request(req);
1653 lockdep_assert_held(q->queue_lock);
1655 blk_req_zone_write_unlock(req);
1656 blk_pm_put_request(req);
1658 elv_completed_request(q, req);
1660 /* this is a bio leak */
1661 WARN_ON(req->bio != NULL);
1663 wbt_done(q->rq_wb, &req->issue_stat);
1666 * Request may not have originated from ll_rw_blk. if not,
1667 * it didn't come out of our reserved rq pools
1669 if (rq_flags & RQF_ALLOCED) {
1670 struct request_list *rl = blk_rq_rl(req);
1671 bool sync = op_is_sync(req->cmd_flags);
1673 BUG_ON(!list_empty(&req->queuelist));
1674 BUG_ON(ELV_ON_HASH(req));
1676 blk_free_request(rl, req);
1677 freed_request(rl, sync, rq_flags);
1682 EXPORT_SYMBOL_GPL(__blk_put_request);
1684 void blk_put_request(struct request *req)
1686 struct request_queue *q = req->q;
1689 blk_mq_free_request(req);
1691 unsigned long flags;
1693 spin_lock_irqsave(q->queue_lock, flags);
1694 __blk_put_request(q, req);
1695 spin_unlock_irqrestore(q->queue_lock, flags);
1698 EXPORT_SYMBOL(blk_put_request);
1700 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1703 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1705 if (!ll_back_merge_fn(q, req, bio))
1708 trace_block_bio_backmerge(q, req, bio);
1710 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1711 blk_rq_set_mixed_merge(req);
1713 req->biotail->bi_next = bio;
1715 req->__data_len += bio->bi_iter.bi_size;
1716 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1718 blk_account_io_start(req, false);
1722 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1725 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1727 if (!ll_front_merge_fn(q, req, bio))
1730 trace_block_bio_frontmerge(q, req, bio);
1732 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1733 blk_rq_set_mixed_merge(req);
1735 bio->bi_next = req->bio;
1738 req->__sector = bio->bi_iter.bi_sector;
1739 req->__data_len += bio->bi_iter.bi_size;
1740 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1742 blk_account_io_start(req, false);
1746 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1749 unsigned short segments = blk_rq_nr_discard_segments(req);
1751 if (segments >= queue_max_discard_segments(q))
1753 if (blk_rq_sectors(req) + bio_sectors(bio) >
1754 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1757 req->biotail->bi_next = bio;
1759 req->__data_len += bio->bi_iter.bi_size;
1760 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1761 req->nr_phys_segments = segments + 1;
1763 blk_account_io_start(req, false);
1766 req_set_nomerge(q, req);
1771 * blk_attempt_plug_merge - try to merge with %current's plugged list
1772 * @q: request_queue new bio is being queued at
1773 * @bio: new bio being queued
1774 * @request_count: out parameter for number of traversed plugged requests
1775 * @same_queue_rq: pointer to &struct request that gets filled in when
1776 * another request associated with @q is found on the plug list
1777 * (optional, may be %NULL)
1779 * Determine whether @bio being queued on @q can be merged with a request
1780 * on %current's plugged list. Returns %true if merge was successful,
1783 * Plugging coalesces IOs from the same issuer for the same purpose without
1784 * going through @q->queue_lock. As such it's more of an issuing mechanism
1785 * than scheduling, and the request, while may have elvpriv data, is not
1786 * added on the elevator at this point. In addition, we don't have
1787 * reliable access to the elevator outside queue lock. Only check basic
1788 * merging parameters without querying the elevator.
1790 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1792 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1793 unsigned int *request_count,
1794 struct request **same_queue_rq)
1796 struct blk_plug *plug;
1798 struct list_head *plug_list;
1800 plug = current->plug;
1806 plug_list = &plug->mq_list;
1808 plug_list = &plug->list;
1810 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1811 bool merged = false;
1816 * Only blk-mq multiple hardware queues case checks the
1817 * rq in the same queue, there should be only one such
1821 *same_queue_rq = rq;
1824 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1827 switch (blk_try_merge(rq, bio)) {
1828 case ELEVATOR_BACK_MERGE:
1829 merged = bio_attempt_back_merge(q, rq, bio);
1831 case ELEVATOR_FRONT_MERGE:
1832 merged = bio_attempt_front_merge(q, rq, bio);
1834 case ELEVATOR_DISCARD_MERGE:
1835 merged = bio_attempt_discard_merge(q, rq, bio);
1848 unsigned int blk_plug_queued_count(struct request_queue *q)
1850 struct blk_plug *plug;
1852 struct list_head *plug_list;
1853 unsigned int ret = 0;
1855 plug = current->plug;
1860 plug_list = &plug->mq_list;
1862 plug_list = &plug->list;
1864 list_for_each_entry(rq, plug_list, queuelist) {
1872 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1874 struct io_context *ioc = rq_ioc(bio);
1876 if (bio->bi_opf & REQ_RAHEAD)
1877 req->cmd_flags |= REQ_FAILFAST_MASK;
1879 req->__sector = bio->bi_iter.bi_sector;
1880 if (ioprio_valid(bio_prio(bio)))
1881 req->ioprio = bio_prio(bio);
1883 req->ioprio = ioc->ioprio;
1885 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1886 req->write_hint = bio->bi_write_hint;
1887 blk_rq_bio_prep(req->q, req, bio);
1889 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1891 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1893 struct blk_plug *plug;
1894 int where = ELEVATOR_INSERT_SORT;
1895 struct request *req, *free;
1896 unsigned int request_count = 0;
1897 unsigned int wb_acct;
1900 * low level driver can indicate that it wants pages above a
1901 * certain limit bounced to low memory (ie for highmem, or even
1902 * ISA dma in theory)
1904 blk_queue_bounce(q, &bio);
1906 blk_queue_split(q, &bio);
1908 if (!bio_integrity_prep(bio))
1909 return BLK_QC_T_NONE;
1911 if (op_is_flush(bio->bi_opf)) {
1912 spin_lock_irq(q->queue_lock);
1913 where = ELEVATOR_INSERT_FLUSH;
1918 * Check if we can merge with the plugged list before grabbing
1921 if (!blk_queue_nomerges(q)) {
1922 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1923 return BLK_QC_T_NONE;
1925 request_count = blk_plug_queued_count(q);
1927 spin_lock_irq(q->queue_lock);
1929 switch (elv_merge(q, &req, bio)) {
1930 case ELEVATOR_BACK_MERGE:
1931 if (!bio_attempt_back_merge(q, req, bio))
1933 elv_bio_merged(q, req, bio);
1934 free = attempt_back_merge(q, req);
1936 __blk_put_request(q, free);
1938 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1940 case ELEVATOR_FRONT_MERGE:
1941 if (!bio_attempt_front_merge(q, req, bio))
1943 elv_bio_merged(q, req, bio);
1944 free = attempt_front_merge(q, req);
1946 __blk_put_request(q, free);
1948 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1955 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1958 * Grab a free request. This is might sleep but can not fail.
1959 * Returns with the queue unlocked.
1961 blk_queue_enter_live(q);
1962 req = get_request(q, bio->bi_opf, bio, 0);
1965 __wbt_done(q->rq_wb, wb_acct);
1966 if (PTR_ERR(req) == -ENOMEM)
1967 bio->bi_status = BLK_STS_RESOURCE;
1969 bio->bi_status = BLK_STS_IOERR;
1974 wbt_track(&req->issue_stat, wb_acct);
1977 * After dropping the lock and possibly sleeping here, our request
1978 * may now be mergeable after it had proven unmergeable (above).
1979 * We don't worry about that case for efficiency. It won't happen
1980 * often, and the elevators are able to handle it.
1982 blk_init_request_from_bio(req, bio);
1984 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1985 req->cpu = raw_smp_processor_id();
1987 plug = current->plug;
1990 * If this is the first request added after a plug, fire
1993 * @request_count may become stale because of schedule
1994 * out, so check plug list again.
1996 if (!request_count || list_empty(&plug->list))
1997 trace_block_plug(q);
1999 struct request *last = list_entry_rq(plug->list.prev);
2000 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2001 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2002 blk_flush_plug_list(plug, false);
2003 trace_block_plug(q);
2006 list_add_tail(&req->queuelist, &plug->list);
2007 blk_account_io_start(req, true);
2009 spin_lock_irq(q->queue_lock);
2010 add_acct_request(q, req, where);
2013 spin_unlock_irq(q->queue_lock);
2016 return BLK_QC_T_NONE;
2019 static void handle_bad_sector(struct bio *bio)
2021 char b[BDEVNAME_SIZE];
2023 printk(KERN_INFO "attempt to access beyond end of device\n");
2024 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2025 bio_devname(bio, b), bio->bi_opf,
2026 (unsigned long long)bio_end_sector(bio),
2027 (long long)get_capacity(bio->bi_disk));
2030 #ifdef CONFIG_FAIL_MAKE_REQUEST
2032 static DECLARE_FAULT_ATTR(fail_make_request);
2034 static int __init setup_fail_make_request(char *str)
2036 return setup_fault_attr(&fail_make_request, str);
2038 __setup("fail_make_request=", setup_fail_make_request);
2040 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2042 return part->make_it_fail && should_fail(&fail_make_request, bytes);
2045 static int __init fail_make_request_debugfs(void)
2047 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2048 NULL, &fail_make_request);
2050 return PTR_ERR_OR_ZERO(dir);
2053 late_initcall(fail_make_request_debugfs);
2055 #else /* CONFIG_FAIL_MAKE_REQUEST */
2057 static inline bool should_fail_request(struct hd_struct *part,
2063 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2066 * Remap block n of partition p to block n+start(p) of the disk.
2068 static inline int blk_partition_remap(struct bio *bio)
2070 struct hd_struct *p;
2074 * Zone reset does not include bi_size so bio_sectors() is always 0.
2075 * Include a test for the reset op code and perform the remap if needed.
2077 if (!bio->bi_partno ||
2078 (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET))
2082 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2083 if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) {
2084 bio->bi_iter.bi_sector += p->start_sect;
2086 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2087 bio->bi_iter.bi_sector - p->start_sect);
2089 printk("%s: fail for partition %d\n", __func__, bio->bi_partno);
2098 * Check whether this bio extends beyond the end of the device.
2100 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
2107 /* Test device or partition size, when known. */
2108 maxsector = get_capacity(bio->bi_disk);
2110 sector_t sector = bio->bi_iter.bi_sector;
2112 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2114 * This may well happen - the kernel calls bread()
2115 * without checking the size of the device, e.g., when
2116 * mounting a device.
2118 handle_bad_sector(bio);
2126 static noinline_for_stack bool
2127 generic_make_request_checks(struct bio *bio)
2129 struct request_queue *q;
2130 int nr_sectors = bio_sectors(bio);
2131 blk_status_t status = BLK_STS_IOERR;
2132 char b[BDEVNAME_SIZE];
2136 if (bio_check_eod(bio, nr_sectors))
2139 q = bio->bi_disk->queue;
2142 "generic_make_request: Trying to access "
2143 "nonexistent block-device %s (%Lu)\n",
2144 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2149 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2150 * if queue is not a request based queue.
2153 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2156 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2159 if (blk_partition_remap(bio))
2162 if (bio_check_eod(bio, nr_sectors))
2166 * Filter flush bio's early so that make_request based
2167 * drivers without flush support don't have to worry
2170 if (op_is_flush(bio->bi_opf) &&
2171 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2172 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2174 status = BLK_STS_OK;
2179 switch (bio_op(bio)) {
2180 case REQ_OP_DISCARD:
2181 if (!blk_queue_discard(q))
2184 case REQ_OP_SECURE_ERASE:
2185 if (!blk_queue_secure_erase(q))
2188 case REQ_OP_WRITE_SAME:
2189 if (!q->limits.max_write_same_sectors)
2192 case REQ_OP_ZONE_REPORT:
2193 case REQ_OP_ZONE_RESET:
2194 if (!blk_queue_is_zoned(q))
2197 case REQ_OP_WRITE_ZEROES:
2198 if (!q->limits.max_write_zeroes_sectors)
2206 * Various block parts want %current->io_context and lazy ioc
2207 * allocation ends up trading a lot of pain for a small amount of
2208 * memory. Just allocate it upfront. This may fail and block
2209 * layer knows how to live with it.
2211 create_io_context(GFP_ATOMIC, q->node);
2213 if (!blkcg_bio_issue_check(q, bio))
2216 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2217 trace_block_bio_queue(q, bio);
2218 /* Now that enqueuing has been traced, we need to trace
2219 * completion as well.
2221 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2226 status = BLK_STS_NOTSUPP;
2228 bio->bi_status = status;
2234 * generic_make_request - hand a buffer to its device driver for I/O
2235 * @bio: The bio describing the location in memory and on the device.
2237 * generic_make_request() is used to make I/O requests of block
2238 * devices. It is passed a &struct bio, which describes the I/O that needs
2241 * generic_make_request() does not return any status. The
2242 * success/failure status of the request, along with notification of
2243 * completion, is delivered asynchronously through the bio->bi_end_io
2244 * function described (one day) else where.
2246 * The caller of generic_make_request must make sure that bi_io_vec
2247 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2248 * set to describe the device address, and the
2249 * bi_end_io and optionally bi_private are set to describe how
2250 * completion notification should be signaled.
2252 * generic_make_request and the drivers it calls may use bi_next if this
2253 * bio happens to be merged with someone else, and may resubmit the bio to
2254 * a lower device by calling into generic_make_request recursively, which
2255 * means the bio should NOT be touched after the call to ->make_request_fn.
2257 blk_qc_t generic_make_request(struct bio *bio)
2260 * bio_list_on_stack[0] contains bios submitted by the current
2262 * bio_list_on_stack[1] contains bios that were submitted before
2263 * the current make_request_fn, but that haven't been processed
2266 struct bio_list bio_list_on_stack[2];
2267 blk_qc_t ret = BLK_QC_T_NONE;
2269 if (!generic_make_request_checks(bio))
2273 * We only want one ->make_request_fn to be active at a time, else
2274 * stack usage with stacked devices could be a problem. So use
2275 * current->bio_list to keep a list of requests submited by a
2276 * make_request_fn function. current->bio_list is also used as a
2277 * flag to say if generic_make_request is currently active in this
2278 * task or not. If it is NULL, then no make_request is active. If
2279 * it is non-NULL, then a make_request is active, and new requests
2280 * should be added at the tail
2282 if (current->bio_list) {
2283 bio_list_add(¤t->bio_list[0], bio);
2287 /* following loop may be a bit non-obvious, and so deserves some
2289 * Before entering the loop, bio->bi_next is NULL (as all callers
2290 * ensure that) so we have a list with a single bio.
2291 * We pretend that we have just taken it off a longer list, so
2292 * we assign bio_list to a pointer to the bio_list_on_stack,
2293 * thus initialising the bio_list of new bios to be
2294 * added. ->make_request() may indeed add some more bios
2295 * through a recursive call to generic_make_request. If it
2296 * did, we find a non-NULL value in bio_list and re-enter the loop
2297 * from the top. In this case we really did just take the bio
2298 * of the top of the list (no pretending) and so remove it from
2299 * bio_list, and call into ->make_request() again.
2301 BUG_ON(bio->bi_next);
2302 bio_list_init(&bio_list_on_stack[0]);
2303 current->bio_list = bio_list_on_stack;
2305 struct request_queue *q = bio->bi_disk->queue;
2306 blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
2307 BLK_MQ_REQ_NOWAIT : 0;
2309 if (likely(blk_queue_enter(q, flags) == 0)) {
2310 struct bio_list lower, same;
2312 /* Create a fresh bio_list for all subordinate requests */
2313 bio_list_on_stack[1] = bio_list_on_stack[0];
2314 bio_list_init(&bio_list_on_stack[0]);
2315 ret = q->make_request_fn(q, bio);
2319 /* sort new bios into those for a lower level
2320 * and those for the same level
2322 bio_list_init(&lower);
2323 bio_list_init(&same);
2324 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2325 if (q == bio->bi_disk->queue)
2326 bio_list_add(&same, bio);
2328 bio_list_add(&lower, bio);
2329 /* now assemble so we handle the lowest level first */
2330 bio_list_merge(&bio_list_on_stack[0], &lower);
2331 bio_list_merge(&bio_list_on_stack[0], &same);
2332 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2334 if (unlikely(!blk_queue_dying(q) &&
2335 (bio->bi_opf & REQ_NOWAIT)))
2336 bio_wouldblock_error(bio);
2340 bio = bio_list_pop(&bio_list_on_stack[0]);
2342 current->bio_list = NULL; /* deactivate */
2347 EXPORT_SYMBOL(generic_make_request);
2350 * direct_make_request - hand a buffer directly to its device driver for I/O
2351 * @bio: The bio describing the location in memory and on the device.
2353 * This function behaves like generic_make_request(), but does not protect
2354 * against recursion. Must only be used if the called driver is known
2355 * to not call generic_make_request (or direct_make_request) again from
2356 * its make_request function. (Calling direct_make_request again from
2357 * a workqueue is perfectly fine as that doesn't recurse).
2359 blk_qc_t direct_make_request(struct bio *bio)
2361 struct request_queue *q = bio->bi_disk->queue;
2362 bool nowait = bio->bi_opf & REQ_NOWAIT;
2365 if (!generic_make_request_checks(bio))
2366 return BLK_QC_T_NONE;
2368 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2369 if (nowait && !blk_queue_dying(q))
2370 bio->bi_status = BLK_STS_AGAIN;
2372 bio->bi_status = BLK_STS_IOERR;
2374 return BLK_QC_T_NONE;
2377 ret = q->make_request_fn(q, bio);
2381 EXPORT_SYMBOL_GPL(direct_make_request);
2384 * submit_bio - submit a bio to the block device layer for I/O
2385 * @bio: The &struct bio which describes the I/O
2387 * submit_bio() is very similar in purpose to generic_make_request(), and
2388 * uses that function to do most of the work. Both are fairly rough
2389 * interfaces; @bio must be presetup and ready for I/O.
2392 blk_qc_t submit_bio(struct bio *bio)
2395 * If it's a regular read/write or a barrier with data attached,
2396 * go through the normal accounting stuff before submission.
2398 if (bio_has_data(bio)) {
2401 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2402 count = queue_logical_block_size(bio->bi_disk->queue);
2404 count = bio_sectors(bio);
2406 if (op_is_write(bio_op(bio))) {
2407 count_vm_events(PGPGOUT, count);
2409 task_io_account_read(bio->bi_iter.bi_size);
2410 count_vm_events(PGPGIN, count);
2413 if (unlikely(block_dump)) {
2414 char b[BDEVNAME_SIZE];
2415 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2416 current->comm, task_pid_nr(current),
2417 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2418 (unsigned long long)bio->bi_iter.bi_sector,
2419 bio_devname(bio, b), count);
2423 return generic_make_request(bio);
2425 EXPORT_SYMBOL(submit_bio);
2427 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2429 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2433 blk_flush_plug_list(current->plug, false);
2434 return q->poll_fn(q, cookie);
2436 EXPORT_SYMBOL_GPL(blk_poll);
2439 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2440 * for new the queue limits
2442 * @rq: the request being checked
2445 * @rq may have been made based on weaker limitations of upper-level queues
2446 * in request stacking drivers, and it may violate the limitation of @q.
2447 * Since the block layer and the underlying device driver trust @rq
2448 * after it is inserted to @q, it should be checked against @q before
2449 * the insertion using this generic function.
2451 * Request stacking drivers like request-based dm may change the queue
2452 * limits when retrying requests on other queues. Those requests need
2453 * to be checked against the new queue limits again during dispatch.
2455 static int blk_cloned_rq_check_limits(struct request_queue *q,
2458 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2459 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2464 * queue's settings related to segment counting like q->bounce_pfn
2465 * may differ from that of other stacking queues.
2466 * Recalculate it to check the request correctly on this queue's
2469 blk_recalc_rq_segments(rq);
2470 if (rq->nr_phys_segments > queue_max_segments(q)) {
2471 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2479 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2480 * @q: the queue to submit the request
2481 * @rq: the request being queued
2483 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2485 unsigned long flags;
2486 int where = ELEVATOR_INSERT_BACK;
2488 if (blk_cloned_rq_check_limits(q, rq))
2489 return BLK_STS_IOERR;
2492 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2493 return BLK_STS_IOERR;
2496 if (blk_queue_io_stat(q))
2497 blk_account_io_start(rq, true);
2499 * Since we have a scheduler attached on the top device,
2500 * bypass a potential scheduler on the bottom device for
2503 blk_mq_request_bypass_insert(rq, true);
2507 spin_lock_irqsave(q->queue_lock, flags);
2508 if (unlikely(blk_queue_dying(q))) {
2509 spin_unlock_irqrestore(q->queue_lock, flags);
2510 return BLK_STS_IOERR;
2514 * Submitting request must be dequeued before calling this function
2515 * because it will be linked to another request_queue
2517 BUG_ON(blk_queued_rq(rq));
2519 if (op_is_flush(rq->cmd_flags))
2520 where = ELEVATOR_INSERT_FLUSH;
2522 add_acct_request(q, rq, where);
2523 if (where == ELEVATOR_INSERT_FLUSH)
2525 spin_unlock_irqrestore(q->queue_lock, flags);
2529 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2532 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2533 * @rq: request to examine
2536 * A request could be merge of IOs which require different failure
2537 * handling. This function determines the number of bytes which
2538 * can be failed from the beginning of the request without
2539 * crossing into area which need to be retried further.
2542 * The number of bytes to fail.
2544 unsigned int blk_rq_err_bytes(const struct request *rq)
2546 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2547 unsigned int bytes = 0;
2550 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2551 return blk_rq_bytes(rq);
2554 * Currently the only 'mixing' which can happen is between
2555 * different fastfail types. We can safely fail portions
2556 * which have all the failfast bits that the first one has -
2557 * the ones which are at least as eager to fail as the first
2560 for (bio = rq->bio; bio; bio = bio->bi_next) {
2561 if ((bio->bi_opf & ff) != ff)
2563 bytes += bio->bi_iter.bi_size;
2566 /* this could lead to infinite loop */
2567 BUG_ON(blk_rq_bytes(rq) && !bytes);
2570 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2572 void blk_account_io_completion(struct request *req, unsigned int bytes)
2574 if (blk_do_io_stat(req)) {
2575 const int rw = rq_data_dir(req);
2576 struct hd_struct *part;
2579 cpu = part_stat_lock();
2581 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2586 void blk_account_io_done(struct request *req)
2589 * Account IO completion. flush_rq isn't accounted as a
2590 * normal IO on queueing nor completion. Accounting the
2591 * containing request is enough.
2593 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2594 unsigned long duration = jiffies - req->start_time;
2595 const int rw = rq_data_dir(req);
2596 struct hd_struct *part;
2599 cpu = part_stat_lock();
2602 part_stat_inc(cpu, part, ios[rw]);
2603 part_stat_add(cpu, part, ticks[rw], duration);
2604 part_round_stats(req->q, cpu, part);
2605 part_dec_in_flight(req->q, part, rw);
2607 hd_struct_put(part);
2614 * Don't process normal requests when queue is suspended
2615 * or in the process of suspending/resuming
2617 static bool blk_pm_allow_request(struct request *rq)
2619 switch (rq->q->rpm_status) {
2621 case RPM_SUSPENDING:
2622 return rq->rq_flags & RQF_PM;
2630 static bool blk_pm_allow_request(struct request *rq)
2636 void blk_account_io_start(struct request *rq, bool new_io)
2638 struct hd_struct *part;
2639 int rw = rq_data_dir(rq);
2642 if (!blk_do_io_stat(rq))
2645 cpu = part_stat_lock();
2649 part_stat_inc(cpu, part, merges[rw]);
2651 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2652 if (!hd_struct_try_get(part)) {
2654 * The partition is already being removed,
2655 * the request will be accounted on the disk only
2657 * We take a reference on disk->part0 although that
2658 * partition will never be deleted, so we can treat
2659 * it as any other partition.
2661 part = &rq->rq_disk->part0;
2662 hd_struct_get(part);
2664 part_round_stats(rq->q, cpu, part);
2665 part_inc_in_flight(rq->q, part, rw);
2672 static struct request *elv_next_request(struct request_queue *q)
2675 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2677 WARN_ON_ONCE(q->mq_ops);
2680 list_for_each_entry(rq, &q->queue_head, queuelist) {
2681 if (blk_pm_allow_request(rq))
2684 if (rq->rq_flags & RQF_SOFTBARRIER)
2689 * Flush request is running and flush request isn't queueable
2690 * in the drive, we can hold the queue till flush request is
2691 * finished. Even we don't do this, driver can't dispatch next
2692 * requests and will requeue them. And this can improve
2693 * throughput too. For example, we have request flush1, write1,
2694 * flush 2. flush1 is dispatched, then queue is hold, write1
2695 * isn't inserted to queue. After flush1 is finished, flush2
2696 * will be dispatched. Since disk cache is already clean,
2697 * flush2 will be finished very soon, so looks like flush2 is
2699 * Since the queue is hold, a flag is set to indicate the queue
2700 * should be restarted later. Please see flush_end_io() for
2703 if (fq->flush_pending_idx != fq->flush_running_idx &&
2704 !queue_flush_queueable(q)) {
2705 fq->flush_queue_delayed = 1;
2708 if (unlikely(blk_queue_bypass(q)) ||
2709 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2715 * blk_peek_request - peek at the top of a request queue
2716 * @q: request queue to peek at
2719 * Return the request at the top of @q. The returned request
2720 * should be started using blk_start_request() before LLD starts
2724 * Pointer to the request at the top of @q if available. Null
2727 struct request *blk_peek_request(struct request_queue *q)
2732 lockdep_assert_held(q->queue_lock);
2733 WARN_ON_ONCE(q->mq_ops);
2735 while ((rq = elv_next_request(q)) != NULL) {
2736 if (!(rq->rq_flags & RQF_STARTED)) {
2738 * This is the first time the device driver
2739 * sees this request (possibly after
2740 * requeueing). Notify IO scheduler.
2742 if (rq->rq_flags & RQF_SORTED)
2743 elv_activate_rq(q, rq);
2746 * just mark as started even if we don't start
2747 * it, a request that has been delayed should
2748 * not be passed by new incoming requests
2750 rq->rq_flags |= RQF_STARTED;
2751 trace_block_rq_issue(q, rq);
2754 if (!q->boundary_rq || q->boundary_rq == rq) {
2755 q->end_sector = rq_end_sector(rq);
2756 q->boundary_rq = NULL;
2759 if (rq->rq_flags & RQF_DONTPREP)
2762 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2764 * make sure space for the drain appears we
2765 * know we can do this because max_hw_segments
2766 * has been adjusted to be one fewer than the
2769 rq->nr_phys_segments++;
2775 ret = q->prep_rq_fn(q, rq);
2776 if (ret == BLKPREP_OK) {
2778 } else if (ret == BLKPREP_DEFER) {
2780 * the request may have been (partially) prepped.
2781 * we need to keep this request in the front to
2782 * avoid resource deadlock. RQF_STARTED will
2783 * prevent other fs requests from passing this one.
2785 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2786 !(rq->rq_flags & RQF_DONTPREP)) {
2788 * remove the space for the drain we added
2789 * so that we don't add it again
2791 --rq->nr_phys_segments;
2796 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2797 rq->rq_flags |= RQF_QUIET;
2799 * Mark this request as started so we don't trigger
2800 * any debug logic in the end I/O path.
2802 blk_start_request(rq);
2803 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2804 BLK_STS_TARGET : BLK_STS_IOERR);
2806 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2813 EXPORT_SYMBOL(blk_peek_request);
2815 static void blk_dequeue_request(struct request *rq)
2817 struct request_queue *q = rq->q;
2819 BUG_ON(list_empty(&rq->queuelist));
2820 BUG_ON(ELV_ON_HASH(rq));
2822 list_del_init(&rq->queuelist);
2825 * the time frame between a request being removed from the lists
2826 * and to it is freed is accounted as io that is in progress at
2829 if (blk_account_rq(rq)) {
2830 q->in_flight[rq_is_sync(rq)]++;
2831 set_io_start_time_ns(rq);
2836 * blk_start_request - start request processing on the driver
2837 * @req: request to dequeue
2840 * Dequeue @req and start timeout timer on it. This hands off the
2841 * request to the driver.
2843 void blk_start_request(struct request *req)
2845 lockdep_assert_held(req->q->queue_lock);
2846 WARN_ON_ONCE(req->q->mq_ops);
2848 blk_dequeue_request(req);
2850 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2851 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2852 req->rq_flags |= RQF_STATS;
2853 wbt_issue(req->q->rq_wb, &req->issue_stat);
2856 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2859 EXPORT_SYMBOL(blk_start_request);
2862 * blk_fetch_request - fetch a request from a request queue
2863 * @q: request queue to fetch a request from
2866 * Return the request at the top of @q. The request is started on
2867 * return and LLD can start processing it immediately.
2870 * Pointer to the request at the top of @q if available. Null
2873 struct request *blk_fetch_request(struct request_queue *q)
2877 lockdep_assert_held(q->queue_lock);
2878 WARN_ON_ONCE(q->mq_ops);
2880 rq = blk_peek_request(q);
2882 blk_start_request(rq);
2885 EXPORT_SYMBOL(blk_fetch_request);
2888 * Steal bios from a request and add them to a bio list.
2889 * The request must not have been partially completed before.
2891 void blk_steal_bios(struct bio_list *list, struct request *rq)
2895 list->tail->bi_next = rq->bio;
2897 list->head = rq->bio;
2898 list->tail = rq->biotail;
2906 EXPORT_SYMBOL_GPL(blk_steal_bios);
2909 * blk_update_request - Special helper function for request stacking drivers
2910 * @req: the request being processed
2911 * @error: block status code
2912 * @nr_bytes: number of bytes to complete @req
2915 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2916 * the request structure even if @req doesn't have leftover.
2917 * If @req has leftover, sets it up for the next range of segments.
2919 * This special helper function is only for request stacking drivers
2920 * (e.g. request-based dm) so that they can handle partial completion.
2921 * Actual device drivers should use blk_end_request instead.
2923 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2924 * %false return from this function.
2927 * %false - this request doesn't have any more data
2928 * %true - this request has more data
2930 bool blk_update_request(struct request *req, blk_status_t error,
2931 unsigned int nr_bytes)
2935 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
2940 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2941 !(req->rq_flags & RQF_QUIET)))
2942 print_req_error(req, error);
2944 blk_account_io_completion(req, nr_bytes);
2948 struct bio *bio = req->bio;
2949 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2951 if (bio_bytes == bio->bi_iter.bi_size)
2952 req->bio = bio->bi_next;
2954 /* Completion has already been traced */
2955 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
2956 req_bio_endio(req, bio, bio_bytes, error);
2958 total_bytes += bio_bytes;
2959 nr_bytes -= bio_bytes;
2970 * Reset counters so that the request stacking driver
2971 * can find how many bytes remain in the request
2974 req->__data_len = 0;
2978 req->__data_len -= total_bytes;
2980 /* update sector only for requests with clear definition of sector */
2981 if (!blk_rq_is_passthrough(req))
2982 req->__sector += total_bytes >> 9;
2984 /* mixed attributes always follow the first bio */
2985 if (req->rq_flags & RQF_MIXED_MERGE) {
2986 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2987 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2990 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
2992 * If total number of sectors is less than the first segment
2993 * size, something has gone terribly wrong.
2995 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2996 blk_dump_rq_flags(req, "request botched");
2997 req->__data_len = blk_rq_cur_bytes(req);
3000 /* recalculate the number of segments */
3001 blk_recalc_rq_segments(req);
3006 EXPORT_SYMBOL_GPL(blk_update_request);
3008 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3009 unsigned int nr_bytes,
3010 unsigned int bidi_bytes)
3012 if (blk_update_request(rq, error, nr_bytes))
3015 /* Bidi request must be completed as a whole */
3016 if (unlikely(blk_bidi_rq(rq)) &&
3017 blk_update_request(rq->next_rq, error, bidi_bytes))
3020 if (blk_queue_add_random(rq->q))
3021 add_disk_randomness(rq->rq_disk);
3027 * blk_unprep_request - unprepare a request
3030 * This function makes a request ready for complete resubmission (or
3031 * completion). It happens only after all error handling is complete,
3032 * so represents the appropriate moment to deallocate any resources
3033 * that were allocated to the request in the prep_rq_fn. The queue
3034 * lock is held when calling this.
3036 void blk_unprep_request(struct request *req)
3038 struct request_queue *q = req->q;
3040 req->rq_flags &= ~RQF_DONTPREP;
3041 if (q->unprep_rq_fn)
3042 q->unprep_rq_fn(q, req);
3044 EXPORT_SYMBOL_GPL(blk_unprep_request);
3046 void blk_finish_request(struct request *req, blk_status_t error)
3048 struct request_queue *q = req->q;
3050 lockdep_assert_held(req->q->queue_lock);
3051 WARN_ON_ONCE(q->mq_ops);
3053 if (req->rq_flags & RQF_STATS)
3056 if (req->rq_flags & RQF_QUEUED)
3057 blk_queue_end_tag(q, req);
3059 BUG_ON(blk_queued_rq(req));
3061 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3062 laptop_io_completion(req->q->backing_dev_info);
3064 blk_delete_timer(req);
3066 if (req->rq_flags & RQF_DONTPREP)
3067 blk_unprep_request(req);
3069 blk_account_io_done(req);
3072 wbt_done(req->q->rq_wb, &req->issue_stat);
3073 req->end_io(req, error);
3075 if (blk_bidi_rq(req))
3076 __blk_put_request(req->next_rq->q, req->next_rq);
3078 __blk_put_request(q, req);
3081 EXPORT_SYMBOL(blk_finish_request);
3084 * blk_end_bidi_request - Complete a bidi request
3085 * @rq: the request to complete
3086 * @error: block status code
3087 * @nr_bytes: number of bytes to complete @rq
3088 * @bidi_bytes: number of bytes to complete @rq->next_rq
3091 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3092 * Drivers that supports bidi can safely call this member for any
3093 * type of request, bidi or uni. In the later case @bidi_bytes is
3097 * %false - we are done with this request
3098 * %true - still buffers pending for this request
3100 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3101 unsigned int nr_bytes, unsigned int bidi_bytes)
3103 struct request_queue *q = rq->q;
3104 unsigned long flags;
3106 WARN_ON_ONCE(q->mq_ops);
3108 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3111 spin_lock_irqsave(q->queue_lock, flags);
3112 blk_finish_request(rq, error);
3113 spin_unlock_irqrestore(q->queue_lock, flags);
3119 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3120 * @rq: the request to complete
3121 * @error: block status code
3122 * @nr_bytes: number of bytes to complete @rq
3123 * @bidi_bytes: number of bytes to complete @rq->next_rq
3126 * Identical to blk_end_bidi_request() except that queue lock is
3127 * assumed to be locked on entry and remains so on return.
3130 * %false - we are done with this request
3131 * %true - still buffers pending for this request
3133 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3134 unsigned int nr_bytes, unsigned int bidi_bytes)
3136 lockdep_assert_held(rq->q->queue_lock);
3137 WARN_ON_ONCE(rq->q->mq_ops);
3139 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3142 blk_finish_request(rq, error);
3148 * blk_end_request - Helper function for drivers to complete the request.
3149 * @rq: the request being processed
3150 * @error: block status code
3151 * @nr_bytes: number of bytes to complete
3154 * Ends I/O on a number of bytes attached to @rq.
3155 * If @rq has leftover, sets it up for the next range of segments.
3158 * %false - we are done with this request
3159 * %true - still buffers pending for this request
3161 bool blk_end_request(struct request *rq, blk_status_t error,
3162 unsigned int nr_bytes)
3164 WARN_ON_ONCE(rq->q->mq_ops);
3165 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3167 EXPORT_SYMBOL(blk_end_request);
3170 * blk_end_request_all - Helper function for drives to finish the request.
3171 * @rq: the request to finish
3172 * @error: block status code
3175 * Completely finish @rq.
3177 void blk_end_request_all(struct request *rq, blk_status_t error)
3180 unsigned int bidi_bytes = 0;
3182 if (unlikely(blk_bidi_rq(rq)))
3183 bidi_bytes = blk_rq_bytes(rq->next_rq);
3185 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3188 EXPORT_SYMBOL(blk_end_request_all);
3191 * __blk_end_request - Helper function for drivers to complete the request.
3192 * @rq: the request being processed
3193 * @error: block status code
3194 * @nr_bytes: number of bytes to complete
3197 * Must be called with queue lock held unlike blk_end_request().
3200 * %false - we are done with this request
3201 * %true - still buffers pending for this request
3203 bool __blk_end_request(struct request *rq, blk_status_t error,
3204 unsigned int nr_bytes)
3206 lockdep_assert_held(rq->q->queue_lock);
3207 WARN_ON_ONCE(rq->q->mq_ops);
3209 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3211 EXPORT_SYMBOL(__blk_end_request);
3214 * __blk_end_request_all - Helper function for drives to finish the request.
3215 * @rq: the request to finish
3216 * @error: block status code
3219 * Completely finish @rq. Must be called with queue lock held.
3221 void __blk_end_request_all(struct request *rq, blk_status_t error)
3224 unsigned int bidi_bytes = 0;
3226 lockdep_assert_held(rq->q->queue_lock);
3227 WARN_ON_ONCE(rq->q->mq_ops);
3229 if (unlikely(blk_bidi_rq(rq)))
3230 bidi_bytes = blk_rq_bytes(rq->next_rq);
3232 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3235 EXPORT_SYMBOL(__blk_end_request_all);
3238 * __blk_end_request_cur - Helper function to finish the current request chunk.
3239 * @rq: the request to finish the current chunk for
3240 * @error: block status code
3243 * Complete the current consecutively mapped chunk from @rq. Must
3244 * be called with queue lock held.
3247 * %false - we are done with this request
3248 * %true - still buffers pending for this request
3250 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3252 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3254 EXPORT_SYMBOL(__blk_end_request_cur);
3256 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3259 if (bio_has_data(bio))
3260 rq->nr_phys_segments = bio_phys_segments(q, bio);
3262 rq->__data_len = bio->bi_iter.bi_size;
3263 rq->bio = rq->biotail = bio;
3266 rq->rq_disk = bio->bi_disk;
3269 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3271 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3272 * @rq: the request to be flushed
3275 * Flush all pages in @rq.
3277 void rq_flush_dcache_pages(struct request *rq)
3279 struct req_iterator iter;
3280 struct bio_vec bvec;
3282 rq_for_each_segment(bvec, rq, iter)
3283 flush_dcache_page(bvec.bv_page);
3285 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3289 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3290 * @q : the queue of the device being checked
3293 * Check if underlying low-level drivers of a device are busy.
3294 * If the drivers want to export their busy state, they must set own
3295 * exporting function using blk_queue_lld_busy() first.
3297 * Basically, this function is used only by request stacking drivers
3298 * to stop dispatching requests to underlying devices when underlying
3299 * devices are busy. This behavior helps more I/O merging on the queue
3300 * of the request stacking driver and prevents I/O throughput regression
3301 * on burst I/O load.
3304 * 0 - Not busy (The request stacking driver should dispatch request)
3305 * 1 - Busy (The request stacking driver should stop dispatching request)
3307 int blk_lld_busy(struct request_queue *q)
3310 return q->lld_busy_fn(q);
3314 EXPORT_SYMBOL_GPL(blk_lld_busy);
3317 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3318 * @rq: the clone request to be cleaned up
3321 * Free all bios in @rq for a cloned request.
3323 void blk_rq_unprep_clone(struct request *rq)
3327 while ((bio = rq->bio) != NULL) {
3328 rq->bio = bio->bi_next;
3333 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3336 * Copy attributes of the original request to the clone request.
3337 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3339 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3341 dst->cpu = src->cpu;
3342 dst->__sector = blk_rq_pos(src);
3343 dst->__data_len = blk_rq_bytes(src);
3344 dst->nr_phys_segments = src->nr_phys_segments;
3345 dst->ioprio = src->ioprio;
3346 dst->extra_len = src->extra_len;
3350 * blk_rq_prep_clone - Helper function to setup clone request
3351 * @rq: the request to be setup
3352 * @rq_src: original request to be cloned
3353 * @bs: bio_set that bios for clone are allocated from
3354 * @gfp_mask: memory allocation mask for bio
3355 * @bio_ctr: setup function to be called for each clone bio.
3356 * Returns %0 for success, non %0 for failure.
3357 * @data: private data to be passed to @bio_ctr
3360 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3361 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3362 * are not copied, and copying such parts is the caller's responsibility.
3363 * Also, pages which the original bios are pointing to are not copied
3364 * and the cloned bios just point same pages.
3365 * So cloned bios must be completed before original bios, which means
3366 * the caller must complete @rq before @rq_src.
3368 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3369 struct bio_set *bs, gfp_t gfp_mask,
3370 int (*bio_ctr)(struct bio *, struct bio *, void *),
3373 struct bio *bio, *bio_src;
3378 __rq_for_each_bio(bio_src, rq_src) {
3379 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3383 if (bio_ctr && bio_ctr(bio, bio_src, data))
3387 rq->biotail->bi_next = bio;
3390 rq->bio = rq->biotail = bio;
3393 __blk_rq_prep_clone(rq, rq_src);
3400 blk_rq_unprep_clone(rq);
3404 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3406 int kblockd_schedule_work(struct work_struct *work)
3408 return queue_work(kblockd_workqueue, work);
3410 EXPORT_SYMBOL(kblockd_schedule_work);
3412 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3414 return queue_work_on(cpu, kblockd_workqueue, work);
3416 EXPORT_SYMBOL(kblockd_schedule_work_on);
3418 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3419 unsigned long delay)
3421 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3423 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3425 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3426 unsigned long delay)
3428 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3430 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3432 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3433 unsigned long delay)
3435 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3437 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3440 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3441 * @plug: The &struct blk_plug that needs to be initialized
3444 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3445 * pending I/O should the task end up blocking between blk_start_plug() and
3446 * blk_finish_plug(). This is important from a performance perspective, but
3447 * also ensures that we don't deadlock. For instance, if the task is blocking
3448 * for a memory allocation, memory reclaim could end up wanting to free a
3449 * page belonging to that request that is currently residing in our private
3450 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3451 * this kind of deadlock.
3453 void blk_start_plug(struct blk_plug *plug)
3455 struct task_struct *tsk = current;
3458 * If this is a nested plug, don't actually assign it.
3463 INIT_LIST_HEAD(&plug->list);
3464 INIT_LIST_HEAD(&plug->mq_list);
3465 INIT_LIST_HEAD(&plug->cb_list);
3467 * Store ordering should not be needed here, since a potential
3468 * preempt will imply a full memory barrier
3472 EXPORT_SYMBOL(blk_start_plug);
3474 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3476 struct request *rqa = container_of(a, struct request, queuelist);
3477 struct request *rqb = container_of(b, struct request, queuelist);
3479 return !(rqa->q < rqb->q ||
3480 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3484 * If 'from_schedule' is true, then postpone the dispatch of requests
3485 * until a safe kblockd context. We due this to avoid accidental big
3486 * additional stack usage in driver dispatch, in places where the originally
3487 * plugger did not intend it.
3489 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3491 __releases(q->queue_lock)
3493 lockdep_assert_held(q->queue_lock);
3495 trace_block_unplug(q, depth, !from_schedule);
3498 blk_run_queue_async(q);
3501 spin_unlock(q->queue_lock);
3504 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3506 LIST_HEAD(callbacks);
3508 while (!list_empty(&plug->cb_list)) {
3509 list_splice_init(&plug->cb_list, &callbacks);
3511 while (!list_empty(&callbacks)) {
3512 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3515 list_del(&cb->list);
3516 cb->callback(cb, from_schedule);
3521 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3524 struct blk_plug *plug = current->plug;
3525 struct blk_plug_cb *cb;
3530 list_for_each_entry(cb, &plug->cb_list, list)
3531 if (cb->callback == unplug && cb->data == data)
3534 /* Not currently on the callback list */
3535 BUG_ON(size < sizeof(*cb));
3536 cb = kzalloc(size, GFP_ATOMIC);
3539 cb->callback = unplug;
3540 list_add(&cb->list, &plug->cb_list);
3544 EXPORT_SYMBOL(blk_check_plugged);
3546 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3548 struct request_queue *q;
3549 unsigned long flags;
3554 flush_plug_callbacks(plug, from_schedule);
3556 if (!list_empty(&plug->mq_list))
3557 blk_mq_flush_plug_list(plug, from_schedule);
3559 if (list_empty(&plug->list))
3562 list_splice_init(&plug->list, &list);
3564 list_sort(NULL, &list, plug_rq_cmp);
3570 * Save and disable interrupts here, to avoid doing it for every
3571 * queue lock we have to take.
3573 local_irq_save(flags);
3574 while (!list_empty(&list)) {
3575 rq = list_entry_rq(list.next);
3576 list_del_init(&rq->queuelist);
3580 * This drops the queue lock
3583 queue_unplugged(q, depth, from_schedule);
3586 spin_lock(q->queue_lock);
3590 * Short-circuit if @q is dead
3592 if (unlikely(blk_queue_dying(q))) {
3593 __blk_end_request_all(rq, BLK_STS_IOERR);
3598 * rq is already accounted, so use raw insert
3600 if (op_is_flush(rq->cmd_flags))
3601 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3603 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3609 * This drops the queue lock
3612 queue_unplugged(q, depth, from_schedule);
3614 local_irq_restore(flags);
3617 void blk_finish_plug(struct blk_plug *plug)
3619 if (plug != current->plug)
3621 blk_flush_plug_list(plug, false);
3623 current->plug = NULL;
3625 EXPORT_SYMBOL(blk_finish_plug);
3629 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3630 * @q: the queue of the device
3631 * @dev: the device the queue belongs to
3634 * Initialize runtime-PM-related fields for @q and start auto suspend for
3635 * @dev. Drivers that want to take advantage of request-based runtime PM
3636 * should call this function after @dev has been initialized, and its
3637 * request queue @q has been allocated, and runtime PM for it can not happen
3638 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3639 * cases, driver should call this function before any I/O has taken place.
3641 * This function takes care of setting up using auto suspend for the device,
3642 * the autosuspend delay is set to -1 to make runtime suspend impossible
3643 * until an updated value is either set by user or by driver. Drivers do
3644 * not need to touch other autosuspend settings.
3646 * The block layer runtime PM is request based, so only works for drivers
3647 * that use request as their IO unit instead of those directly use bio's.
3649 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3651 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3656 q->rpm_status = RPM_ACTIVE;
3657 pm_runtime_set_autosuspend_delay(q->dev, -1);
3658 pm_runtime_use_autosuspend(q->dev);
3660 EXPORT_SYMBOL(blk_pm_runtime_init);
3663 * blk_pre_runtime_suspend - Pre runtime suspend check
3664 * @q: the queue of the device
3667 * This function will check if runtime suspend is allowed for the device
3668 * by examining if there are any requests pending in the queue. If there
3669 * are requests pending, the device can not be runtime suspended; otherwise,
3670 * the queue's status will be updated to SUSPENDING and the driver can
3671 * proceed to suspend the device.
3673 * For the not allowed case, we mark last busy for the device so that
3674 * runtime PM core will try to autosuspend it some time later.
3676 * This function should be called near the start of the device's
3677 * runtime_suspend callback.
3680 * 0 - OK to runtime suspend the device
3681 * -EBUSY - Device should not be runtime suspended
3683 int blk_pre_runtime_suspend(struct request_queue *q)
3690 spin_lock_irq(q->queue_lock);
3691 if (q->nr_pending) {
3693 pm_runtime_mark_last_busy(q->dev);
3695 q->rpm_status = RPM_SUSPENDING;
3697 spin_unlock_irq(q->queue_lock);
3700 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3703 * blk_post_runtime_suspend - Post runtime suspend processing
3704 * @q: the queue of the device
3705 * @err: return value of the device's runtime_suspend function
3708 * Update the queue's runtime status according to the return value of the
3709 * device's runtime suspend function and mark last busy for the device so
3710 * that PM core will try to auto suspend the device at a later time.
3712 * This function should be called near the end of the device's
3713 * runtime_suspend callback.
3715 void blk_post_runtime_suspend(struct request_queue *q, int err)
3720 spin_lock_irq(q->queue_lock);
3722 q->rpm_status = RPM_SUSPENDED;
3724 q->rpm_status = RPM_ACTIVE;
3725 pm_runtime_mark_last_busy(q->dev);
3727 spin_unlock_irq(q->queue_lock);
3729 EXPORT_SYMBOL(blk_post_runtime_suspend);
3732 * blk_pre_runtime_resume - Pre runtime resume processing
3733 * @q: the queue of the device
3736 * Update the queue's runtime status to RESUMING in preparation for the
3737 * runtime resume of the device.
3739 * This function should be called near the start of the device's
3740 * runtime_resume callback.
3742 void blk_pre_runtime_resume(struct request_queue *q)
3747 spin_lock_irq(q->queue_lock);
3748 q->rpm_status = RPM_RESUMING;
3749 spin_unlock_irq(q->queue_lock);
3751 EXPORT_SYMBOL(blk_pre_runtime_resume);
3754 * blk_post_runtime_resume - Post runtime resume processing
3755 * @q: the queue of the device
3756 * @err: return value of the device's runtime_resume function
3759 * Update the queue's runtime status according to the return value of the
3760 * device's runtime_resume function. If it is successfully resumed, process
3761 * the requests that are queued into the device's queue when it is resuming
3762 * and then mark last busy and initiate autosuspend for it.
3764 * This function should be called near the end of the device's
3765 * runtime_resume callback.
3767 void blk_post_runtime_resume(struct request_queue *q, int err)
3772 spin_lock_irq(q->queue_lock);
3774 q->rpm_status = RPM_ACTIVE;
3776 pm_runtime_mark_last_busy(q->dev);
3777 pm_request_autosuspend(q->dev);
3779 q->rpm_status = RPM_SUSPENDED;
3781 spin_unlock_irq(q->queue_lock);
3783 EXPORT_SYMBOL(blk_post_runtime_resume);
3786 * blk_set_runtime_active - Force runtime status of the queue to be active
3787 * @q: the queue of the device
3789 * If the device is left runtime suspended during system suspend the resume
3790 * hook typically resumes the device and corrects runtime status
3791 * accordingly. However, that does not affect the queue runtime PM status
3792 * which is still "suspended". This prevents processing requests from the
3795 * This function can be used in driver's resume hook to correct queue
3796 * runtime PM status and re-enable peeking requests from the queue. It
3797 * should be called before first request is added to the queue.
3799 void blk_set_runtime_active(struct request_queue *q)
3801 spin_lock_irq(q->queue_lock);
3802 q->rpm_status = RPM_ACTIVE;
3803 pm_runtime_mark_last_busy(q->dev);
3804 pm_request_autosuspend(q->dev);
3805 spin_unlock_irq(q->queue_lock);
3807 EXPORT_SYMBOL(blk_set_runtime_active);
3810 int __init blk_dev_init(void)
3812 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3813 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3814 FIELD_SIZEOF(struct request, cmd_flags));
3815 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3816 FIELD_SIZEOF(struct bio, bi_opf));
3818 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3819 kblockd_workqueue = alloc_workqueue("kblockd",
3820 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3821 if (!kblockd_workqueue)
3822 panic("Failed to create kblockd\n");
3824 request_cachep = kmem_cache_create("blkdev_requests",
3825 sizeof(struct request), 0, SLAB_PANIC, NULL);
3827 blk_requestq_cachep = kmem_cache_create("request_queue",
3828 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3830 #ifdef CONFIG_DEBUG_FS
3831 blk_debugfs_root = debugfs_create_dir("block", NULL);