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]);
567 void blk_drain_queue(struct request_queue *q)
569 spin_lock_irq(q->queue_lock);
570 __blk_drain_queue(q, true);
571 spin_unlock_irq(q->queue_lock);
575 * blk_queue_bypass_start - enter queue bypass mode
576 * @q: queue of interest
578 * In bypass mode, only the dispatch FIFO queue of @q is used. This
579 * function makes @q enter bypass mode and drains all requests which were
580 * throttled or issued before. On return, it's guaranteed that no request
581 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
582 * inside queue or RCU read lock.
584 void blk_queue_bypass_start(struct request_queue *q)
586 WARN_ON_ONCE(q->mq_ops);
588 spin_lock_irq(q->queue_lock);
590 queue_flag_set(QUEUE_FLAG_BYPASS, q);
591 spin_unlock_irq(q->queue_lock);
594 * Queues start drained. Skip actual draining till init is
595 * complete. This avoids lenghty delays during queue init which
596 * can happen many times during boot.
598 if (blk_queue_init_done(q)) {
599 spin_lock_irq(q->queue_lock);
600 __blk_drain_queue(q, false);
601 spin_unlock_irq(q->queue_lock);
603 /* ensure blk_queue_bypass() is %true inside RCU read lock */
607 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
610 * blk_queue_bypass_end - leave queue bypass mode
611 * @q: queue of interest
613 * Leave bypass mode and restore the normal queueing behavior.
615 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
616 * this function is called for both blk-sq and blk-mq queues.
618 void blk_queue_bypass_end(struct request_queue *q)
620 spin_lock_irq(q->queue_lock);
621 if (!--q->bypass_depth)
622 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
623 WARN_ON_ONCE(q->bypass_depth < 0);
624 spin_unlock_irq(q->queue_lock);
626 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
628 void blk_set_queue_dying(struct request_queue *q)
630 spin_lock_irq(q->queue_lock);
631 queue_flag_set(QUEUE_FLAG_DYING, q);
632 spin_unlock_irq(q->queue_lock);
635 * When queue DYING flag is set, we need to block new req
636 * entering queue, so we call blk_freeze_queue_start() to
637 * prevent I/O from crossing blk_queue_enter().
639 blk_freeze_queue_start(q);
642 blk_mq_wake_waiters(q);
644 struct request_list *rl;
646 spin_lock_irq(q->queue_lock);
647 blk_queue_for_each_rl(rl, q) {
649 wake_up_all(&rl->wait[BLK_RW_SYNC]);
650 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
653 spin_unlock_irq(q->queue_lock);
656 /* Make blk_queue_enter() reexamine the DYING flag. */
657 wake_up_all(&q->mq_freeze_wq);
659 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
662 * blk_cleanup_queue - shutdown a request queue
663 * @q: request queue to shutdown
665 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
666 * put it. All future requests will be failed immediately with -ENODEV.
668 void blk_cleanup_queue(struct request_queue *q)
670 spinlock_t *lock = q->queue_lock;
672 /* mark @q DYING, no new request or merges will be allowed afterwards */
673 mutex_lock(&q->sysfs_lock);
674 blk_set_queue_dying(q);
678 * A dying queue is permanently in bypass mode till released. Note
679 * that, unlike blk_queue_bypass_start(), we aren't performing
680 * synchronize_rcu() after entering bypass mode to avoid the delay
681 * as some drivers create and destroy a lot of queues while
682 * probing. This is still safe because blk_release_queue() will be
683 * called only after the queue refcnt drops to zero and nothing,
684 * RCU or not, would be traversing the queue by then.
687 queue_flag_set(QUEUE_FLAG_BYPASS, q);
689 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
690 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
691 queue_flag_set(QUEUE_FLAG_DYING, q);
692 spin_unlock_irq(lock);
693 mutex_unlock(&q->sysfs_lock);
696 * Drain all requests queued before DYING marking. Set DEAD flag to
697 * prevent that q->request_fn() gets invoked after draining finished.
701 queue_flag_set(QUEUE_FLAG_DEAD, q);
702 spin_unlock_irq(lock);
705 * make sure all in-progress dispatch are completed because
706 * blk_freeze_queue() can only complete all requests, and
707 * dispatch may still be in-progress since we dispatch requests
708 * from more than one contexts
711 blk_mq_quiesce_queue(q);
713 /* for synchronous bio-based driver finish in-flight integrity i/o */
714 blk_flush_integrity();
716 /* @q won't process any more request, flush async actions */
717 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
721 blk_mq_free_queue(q);
722 percpu_ref_exit(&q->q_usage_counter);
725 if (q->queue_lock != &q->__queue_lock)
726 q->queue_lock = &q->__queue_lock;
727 spin_unlock_irq(lock);
729 /* @q is and will stay empty, shutdown and put */
732 EXPORT_SYMBOL(blk_cleanup_queue);
734 /* Allocate memory local to the request queue */
735 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
737 struct request_queue *q = data;
739 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
742 static void free_request_simple(void *element, void *data)
744 kmem_cache_free(request_cachep, element);
747 static void *alloc_request_size(gfp_t gfp_mask, void *data)
749 struct request_queue *q = data;
752 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
754 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
761 static void free_request_size(void *element, void *data)
763 struct request_queue *q = data;
766 q->exit_rq_fn(q, element);
770 int blk_init_rl(struct request_list *rl, struct request_queue *q,
773 if (unlikely(rl->rq_pool) || q->mq_ops)
777 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
778 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
779 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
780 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
783 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
784 alloc_request_size, free_request_size,
785 q, gfp_mask, q->node);
787 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
788 alloc_request_simple, free_request_simple,
789 q, gfp_mask, q->node);
794 if (rl != &q->root_rl)
795 WARN_ON_ONCE(!blk_get_queue(q));
800 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
803 mempool_destroy(rl->rq_pool);
804 if (rl != &q->root_rl)
809 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
811 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
813 EXPORT_SYMBOL(blk_alloc_queue);
816 * blk_queue_enter() - try to increase q->q_usage_counter
817 * @q: request queue pointer
818 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
820 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
822 const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
825 bool success = false;
828 rcu_read_lock_sched();
829 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
831 * The code that sets the PREEMPT_ONLY flag is
832 * responsible for ensuring that that flag is globally
833 * visible before the queue is unfrozen.
835 if (preempt || !blk_queue_preempt_only(q)) {
838 percpu_ref_put(&q->q_usage_counter);
841 rcu_read_unlock_sched();
846 if (flags & BLK_MQ_REQ_NOWAIT)
850 * read pair of barrier in blk_freeze_queue_start(),
851 * we need to order reading __PERCPU_REF_DEAD flag of
852 * .q_usage_counter and reading .mq_freeze_depth or
853 * queue dying flag, otherwise the following wait may
854 * never return if the two reads are reordered.
858 ret = wait_event_interruptible(q->mq_freeze_wq,
859 (atomic_read(&q->mq_freeze_depth) == 0 &&
860 (preempt || !blk_queue_preempt_only(q))) ||
862 if (blk_queue_dying(q))
869 void blk_queue_exit(struct request_queue *q)
871 percpu_ref_put(&q->q_usage_counter);
874 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
876 struct request_queue *q =
877 container_of(ref, struct request_queue, q_usage_counter);
879 wake_up_all(&q->mq_freeze_wq);
882 static void blk_rq_timed_out_timer(struct timer_list *t)
884 struct request_queue *q = from_timer(q, t, timeout);
886 kblockd_schedule_work(&q->timeout_work);
889 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
891 struct request_queue *q;
893 q = kmem_cache_alloc_node(blk_requestq_cachep,
894 gfp_mask | __GFP_ZERO, node_id);
898 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
902 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
906 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
907 if (!q->backing_dev_info)
910 q->stats = blk_alloc_queue_stats();
914 q->backing_dev_info->ra_pages =
915 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
916 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
917 q->backing_dev_info->name = "block";
920 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
921 laptop_mode_timer_fn, 0);
922 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
923 INIT_WORK(&q->timeout_work, NULL);
924 INIT_LIST_HEAD(&q->queue_head);
925 INIT_LIST_HEAD(&q->timeout_list);
926 INIT_LIST_HEAD(&q->icq_list);
927 #ifdef CONFIG_BLK_CGROUP
928 INIT_LIST_HEAD(&q->blkg_list);
930 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
932 kobject_init(&q->kobj, &blk_queue_ktype);
934 #ifdef CONFIG_BLK_DEV_IO_TRACE
935 mutex_init(&q->blk_trace_mutex);
937 mutex_init(&q->sysfs_lock);
938 spin_lock_init(&q->__queue_lock);
941 * By default initialize queue_lock to internal lock and driver can
942 * override it later if need be.
944 q->queue_lock = &q->__queue_lock;
947 * A queue starts its life with bypass turned on to avoid
948 * unnecessary bypass on/off overhead and nasty surprises during
949 * init. The initial bypass will be finished when the queue is
950 * registered by blk_register_queue().
953 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
955 init_waitqueue_head(&q->mq_freeze_wq);
958 * Init percpu_ref in atomic mode so that it's faster to shutdown.
959 * See blk_register_queue() for details.
961 if (percpu_ref_init(&q->q_usage_counter,
962 blk_queue_usage_counter_release,
963 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
966 if (blkcg_init_queue(q))
972 percpu_ref_exit(&q->q_usage_counter);
974 blk_free_queue_stats(q->stats);
976 bdi_put(q->backing_dev_info);
978 bioset_free(q->bio_split);
980 ida_simple_remove(&blk_queue_ida, q->id);
982 kmem_cache_free(blk_requestq_cachep, q);
985 EXPORT_SYMBOL(blk_alloc_queue_node);
988 * blk_init_queue - prepare a request queue for use with a block device
989 * @rfn: The function to be called to process requests that have been
990 * placed on the queue.
991 * @lock: Request queue spin lock
994 * If a block device wishes to use the standard request handling procedures,
995 * which sorts requests and coalesces adjacent requests, then it must
996 * call blk_init_queue(). The function @rfn will be called when there
997 * are requests on the queue that need to be processed. If the device
998 * supports plugging, then @rfn may not be called immediately when requests
999 * are available on the queue, but may be called at some time later instead.
1000 * Plugged queues are generally unplugged when a buffer belonging to one
1001 * of the requests on the queue is needed, or due to memory pressure.
1003 * @rfn is not required, or even expected, to remove all requests off the
1004 * queue, but only as many as it can handle at a time. If it does leave
1005 * requests on the queue, it is responsible for arranging that the requests
1006 * get dealt with eventually.
1008 * The queue spin lock must be held while manipulating the requests on the
1009 * request queue; this lock will be taken also from interrupt context, so irq
1010 * disabling is needed for it.
1012 * Function returns a pointer to the initialized request queue, or %NULL if
1013 * it didn't succeed.
1016 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1017 * when the block device is deactivated (such as at module unload).
1020 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1022 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1024 EXPORT_SYMBOL(blk_init_queue);
1026 struct request_queue *
1027 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1029 struct request_queue *q;
1031 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1035 q->request_fn = rfn;
1037 q->queue_lock = lock;
1038 if (blk_init_allocated_queue(q) < 0) {
1039 blk_cleanup_queue(q);
1045 EXPORT_SYMBOL(blk_init_queue_node);
1047 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1050 int blk_init_allocated_queue(struct request_queue *q)
1052 WARN_ON_ONCE(q->mq_ops);
1054 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
1058 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1059 goto out_free_flush_queue;
1061 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1062 goto out_exit_flush_rq;
1064 INIT_WORK(&q->timeout_work, blk_timeout_work);
1065 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1068 * This also sets hw/phys segments, boundary and size
1070 blk_queue_make_request(q, blk_queue_bio);
1072 q->sg_reserved_size = INT_MAX;
1074 /* Protect q->elevator from elevator_change */
1075 mutex_lock(&q->sysfs_lock);
1078 if (elevator_init(q, NULL)) {
1079 mutex_unlock(&q->sysfs_lock);
1080 goto out_exit_flush_rq;
1083 mutex_unlock(&q->sysfs_lock);
1088 q->exit_rq_fn(q, q->fq->flush_rq);
1089 out_free_flush_queue:
1090 blk_free_flush_queue(q->fq);
1093 EXPORT_SYMBOL(blk_init_allocated_queue);
1095 bool blk_get_queue(struct request_queue *q)
1097 if (likely(!blk_queue_dying(q))) {
1104 EXPORT_SYMBOL(blk_get_queue);
1106 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1108 if (rq->rq_flags & RQF_ELVPRIV) {
1109 elv_put_request(rl->q, rq);
1111 put_io_context(rq->elv.icq->ioc);
1114 mempool_free(rq, rl->rq_pool);
1118 * ioc_batching returns true if the ioc is a valid batching request and
1119 * should be given priority access to a request.
1121 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1127 * Make sure the process is able to allocate at least 1 request
1128 * even if the batch times out, otherwise we could theoretically
1131 return ioc->nr_batch_requests == q->nr_batching ||
1132 (ioc->nr_batch_requests > 0
1133 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1137 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1138 * will cause the process to be a "batcher" on all queues in the system. This
1139 * is the behaviour we want though - once it gets a wakeup it should be given
1142 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1144 if (!ioc || ioc_batching(q, ioc))
1147 ioc->nr_batch_requests = q->nr_batching;
1148 ioc->last_waited = jiffies;
1151 static void __freed_request(struct request_list *rl, int sync)
1153 struct request_queue *q = rl->q;
1155 if (rl->count[sync] < queue_congestion_off_threshold(q))
1156 blk_clear_congested(rl, sync);
1158 if (rl->count[sync] + 1 <= q->nr_requests) {
1159 if (waitqueue_active(&rl->wait[sync]))
1160 wake_up(&rl->wait[sync]);
1162 blk_clear_rl_full(rl, sync);
1167 * A request has just been released. Account for it, update the full and
1168 * congestion status, wake up any waiters. Called under q->queue_lock.
1170 static void freed_request(struct request_list *rl, bool sync,
1171 req_flags_t rq_flags)
1173 struct request_queue *q = rl->q;
1177 if (rq_flags & RQF_ELVPRIV)
1178 q->nr_rqs_elvpriv--;
1180 __freed_request(rl, sync);
1182 if (unlikely(rl->starved[sync ^ 1]))
1183 __freed_request(rl, sync ^ 1);
1186 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1188 struct request_list *rl;
1189 int on_thresh, off_thresh;
1191 WARN_ON_ONCE(q->mq_ops);
1193 spin_lock_irq(q->queue_lock);
1194 q->nr_requests = nr;
1195 blk_queue_congestion_threshold(q);
1196 on_thresh = queue_congestion_on_threshold(q);
1197 off_thresh = queue_congestion_off_threshold(q);
1199 blk_queue_for_each_rl(rl, q) {
1200 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1201 blk_set_congested(rl, BLK_RW_SYNC);
1202 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1203 blk_clear_congested(rl, BLK_RW_SYNC);
1205 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1206 blk_set_congested(rl, BLK_RW_ASYNC);
1207 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1208 blk_clear_congested(rl, BLK_RW_ASYNC);
1210 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1211 blk_set_rl_full(rl, BLK_RW_SYNC);
1213 blk_clear_rl_full(rl, BLK_RW_SYNC);
1214 wake_up(&rl->wait[BLK_RW_SYNC]);
1217 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1218 blk_set_rl_full(rl, BLK_RW_ASYNC);
1220 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1221 wake_up(&rl->wait[BLK_RW_ASYNC]);
1225 spin_unlock_irq(q->queue_lock);
1230 * __get_request - get a free request
1231 * @rl: request list to allocate from
1232 * @op: operation and flags
1233 * @bio: bio to allocate request for (can be %NULL)
1234 * @flags: BLQ_MQ_REQ_* flags
1236 * Get a free request from @q. This function may fail under memory
1237 * pressure or if @q is dead.
1239 * Must be called with @q->queue_lock held and,
1240 * Returns ERR_PTR on failure, with @q->queue_lock held.
1241 * Returns request pointer on success, with @q->queue_lock *not held*.
1243 static struct request *__get_request(struct request_list *rl, unsigned int op,
1244 struct bio *bio, blk_mq_req_flags_t flags)
1246 struct request_queue *q = rl->q;
1248 struct elevator_type *et = q->elevator->type;
1249 struct io_context *ioc = rq_ioc(bio);
1250 struct io_cq *icq = NULL;
1251 const bool is_sync = op_is_sync(op);
1253 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1254 __GFP_DIRECT_RECLAIM;
1255 req_flags_t rq_flags = RQF_ALLOCED;
1257 lockdep_assert_held(q->queue_lock);
1259 if (unlikely(blk_queue_dying(q)))
1260 return ERR_PTR(-ENODEV);
1262 may_queue = elv_may_queue(q, op);
1263 if (may_queue == ELV_MQUEUE_NO)
1266 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1267 if (rl->count[is_sync]+1 >= q->nr_requests) {
1269 * The queue will fill after this allocation, so set
1270 * it as full, and mark this process as "batching".
1271 * This process will be allowed to complete a batch of
1272 * requests, others will be blocked.
1274 if (!blk_rl_full(rl, is_sync)) {
1275 ioc_set_batching(q, ioc);
1276 blk_set_rl_full(rl, is_sync);
1278 if (may_queue != ELV_MQUEUE_MUST
1279 && !ioc_batching(q, ioc)) {
1281 * The queue is full and the allocating
1282 * process is not a "batcher", and not
1283 * exempted by the IO scheduler
1285 return ERR_PTR(-ENOMEM);
1289 blk_set_congested(rl, is_sync);
1293 * Only allow batching queuers to allocate up to 50% over the defined
1294 * limit of requests, otherwise we could have thousands of requests
1295 * allocated with any setting of ->nr_requests
1297 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1298 return ERR_PTR(-ENOMEM);
1300 q->nr_rqs[is_sync]++;
1301 rl->count[is_sync]++;
1302 rl->starved[is_sync] = 0;
1305 * Decide whether the new request will be managed by elevator. If
1306 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1307 * prevent the current elevator from being destroyed until the new
1308 * request is freed. This guarantees icq's won't be destroyed and
1309 * makes creating new ones safe.
1311 * Flush requests do not use the elevator so skip initialization.
1312 * This allows a request to share the flush and elevator data.
1314 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1315 * it will be created after releasing queue_lock.
1317 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1318 rq_flags |= RQF_ELVPRIV;
1319 q->nr_rqs_elvpriv++;
1320 if (et->icq_cache && ioc)
1321 icq = ioc_lookup_icq(ioc, q);
1324 if (blk_queue_io_stat(q))
1325 rq_flags |= RQF_IO_STAT;
1326 spin_unlock_irq(q->queue_lock);
1328 /* allocate and init request */
1329 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1334 blk_rq_set_rl(rq, rl);
1336 rq->rq_flags = rq_flags;
1337 if (flags & BLK_MQ_REQ_PREEMPT)
1338 rq->rq_flags |= RQF_PREEMPT;
1341 if (rq_flags & RQF_ELVPRIV) {
1342 if (unlikely(et->icq_cache && !icq)) {
1344 icq = ioc_create_icq(ioc, q, gfp_mask);
1350 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1353 /* @rq->elv.icq holds io_context until @rq is freed */
1355 get_io_context(icq->ioc);
1359 * ioc may be NULL here, and ioc_batching will be false. That's
1360 * OK, if the queue is under the request limit then requests need
1361 * not count toward the nr_batch_requests limit. There will always
1362 * be some limit enforced by BLK_BATCH_TIME.
1364 if (ioc_batching(q, ioc))
1365 ioc->nr_batch_requests--;
1367 trace_block_getrq(q, bio, op);
1372 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1373 * and may fail indefinitely under memory pressure and thus
1374 * shouldn't stall IO. Treat this request as !elvpriv. This will
1375 * disturb iosched and blkcg but weird is bettern than dead.
1377 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1378 __func__, dev_name(q->backing_dev_info->dev));
1380 rq->rq_flags &= ~RQF_ELVPRIV;
1383 spin_lock_irq(q->queue_lock);
1384 q->nr_rqs_elvpriv--;
1385 spin_unlock_irq(q->queue_lock);
1390 * Allocation failed presumably due to memory. Undo anything we
1391 * might have messed up.
1393 * Allocating task should really be put onto the front of the wait
1394 * queue, but this is pretty rare.
1396 spin_lock_irq(q->queue_lock);
1397 freed_request(rl, is_sync, rq_flags);
1400 * in the very unlikely event that allocation failed and no
1401 * requests for this direction was pending, mark us starved so that
1402 * freeing of a request in the other direction will notice
1403 * us. another possible fix would be to split the rq mempool into
1407 if (unlikely(rl->count[is_sync] == 0))
1408 rl->starved[is_sync] = 1;
1409 return ERR_PTR(-ENOMEM);
1413 * get_request - get a free request
1414 * @q: request_queue to allocate request from
1415 * @op: operation and flags
1416 * @bio: bio to allocate request for (can be %NULL)
1417 * @flags: BLK_MQ_REQ_* flags.
1419 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1420 * this function keeps retrying under memory pressure and fails iff @q is dead.
1422 * Must be called with @q->queue_lock held and,
1423 * Returns ERR_PTR on failure, with @q->queue_lock held.
1424 * Returns request pointer on success, with @q->queue_lock *not held*.
1426 static struct request *get_request(struct request_queue *q, unsigned int op,
1427 struct bio *bio, blk_mq_req_flags_t flags)
1429 const bool is_sync = op_is_sync(op);
1431 struct request_list *rl;
1434 lockdep_assert_held(q->queue_lock);
1435 WARN_ON_ONCE(q->mq_ops);
1437 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1439 rq = __get_request(rl, op, bio, flags);
1443 if (op & REQ_NOWAIT) {
1445 return ERR_PTR(-EAGAIN);
1448 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1453 /* wait on @rl and retry */
1454 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1455 TASK_UNINTERRUPTIBLE);
1457 trace_block_sleeprq(q, bio, op);
1459 spin_unlock_irq(q->queue_lock);
1463 * After sleeping, we become a "batching" process and will be able
1464 * to allocate at least one request, and up to a big batch of them
1465 * for a small period time. See ioc_batching, ioc_set_batching
1467 ioc_set_batching(q, current->io_context);
1469 spin_lock_irq(q->queue_lock);
1470 finish_wait(&rl->wait[is_sync], &wait);
1475 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1476 static struct request *blk_old_get_request(struct request_queue *q,
1477 unsigned int op, blk_mq_req_flags_t flags)
1480 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1481 __GFP_DIRECT_RECLAIM;
1484 WARN_ON_ONCE(q->mq_ops);
1486 /* create ioc upfront */
1487 create_io_context(gfp_mask, q->node);
1489 ret = blk_queue_enter(q, flags);
1491 return ERR_PTR(ret);
1492 spin_lock_irq(q->queue_lock);
1493 rq = get_request(q, op, NULL, flags);
1495 spin_unlock_irq(q->queue_lock);
1500 /* q->queue_lock is unlocked at this point */
1502 rq->__sector = (sector_t) -1;
1503 rq->bio = rq->biotail = NULL;
1508 * blk_get_request_flags - allocate a request
1509 * @q: request queue to allocate a request for
1510 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1511 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1513 struct request *blk_get_request_flags(struct request_queue *q, unsigned int op,
1514 blk_mq_req_flags_t flags)
1516 struct request *req;
1518 WARN_ON_ONCE(op & REQ_NOWAIT);
1519 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1522 req = blk_mq_alloc_request(q, op, flags);
1523 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1524 q->mq_ops->initialize_rq_fn(req);
1526 req = blk_old_get_request(q, op, flags);
1527 if (!IS_ERR(req) && q->initialize_rq_fn)
1528 q->initialize_rq_fn(req);
1533 EXPORT_SYMBOL(blk_get_request_flags);
1535 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1538 return blk_get_request_flags(q, op, gfp_mask & __GFP_DIRECT_RECLAIM ?
1539 0 : BLK_MQ_REQ_NOWAIT);
1541 EXPORT_SYMBOL(blk_get_request);
1544 * blk_requeue_request - put a request back on queue
1545 * @q: request queue where request should be inserted
1546 * @rq: request to be inserted
1549 * Drivers often keep queueing requests until the hardware cannot accept
1550 * more, when that condition happens we need to put the request back
1551 * on the queue. Must be called with queue lock held.
1553 void blk_requeue_request(struct request_queue *q, struct request *rq)
1555 lockdep_assert_held(q->queue_lock);
1556 WARN_ON_ONCE(q->mq_ops);
1558 blk_delete_timer(rq);
1559 blk_clear_rq_complete(rq);
1560 trace_block_rq_requeue(q, rq);
1561 wbt_requeue(q->rq_wb, &rq->issue_stat);
1563 if (rq->rq_flags & RQF_QUEUED)
1564 blk_queue_end_tag(q, rq);
1566 BUG_ON(blk_queued_rq(rq));
1568 elv_requeue_request(q, rq);
1570 EXPORT_SYMBOL(blk_requeue_request);
1572 static void add_acct_request(struct request_queue *q, struct request *rq,
1575 blk_account_io_start(rq, true);
1576 __elv_add_request(q, rq, where);
1579 static void part_round_stats_single(struct request_queue *q, int cpu,
1580 struct hd_struct *part, unsigned long now,
1581 unsigned int inflight)
1584 __part_stat_add(cpu, part, time_in_queue,
1585 inflight * (now - part->stamp));
1586 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1592 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1593 * @q: target block queue
1594 * @cpu: cpu number for stats access
1595 * @part: target partition
1597 * The average IO queue length and utilisation statistics are maintained
1598 * by observing the current state of the queue length and the amount of
1599 * time it has been in this state for.
1601 * Normally, that accounting is done on IO completion, but that can result
1602 * in more than a second's worth of IO being accounted for within any one
1603 * second, leading to >100% utilisation. To deal with that, we call this
1604 * function to do a round-off before returning the results when reading
1605 * /proc/diskstats. This accounts immediately for all queue usage up to
1606 * the current jiffies and restarts the counters again.
1608 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1610 struct hd_struct *part2 = NULL;
1611 unsigned long now = jiffies;
1612 unsigned int inflight[2];
1615 if (part->stamp != now)
1619 part2 = &part_to_disk(part)->part0;
1620 if (part2->stamp != now)
1627 part_in_flight(q, part, inflight);
1630 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1632 part_round_stats_single(q, cpu, part, now, inflight[0]);
1634 EXPORT_SYMBOL_GPL(part_round_stats);
1637 static void blk_pm_put_request(struct request *rq)
1639 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1640 pm_runtime_mark_last_busy(rq->q->dev);
1643 static inline void blk_pm_put_request(struct request *rq) {}
1646 void __blk_put_request(struct request_queue *q, struct request *req)
1648 req_flags_t rq_flags = req->rq_flags;
1654 blk_mq_free_request(req);
1658 lockdep_assert_held(q->queue_lock);
1660 blk_req_zone_write_unlock(req);
1661 blk_pm_put_request(req);
1663 elv_completed_request(q, req);
1665 /* this is a bio leak */
1666 WARN_ON(req->bio != NULL);
1668 wbt_done(q->rq_wb, &req->issue_stat);
1671 * Request may not have originated from ll_rw_blk. if not,
1672 * it didn't come out of our reserved rq pools
1674 if (rq_flags & RQF_ALLOCED) {
1675 struct request_list *rl = blk_rq_rl(req);
1676 bool sync = op_is_sync(req->cmd_flags);
1678 BUG_ON(!list_empty(&req->queuelist));
1679 BUG_ON(ELV_ON_HASH(req));
1681 blk_free_request(rl, req);
1682 freed_request(rl, sync, rq_flags);
1687 EXPORT_SYMBOL_GPL(__blk_put_request);
1689 void blk_put_request(struct request *req)
1691 struct request_queue *q = req->q;
1694 blk_mq_free_request(req);
1696 unsigned long flags;
1698 spin_lock_irqsave(q->queue_lock, flags);
1699 __blk_put_request(q, req);
1700 spin_unlock_irqrestore(q->queue_lock, flags);
1703 EXPORT_SYMBOL(blk_put_request);
1705 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1708 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1710 if (!ll_back_merge_fn(q, req, bio))
1713 trace_block_bio_backmerge(q, req, bio);
1715 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1716 blk_rq_set_mixed_merge(req);
1718 req->biotail->bi_next = bio;
1720 req->__data_len += bio->bi_iter.bi_size;
1721 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1723 blk_account_io_start(req, false);
1727 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1730 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1732 if (!ll_front_merge_fn(q, req, bio))
1735 trace_block_bio_frontmerge(q, req, bio);
1737 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1738 blk_rq_set_mixed_merge(req);
1740 bio->bi_next = req->bio;
1743 req->__sector = bio->bi_iter.bi_sector;
1744 req->__data_len += bio->bi_iter.bi_size;
1745 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1747 blk_account_io_start(req, false);
1751 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1754 unsigned short segments = blk_rq_nr_discard_segments(req);
1756 if (segments >= queue_max_discard_segments(q))
1758 if (blk_rq_sectors(req) + bio_sectors(bio) >
1759 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1762 req->biotail->bi_next = bio;
1764 req->__data_len += bio->bi_iter.bi_size;
1765 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1766 req->nr_phys_segments = segments + 1;
1768 blk_account_io_start(req, false);
1771 req_set_nomerge(q, req);
1776 * blk_attempt_plug_merge - try to merge with %current's plugged list
1777 * @q: request_queue new bio is being queued at
1778 * @bio: new bio being queued
1779 * @request_count: out parameter for number of traversed plugged requests
1780 * @same_queue_rq: pointer to &struct request that gets filled in when
1781 * another request associated with @q is found on the plug list
1782 * (optional, may be %NULL)
1784 * Determine whether @bio being queued on @q can be merged with a request
1785 * on %current's plugged list. Returns %true if merge was successful,
1788 * Plugging coalesces IOs from the same issuer for the same purpose without
1789 * going through @q->queue_lock. As such it's more of an issuing mechanism
1790 * than scheduling, and the request, while may have elvpriv data, is not
1791 * added on the elevator at this point. In addition, we don't have
1792 * reliable access to the elevator outside queue lock. Only check basic
1793 * merging parameters without querying the elevator.
1795 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1797 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1798 unsigned int *request_count,
1799 struct request **same_queue_rq)
1801 struct blk_plug *plug;
1803 struct list_head *plug_list;
1805 plug = current->plug;
1811 plug_list = &plug->mq_list;
1813 plug_list = &plug->list;
1815 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1816 bool merged = false;
1821 * Only blk-mq multiple hardware queues case checks the
1822 * rq in the same queue, there should be only one such
1826 *same_queue_rq = rq;
1829 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1832 switch (blk_try_merge(rq, bio)) {
1833 case ELEVATOR_BACK_MERGE:
1834 merged = bio_attempt_back_merge(q, rq, bio);
1836 case ELEVATOR_FRONT_MERGE:
1837 merged = bio_attempt_front_merge(q, rq, bio);
1839 case ELEVATOR_DISCARD_MERGE:
1840 merged = bio_attempt_discard_merge(q, rq, bio);
1853 unsigned int blk_plug_queued_count(struct request_queue *q)
1855 struct blk_plug *plug;
1857 struct list_head *plug_list;
1858 unsigned int ret = 0;
1860 plug = current->plug;
1865 plug_list = &plug->mq_list;
1867 plug_list = &plug->list;
1869 list_for_each_entry(rq, plug_list, queuelist) {
1877 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1879 struct io_context *ioc = rq_ioc(bio);
1881 if (bio->bi_opf & REQ_RAHEAD)
1882 req->cmd_flags |= REQ_FAILFAST_MASK;
1884 req->__sector = bio->bi_iter.bi_sector;
1885 if (ioprio_valid(bio_prio(bio)))
1886 req->ioprio = bio_prio(bio);
1888 req->ioprio = ioc->ioprio;
1890 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1891 req->write_hint = bio->bi_write_hint;
1892 blk_rq_bio_prep(req->q, req, bio);
1894 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1896 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1898 struct blk_plug *plug;
1899 int where = ELEVATOR_INSERT_SORT;
1900 struct request *req, *free;
1901 unsigned int request_count = 0;
1902 unsigned int wb_acct;
1905 * low level driver can indicate that it wants pages above a
1906 * certain limit bounced to low memory (ie for highmem, or even
1907 * ISA dma in theory)
1909 blk_queue_bounce(q, &bio);
1911 blk_queue_split(q, &bio);
1913 if (!bio_integrity_prep(bio))
1914 return BLK_QC_T_NONE;
1916 if (op_is_flush(bio->bi_opf)) {
1917 spin_lock_irq(q->queue_lock);
1918 where = ELEVATOR_INSERT_FLUSH;
1923 * Check if we can merge with the plugged list before grabbing
1926 if (!blk_queue_nomerges(q)) {
1927 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1928 return BLK_QC_T_NONE;
1930 request_count = blk_plug_queued_count(q);
1932 spin_lock_irq(q->queue_lock);
1934 switch (elv_merge(q, &req, bio)) {
1935 case ELEVATOR_BACK_MERGE:
1936 if (!bio_attempt_back_merge(q, req, bio))
1938 elv_bio_merged(q, req, bio);
1939 free = attempt_back_merge(q, req);
1941 __blk_put_request(q, free);
1943 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1945 case ELEVATOR_FRONT_MERGE:
1946 if (!bio_attempt_front_merge(q, req, bio))
1948 elv_bio_merged(q, req, bio);
1949 free = attempt_front_merge(q, req);
1951 __blk_put_request(q, free);
1953 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1960 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1963 * Grab a free request. This is might sleep but can not fail.
1964 * Returns with the queue unlocked.
1966 blk_queue_enter_live(q);
1967 req = get_request(q, bio->bi_opf, bio, 0);
1970 __wbt_done(q->rq_wb, wb_acct);
1971 if (PTR_ERR(req) == -ENOMEM)
1972 bio->bi_status = BLK_STS_RESOURCE;
1974 bio->bi_status = BLK_STS_IOERR;
1979 wbt_track(&req->issue_stat, wb_acct);
1982 * After dropping the lock and possibly sleeping here, our request
1983 * may now be mergeable after it had proven unmergeable (above).
1984 * We don't worry about that case for efficiency. It won't happen
1985 * often, and the elevators are able to handle it.
1987 blk_init_request_from_bio(req, bio);
1989 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1990 req->cpu = raw_smp_processor_id();
1992 plug = current->plug;
1995 * If this is the first request added after a plug, fire
1998 * @request_count may become stale because of schedule
1999 * out, so check plug list again.
2001 if (!request_count || list_empty(&plug->list))
2002 trace_block_plug(q);
2004 struct request *last = list_entry_rq(plug->list.prev);
2005 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2006 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2007 blk_flush_plug_list(plug, false);
2008 trace_block_plug(q);
2011 list_add_tail(&req->queuelist, &plug->list);
2012 blk_account_io_start(req, true);
2014 spin_lock_irq(q->queue_lock);
2015 add_acct_request(q, req, where);
2018 spin_unlock_irq(q->queue_lock);
2021 return BLK_QC_T_NONE;
2024 static void handle_bad_sector(struct bio *bio)
2026 char b[BDEVNAME_SIZE];
2028 printk(KERN_INFO "attempt to access beyond end of device\n");
2029 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2030 bio_devname(bio, b), bio->bi_opf,
2031 (unsigned long long)bio_end_sector(bio),
2032 (long long)get_capacity(bio->bi_disk));
2035 #ifdef CONFIG_FAIL_MAKE_REQUEST
2037 static DECLARE_FAULT_ATTR(fail_make_request);
2039 static int __init setup_fail_make_request(char *str)
2041 return setup_fault_attr(&fail_make_request, str);
2043 __setup("fail_make_request=", setup_fail_make_request);
2045 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2047 return part->make_it_fail && should_fail(&fail_make_request, bytes);
2050 static int __init fail_make_request_debugfs(void)
2052 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2053 NULL, &fail_make_request);
2055 return PTR_ERR_OR_ZERO(dir);
2058 late_initcall(fail_make_request_debugfs);
2060 #else /* CONFIG_FAIL_MAKE_REQUEST */
2062 static inline bool should_fail_request(struct hd_struct *part,
2068 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2070 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2072 if (part->policy && op_is_write(bio_op(bio))) {
2073 char b[BDEVNAME_SIZE];
2076 "generic_make_request: Trying to write "
2077 "to read-only block-device %s (partno %d)\n",
2078 bio_devname(bio, b), part->partno);
2086 * Remap block n of partition p to block n+start(p) of the disk.
2088 static inline int blk_partition_remap(struct bio *bio)
2090 struct hd_struct *p;
2094 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2095 if (unlikely(!p || should_fail_request(p, bio->bi_iter.bi_size) ||
2096 bio_check_ro(bio, p))) {
2102 * Zone reset does not include bi_size so bio_sectors() is always 0.
2103 * Include a test for the reset op code and perform the remap if needed.
2105 if (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET)
2108 bio->bi_iter.bi_sector += p->start_sect;
2110 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2111 bio->bi_iter.bi_sector - p->start_sect);
2119 * Check whether this bio extends beyond the end of the device.
2121 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
2128 /* Test device or partition size, when known. */
2129 maxsector = get_capacity(bio->bi_disk);
2131 sector_t sector = bio->bi_iter.bi_sector;
2133 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2135 * This may well happen - the kernel calls bread()
2136 * without checking the size of the device, e.g., when
2137 * mounting a device.
2139 handle_bad_sector(bio);
2147 static noinline_for_stack bool
2148 generic_make_request_checks(struct bio *bio)
2150 struct request_queue *q;
2151 int nr_sectors = bio_sectors(bio);
2152 blk_status_t status = BLK_STS_IOERR;
2153 char b[BDEVNAME_SIZE];
2157 if (bio_check_eod(bio, nr_sectors))
2160 q = bio->bi_disk->queue;
2163 "generic_make_request: Trying to access "
2164 "nonexistent block-device %s (%Lu)\n",
2165 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2170 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2171 * if queue is not a request based queue.
2173 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2176 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2179 if (!bio->bi_partno) {
2180 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2183 if (blk_partition_remap(bio))
2187 if (bio_check_eod(bio, nr_sectors))
2191 * Filter flush bio's early so that make_request based
2192 * drivers without flush support don't have to worry
2195 if (op_is_flush(bio->bi_opf) &&
2196 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2197 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2199 status = BLK_STS_OK;
2204 switch (bio_op(bio)) {
2205 case REQ_OP_DISCARD:
2206 if (!blk_queue_discard(q))
2209 case REQ_OP_SECURE_ERASE:
2210 if (!blk_queue_secure_erase(q))
2213 case REQ_OP_WRITE_SAME:
2214 if (!q->limits.max_write_same_sectors)
2217 case REQ_OP_ZONE_REPORT:
2218 case REQ_OP_ZONE_RESET:
2219 if (!blk_queue_is_zoned(q))
2222 case REQ_OP_WRITE_ZEROES:
2223 if (!q->limits.max_write_zeroes_sectors)
2231 * Various block parts want %current->io_context and lazy ioc
2232 * allocation ends up trading a lot of pain for a small amount of
2233 * memory. Just allocate it upfront. This may fail and block
2234 * layer knows how to live with it.
2236 create_io_context(GFP_ATOMIC, q->node);
2238 if (!blkcg_bio_issue_check(q, bio))
2241 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2242 trace_block_bio_queue(q, bio);
2243 /* Now that enqueuing has been traced, we need to trace
2244 * completion as well.
2246 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2251 status = BLK_STS_NOTSUPP;
2253 bio->bi_status = status;
2259 * generic_make_request - hand a buffer to its device driver for I/O
2260 * @bio: The bio describing the location in memory and on the device.
2262 * generic_make_request() is used to make I/O requests of block
2263 * devices. It is passed a &struct bio, which describes the I/O that needs
2266 * generic_make_request() does not return any status. The
2267 * success/failure status of the request, along with notification of
2268 * completion, is delivered asynchronously through the bio->bi_end_io
2269 * function described (one day) else where.
2271 * The caller of generic_make_request must make sure that bi_io_vec
2272 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2273 * set to describe the device address, and the
2274 * bi_end_io and optionally bi_private are set to describe how
2275 * completion notification should be signaled.
2277 * generic_make_request and the drivers it calls may use bi_next if this
2278 * bio happens to be merged with someone else, and may resubmit the bio to
2279 * a lower device by calling into generic_make_request recursively, which
2280 * means the bio should NOT be touched after the call to ->make_request_fn.
2282 blk_qc_t generic_make_request(struct bio *bio)
2285 * bio_list_on_stack[0] contains bios submitted by the current
2287 * bio_list_on_stack[1] contains bios that were submitted before
2288 * the current make_request_fn, but that haven't been processed
2291 struct bio_list bio_list_on_stack[2];
2292 blk_qc_t ret = BLK_QC_T_NONE;
2294 if (!generic_make_request_checks(bio))
2298 * We only want one ->make_request_fn to be active at a time, else
2299 * stack usage with stacked devices could be a problem. So use
2300 * current->bio_list to keep a list of requests submited by a
2301 * make_request_fn function. current->bio_list is also used as a
2302 * flag to say if generic_make_request is currently active in this
2303 * task or not. If it is NULL, then no make_request is active. If
2304 * it is non-NULL, then a make_request is active, and new requests
2305 * should be added at the tail
2307 if (current->bio_list) {
2308 bio_list_add(¤t->bio_list[0], bio);
2312 /* following loop may be a bit non-obvious, and so deserves some
2314 * Before entering the loop, bio->bi_next is NULL (as all callers
2315 * ensure that) so we have a list with a single bio.
2316 * We pretend that we have just taken it off a longer list, so
2317 * we assign bio_list to a pointer to the bio_list_on_stack,
2318 * thus initialising the bio_list of new bios to be
2319 * added. ->make_request() may indeed add some more bios
2320 * through a recursive call to generic_make_request. If it
2321 * did, we find a non-NULL value in bio_list and re-enter the loop
2322 * from the top. In this case we really did just take the bio
2323 * of the top of the list (no pretending) and so remove it from
2324 * bio_list, and call into ->make_request() again.
2326 BUG_ON(bio->bi_next);
2327 bio_list_init(&bio_list_on_stack[0]);
2328 current->bio_list = bio_list_on_stack;
2330 struct request_queue *q = bio->bi_disk->queue;
2331 blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
2332 BLK_MQ_REQ_NOWAIT : 0;
2334 if (likely(blk_queue_enter(q, flags) == 0)) {
2335 struct bio_list lower, same;
2337 /* Create a fresh bio_list for all subordinate requests */
2338 bio_list_on_stack[1] = bio_list_on_stack[0];
2339 bio_list_init(&bio_list_on_stack[0]);
2340 ret = q->make_request_fn(q, bio);
2344 /* sort new bios into those for a lower level
2345 * and those for the same level
2347 bio_list_init(&lower);
2348 bio_list_init(&same);
2349 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2350 if (q == bio->bi_disk->queue)
2351 bio_list_add(&same, bio);
2353 bio_list_add(&lower, bio);
2354 /* now assemble so we handle the lowest level first */
2355 bio_list_merge(&bio_list_on_stack[0], &lower);
2356 bio_list_merge(&bio_list_on_stack[0], &same);
2357 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2359 if (unlikely(!blk_queue_dying(q) &&
2360 (bio->bi_opf & REQ_NOWAIT)))
2361 bio_wouldblock_error(bio);
2365 bio = bio_list_pop(&bio_list_on_stack[0]);
2367 current->bio_list = NULL; /* deactivate */
2372 EXPORT_SYMBOL(generic_make_request);
2375 * direct_make_request - hand a buffer directly to its device driver for I/O
2376 * @bio: The bio describing the location in memory and on the device.
2378 * This function behaves like generic_make_request(), but does not protect
2379 * against recursion. Must only be used if the called driver is known
2380 * to not call generic_make_request (or direct_make_request) again from
2381 * its make_request function. (Calling direct_make_request again from
2382 * a workqueue is perfectly fine as that doesn't recurse).
2384 blk_qc_t direct_make_request(struct bio *bio)
2386 struct request_queue *q = bio->bi_disk->queue;
2387 bool nowait = bio->bi_opf & REQ_NOWAIT;
2390 if (!generic_make_request_checks(bio))
2391 return BLK_QC_T_NONE;
2393 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2394 if (nowait && !blk_queue_dying(q))
2395 bio->bi_status = BLK_STS_AGAIN;
2397 bio->bi_status = BLK_STS_IOERR;
2399 return BLK_QC_T_NONE;
2402 ret = q->make_request_fn(q, bio);
2406 EXPORT_SYMBOL_GPL(direct_make_request);
2409 * submit_bio - submit a bio to the block device layer for I/O
2410 * @bio: The &struct bio which describes the I/O
2412 * submit_bio() is very similar in purpose to generic_make_request(), and
2413 * uses that function to do most of the work. Both are fairly rough
2414 * interfaces; @bio must be presetup and ready for I/O.
2417 blk_qc_t submit_bio(struct bio *bio)
2420 * If it's a regular read/write or a barrier with data attached,
2421 * go through the normal accounting stuff before submission.
2423 if (bio_has_data(bio)) {
2426 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2427 count = queue_logical_block_size(bio->bi_disk->queue);
2429 count = bio_sectors(bio);
2431 if (op_is_write(bio_op(bio))) {
2432 count_vm_events(PGPGOUT, count);
2434 task_io_account_read(bio->bi_iter.bi_size);
2435 count_vm_events(PGPGIN, count);
2438 if (unlikely(block_dump)) {
2439 char b[BDEVNAME_SIZE];
2440 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2441 current->comm, task_pid_nr(current),
2442 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2443 (unsigned long long)bio->bi_iter.bi_sector,
2444 bio_devname(bio, b), count);
2448 return generic_make_request(bio);
2450 EXPORT_SYMBOL(submit_bio);
2452 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2454 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2458 blk_flush_plug_list(current->plug, false);
2459 return q->poll_fn(q, cookie);
2461 EXPORT_SYMBOL_GPL(blk_poll);
2464 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2465 * for new the queue limits
2467 * @rq: the request being checked
2470 * @rq may have been made based on weaker limitations of upper-level queues
2471 * in request stacking drivers, and it may violate the limitation of @q.
2472 * Since the block layer and the underlying device driver trust @rq
2473 * after it is inserted to @q, it should be checked against @q before
2474 * the insertion using this generic function.
2476 * Request stacking drivers like request-based dm may change the queue
2477 * limits when retrying requests on other queues. Those requests need
2478 * to be checked against the new queue limits again during dispatch.
2480 static int blk_cloned_rq_check_limits(struct request_queue *q,
2483 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2484 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2489 * queue's settings related to segment counting like q->bounce_pfn
2490 * may differ from that of other stacking queues.
2491 * Recalculate it to check the request correctly on this queue's
2494 blk_recalc_rq_segments(rq);
2495 if (rq->nr_phys_segments > queue_max_segments(q)) {
2496 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2504 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2505 * @q: the queue to submit the request
2506 * @rq: the request being queued
2508 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2510 unsigned long flags;
2511 int where = ELEVATOR_INSERT_BACK;
2513 if (blk_cloned_rq_check_limits(q, rq))
2514 return BLK_STS_IOERR;
2517 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2518 return BLK_STS_IOERR;
2521 if (blk_queue_io_stat(q))
2522 blk_account_io_start(rq, true);
2524 * Since we have a scheduler attached on the top device,
2525 * bypass a potential scheduler on the bottom device for
2528 return blk_mq_request_issue_directly(rq);
2531 spin_lock_irqsave(q->queue_lock, flags);
2532 if (unlikely(blk_queue_dying(q))) {
2533 spin_unlock_irqrestore(q->queue_lock, flags);
2534 return BLK_STS_IOERR;
2538 * Submitting request must be dequeued before calling this function
2539 * because it will be linked to another request_queue
2541 BUG_ON(blk_queued_rq(rq));
2543 if (op_is_flush(rq->cmd_flags))
2544 where = ELEVATOR_INSERT_FLUSH;
2546 add_acct_request(q, rq, where);
2547 if (where == ELEVATOR_INSERT_FLUSH)
2549 spin_unlock_irqrestore(q->queue_lock, flags);
2553 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2556 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2557 * @rq: request to examine
2560 * A request could be merge of IOs which require different failure
2561 * handling. This function determines the number of bytes which
2562 * can be failed from the beginning of the request without
2563 * crossing into area which need to be retried further.
2566 * The number of bytes to fail.
2568 unsigned int blk_rq_err_bytes(const struct request *rq)
2570 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2571 unsigned int bytes = 0;
2574 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2575 return blk_rq_bytes(rq);
2578 * Currently the only 'mixing' which can happen is between
2579 * different fastfail types. We can safely fail portions
2580 * which have all the failfast bits that the first one has -
2581 * the ones which are at least as eager to fail as the first
2584 for (bio = rq->bio; bio; bio = bio->bi_next) {
2585 if ((bio->bi_opf & ff) != ff)
2587 bytes += bio->bi_iter.bi_size;
2590 /* this could lead to infinite loop */
2591 BUG_ON(blk_rq_bytes(rq) && !bytes);
2594 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2596 void blk_account_io_completion(struct request *req, unsigned int bytes)
2598 if (blk_do_io_stat(req)) {
2599 const int rw = rq_data_dir(req);
2600 struct hd_struct *part;
2603 cpu = part_stat_lock();
2605 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2610 void blk_account_io_done(struct request *req)
2613 * Account IO completion. flush_rq isn't accounted as a
2614 * normal IO on queueing nor completion. Accounting the
2615 * containing request is enough.
2617 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2618 unsigned long duration = jiffies - req->start_time;
2619 const int rw = rq_data_dir(req);
2620 struct hd_struct *part;
2623 cpu = part_stat_lock();
2626 part_stat_inc(cpu, part, ios[rw]);
2627 part_stat_add(cpu, part, ticks[rw], duration);
2628 part_round_stats(req->q, cpu, part);
2629 part_dec_in_flight(req->q, part, rw);
2631 hd_struct_put(part);
2638 * Don't process normal requests when queue is suspended
2639 * or in the process of suspending/resuming
2641 static bool blk_pm_allow_request(struct request *rq)
2643 switch (rq->q->rpm_status) {
2645 case RPM_SUSPENDING:
2646 return rq->rq_flags & RQF_PM;
2654 static bool blk_pm_allow_request(struct request *rq)
2660 void blk_account_io_start(struct request *rq, bool new_io)
2662 struct hd_struct *part;
2663 int rw = rq_data_dir(rq);
2666 if (!blk_do_io_stat(rq))
2669 cpu = part_stat_lock();
2673 part_stat_inc(cpu, part, merges[rw]);
2675 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2676 if (!hd_struct_try_get(part)) {
2678 * The partition is already being removed,
2679 * the request will be accounted on the disk only
2681 * We take a reference on disk->part0 although that
2682 * partition will never be deleted, so we can treat
2683 * it as any other partition.
2685 part = &rq->rq_disk->part0;
2686 hd_struct_get(part);
2688 part_round_stats(rq->q, cpu, part);
2689 part_inc_in_flight(rq->q, part, rw);
2696 static struct request *elv_next_request(struct request_queue *q)
2699 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2701 WARN_ON_ONCE(q->mq_ops);
2704 list_for_each_entry(rq, &q->queue_head, queuelist) {
2705 if (blk_pm_allow_request(rq))
2708 if (rq->rq_flags & RQF_SOFTBARRIER)
2713 * Flush request is running and flush request isn't queueable
2714 * in the drive, we can hold the queue till flush request is
2715 * finished. Even we don't do this, driver can't dispatch next
2716 * requests and will requeue them. And this can improve
2717 * throughput too. For example, we have request flush1, write1,
2718 * flush 2. flush1 is dispatched, then queue is hold, write1
2719 * isn't inserted to queue. After flush1 is finished, flush2
2720 * will be dispatched. Since disk cache is already clean,
2721 * flush2 will be finished very soon, so looks like flush2 is
2723 * Since the queue is hold, a flag is set to indicate the queue
2724 * should be restarted later. Please see flush_end_io() for
2727 if (fq->flush_pending_idx != fq->flush_running_idx &&
2728 !queue_flush_queueable(q)) {
2729 fq->flush_queue_delayed = 1;
2732 if (unlikely(blk_queue_bypass(q)) ||
2733 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2739 * blk_peek_request - peek at the top of a request queue
2740 * @q: request queue to peek at
2743 * Return the request at the top of @q. The returned request
2744 * should be started using blk_start_request() before LLD starts
2748 * Pointer to the request at the top of @q if available. Null
2751 struct request *blk_peek_request(struct request_queue *q)
2756 lockdep_assert_held(q->queue_lock);
2757 WARN_ON_ONCE(q->mq_ops);
2759 while ((rq = elv_next_request(q)) != NULL) {
2760 if (!(rq->rq_flags & RQF_STARTED)) {
2762 * This is the first time the device driver
2763 * sees this request (possibly after
2764 * requeueing). Notify IO scheduler.
2766 if (rq->rq_flags & RQF_SORTED)
2767 elv_activate_rq(q, rq);
2770 * just mark as started even if we don't start
2771 * it, a request that has been delayed should
2772 * not be passed by new incoming requests
2774 rq->rq_flags |= RQF_STARTED;
2775 trace_block_rq_issue(q, rq);
2778 if (!q->boundary_rq || q->boundary_rq == rq) {
2779 q->end_sector = rq_end_sector(rq);
2780 q->boundary_rq = NULL;
2783 if (rq->rq_flags & RQF_DONTPREP)
2786 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2788 * make sure space for the drain appears we
2789 * know we can do this because max_hw_segments
2790 * has been adjusted to be one fewer than the
2793 rq->nr_phys_segments++;
2799 ret = q->prep_rq_fn(q, rq);
2800 if (ret == BLKPREP_OK) {
2802 } else if (ret == BLKPREP_DEFER) {
2804 * the request may have been (partially) prepped.
2805 * we need to keep this request in the front to
2806 * avoid resource deadlock. RQF_STARTED will
2807 * prevent other fs requests from passing this one.
2809 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2810 !(rq->rq_flags & RQF_DONTPREP)) {
2812 * remove the space for the drain we added
2813 * so that we don't add it again
2815 --rq->nr_phys_segments;
2820 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2821 rq->rq_flags |= RQF_QUIET;
2823 * Mark this request as started so we don't trigger
2824 * any debug logic in the end I/O path.
2826 blk_start_request(rq);
2827 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2828 BLK_STS_TARGET : BLK_STS_IOERR);
2830 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2837 EXPORT_SYMBOL(blk_peek_request);
2839 static void blk_dequeue_request(struct request *rq)
2841 struct request_queue *q = rq->q;
2843 BUG_ON(list_empty(&rq->queuelist));
2844 BUG_ON(ELV_ON_HASH(rq));
2846 list_del_init(&rq->queuelist);
2849 * the time frame between a request being removed from the lists
2850 * and to it is freed is accounted as io that is in progress at
2853 if (blk_account_rq(rq)) {
2854 q->in_flight[rq_is_sync(rq)]++;
2855 set_io_start_time_ns(rq);
2860 * blk_start_request - start request processing on the driver
2861 * @req: request to dequeue
2864 * Dequeue @req and start timeout timer on it. This hands off the
2865 * request to the driver.
2867 void blk_start_request(struct request *req)
2869 lockdep_assert_held(req->q->queue_lock);
2870 WARN_ON_ONCE(req->q->mq_ops);
2872 blk_dequeue_request(req);
2874 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2875 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2876 req->rq_flags |= RQF_STATS;
2877 wbt_issue(req->q->rq_wb, &req->issue_stat);
2880 BUG_ON(blk_rq_is_complete(req));
2883 EXPORT_SYMBOL(blk_start_request);
2886 * blk_fetch_request - fetch a request from a request queue
2887 * @q: request queue to fetch a request from
2890 * Return the request at the top of @q. The request is started on
2891 * return and LLD can start processing it immediately.
2894 * Pointer to the request at the top of @q if available. Null
2897 struct request *blk_fetch_request(struct request_queue *q)
2901 lockdep_assert_held(q->queue_lock);
2902 WARN_ON_ONCE(q->mq_ops);
2904 rq = blk_peek_request(q);
2906 blk_start_request(rq);
2909 EXPORT_SYMBOL(blk_fetch_request);
2912 * Steal bios from a request and add them to a bio list.
2913 * The request must not have been partially completed before.
2915 void blk_steal_bios(struct bio_list *list, struct request *rq)
2919 list->tail->bi_next = rq->bio;
2921 list->head = rq->bio;
2922 list->tail = rq->biotail;
2930 EXPORT_SYMBOL_GPL(blk_steal_bios);
2933 * blk_update_request - Special helper function for request stacking drivers
2934 * @req: the request being processed
2935 * @error: block status code
2936 * @nr_bytes: number of bytes to complete @req
2939 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2940 * the request structure even if @req doesn't have leftover.
2941 * If @req has leftover, sets it up for the next range of segments.
2943 * This special helper function is only for request stacking drivers
2944 * (e.g. request-based dm) so that they can handle partial completion.
2945 * Actual device drivers should use blk_end_request instead.
2947 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2948 * %false return from this function.
2951 * %false - this request doesn't have any more data
2952 * %true - this request has more data
2954 bool blk_update_request(struct request *req, blk_status_t error,
2955 unsigned int nr_bytes)
2959 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
2964 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2965 !(req->rq_flags & RQF_QUIET)))
2966 print_req_error(req, error);
2968 blk_account_io_completion(req, nr_bytes);
2972 struct bio *bio = req->bio;
2973 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2975 if (bio_bytes == bio->bi_iter.bi_size)
2976 req->bio = bio->bi_next;
2978 /* Completion has already been traced */
2979 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
2980 req_bio_endio(req, bio, bio_bytes, error);
2982 total_bytes += bio_bytes;
2983 nr_bytes -= bio_bytes;
2994 * Reset counters so that the request stacking driver
2995 * can find how many bytes remain in the request
2998 req->__data_len = 0;
3002 req->__data_len -= total_bytes;
3004 /* update sector only for requests with clear definition of sector */
3005 if (!blk_rq_is_passthrough(req))
3006 req->__sector += total_bytes >> 9;
3008 /* mixed attributes always follow the first bio */
3009 if (req->rq_flags & RQF_MIXED_MERGE) {
3010 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3011 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3014 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3016 * If total number of sectors is less than the first segment
3017 * size, something has gone terribly wrong.
3019 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3020 blk_dump_rq_flags(req, "request botched");
3021 req->__data_len = blk_rq_cur_bytes(req);
3024 /* recalculate the number of segments */
3025 blk_recalc_rq_segments(req);
3030 EXPORT_SYMBOL_GPL(blk_update_request);
3032 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3033 unsigned int nr_bytes,
3034 unsigned int bidi_bytes)
3036 if (blk_update_request(rq, error, nr_bytes))
3039 /* Bidi request must be completed as a whole */
3040 if (unlikely(blk_bidi_rq(rq)) &&
3041 blk_update_request(rq->next_rq, error, bidi_bytes))
3044 if (blk_queue_add_random(rq->q))
3045 add_disk_randomness(rq->rq_disk);
3051 * blk_unprep_request - unprepare a request
3054 * This function makes a request ready for complete resubmission (or
3055 * completion). It happens only after all error handling is complete,
3056 * so represents the appropriate moment to deallocate any resources
3057 * that were allocated to the request in the prep_rq_fn. The queue
3058 * lock is held when calling this.
3060 void blk_unprep_request(struct request *req)
3062 struct request_queue *q = req->q;
3064 req->rq_flags &= ~RQF_DONTPREP;
3065 if (q->unprep_rq_fn)
3066 q->unprep_rq_fn(q, req);
3068 EXPORT_SYMBOL_GPL(blk_unprep_request);
3070 void blk_finish_request(struct request *req, blk_status_t error)
3072 struct request_queue *q = req->q;
3074 lockdep_assert_held(req->q->queue_lock);
3075 WARN_ON_ONCE(q->mq_ops);
3077 if (req->rq_flags & RQF_STATS)
3080 if (req->rq_flags & RQF_QUEUED)
3081 blk_queue_end_tag(q, req);
3083 BUG_ON(blk_queued_rq(req));
3085 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3086 laptop_io_completion(req->q->backing_dev_info);
3088 blk_delete_timer(req);
3090 if (req->rq_flags & RQF_DONTPREP)
3091 blk_unprep_request(req);
3093 blk_account_io_done(req);
3096 wbt_done(req->q->rq_wb, &req->issue_stat);
3097 req->end_io(req, error);
3099 if (blk_bidi_rq(req))
3100 __blk_put_request(req->next_rq->q, req->next_rq);
3102 __blk_put_request(q, req);
3105 EXPORT_SYMBOL(blk_finish_request);
3108 * blk_end_bidi_request - Complete a bidi request
3109 * @rq: the request to complete
3110 * @error: block status code
3111 * @nr_bytes: number of bytes to complete @rq
3112 * @bidi_bytes: number of bytes to complete @rq->next_rq
3115 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3116 * Drivers that supports bidi can safely call this member for any
3117 * type of request, bidi or uni. In the later case @bidi_bytes is
3121 * %false - we are done with this request
3122 * %true - still buffers pending for this request
3124 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3125 unsigned int nr_bytes, unsigned int bidi_bytes)
3127 struct request_queue *q = rq->q;
3128 unsigned long flags;
3130 WARN_ON_ONCE(q->mq_ops);
3132 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3135 spin_lock_irqsave(q->queue_lock, flags);
3136 blk_finish_request(rq, error);
3137 spin_unlock_irqrestore(q->queue_lock, flags);
3143 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3144 * @rq: the request to complete
3145 * @error: block status code
3146 * @nr_bytes: number of bytes to complete @rq
3147 * @bidi_bytes: number of bytes to complete @rq->next_rq
3150 * Identical to blk_end_bidi_request() except that queue lock is
3151 * assumed to be locked on entry and remains so on return.
3154 * %false - we are done with this request
3155 * %true - still buffers pending for this request
3157 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3158 unsigned int nr_bytes, unsigned int bidi_bytes)
3160 lockdep_assert_held(rq->q->queue_lock);
3161 WARN_ON_ONCE(rq->q->mq_ops);
3163 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3166 blk_finish_request(rq, error);
3172 * blk_end_request - Helper function for drivers to complete the request.
3173 * @rq: the request being processed
3174 * @error: block status code
3175 * @nr_bytes: number of bytes to complete
3178 * Ends I/O on a number of bytes attached to @rq.
3179 * If @rq has leftover, sets it up for the next range of segments.
3182 * %false - we are done with this request
3183 * %true - still buffers pending for this request
3185 bool blk_end_request(struct request *rq, blk_status_t error,
3186 unsigned int nr_bytes)
3188 WARN_ON_ONCE(rq->q->mq_ops);
3189 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3191 EXPORT_SYMBOL(blk_end_request);
3194 * blk_end_request_all - Helper function for drives to finish the request.
3195 * @rq: the request to finish
3196 * @error: block status code
3199 * Completely finish @rq.
3201 void blk_end_request_all(struct request *rq, blk_status_t error)
3204 unsigned int bidi_bytes = 0;
3206 if (unlikely(blk_bidi_rq(rq)))
3207 bidi_bytes = blk_rq_bytes(rq->next_rq);
3209 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3212 EXPORT_SYMBOL(blk_end_request_all);
3215 * __blk_end_request - Helper function for drivers to complete the request.
3216 * @rq: the request being processed
3217 * @error: block status code
3218 * @nr_bytes: number of bytes to complete
3221 * Must be called with queue lock held unlike blk_end_request().
3224 * %false - we are done with this request
3225 * %true - still buffers pending for this request
3227 bool __blk_end_request(struct request *rq, blk_status_t error,
3228 unsigned int nr_bytes)
3230 lockdep_assert_held(rq->q->queue_lock);
3231 WARN_ON_ONCE(rq->q->mq_ops);
3233 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3235 EXPORT_SYMBOL(__blk_end_request);
3238 * __blk_end_request_all - Helper function for drives to finish the request.
3239 * @rq: the request to finish
3240 * @error: block status code
3243 * Completely finish @rq. Must be called with queue lock held.
3245 void __blk_end_request_all(struct request *rq, blk_status_t error)
3248 unsigned int bidi_bytes = 0;
3250 lockdep_assert_held(rq->q->queue_lock);
3251 WARN_ON_ONCE(rq->q->mq_ops);
3253 if (unlikely(blk_bidi_rq(rq)))
3254 bidi_bytes = blk_rq_bytes(rq->next_rq);
3256 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3259 EXPORT_SYMBOL(__blk_end_request_all);
3262 * __blk_end_request_cur - Helper function to finish the current request chunk.
3263 * @rq: the request to finish the current chunk for
3264 * @error: block status code
3267 * Complete the current consecutively mapped chunk from @rq. Must
3268 * be called with queue lock held.
3271 * %false - we are done with this request
3272 * %true - still buffers pending for this request
3274 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3276 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3278 EXPORT_SYMBOL(__blk_end_request_cur);
3280 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3283 if (bio_has_data(bio))
3284 rq->nr_phys_segments = bio_phys_segments(q, bio);
3286 rq->__data_len = bio->bi_iter.bi_size;
3287 rq->bio = rq->biotail = bio;
3290 rq->rq_disk = bio->bi_disk;
3293 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3295 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3296 * @rq: the request to be flushed
3299 * Flush all pages in @rq.
3301 void rq_flush_dcache_pages(struct request *rq)
3303 struct req_iterator iter;
3304 struct bio_vec bvec;
3306 rq_for_each_segment(bvec, rq, iter)
3307 flush_dcache_page(bvec.bv_page);
3309 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3313 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3314 * @q : the queue of the device being checked
3317 * Check if underlying low-level drivers of a device are busy.
3318 * If the drivers want to export their busy state, they must set own
3319 * exporting function using blk_queue_lld_busy() first.
3321 * Basically, this function is used only by request stacking drivers
3322 * to stop dispatching requests to underlying devices when underlying
3323 * devices are busy. This behavior helps more I/O merging on the queue
3324 * of the request stacking driver and prevents I/O throughput regression
3325 * on burst I/O load.
3328 * 0 - Not busy (The request stacking driver should dispatch request)
3329 * 1 - Busy (The request stacking driver should stop dispatching request)
3331 int blk_lld_busy(struct request_queue *q)
3334 return q->lld_busy_fn(q);
3338 EXPORT_SYMBOL_GPL(blk_lld_busy);
3341 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3342 * @rq: the clone request to be cleaned up
3345 * Free all bios in @rq for a cloned request.
3347 void blk_rq_unprep_clone(struct request *rq)
3351 while ((bio = rq->bio) != NULL) {
3352 rq->bio = bio->bi_next;
3357 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3360 * Copy attributes of the original request to the clone request.
3361 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3363 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3365 dst->cpu = src->cpu;
3366 dst->__sector = blk_rq_pos(src);
3367 dst->__data_len = blk_rq_bytes(src);
3368 dst->nr_phys_segments = src->nr_phys_segments;
3369 dst->ioprio = src->ioprio;
3370 dst->extra_len = src->extra_len;
3374 * blk_rq_prep_clone - Helper function to setup clone request
3375 * @rq: the request to be setup
3376 * @rq_src: original request to be cloned
3377 * @bs: bio_set that bios for clone are allocated from
3378 * @gfp_mask: memory allocation mask for bio
3379 * @bio_ctr: setup function to be called for each clone bio.
3380 * Returns %0 for success, non %0 for failure.
3381 * @data: private data to be passed to @bio_ctr
3384 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3385 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3386 * are not copied, and copying such parts is the caller's responsibility.
3387 * Also, pages which the original bios are pointing to are not copied
3388 * and the cloned bios just point same pages.
3389 * So cloned bios must be completed before original bios, which means
3390 * the caller must complete @rq before @rq_src.
3392 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3393 struct bio_set *bs, gfp_t gfp_mask,
3394 int (*bio_ctr)(struct bio *, struct bio *, void *),
3397 struct bio *bio, *bio_src;
3402 __rq_for_each_bio(bio_src, rq_src) {
3403 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3407 if (bio_ctr && bio_ctr(bio, bio_src, data))
3411 rq->biotail->bi_next = bio;
3414 rq->bio = rq->biotail = bio;
3417 __blk_rq_prep_clone(rq, rq_src);
3424 blk_rq_unprep_clone(rq);
3428 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3430 int kblockd_schedule_work(struct work_struct *work)
3432 return queue_work(kblockd_workqueue, work);
3434 EXPORT_SYMBOL(kblockd_schedule_work);
3436 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3438 return queue_work_on(cpu, kblockd_workqueue, work);
3440 EXPORT_SYMBOL(kblockd_schedule_work_on);
3442 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3443 unsigned long delay)
3445 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3447 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3450 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3451 * @plug: The &struct blk_plug that needs to be initialized
3454 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3455 * pending I/O should the task end up blocking between blk_start_plug() and
3456 * blk_finish_plug(). This is important from a performance perspective, but
3457 * also ensures that we don't deadlock. For instance, if the task is blocking
3458 * for a memory allocation, memory reclaim could end up wanting to free a
3459 * page belonging to that request that is currently residing in our private
3460 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3461 * this kind of deadlock.
3463 void blk_start_plug(struct blk_plug *plug)
3465 struct task_struct *tsk = current;
3468 * If this is a nested plug, don't actually assign it.
3473 INIT_LIST_HEAD(&plug->list);
3474 INIT_LIST_HEAD(&plug->mq_list);
3475 INIT_LIST_HEAD(&plug->cb_list);
3477 * Store ordering should not be needed here, since a potential
3478 * preempt will imply a full memory barrier
3482 EXPORT_SYMBOL(blk_start_plug);
3484 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3486 struct request *rqa = container_of(a, struct request, queuelist);
3487 struct request *rqb = container_of(b, struct request, queuelist);
3489 return !(rqa->q < rqb->q ||
3490 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3494 * If 'from_schedule' is true, then postpone the dispatch of requests
3495 * until a safe kblockd context. We due this to avoid accidental big
3496 * additional stack usage in driver dispatch, in places where the originally
3497 * plugger did not intend it.
3499 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3501 __releases(q->queue_lock)
3503 lockdep_assert_held(q->queue_lock);
3505 trace_block_unplug(q, depth, !from_schedule);
3508 blk_run_queue_async(q);
3511 spin_unlock(q->queue_lock);
3514 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3516 LIST_HEAD(callbacks);
3518 while (!list_empty(&plug->cb_list)) {
3519 list_splice_init(&plug->cb_list, &callbacks);
3521 while (!list_empty(&callbacks)) {
3522 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3525 list_del(&cb->list);
3526 cb->callback(cb, from_schedule);
3531 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3534 struct blk_plug *plug = current->plug;
3535 struct blk_plug_cb *cb;
3540 list_for_each_entry(cb, &plug->cb_list, list)
3541 if (cb->callback == unplug && cb->data == data)
3544 /* Not currently on the callback list */
3545 BUG_ON(size < sizeof(*cb));
3546 cb = kzalloc(size, GFP_ATOMIC);
3549 cb->callback = unplug;
3550 list_add(&cb->list, &plug->cb_list);
3554 EXPORT_SYMBOL(blk_check_plugged);
3556 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3558 struct request_queue *q;
3559 unsigned long flags;
3564 flush_plug_callbacks(plug, from_schedule);
3566 if (!list_empty(&plug->mq_list))
3567 blk_mq_flush_plug_list(plug, from_schedule);
3569 if (list_empty(&plug->list))
3572 list_splice_init(&plug->list, &list);
3574 list_sort(NULL, &list, plug_rq_cmp);
3580 * Save and disable interrupts here, to avoid doing it for every
3581 * queue lock we have to take.
3583 local_irq_save(flags);
3584 while (!list_empty(&list)) {
3585 rq = list_entry_rq(list.next);
3586 list_del_init(&rq->queuelist);
3590 * This drops the queue lock
3593 queue_unplugged(q, depth, from_schedule);
3596 spin_lock(q->queue_lock);
3600 * Short-circuit if @q is dead
3602 if (unlikely(blk_queue_dying(q))) {
3603 __blk_end_request_all(rq, BLK_STS_IOERR);
3608 * rq is already accounted, so use raw insert
3610 if (op_is_flush(rq->cmd_flags))
3611 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3613 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3619 * This drops the queue lock
3622 queue_unplugged(q, depth, from_schedule);
3624 local_irq_restore(flags);
3627 void blk_finish_plug(struct blk_plug *plug)
3629 if (plug != current->plug)
3631 blk_flush_plug_list(plug, false);
3633 current->plug = NULL;
3635 EXPORT_SYMBOL(blk_finish_plug);
3639 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3640 * @q: the queue of the device
3641 * @dev: the device the queue belongs to
3644 * Initialize runtime-PM-related fields for @q and start auto suspend for
3645 * @dev. Drivers that want to take advantage of request-based runtime PM
3646 * should call this function after @dev has been initialized, and its
3647 * request queue @q has been allocated, and runtime PM for it can not happen
3648 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3649 * cases, driver should call this function before any I/O has taken place.
3651 * This function takes care of setting up using auto suspend for the device,
3652 * the autosuspend delay is set to -1 to make runtime suspend impossible
3653 * until an updated value is either set by user or by driver. Drivers do
3654 * not need to touch other autosuspend settings.
3656 * The block layer runtime PM is request based, so only works for drivers
3657 * that use request as their IO unit instead of those directly use bio's.
3659 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3661 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3666 q->rpm_status = RPM_ACTIVE;
3667 pm_runtime_set_autosuspend_delay(q->dev, -1);
3668 pm_runtime_use_autosuspend(q->dev);
3670 EXPORT_SYMBOL(blk_pm_runtime_init);
3673 * blk_pre_runtime_suspend - Pre runtime suspend check
3674 * @q: the queue of the device
3677 * This function will check if runtime suspend is allowed for the device
3678 * by examining if there are any requests pending in the queue. If there
3679 * are requests pending, the device can not be runtime suspended; otherwise,
3680 * the queue's status will be updated to SUSPENDING and the driver can
3681 * proceed to suspend the device.
3683 * For the not allowed case, we mark last busy for the device so that
3684 * runtime PM core will try to autosuspend it some time later.
3686 * This function should be called near the start of the device's
3687 * runtime_suspend callback.
3690 * 0 - OK to runtime suspend the device
3691 * -EBUSY - Device should not be runtime suspended
3693 int blk_pre_runtime_suspend(struct request_queue *q)
3700 spin_lock_irq(q->queue_lock);
3701 if (q->nr_pending) {
3703 pm_runtime_mark_last_busy(q->dev);
3705 q->rpm_status = RPM_SUSPENDING;
3707 spin_unlock_irq(q->queue_lock);
3710 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3713 * blk_post_runtime_suspend - Post runtime suspend processing
3714 * @q: the queue of the device
3715 * @err: return value of the device's runtime_suspend function
3718 * Update the queue's runtime status according to the return value of the
3719 * device's runtime suspend function and mark last busy for the device so
3720 * that PM core will try to auto suspend the device at a later time.
3722 * This function should be called near the end of the device's
3723 * runtime_suspend callback.
3725 void blk_post_runtime_suspend(struct request_queue *q, int err)
3730 spin_lock_irq(q->queue_lock);
3732 q->rpm_status = RPM_SUSPENDED;
3734 q->rpm_status = RPM_ACTIVE;
3735 pm_runtime_mark_last_busy(q->dev);
3737 spin_unlock_irq(q->queue_lock);
3739 EXPORT_SYMBOL(blk_post_runtime_suspend);
3742 * blk_pre_runtime_resume - Pre runtime resume processing
3743 * @q: the queue of the device
3746 * Update the queue's runtime status to RESUMING in preparation for the
3747 * runtime resume of the device.
3749 * This function should be called near the start of the device's
3750 * runtime_resume callback.
3752 void blk_pre_runtime_resume(struct request_queue *q)
3757 spin_lock_irq(q->queue_lock);
3758 q->rpm_status = RPM_RESUMING;
3759 spin_unlock_irq(q->queue_lock);
3761 EXPORT_SYMBOL(blk_pre_runtime_resume);
3764 * blk_post_runtime_resume - Post runtime resume processing
3765 * @q: the queue of the device
3766 * @err: return value of the device's runtime_resume function
3769 * Update the queue's runtime status according to the return value of the
3770 * device's runtime_resume function. If it is successfully resumed, process
3771 * the requests that are queued into the device's queue when it is resuming
3772 * and then mark last busy and initiate autosuspend for it.
3774 * This function should be called near the end of the device's
3775 * runtime_resume callback.
3777 void blk_post_runtime_resume(struct request_queue *q, int err)
3782 spin_lock_irq(q->queue_lock);
3784 q->rpm_status = RPM_ACTIVE;
3786 pm_runtime_mark_last_busy(q->dev);
3787 pm_request_autosuspend(q->dev);
3789 q->rpm_status = RPM_SUSPENDED;
3791 spin_unlock_irq(q->queue_lock);
3793 EXPORT_SYMBOL(blk_post_runtime_resume);
3796 * blk_set_runtime_active - Force runtime status of the queue to be active
3797 * @q: the queue of the device
3799 * If the device is left runtime suspended during system suspend the resume
3800 * hook typically resumes the device and corrects runtime status
3801 * accordingly. However, that does not affect the queue runtime PM status
3802 * which is still "suspended". This prevents processing requests from the
3805 * This function can be used in driver's resume hook to correct queue
3806 * runtime PM status and re-enable peeking requests from the queue. It
3807 * should be called before first request is added to the queue.
3809 void blk_set_runtime_active(struct request_queue *q)
3811 spin_lock_irq(q->queue_lock);
3812 q->rpm_status = RPM_ACTIVE;
3813 pm_runtime_mark_last_busy(q->dev);
3814 pm_request_autosuspend(q->dev);
3815 spin_unlock_irq(q->queue_lock);
3817 EXPORT_SYMBOL(blk_set_runtime_active);
3820 int __init blk_dev_init(void)
3822 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3823 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3824 FIELD_SIZEOF(struct request, cmd_flags));
3825 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3826 FIELD_SIZEOF(struct bio, bi_opf));
3828 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3829 kblockd_workqueue = alloc_workqueue("kblockd",
3830 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3831 if (!kblockd_workqueue)
3832 panic("Failed to create kblockd\n");
3834 request_cachep = kmem_cache_create("blkdev_requests",
3835 sizeof(struct request), 0, SLAB_PANIC, NULL);
3837 blk_requestq_cachep = kmem_cache_create("request_queue",
3838 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3840 #ifdef CONFIG_DEBUG_FS
3841 blk_debugfs_root = debugfs_create_dir("block", NULL);