2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/block.h>
43 #include "blk-mq-sched.h"
46 #ifdef CONFIG_DEBUG_FS
47 struct dentry *blk_debugfs_root;
50 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
51 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
56 DEFINE_IDA(blk_queue_ida);
59 * For the allocated request tables
61 struct kmem_cache *request_cachep;
64 * For queue allocation
66 struct kmem_cache *blk_requestq_cachep;
69 * Controlling structure to kblockd
71 static struct workqueue_struct *kblockd_workqueue;
73 static void blk_clear_congested(struct request_list *rl, int sync)
75 #ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl->blkg->wb_congested, sync);
79 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
80 * flip its congestion state for events on other blkcgs.
82 if (rl == &rl->q->root_rl)
83 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
87 static void blk_set_congested(struct request_list *rl, int sync)
89 #ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl->blkg->wb_congested, sync);
92 /* see blk_clear_congested() */
93 if (rl == &rl->q->root_rl)
94 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
98 void blk_queue_congestion_threshold(struct request_queue *q)
102 nr = q->nr_requests - (q->nr_requests / 8) + 1;
103 if (nr > q->nr_requests)
105 q->nr_congestion_on = nr;
107 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
110 q->nr_congestion_off = nr;
113 void blk_rq_init(struct request_queue *q, struct request *rq)
115 memset(rq, 0, sizeof(*rq));
117 INIT_LIST_HEAD(&rq->queuelist);
118 INIT_LIST_HEAD(&rq->timeout_list);
121 rq->__sector = (sector_t) -1;
122 INIT_HLIST_NODE(&rq->hash);
123 RB_CLEAR_NODE(&rq->rb_node);
125 rq->internal_tag = -1;
126 rq->start_time = jiffies;
127 set_start_time_ns(rq);
130 EXPORT_SYMBOL(blk_rq_init);
132 static const struct {
136 [BLK_STS_OK] = { 0, "" },
137 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
138 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
139 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
140 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
141 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
142 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
143 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
144 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
145 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
146 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
148 /* device mapper special case, should not leak out: */
149 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
151 /* everything else not covered above: */
152 [BLK_STS_IOERR] = { -EIO, "I/O" },
155 blk_status_t errno_to_blk_status(int errno)
159 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
160 if (blk_errors[i].errno == errno)
161 return (__force blk_status_t)i;
164 return BLK_STS_IOERR;
166 EXPORT_SYMBOL_GPL(errno_to_blk_status);
168 int blk_status_to_errno(blk_status_t status)
170 int idx = (__force int)status;
172 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
174 return blk_errors[idx].errno;
176 EXPORT_SYMBOL_GPL(blk_status_to_errno);
178 static void print_req_error(struct request *req, blk_status_t status)
180 int idx = (__force int)status;
182 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
185 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
186 __func__, blk_errors[idx].name, req->rq_disk ?
187 req->rq_disk->disk_name : "?",
188 (unsigned long long)blk_rq_pos(req));
191 static void req_bio_endio(struct request *rq, struct bio *bio,
192 unsigned int nbytes, blk_status_t error)
195 bio->bi_status = error;
197 if (unlikely(rq->rq_flags & RQF_QUIET))
198 bio_set_flag(bio, BIO_QUIET);
200 bio_advance(bio, nbytes);
202 /* don't actually finish bio if it's part of flush sequence */
203 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
207 void blk_dump_rq_flags(struct request *rq, char *msg)
209 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
210 rq->rq_disk ? rq->rq_disk->disk_name : "?",
211 (unsigned long long) rq->cmd_flags);
213 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
214 (unsigned long long)blk_rq_pos(rq),
215 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
216 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
217 rq->bio, rq->biotail, blk_rq_bytes(rq));
219 EXPORT_SYMBOL(blk_dump_rq_flags);
221 static void blk_delay_work(struct work_struct *work)
223 struct request_queue *q;
225 q = container_of(work, struct request_queue, delay_work.work);
226 spin_lock_irq(q->queue_lock);
228 spin_unlock_irq(q->queue_lock);
232 * blk_delay_queue - restart queueing after defined interval
233 * @q: The &struct request_queue in question
234 * @msecs: Delay in msecs
237 * Sometimes queueing needs to be postponed for a little while, to allow
238 * resources to come back. This function will make sure that queueing is
239 * restarted around the specified time.
241 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
243 lockdep_assert_held(q->queue_lock);
244 WARN_ON_ONCE(q->mq_ops);
246 if (likely(!blk_queue_dead(q)))
247 queue_delayed_work(kblockd_workqueue, &q->delay_work,
248 msecs_to_jiffies(msecs));
250 EXPORT_SYMBOL(blk_delay_queue);
253 * blk_start_queue_async - asynchronously restart a previously stopped queue
254 * @q: The &struct request_queue in question
257 * blk_start_queue_async() will clear the stop flag on the queue, and
258 * ensure that the request_fn for the queue is run from an async
261 void blk_start_queue_async(struct request_queue *q)
263 lockdep_assert_held(q->queue_lock);
264 WARN_ON_ONCE(q->mq_ops);
266 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
267 blk_run_queue_async(q);
269 EXPORT_SYMBOL(blk_start_queue_async);
272 * blk_start_queue - restart a previously stopped queue
273 * @q: The &struct request_queue in question
276 * blk_start_queue() will clear the stop flag on the queue, and call
277 * the request_fn for the queue if it was in a stopped state when
278 * entered. Also see blk_stop_queue().
280 void blk_start_queue(struct request_queue *q)
282 lockdep_assert_held(q->queue_lock);
283 WARN_ON(!in_interrupt() && !irqs_disabled());
284 WARN_ON_ONCE(q->mq_ops);
286 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
289 EXPORT_SYMBOL(blk_start_queue);
292 * blk_stop_queue - stop a queue
293 * @q: The &struct request_queue in question
296 * The Linux block layer assumes that a block driver will consume all
297 * entries on the request queue when the request_fn strategy is called.
298 * Often this will not happen, because of hardware limitations (queue
299 * depth settings). If a device driver gets a 'queue full' response,
300 * or if it simply chooses not to queue more I/O at one point, it can
301 * call this function to prevent the request_fn from being called until
302 * the driver has signalled it's ready to go again. This happens by calling
303 * blk_start_queue() to restart queue operations.
305 void blk_stop_queue(struct request_queue *q)
307 lockdep_assert_held(q->queue_lock);
308 WARN_ON_ONCE(q->mq_ops);
310 cancel_delayed_work(&q->delay_work);
311 queue_flag_set(QUEUE_FLAG_STOPPED, q);
313 EXPORT_SYMBOL(blk_stop_queue);
316 * blk_sync_queue - cancel any pending callbacks on a queue
320 * The block layer may perform asynchronous callback activity
321 * on a queue, such as calling the unplug function after a timeout.
322 * A block device may call blk_sync_queue to ensure that any
323 * such activity is cancelled, thus allowing it to release resources
324 * that the callbacks might use. The caller must already have made sure
325 * that its ->make_request_fn will not re-add plugging prior to calling
328 * This function does not cancel any asynchronous activity arising
329 * out of elevator or throttling code. That would require elevator_exit()
330 * and blkcg_exit_queue() to be called with queue lock initialized.
333 void blk_sync_queue(struct request_queue *q)
335 del_timer_sync(&q->timeout);
336 cancel_work_sync(&q->timeout_work);
339 struct blk_mq_hw_ctx *hctx;
342 cancel_delayed_work_sync(&q->requeue_work);
343 queue_for_each_hw_ctx(q, hctx, i)
344 cancel_delayed_work_sync(&hctx->run_work);
346 cancel_delayed_work_sync(&q->delay_work);
349 EXPORT_SYMBOL(blk_sync_queue);
352 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
353 * @q: The queue to run
356 * Invoke request handling on a queue if there are any pending requests.
357 * May be used to restart request handling after a request has completed.
358 * This variant runs the queue whether or not the queue has been
359 * stopped. Must be called with the queue lock held and interrupts
360 * disabled. See also @blk_run_queue.
362 inline void __blk_run_queue_uncond(struct request_queue *q)
364 lockdep_assert_held(q->queue_lock);
365 WARN_ON_ONCE(q->mq_ops);
367 if (unlikely(blk_queue_dead(q)))
371 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
372 * the queue lock internally. As a result multiple threads may be
373 * running such a request function concurrently. Keep track of the
374 * number of active request_fn invocations such that blk_drain_queue()
375 * can wait until all these request_fn calls have finished.
377 q->request_fn_active++;
379 q->request_fn_active--;
381 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
384 * __blk_run_queue - run a single device queue
385 * @q: The queue to run
388 * See @blk_run_queue.
390 void __blk_run_queue(struct request_queue *q)
392 lockdep_assert_held(q->queue_lock);
393 WARN_ON_ONCE(q->mq_ops);
395 if (unlikely(blk_queue_stopped(q)))
398 __blk_run_queue_uncond(q);
400 EXPORT_SYMBOL(__blk_run_queue);
403 * blk_run_queue_async - run a single device queue in workqueue context
404 * @q: The queue to run
407 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
411 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
412 * has canceled q->delay_work, callers must hold the queue lock to avoid
413 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
415 void blk_run_queue_async(struct request_queue *q)
417 lockdep_assert_held(q->queue_lock);
418 WARN_ON_ONCE(q->mq_ops);
420 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
421 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
423 EXPORT_SYMBOL(blk_run_queue_async);
426 * blk_run_queue - run a single device queue
427 * @q: The queue to run
430 * Invoke request handling on this queue, if it has pending work to do.
431 * May be used to restart queueing when a request has completed.
433 void blk_run_queue(struct request_queue *q)
437 WARN_ON_ONCE(q->mq_ops);
439 spin_lock_irqsave(q->queue_lock, flags);
441 spin_unlock_irqrestore(q->queue_lock, flags);
443 EXPORT_SYMBOL(blk_run_queue);
445 void blk_put_queue(struct request_queue *q)
447 kobject_put(&q->kobj);
449 EXPORT_SYMBOL(blk_put_queue);
452 * __blk_drain_queue - drain requests from request_queue
454 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
456 * Drain requests from @q. If @drain_all is set, all requests are drained.
457 * If not, only ELVPRIV requests are drained. The caller is responsible
458 * for ensuring that no new requests which need to be drained are queued.
460 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
461 __releases(q->queue_lock)
462 __acquires(q->queue_lock)
466 lockdep_assert_held(q->queue_lock);
467 WARN_ON_ONCE(q->mq_ops);
473 * The caller might be trying to drain @q before its
474 * elevator is initialized.
477 elv_drain_elevator(q);
479 blkcg_drain_queue(q);
482 * This function might be called on a queue which failed
483 * driver init after queue creation or is not yet fully
484 * active yet. Some drivers (e.g. fd and loop) get unhappy
485 * in such cases. Kick queue iff dispatch queue has
486 * something on it and @q has request_fn set.
488 if (!list_empty(&q->queue_head) && q->request_fn)
491 drain |= q->nr_rqs_elvpriv;
492 drain |= q->request_fn_active;
495 * Unfortunately, requests are queued at and tracked from
496 * multiple places and there's no single counter which can
497 * be drained. Check all the queues and counters.
500 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
501 drain |= !list_empty(&q->queue_head);
502 for (i = 0; i < 2; i++) {
503 drain |= q->nr_rqs[i];
504 drain |= q->in_flight[i];
506 drain |= !list_empty(&fq->flush_queue[i]);
513 spin_unlock_irq(q->queue_lock);
517 spin_lock_irq(q->queue_lock);
521 * With queue marked dead, any woken up waiter will fail the
522 * allocation path, so the wakeup chaining is lost and we're
523 * left with hung waiters. We need to wake up those waiters.
526 struct request_list *rl;
528 blk_queue_for_each_rl(rl, q)
529 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
530 wake_up_all(&rl->wait[i]);
534 void blk_drain_queue(struct request_queue *q)
536 spin_lock_irq(q->queue_lock);
537 __blk_drain_queue(q, true);
538 spin_unlock_irq(q->queue_lock);
542 * blk_queue_bypass_start - enter queue bypass mode
543 * @q: queue of interest
545 * In bypass mode, only the dispatch FIFO queue of @q is used. This
546 * function makes @q enter bypass mode and drains all requests which were
547 * throttled or issued before. On return, it's guaranteed that no request
548 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
549 * inside queue or RCU read lock.
551 void blk_queue_bypass_start(struct request_queue *q)
553 WARN_ON_ONCE(q->mq_ops);
555 spin_lock_irq(q->queue_lock);
557 queue_flag_set(QUEUE_FLAG_BYPASS, q);
558 spin_unlock_irq(q->queue_lock);
561 * Queues start drained. Skip actual draining till init is
562 * complete. This avoids lenghty delays during queue init which
563 * can happen many times during boot.
565 if (blk_queue_init_done(q)) {
566 spin_lock_irq(q->queue_lock);
567 __blk_drain_queue(q, false);
568 spin_unlock_irq(q->queue_lock);
570 /* ensure blk_queue_bypass() is %true inside RCU read lock */
574 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
577 * blk_queue_bypass_end - leave queue bypass mode
578 * @q: queue of interest
580 * Leave bypass mode and restore the normal queueing behavior.
582 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
583 * this function is called for both blk-sq and blk-mq queues.
585 void blk_queue_bypass_end(struct request_queue *q)
587 spin_lock_irq(q->queue_lock);
588 if (!--q->bypass_depth)
589 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
590 WARN_ON_ONCE(q->bypass_depth < 0);
591 spin_unlock_irq(q->queue_lock);
593 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
595 void blk_set_queue_dying(struct request_queue *q)
597 spin_lock_irq(q->queue_lock);
598 queue_flag_set(QUEUE_FLAG_DYING, q);
599 spin_unlock_irq(q->queue_lock);
602 * When queue DYING flag is set, we need to block new req
603 * entering queue, so we call blk_freeze_queue_start() to
604 * prevent I/O from crossing blk_queue_enter().
606 blk_freeze_queue_start(q);
609 blk_mq_wake_waiters(q);
611 struct request_list *rl;
613 spin_lock_irq(q->queue_lock);
614 blk_queue_for_each_rl(rl, q) {
616 wake_up_all(&rl->wait[BLK_RW_SYNC]);
617 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
620 spin_unlock_irq(q->queue_lock);
623 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
626 * blk_cleanup_queue - shutdown a request queue
627 * @q: request queue to shutdown
629 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
630 * put it. All future requests will be failed immediately with -ENODEV.
632 void blk_cleanup_queue(struct request_queue *q)
634 spinlock_t *lock = q->queue_lock;
636 /* mark @q DYING, no new request or merges will be allowed afterwards */
637 mutex_lock(&q->sysfs_lock);
638 blk_set_queue_dying(q);
642 * A dying queue is permanently in bypass mode till released. Note
643 * that, unlike blk_queue_bypass_start(), we aren't performing
644 * synchronize_rcu() after entering bypass mode to avoid the delay
645 * as some drivers create and destroy a lot of queues while
646 * probing. This is still safe because blk_release_queue() will be
647 * called only after the queue refcnt drops to zero and nothing,
648 * RCU or not, would be traversing the queue by then.
651 queue_flag_set(QUEUE_FLAG_BYPASS, q);
653 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
654 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
655 queue_flag_set(QUEUE_FLAG_DYING, q);
656 spin_unlock_irq(lock);
657 mutex_unlock(&q->sysfs_lock);
660 * Drain all requests queued before DYING marking. Set DEAD flag to
661 * prevent that q->request_fn() gets invoked after draining finished.
665 queue_flag_set(QUEUE_FLAG_DEAD, q);
666 spin_unlock_irq(lock);
669 * make sure all in-progress dispatch are completed because
670 * blk_freeze_queue() can only complete all requests, and
671 * dispatch may still be in-progress since we dispatch requests
672 * from more than one contexts
675 blk_mq_quiesce_queue(q);
677 /* for synchronous bio-based driver finish in-flight integrity i/o */
678 blk_flush_integrity();
680 /* @q won't process any more request, flush async actions */
681 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
685 blk_mq_free_queue(q);
686 percpu_ref_exit(&q->q_usage_counter);
689 if (q->queue_lock != &q->__queue_lock)
690 q->queue_lock = &q->__queue_lock;
691 spin_unlock_irq(lock);
693 /* @q is and will stay empty, shutdown and put */
696 EXPORT_SYMBOL(blk_cleanup_queue);
698 /* Allocate memory local to the request queue */
699 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
701 struct request_queue *q = data;
703 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
706 static void free_request_simple(void *element, void *data)
708 kmem_cache_free(request_cachep, element);
711 static void *alloc_request_size(gfp_t gfp_mask, void *data)
713 struct request_queue *q = data;
716 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
718 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
725 static void free_request_size(void *element, void *data)
727 struct request_queue *q = data;
730 q->exit_rq_fn(q, element);
734 int blk_init_rl(struct request_list *rl, struct request_queue *q,
737 if (unlikely(rl->rq_pool))
741 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
742 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
743 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
744 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
747 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
748 alloc_request_size, free_request_size,
749 q, gfp_mask, q->node);
751 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
752 alloc_request_simple, free_request_simple,
753 q, gfp_mask, q->node);
758 if (rl != &q->root_rl)
759 WARN_ON_ONCE(!blk_get_queue(q));
764 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
767 mempool_destroy(rl->rq_pool);
768 if (rl != &q->root_rl)
773 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
775 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
777 EXPORT_SYMBOL(blk_alloc_queue);
779 int blk_queue_enter(struct request_queue *q, bool nowait)
784 if (percpu_ref_tryget_live(&q->q_usage_counter))
791 * read pair of barrier in blk_freeze_queue_start(),
792 * we need to order reading __PERCPU_REF_DEAD flag of
793 * .q_usage_counter and reading .mq_freeze_depth or
794 * queue dying flag, otherwise the following wait may
795 * never return if the two reads are reordered.
799 ret = wait_event_interruptible(q->mq_freeze_wq,
800 !atomic_read(&q->mq_freeze_depth) ||
802 if (blk_queue_dying(q))
809 void blk_queue_exit(struct request_queue *q)
811 percpu_ref_put(&q->q_usage_counter);
814 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
816 struct request_queue *q =
817 container_of(ref, struct request_queue, q_usage_counter);
819 wake_up_all(&q->mq_freeze_wq);
822 static void blk_rq_timed_out_timer(unsigned long data)
824 struct request_queue *q = (struct request_queue *)data;
826 kblockd_schedule_work(&q->timeout_work);
829 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
831 struct request_queue *q;
833 q = kmem_cache_alloc_node(blk_requestq_cachep,
834 gfp_mask | __GFP_ZERO, node_id);
838 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
842 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
846 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
847 if (!q->backing_dev_info)
850 q->stats = blk_alloc_queue_stats();
854 q->backing_dev_info->ra_pages =
855 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
856 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
857 q->backing_dev_info->name = "block";
860 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
861 laptop_mode_timer_fn, (unsigned long) q);
862 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
863 INIT_WORK(&q->timeout_work, NULL);
864 INIT_LIST_HEAD(&q->queue_head);
865 INIT_LIST_HEAD(&q->timeout_list);
866 INIT_LIST_HEAD(&q->icq_list);
867 #ifdef CONFIG_BLK_CGROUP
868 INIT_LIST_HEAD(&q->blkg_list);
870 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
872 kobject_init(&q->kobj, &blk_queue_ktype);
874 #ifdef CONFIG_BLK_DEV_IO_TRACE
875 mutex_init(&q->blk_trace_mutex);
877 mutex_init(&q->sysfs_lock);
878 spin_lock_init(&q->__queue_lock);
881 * By default initialize queue_lock to internal lock and driver can
882 * override it later if need be.
884 q->queue_lock = &q->__queue_lock;
887 * A queue starts its life with bypass turned on to avoid
888 * unnecessary bypass on/off overhead and nasty surprises during
889 * init. The initial bypass will be finished when the queue is
890 * registered by blk_register_queue().
893 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
895 init_waitqueue_head(&q->mq_freeze_wq);
898 * Init percpu_ref in atomic mode so that it's faster to shutdown.
899 * See blk_register_queue() for details.
901 if (percpu_ref_init(&q->q_usage_counter,
902 blk_queue_usage_counter_release,
903 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
906 if (blkcg_init_queue(q))
912 percpu_ref_exit(&q->q_usage_counter);
914 blk_free_queue_stats(q->stats);
916 bdi_put(q->backing_dev_info);
918 bioset_free(q->bio_split);
920 ida_simple_remove(&blk_queue_ida, q->id);
922 kmem_cache_free(blk_requestq_cachep, q);
925 EXPORT_SYMBOL(blk_alloc_queue_node);
928 * blk_init_queue - prepare a request queue for use with a block device
929 * @rfn: The function to be called to process requests that have been
930 * placed on the queue.
931 * @lock: Request queue spin lock
934 * If a block device wishes to use the standard request handling procedures,
935 * which sorts requests and coalesces adjacent requests, then it must
936 * call blk_init_queue(). The function @rfn will be called when there
937 * are requests on the queue that need to be processed. If the device
938 * supports plugging, then @rfn may not be called immediately when requests
939 * are available on the queue, but may be called at some time later instead.
940 * Plugged queues are generally unplugged when a buffer belonging to one
941 * of the requests on the queue is needed, or due to memory pressure.
943 * @rfn is not required, or even expected, to remove all requests off the
944 * queue, but only as many as it can handle at a time. If it does leave
945 * requests on the queue, it is responsible for arranging that the requests
946 * get dealt with eventually.
948 * The queue spin lock must be held while manipulating the requests on the
949 * request queue; this lock will be taken also from interrupt context, so irq
950 * disabling is needed for it.
952 * Function returns a pointer to the initialized request queue, or %NULL if
956 * blk_init_queue() must be paired with a blk_cleanup_queue() call
957 * when the block device is deactivated (such as at module unload).
960 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
962 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
964 EXPORT_SYMBOL(blk_init_queue);
966 struct request_queue *
967 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
969 struct request_queue *q;
971 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
977 q->queue_lock = lock;
978 if (blk_init_allocated_queue(q) < 0) {
979 blk_cleanup_queue(q);
985 EXPORT_SYMBOL(blk_init_queue_node);
987 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
990 int blk_init_allocated_queue(struct request_queue *q)
992 WARN_ON_ONCE(q->mq_ops);
994 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
998 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
999 goto out_free_flush_queue;
1001 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1002 goto out_exit_flush_rq;
1004 INIT_WORK(&q->timeout_work, blk_timeout_work);
1005 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1008 * This also sets hw/phys segments, boundary and size
1010 blk_queue_make_request(q, blk_queue_bio);
1012 q->sg_reserved_size = INT_MAX;
1014 /* Protect q->elevator from elevator_change */
1015 mutex_lock(&q->sysfs_lock);
1018 if (elevator_init(q, NULL)) {
1019 mutex_unlock(&q->sysfs_lock);
1020 goto out_exit_flush_rq;
1023 mutex_unlock(&q->sysfs_lock);
1028 q->exit_rq_fn(q, q->fq->flush_rq);
1029 out_free_flush_queue:
1030 blk_free_flush_queue(q->fq);
1033 EXPORT_SYMBOL(blk_init_allocated_queue);
1035 bool blk_get_queue(struct request_queue *q)
1037 if (likely(!blk_queue_dying(q))) {
1044 EXPORT_SYMBOL(blk_get_queue);
1046 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1048 if (rq->rq_flags & RQF_ELVPRIV) {
1049 elv_put_request(rl->q, rq);
1051 put_io_context(rq->elv.icq->ioc);
1054 mempool_free(rq, rl->rq_pool);
1058 * ioc_batching returns true if the ioc is a valid batching request and
1059 * should be given priority access to a request.
1061 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1067 * Make sure the process is able to allocate at least 1 request
1068 * even if the batch times out, otherwise we could theoretically
1071 return ioc->nr_batch_requests == q->nr_batching ||
1072 (ioc->nr_batch_requests > 0
1073 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1077 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1078 * will cause the process to be a "batcher" on all queues in the system. This
1079 * is the behaviour we want though - once it gets a wakeup it should be given
1082 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1084 if (!ioc || ioc_batching(q, ioc))
1087 ioc->nr_batch_requests = q->nr_batching;
1088 ioc->last_waited = jiffies;
1091 static void __freed_request(struct request_list *rl, int sync)
1093 struct request_queue *q = rl->q;
1095 if (rl->count[sync] < queue_congestion_off_threshold(q))
1096 blk_clear_congested(rl, sync);
1098 if (rl->count[sync] + 1 <= q->nr_requests) {
1099 if (waitqueue_active(&rl->wait[sync]))
1100 wake_up(&rl->wait[sync]);
1102 blk_clear_rl_full(rl, sync);
1107 * A request has just been released. Account for it, update the full and
1108 * congestion status, wake up any waiters. Called under q->queue_lock.
1110 static void freed_request(struct request_list *rl, bool sync,
1111 req_flags_t rq_flags)
1113 struct request_queue *q = rl->q;
1117 if (rq_flags & RQF_ELVPRIV)
1118 q->nr_rqs_elvpriv--;
1120 __freed_request(rl, sync);
1122 if (unlikely(rl->starved[sync ^ 1]))
1123 __freed_request(rl, sync ^ 1);
1126 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1128 struct request_list *rl;
1129 int on_thresh, off_thresh;
1131 WARN_ON_ONCE(q->mq_ops);
1133 spin_lock_irq(q->queue_lock);
1134 q->nr_requests = nr;
1135 blk_queue_congestion_threshold(q);
1136 on_thresh = queue_congestion_on_threshold(q);
1137 off_thresh = queue_congestion_off_threshold(q);
1139 blk_queue_for_each_rl(rl, q) {
1140 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1141 blk_set_congested(rl, BLK_RW_SYNC);
1142 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1143 blk_clear_congested(rl, BLK_RW_SYNC);
1145 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1146 blk_set_congested(rl, BLK_RW_ASYNC);
1147 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1148 blk_clear_congested(rl, BLK_RW_ASYNC);
1150 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1151 blk_set_rl_full(rl, BLK_RW_SYNC);
1153 blk_clear_rl_full(rl, BLK_RW_SYNC);
1154 wake_up(&rl->wait[BLK_RW_SYNC]);
1157 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1158 blk_set_rl_full(rl, BLK_RW_ASYNC);
1160 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1161 wake_up(&rl->wait[BLK_RW_ASYNC]);
1165 spin_unlock_irq(q->queue_lock);
1170 * __get_request - get a free request
1171 * @rl: request list to allocate from
1172 * @op: operation and flags
1173 * @bio: bio to allocate request for (can be %NULL)
1174 * @gfp_mask: allocation mask
1176 * Get a free request from @q. This function may fail under memory
1177 * pressure or if @q is dead.
1179 * Must be called with @q->queue_lock held and,
1180 * Returns ERR_PTR on failure, with @q->queue_lock held.
1181 * Returns request pointer on success, with @q->queue_lock *not held*.
1183 static struct request *__get_request(struct request_list *rl, unsigned int op,
1184 struct bio *bio, gfp_t gfp_mask)
1186 struct request_queue *q = rl->q;
1188 struct elevator_type *et = q->elevator->type;
1189 struct io_context *ioc = rq_ioc(bio);
1190 struct io_cq *icq = NULL;
1191 const bool is_sync = op_is_sync(op);
1193 req_flags_t rq_flags = RQF_ALLOCED;
1195 lockdep_assert_held(q->queue_lock);
1197 if (unlikely(blk_queue_dying(q)))
1198 return ERR_PTR(-ENODEV);
1200 may_queue = elv_may_queue(q, op);
1201 if (may_queue == ELV_MQUEUE_NO)
1204 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1205 if (rl->count[is_sync]+1 >= q->nr_requests) {
1207 * The queue will fill after this allocation, so set
1208 * it as full, and mark this process as "batching".
1209 * This process will be allowed to complete a batch of
1210 * requests, others will be blocked.
1212 if (!blk_rl_full(rl, is_sync)) {
1213 ioc_set_batching(q, ioc);
1214 blk_set_rl_full(rl, is_sync);
1216 if (may_queue != ELV_MQUEUE_MUST
1217 && !ioc_batching(q, ioc)) {
1219 * The queue is full and the allocating
1220 * process is not a "batcher", and not
1221 * exempted by the IO scheduler
1223 return ERR_PTR(-ENOMEM);
1227 blk_set_congested(rl, is_sync);
1231 * Only allow batching queuers to allocate up to 50% over the defined
1232 * limit of requests, otherwise we could have thousands of requests
1233 * allocated with any setting of ->nr_requests
1235 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1236 return ERR_PTR(-ENOMEM);
1238 q->nr_rqs[is_sync]++;
1239 rl->count[is_sync]++;
1240 rl->starved[is_sync] = 0;
1243 * Decide whether the new request will be managed by elevator. If
1244 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1245 * prevent the current elevator from being destroyed until the new
1246 * request is freed. This guarantees icq's won't be destroyed and
1247 * makes creating new ones safe.
1249 * Flush requests do not use the elevator so skip initialization.
1250 * This allows a request to share the flush and elevator data.
1252 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1253 * it will be created after releasing queue_lock.
1255 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1256 rq_flags |= RQF_ELVPRIV;
1257 q->nr_rqs_elvpriv++;
1258 if (et->icq_cache && ioc)
1259 icq = ioc_lookup_icq(ioc, q);
1262 if (blk_queue_io_stat(q))
1263 rq_flags |= RQF_IO_STAT;
1264 spin_unlock_irq(q->queue_lock);
1266 /* allocate and init request */
1267 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1272 blk_rq_set_rl(rq, rl);
1274 rq->rq_flags = rq_flags;
1277 if (rq_flags & RQF_ELVPRIV) {
1278 if (unlikely(et->icq_cache && !icq)) {
1280 icq = ioc_create_icq(ioc, q, gfp_mask);
1286 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1289 /* @rq->elv.icq holds io_context until @rq is freed */
1291 get_io_context(icq->ioc);
1295 * ioc may be NULL here, and ioc_batching will be false. That's
1296 * OK, if the queue is under the request limit then requests need
1297 * not count toward the nr_batch_requests limit. There will always
1298 * be some limit enforced by BLK_BATCH_TIME.
1300 if (ioc_batching(q, ioc))
1301 ioc->nr_batch_requests--;
1303 trace_block_getrq(q, bio, op);
1308 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1309 * and may fail indefinitely under memory pressure and thus
1310 * shouldn't stall IO. Treat this request as !elvpriv. This will
1311 * disturb iosched and blkcg but weird is bettern than dead.
1313 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1314 __func__, dev_name(q->backing_dev_info->dev));
1316 rq->rq_flags &= ~RQF_ELVPRIV;
1319 spin_lock_irq(q->queue_lock);
1320 q->nr_rqs_elvpriv--;
1321 spin_unlock_irq(q->queue_lock);
1326 * Allocation failed presumably due to memory. Undo anything we
1327 * might have messed up.
1329 * Allocating task should really be put onto the front of the wait
1330 * queue, but this is pretty rare.
1332 spin_lock_irq(q->queue_lock);
1333 freed_request(rl, is_sync, rq_flags);
1336 * in the very unlikely event that allocation failed and no
1337 * requests for this direction was pending, mark us starved so that
1338 * freeing of a request in the other direction will notice
1339 * us. another possible fix would be to split the rq mempool into
1343 if (unlikely(rl->count[is_sync] == 0))
1344 rl->starved[is_sync] = 1;
1345 return ERR_PTR(-ENOMEM);
1349 * get_request - get a free request
1350 * @q: request_queue to allocate request from
1351 * @op: operation and flags
1352 * @bio: bio to allocate request for (can be %NULL)
1353 * @gfp_mask: allocation mask
1355 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1356 * this function keeps retrying under memory pressure and fails iff @q is dead.
1358 * Must be called with @q->queue_lock held and,
1359 * Returns ERR_PTR on failure, with @q->queue_lock held.
1360 * Returns request pointer on success, with @q->queue_lock *not held*.
1362 static struct request *get_request(struct request_queue *q, unsigned int op,
1363 struct bio *bio, gfp_t gfp_mask)
1365 const bool is_sync = op_is_sync(op);
1367 struct request_list *rl;
1370 lockdep_assert_held(q->queue_lock);
1371 WARN_ON_ONCE(q->mq_ops);
1373 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1375 rq = __get_request(rl, op, bio, gfp_mask);
1379 if (op & REQ_NOWAIT) {
1381 return ERR_PTR(-EAGAIN);
1384 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1389 /* wait on @rl and retry */
1390 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1391 TASK_UNINTERRUPTIBLE);
1393 trace_block_sleeprq(q, bio, op);
1395 spin_unlock_irq(q->queue_lock);
1399 * After sleeping, we become a "batching" process and will be able
1400 * to allocate at least one request, and up to a big batch of them
1401 * for a small period time. See ioc_batching, ioc_set_batching
1403 ioc_set_batching(q, current->io_context);
1405 spin_lock_irq(q->queue_lock);
1406 finish_wait(&rl->wait[is_sync], &wait);
1411 static struct request *blk_old_get_request(struct request_queue *q,
1412 unsigned int op, gfp_t gfp_mask)
1416 WARN_ON_ONCE(q->mq_ops);
1418 /* create ioc upfront */
1419 create_io_context(gfp_mask, q->node);
1421 spin_lock_irq(q->queue_lock);
1422 rq = get_request(q, op, NULL, gfp_mask);
1424 spin_unlock_irq(q->queue_lock);
1428 /* q->queue_lock is unlocked at this point */
1430 rq->__sector = (sector_t) -1;
1431 rq->bio = rq->biotail = NULL;
1435 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1438 struct request *req;
1441 req = blk_mq_alloc_request(q, op,
1442 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1443 0 : BLK_MQ_REQ_NOWAIT);
1444 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1445 q->mq_ops->initialize_rq_fn(req);
1447 req = blk_old_get_request(q, op, gfp_mask);
1448 if (!IS_ERR(req) && q->initialize_rq_fn)
1449 q->initialize_rq_fn(req);
1454 EXPORT_SYMBOL(blk_get_request);
1457 * blk_requeue_request - put a request back on queue
1458 * @q: request queue where request should be inserted
1459 * @rq: request to be inserted
1462 * Drivers often keep queueing requests until the hardware cannot accept
1463 * more, when that condition happens we need to put the request back
1464 * on the queue. Must be called with queue lock held.
1466 void blk_requeue_request(struct request_queue *q, struct request *rq)
1468 lockdep_assert_held(q->queue_lock);
1469 WARN_ON_ONCE(q->mq_ops);
1471 blk_delete_timer(rq);
1472 blk_clear_rq_complete(rq);
1473 trace_block_rq_requeue(q, rq);
1474 wbt_requeue(q->rq_wb, &rq->issue_stat);
1476 if (rq->rq_flags & RQF_QUEUED)
1477 blk_queue_end_tag(q, rq);
1479 BUG_ON(blk_queued_rq(rq));
1481 elv_requeue_request(q, rq);
1483 EXPORT_SYMBOL(blk_requeue_request);
1485 static void add_acct_request(struct request_queue *q, struct request *rq,
1488 blk_account_io_start(rq, true);
1489 __elv_add_request(q, rq, where);
1492 static void part_round_stats_single(struct request_queue *q, int cpu,
1493 struct hd_struct *part, unsigned long now,
1494 unsigned int inflight)
1497 __part_stat_add(cpu, part, time_in_queue,
1498 inflight * (now - part->stamp));
1499 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1505 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1506 * @q: target block queue
1507 * @cpu: cpu number for stats access
1508 * @part: target partition
1510 * The average IO queue length and utilisation statistics are maintained
1511 * by observing the current state of the queue length and the amount of
1512 * time it has been in this state for.
1514 * Normally, that accounting is done on IO completion, but that can result
1515 * in more than a second's worth of IO being accounted for within any one
1516 * second, leading to >100% utilisation. To deal with that, we call this
1517 * function to do a round-off before returning the results when reading
1518 * /proc/diskstats. This accounts immediately for all queue usage up to
1519 * the current jiffies and restarts the counters again.
1521 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1523 struct hd_struct *part2 = NULL;
1524 unsigned long now = jiffies;
1525 unsigned int inflight[2];
1528 if (part->stamp != now)
1532 part2 = &part_to_disk(part)->part0;
1533 if (part2->stamp != now)
1540 part_in_flight(q, part, inflight);
1543 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1545 part_round_stats_single(q, cpu, part, now, inflight[0]);
1547 EXPORT_SYMBOL_GPL(part_round_stats);
1550 static void blk_pm_put_request(struct request *rq)
1552 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1553 pm_runtime_mark_last_busy(rq->q->dev);
1556 static inline void blk_pm_put_request(struct request *rq) {}
1559 void __blk_put_request(struct request_queue *q, struct request *req)
1561 req_flags_t rq_flags = req->rq_flags;
1567 blk_mq_free_request(req);
1571 lockdep_assert_held(q->queue_lock);
1573 blk_pm_put_request(req);
1575 elv_completed_request(q, req);
1577 /* this is a bio leak */
1578 WARN_ON(req->bio != NULL);
1580 wbt_done(q->rq_wb, &req->issue_stat);
1583 * Request may not have originated from ll_rw_blk. if not,
1584 * it didn't come out of our reserved rq pools
1586 if (rq_flags & RQF_ALLOCED) {
1587 struct request_list *rl = blk_rq_rl(req);
1588 bool sync = op_is_sync(req->cmd_flags);
1590 BUG_ON(!list_empty(&req->queuelist));
1591 BUG_ON(ELV_ON_HASH(req));
1593 blk_free_request(rl, req);
1594 freed_request(rl, sync, rq_flags);
1598 EXPORT_SYMBOL_GPL(__blk_put_request);
1600 void blk_put_request(struct request *req)
1602 struct request_queue *q = req->q;
1605 blk_mq_free_request(req);
1607 unsigned long flags;
1609 spin_lock_irqsave(q->queue_lock, flags);
1610 __blk_put_request(q, req);
1611 spin_unlock_irqrestore(q->queue_lock, flags);
1614 EXPORT_SYMBOL(blk_put_request);
1616 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1619 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1621 if (!ll_back_merge_fn(q, req, bio))
1624 trace_block_bio_backmerge(q, req, bio);
1626 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1627 blk_rq_set_mixed_merge(req);
1629 req->biotail->bi_next = bio;
1631 req->__data_len += bio->bi_iter.bi_size;
1632 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1634 blk_account_io_start(req, false);
1638 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1641 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1643 if (!ll_front_merge_fn(q, req, bio))
1646 trace_block_bio_frontmerge(q, req, bio);
1648 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1649 blk_rq_set_mixed_merge(req);
1651 bio->bi_next = req->bio;
1654 req->__sector = bio->bi_iter.bi_sector;
1655 req->__data_len += bio->bi_iter.bi_size;
1656 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1658 blk_account_io_start(req, false);
1662 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1665 unsigned short segments = blk_rq_nr_discard_segments(req);
1667 if (segments >= queue_max_discard_segments(q))
1669 if (blk_rq_sectors(req) + bio_sectors(bio) >
1670 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1673 req->biotail->bi_next = bio;
1675 req->__data_len += bio->bi_iter.bi_size;
1676 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1677 req->nr_phys_segments = segments + 1;
1679 blk_account_io_start(req, false);
1682 req_set_nomerge(q, req);
1687 * blk_attempt_plug_merge - try to merge with %current's plugged list
1688 * @q: request_queue new bio is being queued at
1689 * @bio: new bio being queued
1690 * @request_count: out parameter for number of traversed plugged requests
1691 * @same_queue_rq: pointer to &struct request that gets filled in when
1692 * another request associated with @q is found on the plug list
1693 * (optional, may be %NULL)
1695 * Determine whether @bio being queued on @q can be merged with a request
1696 * on %current's plugged list. Returns %true if merge was successful,
1699 * Plugging coalesces IOs from the same issuer for the same purpose without
1700 * going through @q->queue_lock. As such it's more of an issuing mechanism
1701 * than scheduling, and the request, while may have elvpriv data, is not
1702 * added on the elevator at this point. In addition, we don't have
1703 * reliable access to the elevator outside queue lock. Only check basic
1704 * merging parameters without querying the elevator.
1706 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1708 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1709 unsigned int *request_count,
1710 struct request **same_queue_rq)
1712 struct blk_plug *plug;
1714 struct list_head *plug_list;
1716 plug = current->plug;
1722 plug_list = &plug->mq_list;
1724 plug_list = &plug->list;
1726 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1727 bool merged = false;
1732 * Only blk-mq multiple hardware queues case checks the
1733 * rq in the same queue, there should be only one such
1737 *same_queue_rq = rq;
1740 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1743 switch (blk_try_merge(rq, bio)) {
1744 case ELEVATOR_BACK_MERGE:
1745 merged = bio_attempt_back_merge(q, rq, bio);
1747 case ELEVATOR_FRONT_MERGE:
1748 merged = bio_attempt_front_merge(q, rq, bio);
1750 case ELEVATOR_DISCARD_MERGE:
1751 merged = bio_attempt_discard_merge(q, rq, bio);
1764 unsigned int blk_plug_queued_count(struct request_queue *q)
1766 struct blk_plug *plug;
1768 struct list_head *plug_list;
1769 unsigned int ret = 0;
1771 plug = current->plug;
1776 plug_list = &plug->mq_list;
1778 plug_list = &plug->list;
1780 list_for_each_entry(rq, plug_list, queuelist) {
1788 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1790 struct io_context *ioc = rq_ioc(bio);
1792 if (bio->bi_opf & REQ_RAHEAD)
1793 req->cmd_flags |= REQ_FAILFAST_MASK;
1795 req->__sector = bio->bi_iter.bi_sector;
1796 if (ioprio_valid(bio_prio(bio)))
1797 req->ioprio = bio_prio(bio);
1799 req->ioprio = ioc->ioprio;
1801 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1802 req->write_hint = bio->bi_write_hint;
1803 blk_rq_bio_prep(req->q, req, bio);
1805 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1807 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1809 struct blk_plug *plug;
1810 int where = ELEVATOR_INSERT_SORT;
1811 struct request *req, *free;
1812 unsigned int request_count = 0;
1813 unsigned int wb_acct;
1816 * low level driver can indicate that it wants pages above a
1817 * certain limit bounced to low memory (ie for highmem, or even
1818 * ISA dma in theory)
1820 blk_queue_bounce(q, &bio);
1822 blk_queue_split(q, &bio);
1824 if (!bio_integrity_prep(bio))
1825 return BLK_QC_T_NONE;
1827 if (op_is_flush(bio->bi_opf)) {
1828 spin_lock_irq(q->queue_lock);
1829 where = ELEVATOR_INSERT_FLUSH;
1834 * Check if we can merge with the plugged list before grabbing
1837 if (!blk_queue_nomerges(q)) {
1838 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1839 return BLK_QC_T_NONE;
1841 request_count = blk_plug_queued_count(q);
1843 spin_lock_irq(q->queue_lock);
1845 switch (elv_merge(q, &req, bio)) {
1846 case ELEVATOR_BACK_MERGE:
1847 if (!bio_attempt_back_merge(q, req, bio))
1849 elv_bio_merged(q, req, bio);
1850 free = attempt_back_merge(q, req);
1852 __blk_put_request(q, free);
1854 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1856 case ELEVATOR_FRONT_MERGE:
1857 if (!bio_attempt_front_merge(q, req, bio))
1859 elv_bio_merged(q, req, bio);
1860 free = attempt_front_merge(q, req);
1862 __blk_put_request(q, free);
1864 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1871 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1874 * Grab a free request. This is might sleep but can not fail.
1875 * Returns with the queue unlocked.
1877 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1879 __wbt_done(q->rq_wb, wb_acct);
1880 if (PTR_ERR(req) == -ENOMEM)
1881 bio->bi_status = BLK_STS_RESOURCE;
1883 bio->bi_status = BLK_STS_IOERR;
1888 wbt_track(&req->issue_stat, wb_acct);
1891 * After dropping the lock and possibly sleeping here, our request
1892 * may now be mergeable after it had proven unmergeable (above).
1893 * We don't worry about that case for efficiency. It won't happen
1894 * often, and the elevators are able to handle it.
1896 blk_init_request_from_bio(req, bio);
1898 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1899 req->cpu = raw_smp_processor_id();
1901 plug = current->plug;
1904 * If this is the first request added after a plug, fire
1907 * @request_count may become stale because of schedule
1908 * out, so check plug list again.
1910 if (!request_count || list_empty(&plug->list))
1911 trace_block_plug(q);
1913 struct request *last = list_entry_rq(plug->list.prev);
1914 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1915 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1916 blk_flush_plug_list(plug, false);
1917 trace_block_plug(q);
1920 list_add_tail(&req->queuelist, &plug->list);
1921 blk_account_io_start(req, true);
1923 spin_lock_irq(q->queue_lock);
1924 add_acct_request(q, req, where);
1927 spin_unlock_irq(q->queue_lock);
1930 return BLK_QC_T_NONE;
1933 static void handle_bad_sector(struct bio *bio)
1935 char b[BDEVNAME_SIZE];
1937 printk(KERN_INFO "attempt to access beyond end of device\n");
1938 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1939 bio_devname(bio, b), bio->bi_opf,
1940 (unsigned long long)bio_end_sector(bio),
1941 (long long)get_capacity(bio->bi_disk));
1944 #ifdef CONFIG_FAIL_MAKE_REQUEST
1946 static DECLARE_FAULT_ATTR(fail_make_request);
1948 static int __init setup_fail_make_request(char *str)
1950 return setup_fault_attr(&fail_make_request, str);
1952 __setup("fail_make_request=", setup_fail_make_request);
1954 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1956 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1959 static int __init fail_make_request_debugfs(void)
1961 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1962 NULL, &fail_make_request);
1964 return PTR_ERR_OR_ZERO(dir);
1967 late_initcall(fail_make_request_debugfs);
1969 #else /* CONFIG_FAIL_MAKE_REQUEST */
1971 static inline bool should_fail_request(struct hd_struct *part,
1977 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1980 * Remap block n of partition p to block n+start(p) of the disk.
1982 static inline int blk_partition_remap(struct bio *bio)
1984 struct hd_struct *p;
1988 * Zone reset does not include bi_size so bio_sectors() is always 0.
1989 * Include a test for the reset op code and perform the remap if needed.
1991 if (!bio->bi_partno ||
1992 (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET))
1996 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
1997 if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) {
1998 bio->bi_iter.bi_sector += p->start_sect;
2000 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2001 bio->bi_iter.bi_sector - p->start_sect);
2003 printk("%s: fail for partition %d\n", __func__, bio->bi_partno);
2012 * Check whether this bio extends beyond the end of the device.
2014 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
2021 /* Test device or partition size, when known. */
2022 maxsector = get_capacity(bio->bi_disk);
2024 sector_t sector = bio->bi_iter.bi_sector;
2026 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2028 * This may well happen - the kernel calls bread()
2029 * without checking the size of the device, e.g., when
2030 * mounting a device.
2032 handle_bad_sector(bio);
2040 static noinline_for_stack bool
2041 generic_make_request_checks(struct bio *bio)
2043 struct request_queue *q;
2044 int nr_sectors = bio_sectors(bio);
2045 blk_status_t status = BLK_STS_IOERR;
2046 char b[BDEVNAME_SIZE];
2050 if (bio_check_eod(bio, nr_sectors))
2053 q = bio->bi_disk->queue;
2056 "generic_make_request: Trying to access "
2057 "nonexistent block-device %s (%Lu)\n",
2058 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2063 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2064 * if queue is not a request based queue.
2067 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2070 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2073 if (blk_partition_remap(bio))
2076 if (bio_check_eod(bio, nr_sectors))
2080 * Filter flush bio's early so that make_request based
2081 * drivers without flush support don't have to worry
2084 if (op_is_flush(bio->bi_opf) &&
2085 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2086 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2088 status = BLK_STS_OK;
2093 switch (bio_op(bio)) {
2094 case REQ_OP_DISCARD:
2095 if (!blk_queue_discard(q))
2098 case REQ_OP_SECURE_ERASE:
2099 if (!blk_queue_secure_erase(q))
2102 case REQ_OP_WRITE_SAME:
2103 if (!q->limits.max_write_same_sectors)
2106 case REQ_OP_ZONE_REPORT:
2107 case REQ_OP_ZONE_RESET:
2108 if (!blk_queue_is_zoned(q))
2111 case REQ_OP_WRITE_ZEROES:
2112 if (!q->limits.max_write_zeroes_sectors)
2120 * Various block parts want %current->io_context and lazy ioc
2121 * allocation ends up trading a lot of pain for a small amount of
2122 * memory. Just allocate it upfront. This may fail and block
2123 * layer knows how to live with it.
2125 create_io_context(GFP_ATOMIC, q->node);
2127 if (!blkcg_bio_issue_check(q, bio))
2130 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2131 trace_block_bio_queue(q, bio);
2132 /* Now that enqueuing has been traced, we need to trace
2133 * completion as well.
2135 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2140 status = BLK_STS_NOTSUPP;
2142 bio->bi_status = status;
2148 * generic_make_request - hand a buffer to its device driver for I/O
2149 * @bio: The bio describing the location in memory and on the device.
2151 * generic_make_request() is used to make I/O requests of block
2152 * devices. It is passed a &struct bio, which describes the I/O that needs
2155 * generic_make_request() does not return any status. The
2156 * success/failure status of the request, along with notification of
2157 * completion, is delivered asynchronously through the bio->bi_end_io
2158 * function described (one day) else where.
2160 * The caller of generic_make_request must make sure that bi_io_vec
2161 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2162 * set to describe the device address, and the
2163 * bi_end_io and optionally bi_private are set to describe how
2164 * completion notification should be signaled.
2166 * generic_make_request and the drivers it calls may use bi_next if this
2167 * bio happens to be merged with someone else, and may resubmit the bio to
2168 * a lower device by calling into generic_make_request recursively, which
2169 * means the bio should NOT be touched after the call to ->make_request_fn.
2171 blk_qc_t generic_make_request(struct bio *bio)
2174 * bio_list_on_stack[0] contains bios submitted by the current
2176 * bio_list_on_stack[1] contains bios that were submitted before
2177 * the current make_request_fn, but that haven't been processed
2180 struct bio_list bio_list_on_stack[2];
2181 blk_qc_t ret = BLK_QC_T_NONE;
2183 if (!generic_make_request_checks(bio))
2187 * We only want one ->make_request_fn to be active at a time, else
2188 * stack usage with stacked devices could be a problem. So use
2189 * current->bio_list to keep a list of requests submited by a
2190 * make_request_fn function. current->bio_list is also used as a
2191 * flag to say if generic_make_request is currently active in this
2192 * task or not. If it is NULL, then no make_request is active. If
2193 * it is non-NULL, then a make_request is active, and new requests
2194 * should be added at the tail
2196 if (current->bio_list) {
2197 bio_list_add(¤t->bio_list[0], bio);
2201 /* following loop may be a bit non-obvious, and so deserves some
2203 * Before entering the loop, bio->bi_next is NULL (as all callers
2204 * ensure that) so we have a list with a single bio.
2205 * We pretend that we have just taken it off a longer list, so
2206 * we assign bio_list to a pointer to the bio_list_on_stack,
2207 * thus initialising the bio_list of new bios to be
2208 * added. ->make_request() may indeed add some more bios
2209 * through a recursive call to generic_make_request. If it
2210 * did, we find a non-NULL value in bio_list and re-enter the loop
2211 * from the top. In this case we really did just take the bio
2212 * of the top of the list (no pretending) and so remove it from
2213 * bio_list, and call into ->make_request() again.
2215 BUG_ON(bio->bi_next);
2216 bio_list_init(&bio_list_on_stack[0]);
2217 current->bio_list = bio_list_on_stack;
2219 struct request_queue *q = bio->bi_disk->queue;
2221 if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) {
2222 struct bio_list lower, same;
2224 /* Create a fresh bio_list for all subordinate requests */
2225 bio_list_on_stack[1] = bio_list_on_stack[0];
2226 bio_list_init(&bio_list_on_stack[0]);
2227 ret = q->make_request_fn(q, bio);
2231 /* sort new bios into those for a lower level
2232 * and those for the same level
2234 bio_list_init(&lower);
2235 bio_list_init(&same);
2236 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2237 if (q == bio->bi_disk->queue)
2238 bio_list_add(&same, bio);
2240 bio_list_add(&lower, bio);
2241 /* now assemble so we handle the lowest level first */
2242 bio_list_merge(&bio_list_on_stack[0], &lower);
2243 bio_list_merge(&bio_list_on_stack[0], &same);
2244 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2246 if (unlikely(!blk_queue_dying(q) &&
2247 (bio->bi_opf & REQ_NOWAIT)))
2248 bio_wouldblock_error(bio);
2252 bio = bio_list_pop(&bio_list_on_stack[0]);
2254 current->bio_list = NULL; /* deactivate */
2259 EXPORT_SYMBOL(generic_make_request);
2262 * submit_bio - submit a bio to the block device layer for I/O
2263 * @bio: The &struct bio which describes the I/O
2265 * submit_bio() is very similar in purpose to generic_make_request(), and
2266 * uses that function to do most of the work. Both are fairly rough
2267 * interfaces; @bio must be presetup and ready for I/O.
2270 blk_qc_t submit_bio(struct bio *bio)
2273 * If it's a regular read/write or a barrier with data attached,
2274 * go through the normal accounting stuff before submission.
2276 if (bio_has_data(bio)) {
2279 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2280 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2282 count = bio_sectors(bio);
2284 if (op_is_write(bio_op(bio))) {
2285 count_vm_events(PGPGOUT, count);
2287 task_io_account_read(bio->bi_iter.bi_size);
2288 count_vm_events(PGPGIN, count);
2291 if (unlikely(block_dump)) {
2292 char b[BDEVNAME_SIZE];
2293 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2294 current->comm, task_pid_nr(current),
2295 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2296 (unsigned long long)bio->bi_iter.bi_sector,
2297 bio_devname(bio, b), count);
2301 return generic_make_request(bio);
2303 EXPORT_SYMBOL(submit_bio);
2306 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2307 * for new the queue limits
2309 * @rq: the request being checked
2312 * @rq may have been made based on weaker limitations of upper-level queues
2313 * in request stacking drivers, and it may violate the limitation of @q.
2314 * Since the block layer and the underlying device driver trust @rq
2315 * after it is inserted to @q, it should be checked against @q before
2316 * the insertion using this generic function.
2318 * Request stacking drivers like request-based dm may change the queue
2319 * limits when retrying requests on other queues. Those requests need
2320 * to be checked against the new queue limits again during dispatch.
2322 static int blk_cloned_rq_check_limits(struct request_queue *q,
2325 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2326 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2331 * queue's settings related to segment counting like q->bounce_pfn
2332 * may differ from that of other stacking queues.
2333 * Recalculate it to check the request correctly on this queue's
2336 blk_recalc_rq_segments(rq);
2337 if (rq->nr_phys_segments > queue_max_segments(q)) {
2338 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2346 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2347 * @q: the queue to submit the request
2348 * @rq: the request being queued
2350 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2352 unsigned long flags;
2353 int where = ELEVATOR_INSERT_BACK;
2355 if (blk_cloned_rq_check_limits(q, rq))
2356 return BLK_STS_IOERR;
2359 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2360 return BLK_STS_IOERR;
2363 if (blk_queue_io_stat(q))
2364 blk_account_io_start(rq, true);
2366 * Since we have a scheduler attached on the top device,
2367 * bypass a potential scheduler on the bottom device for
2370 blk_mq_request_bypass_insert(rq);
2374 spin_lock_irqsave(q->queue_lock, flags);
2375 if (unlikely(blk_queue_dying(q))) {
2376 spin_unlock_irqrestore(q->queue_lock, flags);
2377 return BLK_STS_IOERR;
2381 * Submitting request must be dequeued before calling this function
2382 * because it will be linked to another request_queue
2384 BUG_ON(blk_queued_rq(rq));
2386 if (op_is_flush(rq->cmd_flags))
2387 where = ELEVATOR_INSERT_FLUSH;
2389 add_acct_request(q, rq, where);
2390 if (where == ELEVATOR_INSERT_FLUSH)
2392 spin_unlock_irqrestore(q->queue_lock, flags);
2396 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2399 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2400 * @rq: request to examine
2403 * A request could be merge of IOs which require different failure
2404 * handling. This function determines the number of bytes which
2405 * can be failed from the beginning of the request without
2406 * crossing into area which need to be retried further.
2409 * The number of bytes to fail.
2411 unsigned int blk_rq_err_bytes(const struct request *rq)
2413 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2414 unsigned int bytes = 0;
2417 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2418 return blk_rq_bytes(rq);
2421 * Currently the only 'mixing' which can happen is between
2422 * different fastfail types. We can safely fail portions
2423 * which have all the failfast bits that the first one has -
2424 * the ones which are at least as eager to fail as the first
2427 for (bio = rq->bio; bio; bio = bio->bi_next) {
2428 if ((bio->bi_opf & ff) != ff)
2430 bytes += bio->bi_iter.bi_size;
2433 /* this could lead to infinite loop */
2434 BUG_ON(blk_rq_bytes(rq) && !bytes);
2437 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2439 void blk_account_io_completion(struct request *req, unsigned int bytes)
2441 if (blk_do_io_stat(req)) {
2442 const int rw = rq_data_dir(req);
2443 struct hd_struct *part;
2446 cpu = part_stat_lock();
2448 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2453 void blk_account_io_done(struct request *req)
2456 * Account IO completion. flush_rq isn't accounted as a
2457 * normal IO on queueing nor completion. Accounting the
2458 * containing request is enough.
2460 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2461 unsigned long duration = jiffies - req->start_time;
2462 const int rw = rq_data_dir(req);
2463 struct hd_struct *part;
2466 cpu = part_stat_lock();
2469 part_stat_inc(cpu, part, ios[rw]);
2470 part_stat_add(cpu, part, ticks[rw], duration);
2471 part_round_stats(req->q, cpu, part);
2472 part_dec_in_flight(req->q, part, rw);
2474 hd_struct_put(part);
2481 * Don't process normal requests when queue is suspended
2482 * or in the process of suspending/resuming
2484 static struct request *blk_pm_peek_request(struct request_queue *q,
2487 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2488 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2494 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2501 void blk_account_io_start(struct request *rq, bool new_io)
2503 struct hd_struct *part;
2504 int rw = rq_data_dir(rq);
2507 if (!blk_do_io_stat(rq))
2510 cpu = part_stat_lock();
2514 part_stat_inc(cpu, part, merges[rw]);
2516 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2517 if (!hd_struct_try_get(part)) {
2519 * The partition is already being removed,
2520 * the request will be accounted on the disk only
2522 * We take a reference on disk->part0 although that
2523 * partition will never be deleted, so we can treat
2524 * it as any other partition.
2526 part = &rq->rq_disk->part0;
2527 hd_struct_get(part);
2529 part_round_stats(rq->q, cpu, part);
2530 part_inc_in_flight(rq->q, part, rw);
2538 * blk_peek_request - peek at the top of a request queue
2539 * @q: request queue to peek at
2542 * Return the request at the top of @q. The returned request
2543 * should be started using blk_start_request() before LLD starts
2547 * Pointer to the request at the top of @q if available. Null
2550 struct request *blk_peek_request(struct request_queue *q)
2555 lockdep_assert_held(q->queue_lock);
2556 WARN_ON_ONCE(q->mq_ops);
2558 while ((rq = __elv_next_request(q)) != NULL) {
2560 rq = blk_pm_peek_request(q, rq);
2564 if (!(rq->rq_flags & RQF_STARTED)) {
2566 * This is the first time the device driver
2567 * sees this request (possibly after
2568 * requeueing). Notify IO scheduler.
2570 if (rq->rq_flags & RQF_SORTED)
2571 elv_activate_rq(q, rq);
2574 * just mark as started even if we don't start
2575 * it, a request that has been delayed should
2576 * not be passed by new incoming requests
2578 rq->rq_flags |= RQF_STARTED;
2579 trace_block_rq_issue(q, rq);
2582 if (!q->boundary_rq || q->boundary_rq == rq) {
2583 q->end_sector = rq_end_sector(rq);
2584 q->boundary_rq = NULL;
2587 if (rq->rq_flags & RQF_DONTPREP)
2590 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2592 * make sure space for the drain appears we
2593 * know we can do this because max_hw_segments
2594 * has been adjusted to be one fewer than the
2597 rq->nr_phys_segments++;
2603 ret = q->prep_rq_fn(q, rq);
2604 if (ret == BLKPREP_OK) {
2606 } else if (ret == BLKPREP_DEFER) {
2608 * the request may have been (partially) prepped.
2609 * we need to keep this request in the front to
2610 * avoid resource deadlock. RQF_STARTED will
2611 * prevent other fs requests from passing this one.
2613 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2614 !(rq->rq_flags & RQF_DONTPREP)) {
2616 * remove the space for the drain we added
2617 * so that we don't add it again
2619 --rq->nr_phys_segments;
2624 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2625 rq->rq_flags |= RQF_QUIET;
2627 * Mark this request as started so we don't trigger
2628 * any debug logic in the end I/O path.
2630 blk_start_request(rq);
2631 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2632 BLK_STS_TARGET : BLK_STS_IOERR);
2634 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2641 EXPORT_SYMBOL(blk_peek_request);
2643 static void blk_dequeue_request(struct request *rq)
2645 struct request_queue *q = rq->q;
2647 BUG_ON(list_empty(&rq->queuelist));
2648 BUG_ON(ELV_ON_HASH(rq));
2650 list_del_init(&rq->queuelist);
2653 * the time frame between a request being removed from the lists
2654 * and to it is freed is accounted as io that is in progress at
2657 if (blk_account_rq(rq)) {
2658 q->in_flight[rq_is_sync(rq)]++;
2659 set_io_start_time_ns(rq);
2664 * blk_start_request - start request processing on the driver
2665 * @req: request to dequeue
2668 * Dequeue @req and start timeout timer on it. This hands off the
2669 * request to the driver.
2671 void blk_start_request(struct request *req)
2673 lockdep_assert_held(req->q->queue_lock);
2674 WARN_ON_ONCE(req->q->mq_ops);
2676 blk_dequeue_request(req);
2678 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2679 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2680 req->rq_flags |= RQF_STATS;
2681 wbt_issue(req->q->rq_wb, &req->issue_stat);
2684 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2687 EXPORT_SYMBOL(blk_start_request);
2690 * blk_fetch_request - fetch a request from a request queue
2691 * @q: request queue to fetch a request from
2694 * Return the request at the top of @q. The request is started on
2695 * return and LLD can start processing it immediately.
2698 * Pointer to the request at the top of @q if available. Null
2701 struct request *blk_fetch_request(struct request_queue *q)
2705 lockdep_assert_held(q->queue_lock);
2706 WARN_ON_ONCE(q->mq_ops);
2708 rq = blk_peek_request(q);
2710 blk_start_request(rq);
2713 EXPORT_SYMBOL(blk_fetch_request);
2716 * blk_update_request - Special helper function for request stacking drivers
2717 * @req: the request being processed
2718 * @error: block status code
2719 * @nr_bytes: number of bytes to complete @req
2722 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2723 * the request structure even if @req doesn't have leftover.
2724 * If @req has leftover, sets it up for the next range of segments.
2726 * This special helper function is only for request stacking drivers
2727 * (e.g. request-based dm) so that they can handle partial completion.
2728 * Actual device drivers should use blk_end_request instead.
2730 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2731 * %false return from this function.
2734 * %false - this request doesn't have any more data
2735 * %true - this request has more data
2737 bool blk_update_request(struct request *req, blk_status_t error,
2738 unsigned int nr_bytes)
2742 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
2747 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2748 !(req->rq_flags & RQF_QUIET)))
2749 print_req_error(req, error);
2751 blk_account_io_completion(req, nr_bytes);
2755 struct bio *bio = req->bio;
2756 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2758 if (bio_bytes == bio->bi_iter.bi_size)
2759 req->bio = bio->bi_next;
2761 /* Completion has already been traced */
2762 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
2763 req_bio_endio(req, bio, bio_bytes, error);
2765 total_bytes += bio_bytes;
2766 nr_bytes -= bio_bytes;
2777 * Reset counters so that the request stacking driver
2778 * can find how many bytes remain in the request
2781 req->__data_len = 0;
2785 req->__data_len -= total_bytes;
2787 /* update sector only for requests with clear definition of sector */
2788 if (!blk_rq_is_passthrough(req))
2789 req->__sector += total_bytes >> 9;
2791 /* mixed attributes always follow the first bio */
2792 if (req->rq_flags & RQF_MIXED_MERGE) {
2793 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2794 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2797 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
2799 * If total number of sectors is less than the first segment
2800 * size, something has gone terribly wrong.
2802 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2803 blk_dump_rq_flags(req, "request botched");
2804 req->__data_len = blk_rq_cur_bytes(req);
2807 /* recalculate the number of segments */
2808 blk_recalc_rq_segments(req);
2813 EXPORT_SYMBOL_GPL(blk_update_request);
2815 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2816 unsigned int nr_bytes,
2817 unsigned int bidi_bytes)
2819 if (blk_update_request(rq, error, nr_bytes))
2822 /* Bidi request must be completed as a whole */
2823 if (unlikely(blk_bidi_rq(rq)) &&
2824 blk_update_request(rq->next_rq, error, bidi_bytes))
2827 if (blk_queue_add_random(rq->q))
2828 add_disk_randomness(rq->rq_disk);
2834 * blk_unprep_request - unprepare a request
2837 * This function makes a request ready for complete resubmission (or
2838 * completion). It happens only after all error handling is complete,
2839 * so represents the appropriate moment to deallocate any resources
2840 * that were allocated to the request in the prep_rq_fn. The queue
2841 * lock is held when calling this.
2843 void blk_unprep_request(struct request *req)
2845 struct request_queue *q = req->q;
2847 req->rq_flags &= ~RQF_DONTPREP;
2848 if (q->unprep_rq_fn)
2849 q->unprep_rq_fn(q, req);
2851 EXPORT_SYMBOL_GPL(blk_unprep_request);
2853 void blk_finish_request(struct request *req, blk_status_t error)
2855 struct request_queue *q = req->q;
2857 lockdep_assert_held(req->q->queue_lock);
2858 WARN_ON_ONCE(q->mq_ops);
2860 if (req->rq_flags & RQF_STATS)
2863 if (req->rq_flags & RQF_QUEUED)
2864 blk_queue_end_tag(q, req);
2866 BUG_ON(blk_queued_rq(req));
2868 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
2869 laptop_io_completion(req->q->backing_dev_info);
2871 blk_delete_timer(req);
2873 if (req->rq_flags & RQF_DONTPREP)
2874 blk_unprep_request(req);
2876 blk_account_io_done(req);
2879 wbt_done(req->q->rq_wb, &req->issue_stat);
2880 req->end_io(req, error);
2882 if (blk_bidi_rq(req))
2883 __blk_put_request(req->next_rq->q, req->next_rq);
2885 __blk_put_request(q, req);
2888 EXPORT_SYMBOL(blk_finish_request);
2891 * blk_end_bidi_request - Complete a bidi request
2892 * @rq: the request to complete
2893 * @error: block status code
2894 * @nr_bytes: number of bytes to complete @rq
2895 * @bidi_bytes: number of bytes to complete @rq->next_rq
2898 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2899 * Drivers that supports bidi can safely call this member for any
2900 * type of request, bidi or uni. In the later case @bidi_bytes is
2904 * %false - we are done with this request
2905 * %true - still buffers pending for this request
2907 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
2908 unsigned int nr_bytes, unsigned int bidi_bytes)
2910 struct request_queue *q = rq->q;
2911 unsigned long flags;
2913 WARN_ON_ONCE(q->mq_ops);
2915 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2918 spin_lock_irqsave(q->queue_lock, flags);
2919 blk_finish_request(rq, error);
2920 spin_unlock_irqrestore(q->queue_lock, flags);
2926 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2927 * @rq: the request to complete
2928 * @error: block status code
2929 * @nr_bytes: number of bytes to complete @rq
2930 * @bidi_bytes: number of bytes to complete @rq->next_rq
2933 * Identical to blk_end_bidi_request() except that queue lock is
2934 * assumed to be locked on entry and remains so on return.
2937 * %false - we are done with this request
2938 * %true - still buffers pending for this request
2940 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
2941 unsigned int nr_bytes, unsigned int bidi_bytes)
2943 lockdep_assert_held(rq->q->queue_lock);
2944 WARN_ON_ONCE(rq->q->mq_ops);
2946 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2949 blk_finish_request(rq, error);
2955 * blk_end_request - Helper function for drivers to complete the request.
2956 * @rq: the request being processed
2957 * @error: block status code
2958 * @nr_bytes: number of bytes to complete
2961 * Ends I/O on a number of bytes attached to @rq.
2962 * If @rq has leftover, sets it up for the next range of segments.
2965 * %false - we are done with this request
2966 * %true - still buffers pending for this request
2968 bool blk_end_request(struct request *rq, blk_status_t error,
2969 unsigned int nr_bytes)
2971 WARN_ON_ONCE(rq->q->mq_ops);
2972 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2974 EXPORT_SYMBOL(blk_end_request);
2977 * blk_end_request_all - Helper function for drives to finish the request.
2978 * @rq: the request to finish
2979 * @error: block status code
2982 * Completely finish @rq.
2984 void blk_end_request_all(struct request *rq, blk_status_t error)
2987 unsigned int bidi_bytes = 0;
2989 if (unlikely(blk_bidi_rq(rq)))
2990 bidi_bytes = blk_rq_bytes(rq->next_rq);
2992 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2995 EXPORT_SYMBOL(blk_end_request_all);
2998 * __blk_end_request - Helper function for drivers to complete the request.
2999 * @rq: the request being processed
3000 * @error: block status code
3001 * @nr_bytes: number of bytes to complete
3004 * Must be called with queue lock held unlike blk_end_request().
3007 * %false - we are done with this request
3008 * %true - still buffers pending for this request
3010 bool __blk_end_request(struct request *rq, blk_status_t error,
3011 unsigned int nr_bytes)
3013 lockdep_assert_held(rq->q->queue_lock);
3014 WARN_ON_ONCE(rq->q->mq_ops);
3016 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3018 EXPORT_SYMBOL(__blk_end_request);
3021 * __blk_end_request_all - Helper function for drives to finish the request.
3022 * @rq: the request to finish
3023 * @error: block status code
3026 * Completely finish @rq. Must be called with queue lock held.
3028 void __blk_end_request_all(struct request *rq, blk_status_t error)
3031 unsigned int bidi_bytes = 0;
3033 lockdep_assert_held(rq->q->queue_lock);
3034 WARN_ON_ONCE(rq->q->mq_ops);
3036 if (unlikely(blk_bidi_rq(rq)))
3037 bidi_bytes = blk_rq_bytes(rq->next_rq);
3039 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3042 EXPORT_SYMBOL(__blk_end_request_all);
3045 * __blk_end_request_cur - Helper function to finish the current request chunk.
3046 * @rq: the request to finish the current chunk for
3047 * @error: block status code
3050 * Complete the current consecutively mapped chunk from @rq. Must
3051 * be called with queue lock held.
3054 * %false - we are done with this request
3055 * %true - still buffers pending for this request
3057 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3059 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3061 EXPORT_SYMBOL(__blk_end_request_cur);
3063 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3066 if (bio_has_data(bio))
3067 rq->nr_phys_segments = bio_phys_segments(q, bio);
3069 rq->__data_len = bio->bi_iter.bi_size;
3070 rq->bio = rq->biotail = bio;
3073 rq->rq_disk = bio->bi_disk;
3076 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3078 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3079 * @rq: the request to be flushed
3082 * Flush all pages in @rq.
3084 void rq_flush_dcache_pages(struct request *rq)
3086 struct req_iterator iter;
3087 struct bio_vec bvec;
3089 rq_for_each_segment(bvec, rq, iter)
3090 flush_dcache_page(bvec.bv_page);
3092 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3096 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3097 * @q : the queue of the device being checked
3100 * Check if underlying low-level drivers of a device are busy.
3101 * If the drivers want to export their busy state, they must set own
3102 * exporting function using blk_queue_lld_busy() first.
3104 * Basically, this function is used only by request stacking drivers
3105 * to stop dispatching requests to underlying devices when underlying
3106 * devices are busy. This behavior helps more I/O merging on the queue
3107 * of the request stacking driver and prevents I/O throughput regression
3108 * on burst I/O load.
3111 * 0 - Not busy (The request stacking driver should dispatch request)
3112 * 1 - Busy (The request stacking driver should stop dispatching request)
3114 int blk_lld_busy(struct request_queue *q)
3117 return q->lld_busy_fn(q);
3121 EXPORT_SYMBOL_GPL(blk_lld_busy);
3124 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3125 * @rq: the clone request to be cleaned up
3128 * Free all bios in @rq for a cloned request.
3130 void blk_rq_unprep_clone(struct request *rq)
3134 while ((bio = rq->bio) != NULL) {
3135 rq->bio = bio->bi_next;
3140 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3143 * Copy attributes of the original request to the clone request.
3144 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3146 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3148 dst->cpu = src->cpu;
3149 dst->__sector = blk_rq_pos(src);
3150 dst->__data_len = blk_rq_bytes(src);
3151 dst->nr_phys_segments = src->nr_phys_segments;
3152 dst->ioprio = src->ioprio;
3153 dst->extra_len = src->extra_len;
3157 * blk_rq_prep_clone - Helper function to setup clone request
3158 * @rq: the request to be setup
3159 * @rq_src: original request to be cloned
3160 * @bs: bio_set that bios for clone are allocated from
3161 * @gfp_mask: memory allocation mask for bio
3162 * @bio_ctr: setup function to be called for each clone bio.
3163 * Returns %0 for success, non %0 for failure.
3164 * @data: private data to be passed to @bio_ctr
3167 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3168 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3169 * are not copied, and copying such parts is the caller's responsibility.
3170 * Also, pages which the original bios are pointing to are not copied
3171 * and the cloned bios just point same pages.
3172 * So cloned bios must be completed before original bios, which means
3173 * the caller must complete @rq before @rq_src.
3175 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3176 struct bio_set *bs, gfp_t gfp_mask,
3177 int (*bio_ctr)(struct bio *, struct bio *, void *),
3180 struct bio *bio, *bio_src;
3185 __rq_for_each_bio(bio_src, rq_src) {
3186 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3190 if (bio_ctr && bio_ctr(bio, bio_src, data))
3194 rq->biotail->bi_next = bio;
3197 rq->bio = rq->biotail = bio;
3200 __blk_rq_prep_clone(rq, rq_src);
3207 blk_rq_unprep_clone(rq);
3211 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3213 int kblockd_schedule_work(struct work_struct *work)
3215 return queue_work(kblockd_workqueue, work);
3217 EXPORT_SYMBOL(kblockd_schedule_work);
3219 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3221 return queue_work_on(cpu, kblockd_workqueue, work);
3223 EXPORT_SYMBOL(kblockd_schedule_work_on);
3225 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3226 unsigned long delay)
3228 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3230 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3232 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3233 unsigned long delay)
3235 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3237 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3239 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3240 unsigned long delay)
3242 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3244 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3247 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3248 * @plug: The &struct blk_plug that needs to be initialized
3251 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3252 * pending I/O should the task end up blocking between blk_start_plug() and
3253 * blk_finish_plug(). This is important from a performance perspective, but
3254 * also ensures that we don't deadlock. For instance, if the task is blocking
3255 * for a memory allocation, memory reclaim could end up wanting to free a
3256 * page belonging to that request that is currently residing in our private
3257 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3258 * this kind of deadlock.
3260 void blk_start_plug(struct blk_plug *plug)
3262 struct task_struct *tsk = current;
3265 * If this is a nested plug, don't actually assign it.
3270 INIT_LIST_HEAD(&plug->list);
3271 INIT_LIST_HEAD(&plug->mq_list);
3272 INIT_LIST_HEAD(&plug->cb_list);
3274 * Store ordering should not be needed here, since a potential
3275 * preempt will imply a full memory barrier
3279 EXPORT_SYMBOL(blk_start_plug);
3281 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3283 struct request *rqa = container_of(a, struct request, queuelist);
3284 struct request *rqb = container_of(b, struct request, queuelist);
3286 return !(rqa->q < rqb->q ||
3287 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3291 * If 'from_schedule' is true, then postpone the dispatch of requests
3292 * until a safe kblockd context. We due this to avoid accidental big
3293 * additional stack usage in driver dispatch, in places where the originally
3294 * plugger did not intend it.
3296 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3298 __releases(q->queue_lock)
3300 lockdep_assert_held(q->queue_lock);
3302 trace_block_unplug(q, depth, !from_schedule);
3305 blk_run_queue_async(q);
3308 spin_unlock(q->queue_lock);
3311 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3313 LIST_HEAD(callbacks);
3315 while (!list_empty(&plug->cb_list)) {
3316 list_splice_init(&plug->cb_list, &callbacks);
3318 while (!list_empty(&callbacks)) {
3319 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3322 list_del(&cb->list);
3323 cb->callback(cb, from_schedule);
3328 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3331 struct blk_plug *plug = current->plug;
3332 struct blk_plug_cb *cb;
3337 list_for_each_entry(cb, &plug->cb_list, list)
3338 if (cb->callback == unplug && cb->data == data)
3341 /* Not currently on the callback list */
3342 BUG_ON(size < sizeof(*cb));
3343 cb = kzalloc(size, GFP_ATOMIC);
3346 cb->callback = unplug;
3347 list_add(&cb->list, &plug->cb_list);
3351 EXPORT_SYMBOL(blk_check_plugged);
3353 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3355 struct request_queue *q;
3356 unsigned long flags;
3361 flush_plug_callbacks(plug, from_schedule);
3363 if (!list_empty(&plug->mq_list))
3364 blk_mq_flush_plug_list(plug, from_schedule);
3366 if (list_empty(&plug->list))
3369 list_splice_init(&plug->list, &list);
3371 list_sort(NULL, &list, plug_rq_cmp);
3377 * Save and disable interrupts here, to avoid doing it for every
3378 * queue lock we have to take.
3380 local_irq_save(flags);
3381 while (!list_empty(&list)) {
3382 rq = list_entry_rq(list.next);
3383 list_del_init(&rq->queuelist);
3387 * This drops the queue lock
3390 queue_unplugged(q, depth, from_schedule);
3393 spin_lock(q->queue_lock);
3397 * Short-circuit if @q is dead
3399 if (unlikely(blk_queue_dying(q))) {
3400 __blk_end_request_all(rq, BLK_STS_IOERR);
3405 * rq is already accounted, so use raw insert
3407 if (op_is_flush(rq->cmd_flags))
3408 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3410 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3416 * This drops the queue lock
3419 queue_unplugged(q, depth, from_schedule);
3421 local_irq_restore(flags);
3424 void blk_finish_plug(struct blk_plug *plug)
3426 if (plug != current->plug)
3428 blk_flush_plug_list(plug, false);
3430 current->plug = NULL;
3432 EXPORT_SYMBOL(blk_finish_plug);
3436 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3437 * @q: the queue of the device
3438 * @dev: the device the queue belongs to
3441 * Initialize runtime-PM-related fields for @q and start auto suspend for
3442 * @dev. Drivers that want to take advantage of request-based runtime PM
3443 * should call this function after @dev has been initialized, and its
3444 * request queue @q has been allocated, and runtime PM for it can not happen
3445 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3446 * cases, driver should call this function before any I/O has taken place.
3448 * This function takes care of setting up using auto suspend for the device,
3449 * the autosuspend delay is set to -1 to make runtime suspend impossible
3450 * until an updated value is either set by user or by driver. Drivers do
3451 * not need to touch other autosuspend settings.
3453 * The block layer runtime PM is request based, so only works for drivers
3454 * that use request as their IO unit instead of those directly use bio's.
3456 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3458 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3463 q->rpm_status = RPM_ACTIVE;
3464 pm_runtime_set_autosuspend_delay(q->dev, -1);
3465 pm_runtime_use_autosuspend(q->dev);
3467 EXPORT_SYMBOL(blk_pm_runtime_init);
3470 * blk_pre_runtime_suspend - Pre runtime suspend check
3471 * @q: the queue of the device
3474 * This function will check if runtime suspend is allowed for the device
3475 * by examining if there are any requests pending in the queue. If there
3476 * are requests pending, the device can not be runtime suspended; otherwise,
3477 * the queue's status will be updated to SUSPENDING and the driver can
3478 * proceed to suspend the device.
3480 * For the not allowed case, we mark last busy for the device so that
3481 * runtime PM core will try to autosuspend it some time later.
3483 * This function should be called near the start of the device's
3484 * runtime_suspend callback.
3487 * 0 - OK to runtime suspend the device
3488 * -EBUSY - Device should not be runtime suspended
3490 int blk_pre_runtime_suspend(struct request_queue *q)
3497 spin_lock_irq(q->queue_lock);
3498 if (q->nr_pending) {
3500 pm_runtime_mark_last_busy(q->dev);
3502 q->rpm_status = RPM_SUSPENDING;
3504 spin_unlock_irq(q->queue_lock);
3507 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3510 * blk_post_runtime_suspend - Post runtime suspend processing
3511 * @q: the queue of the device
3512 * @err: return value of the device's runtime_suspend function
3515 * Update the queue's runtime status according to the return value of the
3516 * device's runtime suspend function and mark last busy for the device so
3517 * that PM core will try to auto suspend the device at a later time.
3519 * This function should be called near the end of the device's
3520 * runtime_suspend callback.
3522 void blk_post_runtime_suspend(struct request_queue *q, int err)
3527 spin_lock_irq(q->queue_lock);
3529 q->rpm_status = RPM_SUSPENDED;
3531 q->rpm_status = RPM_ACTIVE;
3532 pm_runtime_mark_last_busy(q->dev);
3534 spin_unlock_irq(q->queue_lock);
3536 EXPORT_SYMBOL(blk_post_runtime_suspend);
3539 * blk_pre_runtime_resume - Pre runtime resume processing
3540 * @q: the queue of the device
3543 * Update the queue's runtime status to RESUMING in preparation for the
3544 * runtime resume of the device.
3546 * This function should be called near the start of the device's
3547 * runtime_resume callback.
3549 void blk_pre_runtime_resume(struct request_queue *q)
3554 spin_lock_irq(q->queue_lock);
3555 q->rpm_status = RPM_RESUMING;
3556 spin_unlock_irq(q->queue_lock);
3558 EXPORT_SYMBOL(blk_pre_runtime_resume);
3561 * blk_post_runtime_resume - Post runtime resume processing
3562 * @q: the queue of the device
3563 * @err: return value of the device's runtime_resume function
3566 * Update the queue's runtime status according to the return value of the
3567 * device's runtime_resume function. If it is successfully resumed, process
3568 * the requests that are queued into the device's queue when it is resuming
3569 * and then mark last busy and initiate autosuspend for it.
3571 * This function should be called near the end of the device's
3572 * runtime_resume callback.
3574 void blk_post_runtime_resume(struct request_queue *q, int err)
3579 spin_lock_irq(q->queue_lock);
3581 q->rpm_status = RPM_ACTIVE;
3583 pm_runtime_mark_last_busy(q->dev);
3584 pm_request_autosuspend(q->dev);
3586 q->rpm_status = RPM_SUSPENDED;
3588 spin_unlock_irq(q->queue_lock);
3590 EXPORT_SYMBOL(blk_post_runtime_resume);
3593 * blk_set_runtime_active - Force runtime status of the queue to be active
3594 * @q: the queue of the device
3596 * If the device is left runtime suspended during system suspend the resume
3597 * hook typically resumes the device and corrects runtime status
3598 * accordingly. However, that does not affect the queue runtime PM status
3599 * which is still "suspended". This prevents processing requests from the
3602 * This function can be used in driver's resume hook to correct queue
3603 * runtime PM status and re-enable peeking requests from the queue. It
3604 * should be called before first request is added to the queue.
3606 void blk_set_runtime_active(struct request_queue *q)
3608 spin_lock_irq(q->queue_lock);
3609 q->rpm_status = RPM_ACTIVE;
3610 pm_runtime_mark_last_busy(q->dev);
3611 pm_request_autosuspend(q->dev);
3612 spin_unlock_irq(q->queue_lock);
3614 EXPORT_SYMBOL(blk_set_runtime_active);
3617 int __init blk_dev_init(void)
3619 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3620 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3621 FIELD_SIZEOF(struct request, cmd_flags));
3622 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3623 FIELD_SIZEOF(struct bio, bi_opf));
3625 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3626 kblockd_workqueue = alloc_workqueue("kblockd",
3627 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3628 if (!kblockd_workqueue)
3629 panic("Failed to create kblockd\n");
3631 request_cachep = kmem_cache_create("blkdev_requests",
3632 sizeof(struct request), 0, SLAB_PANIC, NULL);
3634 blk_requestq_cachep = kmem_cache_create("request_queue",
3635 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3637 #ifdef CONFIG_DEBUG_FS
3638 blk_debugfs_root = debugfs_create_dir("block", NULL);