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>
37 #include <linux/bpf.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/block.h>
44 #include "blk-mq-sched.h"
45 #include "blk-rq-qos.h"
47 #ifdef CONFIG_DEBUG_FS
48 struct dentry *blk_debugfs_root;
51 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
57 DEFINE_IDA(blk_queue_ida);
60 * For the allocated request tables
62 struct kmem_cache *request_cachep;
65 * For queue allocation
67 struct kmem_cache *blk_requestq_cachep;
70 * Controlling structure to kblockd
72 static struct workqueue_struct *kblockd_workqueue;
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
83 spin_lock_irqsave(q->queue_lock, flags);
84 queue_flag_set(flag, q);
85 spin_unlock_irqrestore(q->queue_lock, flags);
87 EXPORT_SYMBOL(blk_queue_flag_set);
90 * blk_queue_flag_clear - atomically clear a queue flag
91 * @flag: flag to be cleared
94 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
98 spin_lock_irqsave(q->queue_lock, flags);
99 queue_flag_clear(flag, q);
100 spin_unlock_irqrestore(q->queue_lock, flags);
102 EXPORT_SYMBOL(blk_queue_flag_clear);
105 * blk_queue_flag_test_and_set - atomically test and set a queue flag
106 * @flag: flag to be set
109 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
110 * the flag was already set.
112 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
117 spin_lock_irqsave(q->queue_lock, flags);
118 res = queue_flag_test_and_set(flag, q);
119 spin_unlock_irqrestore(q->queue_lock, flags);
123 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
126 * blk_queue_flag_test_and_clear - atomically test and clear a queue flag
127 * @flag: flag to be cleared
130 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
133 bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q)
138 spin_lock_irqsave(q->queue_lock, flags);
139 res = queue_flag_test_and_clear(flag, q);
140 spin_unlock_irqrestore(q->queue_lock, flags);
144 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_clear);
146 static void blk_clear_congested(struct request_list *rl, int sync)
148 #ifdef CONFIG_CGROUP_WRITEBACK
149 clear_wb_congested(rl->blkg->wb_congested, sync);
152 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
153 * flip its congestion state for events on other blkcgs.
155 if (rl == &rl->q->root_rl)
156 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
160 static void blk_set_congested(struct request_list *rl, int sync)
162 #ifdef CONFIG_CGROUP_WRITEBACK
163 set_wb_congested(rl->blkg->wb_congested, sync);
165 /* see blk_clear_congested() */
166 if (rl == &rl->q->root_rl)
167 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
171 void blk_queue_congestion_threshold(struct request_queue *q)
175 nr = q->nr_requests - (q->nr_requests / 8) + 1;
176 if (nr > q->nr_requests)
178 q->nr_congestion_on = nr;
180 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
183 q->nr_congestion_off = nr;
186 void blk_rq_init(struct request_queue *q, struct request *rq)
188 memset(rq, 0, sizeof(*rq));
190 INIT_LIST_HEAD(&rq->queuelist);
191 INIT_LIST_HEAD(&rq->timeout_list);
194 rq->__sector = (sector_t) -1;
195 INIT_HLIST_NODE(&rq->hash);
196 RB_CLEAR_NODE(&rq->rb_node);
198 rq->internal_tag = -1;
199 rq->start_time_ns = ktime_get_ns();
201 refcount_set(&rq->ref, 1);
203 EXPORT_SYMBOL(blk_rq_init);
205 static const struct {
209 [BLK_STS_OK] = { 0, "" },
210 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
211 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
212 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
213 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
214 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
215 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
216 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
217 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
218 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
219 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
220 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
222 /* device mapper special case, should not leak out: */
223 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
225 /* everything else not covered above: */
226 [BLK_STS_IOERR] = { -EIO, "I/O" },
229 blk_status_t errno_to_blk_status(int errno)
233 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
234 if (blk_errors[i].errno == errno)
235 return (__force blk_status_t)i;
238 return BLK_STS_IOERR;
240 EXPORT_SYMBOL_GPL(errno_to_blk_status);
242 int blk_status_to_errno(blk_status_t status)
244 int idx = (__force int)status;
246 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
248 return blk_errors[idx].errno;
250 EXPORT_SYMBOL_GPL(blk_status_to_errno);
252 static void print_req_error(struct request *req, blk_status_t status)
254 int idx = (__force int)status;
256 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
259 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
260 __func__, blk_errors[idx].name, req->rq_disk ?
261 req->rq_disk->disk_name : "?",
262 (unsigned long long)blk_rq_pos(req));
265 static void req_bio_endio(struct request *rq, struct bio *bio,
266 unsigned int nbytes, blk_status_t error)
269 bio->bi_status = error;
271 if (unlikely(rq->rq_flags & RQF_QUIET))
272 bio_set_flag(bio, BIO_QUIET);
274 bio_advance(bio, nbytes);
276 /* don't actually finish bio if it's part of flush sequence */
277 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
281 void blk_dump_rq_flags(struct request *rq, char *msg)
283 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
284 rq->rq_disk ? rq->rq_disk->disk_name : "?",
285 (unsigned long long) rq->cmd_flags);
287 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
288 (unsigned long long)blk_rq_pos(rq),
289 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
290 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
291 rq->bio, rq->biotail, blk_rq_bytes(rq));
293 EXPORT_SYMBOL(blk_dump_rq_flags);
295 static void blk_delay_work(struct work_struct *work)
297 struct request_queue *q;
299 q = container_of(work, struct request_queue, delay_work.work);
300 spin_lock_irq(q->queue_lock);
302 spin_unlock_irq(q->queue_lock);
306 * blk_delay_queue - restart queueing after defined interval
307 * @q: The &struct request_queue in question
308 * @msecs: Delay in msecs
311 * Sometimes queueing needs to be postponed for a little while, to allow
312 * resources to come back. This function will make sure that queueing is
313 * restarted around the specified time.
315 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
317 lockdep_assert_held(q->queue_lock);
318 WARN_ON_ONCE(q->mq_ops);
320 if (likely(!blk_queue_dead(q)))
321 queue_delayed_work(kblockd_workqueue, &q->delay_work,
322 msecs_to_jiffies(msecs));
324 EXPORT_SYMBOL(blk_delay_queue);
327 * blk_start_queue_async - asynchronously restart a previously stopped queue
328 * @q: The &struct request_queue in question
331 * blk_start_queue_async() will clear the stop flag on the queue, and
332 * ensure that the request_fn for the queue is run from an async
335 void blk_start_queue_async(struct request_queue *q)
337 lockdep_assert_held(q->queue_lock);
338 WARN_ON_ONCE(q->mq_ops);
340 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
341 blk_run_queue_async(q);
343 EXPORT_SYMBOL(blk_start_queue_async);
346 * blk_start_queue - restart a previously stopped queue
347 * @q: The &struct request_queue in question
350 * blk_start_queue() will clear the stop flag on the queue, and call
351 * the request_fn for the queue if it was in a stopped state when
352 * entered. Also see blk_stop_queue().
354 void blk_start_queue(struct request_queue *q)
356 lockdep_assert_held(q->queue_lock);
357 WARN_ON_ONCE(q->mq_ops);
359 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
362 EXPORT_SYMBOL(blk_start_queue);
365 * blk_stop_queue - stop a queue
366 * @q: The &struct request_queue in question
369 * The Linux block layer assumes that a block driver will consume all
370 * entries on the request queue when the request_fn strategy is called.
371 * Often this will not happen, because of hardware limitations (queue
372 * depth settings). If a device driver gets a 'queue full' response,
373 * or if it simply chooses not to queue more I/O at one point, it can
374 * call this function to prevent the request_fn from being called until
375 * the driver has signalled it's ready to go again. This happens by calling
376 * blk_start_queue() to restart queue operations.
378 void blk_stop_queue(struct request_queue *q)
380 lockdep_assert_held(q->queue_lock);
381 WARN_ON_ONCE(q->mq_ops);
383 cancel_delayed_work(&q->delay_work);
384 queue_flag_set(QUEUE_FLAG_STOPPED, q);
386 EXPORT_SYMBOL(blk_stop_queue);
389 * blk_sync_queue - cancel any pending callbacks on a queue
393 * The block layer may perform asynchronous callback activity
394 * on a queue, such as calling the unplug function after a timeout.
395 * A block device may call blk_sync_queue to ensure that any
396 * such activity is cancelled, thus allowing it to release resources
397 * that the callbacks might use. The caller must already have made sure
398 * that its ->make_request_fn will not re-add plugging prior to calling
401 * This function does not cancel any asynchronous activity arising
402 * out of elevator or throttling code. That would require elevator_exit()
403 * and blkcg_exit_queue() to be called with queue lock initialized.
406 void blk_sync_queue(struct request_queue *q)
408 del_timer_sync(&q->timeout);
409 cancel_work_sync(&q->timeout_work);
412 struct blk_mq_hw_ctx *hctx;
415 queue_for_each_hw_ctx(q, hctx, i)
416 cancel_delayed_work_sync(&hctx->run_work);
418 cancel_delayed_work_sync(&q->delay_work);
421 EXPORT_SYMBOL(blk_sync_queue);
424 * blk_set_pm_only - increment pm_only counter
425 * @q: request queue pointer
427 void blk_set_pm_only(struct request_queue *q)
429 atomic_inc(&q->pm_only);
431 EXPORT_SYMBOL_GPL(blk_set_pm_only);
433 void blk_clear_pm_only(struct request_queue *q)
437 pm_only = atomic_dec_return(&q->pm_only);
438 WARN_ON_ONCE(pm_only < 0);
440 wake_up_all(&q->mq_freeze_wq);
442 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
445 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
446 * @q: The queue to run
449 * Invoke request handling on a queue if there are any pending requests.
450 * May be used to restart request handling after a request has completed.
451 * This variant runs the queue whether or not the queue has been
452 * stopped. Must be called with the queue lock held and interrupts
453 * disabled. See also @blk_run_queue.
455 inline void __blk_run_queue_uncond(struct request_queue *q)
457 lockdep_assert_held(q->queue_lock);
458 WARN_ON_ONCE(q->mq_ops);
460 if (unlikely(blk_queue_dead(q)))
464 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
465 * the queue lock internally. As a result multiple threads may be
466 * running such a request function concurrently. Keep track of the
467 * number of active request_fn invocations such that blk_drain_queue()
468 * can wait until all these request_fn calls have finished.
470 q->request_fn_active++;
472 q->request_fn_active--;
474 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
477 * __blk_run_queue - run a single device queue
478 * @q: The queue to run
481 * See @blk_run_queue.
483 void __blk_run_queue(struct request_queue *q)
485 lockdep_assert_held(q->queue_lock);
486 WARN_ON_ONCE(q->mq_ops);
488 if (unlikely(blk_queue_stopped(q)))
491 __blk_run_queue_uncond(q);
493 EXPORT_SYMBOL(__blk_run_queue);
496 * blk_run_queue_async - run a single device queue in workqueue context
497 * @q: The queue to run
500 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
504 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
505 * has canceled q->delay_work, callers must hold the queue lock to avoid
506 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
508 void blk_run_queue_async(struct request_queue *q)
510 lockdep_assert_held(q->queue_lock);
511 WARN_ON_ONCE(q->mq_ops);
513 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
514 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
516 EXPORT_SYMBOL(blk_run_queue_async);
519 * blk_run_queue - run a single device queue
520 * @q: The queue to run
523 * Invoke request handling on this queue, if it has pending work to do.
524 * May be used to restart queueing when a request has completed.
526 void blk_run_queue(struct request_queue *q)
530 WARN_ON_ONCE(q->mq_ops);
532 spin_lock_irqsave(q->queue_lock, flags);
534 spin_unlock_irqrestore(q->queue_lock, flags);
536 EXPORT_SYMBOL(blk_run_queue);
538 void blk_put_queue(struct request_queue *q)
540 kobject_put(&q->kobj);
542 EXPORT_SYMBOL(blk_put_queue);
545 * __blk_drain_queue - drain requests from request_queue
547 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
549 * Drain requests from @q. If @drain_all is set, all requests are drained.
550 * If not, only ELVPRIV requests are drained. The caller is responsible
551 * for ensuring that no new requests which need to be drained are queued.
553 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
554 __releases(q->queue_lock)
555 __acquires(q->queue_lock)
559 lockdep_assert_held(q->queue_lock);
560 WARN_ON_ONCE(q->mq_ops);
566 * The caller might be trying to drain @q before its
567 * elevator is initialized.
570 elv_drain_elevator(q);
572 blkcg_drain_queue(q);
575 * This function might be called on a queue which failed
576 * driver init after queue creation or is not yet fully
577 * active yet. Some drivers (e.g. fd and loop) get unhappy
578 * in such cases. Kick queue iff dispatch queue has
579 * something on it and @q has request_fn set.
581 if (!list_empty(&q->queue_head) && q->request_fn)
584 drain |= q->nr_rqs_elvpriv;
585 drain |= q->request_fn_active;
588 * Unfortunately, requests are queued at and tracked from
589 * multiple places and there's no single counter which can
590 * be drained. Check all the queues and counters.
593 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
594 drain |= !list_empty(&q->queue_head);
595 for (i = 0; i < 2; i++) {
596 drain |= q->nr_rqs[i];
597 drain |= q->in_flight[i];
599 drain |= !list_empty(&fq->flush_queue[i]);
606 spin_unlock_irq(q->queue_lock);
610 spin_lock_irq(q->queue_lock);
614 * With queue marked dead, any woken up waiter will fail the
615 * allocation path, so the wakeup chaining is lost and we're
616 * left with hung waiters. We need to wake up those waiters.
619 struct request_list *rl;
621 blk_queue_for_each_rl(rl, q)
622 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
623 wake_up_all(&rl->wait[i]);
627 void blk_drain_queue(struct request_queue *q)
629 spin_lock_irq(q->queue_lock);
630 __blk_drain_queue(q, true);
631 spin_unlock_irq(q->queue_lock);
635 * blk_queue_bypass_start - enter queue bypass mode
636 * @q: queue of interest
638 * In bypass mode, only the dispatch FIFO queue of @q is used. This
639 * function makes @q enter bypass mode and drains all requests which were
640 * throttled or issued before. On return, it's guaranteed that no request
641 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
642 * inside queue or RCU read lock.
644 void blk_queue_bypass_start(struct request_queue *q)
646 WARN_ON_ONCE(q->mq_ops);
648 spin_lock_irq(q->queue_lock);
650 queue_flag_set(QUEUE_FLAG_BYPASS, q);
651 spin_unlock_irq(q->queue_lock);
654 * Queues start drained. Skip actual draining till init is
655 * complete. This avoids lenghty delays during queue init which
656 * can happen many times during boot.
658 if (blk_queue_init_done(q)) {
659 spin_lock_irq(q->queue_lock);
660 __blk_drain_queue(q, false);
661 spin_unlock_irq(q->queue_lock);
663 /* ensure blk_queue_bypass() is %true inside RCU read lock */
667 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
670 * blk_queue_bypass_end - leave queue bypass mode
671 * @q: queue of interest
673 * Leave bypass mode and restore the normal queueing behavior.
675 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
676 * this function is called for both blk-sq and blk-mq queues.
678 void blk_queue_bypass_end(struct request_queue *q)
680 spin_lock_irq(q->queue_lock);
681 if (!--q->bypass_depth)
682 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
683 WARN_ON_ONCE(q->bypass_depth < 0);
684 spin_unlock_irq(q->queue_lock);
686 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
688 void blk_set_queue_dying(struct request_queue *q)
690 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
693 * When queue DYING flag is set, we need to block new req
694 * entering queue, so we call blk_freeze_queue_start() to
695 * prevent I/O from crossing blk_queue_enter().
697 blk_freeze_queue_start(q);
700 blk_mq_wake_waiters(q);
702 struct request_list *rl;
704 spin_lock_irq(q->queue_lock);
705 blk_queue_for_each_rl(rl, q) {
707 wake_up_all(&rl->wait[BLK_RW_SYNC]);
708 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
711 spin_unlock_irq(q->queue_lock);
714 /* Make blk_queue_enter() reexamine the DYING flag. */
715 wake_up_all(&q->mq_freeze_wq);
717 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
719 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
720 void blk_exit_queue(struct request_queue *q)
723 * Since the I/O scheduler exit code may access cgroup information,
724 * perform I/O scheduler exit before disassociating from the block
729 elevator_exit(q, q->elevator);
734 * Remove all references to @q from the block cgroup controller before
735 * restoring @q->queue_lock to avoid that restoring this pointer causes
736 * e.g. blkcg_print_blkgs() to crash.
741 * Since the cgroup code may dereference the @q->backing_dev_info
742 * pointer, only decrease its reference count after having removed the
743 * association with the block cgroup controller.
745 bdi_put(q->backing_dev_info);
749 * blk_cleanup_queue - shutdown a request queue
750 * @q: request queue to shutdown
752 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
753 * put it. All future requests will be failed immediately with -ENODEV.
755 void blk_cleanup_queue(struct request_queue *q)
757 spinlock_t *lock = q->queue_lock;
759 /* mark @q DYING, no new request or merges will be allowed afterwards */
760 mutex_lock(&q->sysfs_lock);
761 blk_set_queue_dying(q);
765 * A dying queue is permanently in bypass mode till released. Note
766 * that, unlike blk_queue_bypass_start(), we aren't performing
767 * synchronize_rcu() after entering bypass mode to avoid the delay
768 * as some drivers create and destroy a lot of queues while
769 * probing. This is still safe because blk_release_queue() will be
770 * called only after the queue refcnt drops to zero and nothing,
771 * RCU or not, would be traversing the queue by then.
774 queue_flag_set(QUEUE_FLAG_BYPASS, q);
776 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
777 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
778 queue_flag_set(QUEUE_FLAG_DYING, q);
779 spin_unlock_irq(lock);
780 mutex_unlock(&q->sysfs_lock);
783 * Drain all requests queued before DYING marking. Set DEAD flag to
784 * prevent that q->request_fn() gets invoked after draining finished.
788 queue_flag_set(QUEUE_FLAG_DEAD, q);
789 spin_unlock_irq(lock);
792 * make sure all in-progress dispatch are completed because
793 * blk_freeze_queue() can only complete all requests, and
794 * dispatch may still be in-progress since we dispatch requests
795 * from more than one contexts.
797 * We rely on driver to deal with the race in case that queue
798 * initialization isn't done.
800 if (q->mq_ops && blk_queue_init_done(q))
801 blk_mq_quiesce_queue(q);
803 /* for synchronous bio-based driver finish in-flight integrity i/o */
804 blk_flush_integrity();
806 /* @q won't process any more request, flush async actions */
807 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
811 * I/O scheduler exit is only safe after the sysfs scheduler attribute
814 WARN_ON_ONCE(q->kobj.state_in_sysfs);
819 blk_mq_exit_queue(q);
821 percpu_ref_exit(&q->q_usage_counter);
824 if (q->queue_lock != &q->__queue_lock)
825 q->queue_lock = &q->__queue_lock;
826 spin_unlock_irq(lock);
828 /* @q is and will stay empty, shutdown and put */
831 EXPORT_SYMBOL(blk_cleanup_queue);
833 /* Allocate memory local to the request queue */
834 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
836 struct request_queue *q = data;
838 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
841 static void free_request_simple(void *element, void *data)
843 kmem_cache_free(request_cachep, element);
846 static void *alloc_request_size(gfp_t gfp_mask, void *data)
848 struct request_queue *q = data;
851 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
853 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
860 static void free_request_size(void *element, void *data)
862 struct request_queue *q = data;
865 q->exit_rq_fn(q, element);
869 int blk_init_rl(struct request_list *rl, struct request_queue *q,
872 if (unlikely(rl->rq_pool) || q->mq_ops)
876 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
877 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
878 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
879 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
882 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
883 alloc_request_size, free_request_size,
884 q, gfp_mask, q->node);
886 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
887 alloc_request_simple, free_request_simple,
888 q, gfp_mask, q->node);
893 if (rl != &q->root_rl)
894 WARN_ON_ONCE(!blk_get_queue(q));
899 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
902 mempool_destroy(rl->rq_pool);
903 if (rl != &q->root_rl)
908 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
910 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL);
912 EXPORT_SYMBOL(blk_alloc_queue);
915 * blk_queue_enter() - try to increase q->q_usage_counter
916 * @q: request queue pointer
917 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
919 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
921 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
924 bool success = false;
927 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
929 * The code that increments the pm_only counter is
930 * responsible for ensuring that that counter is
931 * globally visible before the queue is unfrozen.
933 if (pm || !blk_queue_pm_only(q)) {
936 percpu_ref_put(&q->q_usage_counter);
944 if (flags & BLK_MQ_REQ_NOWAIT)
948 * read pair of barrier in blk_freeze_queue_start(),
949 * we need to order reading __PERCPU_REF_DEAD flag of
950 * .q_usage_counter and reading .mq_freeze_depth or
951 * queue dying flag, otherwise the following wait may
952 * never return if the two reads are reordered.
956 wait_event(q->mq_freeze_wq,
957 (atomic_read(&q->mq_freeze_depth) == 0 &&
958 (pm || !blk_queue_pm_only(q))) ||
960 if (blk_queue_dying(q))
965 void blk_queue_exit(struct request_queue *q)
967 percpu_ref_put(&q->q_usage_counter);
970 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
972 struct request_queue *q =
973 container_of(ref, struct request_queue, q_usage_counter);
975 wake_up_all(&q->mq_freeze_wq);
978 static void blk_rq_timed_out_timer(struct timer_list *t)
980 struct request_queue *q = from_timer(q, t, timeout);
982 kblockd_schedule_work(&q->timeout_work);
985 static void blk_timeout_work_dummy(struct work_struct *work)
990 * blk_alloc_queue_node - allocate a request queue
991 * @gfp_mask: memory allocation flags
992 * @node_id: NUMA node to allocate memory from
993 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
994 * serialize calls to the legacy .request_fn() callback. Ignored for
995 * blk-mq request queues.
997 * Note: pass the queue lock as the third argument to this function instead of
998 * setting the queue lock pointer explicitly to avoid triggering a sporadic
999 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
1000 * the queue lock pointer must be set before blkcg_init_queue() is called.
1002 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
1005 struct request_queue *q;
1008 q = kmem_cache_alloc_node(blk_requestq_cachep,
1009 gfp_mask | __GFP_ZERO, node_id);
1013 INIT_LIST_HEAD(&q->queue_head);
1014 q->last_merge = NULL;
1016 q->boundary_rq = NULL;
1018 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
1022 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1026 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1027 if (!q->backing_dev_info)
1030 q->stats = blk_alloc_queue_stats();
1034 q->backing_dev_info->ra_pages =
1035 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1036 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1037 q->backing_dev_info->name = "block";
1040 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1041 laptop_mode_timer_fn, 0);
1042 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
1043 INIT_WORK(&q->timeout_work, blk_timeout_work_dummy);
1044 INIT_LIST_HEAD(&q->timeout_list);
1045 INIT_LIST_HEAD(&q->icq_list);
1046 #ifdef CONFIG_BLK_CGROUP
1047 INIT_LIST_HEAD(&q->blkg_list);
1049 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
1051 kobject_init(&q->kobj, &blk_queue_ktype);
1053 #ifdef CONFIG_BLK_DEV_IO_TRACE
1054 mutex_init(&q->blk_trace_mutex);
1056 mutex_init(&q->sysfs_lock);
1057 spin_lock_init(&q->__queue_lock);
1060 q->queue_lock = lock ? : &q->__queue_lock;
1063 * A queue starts its life with bypass turned on to avoid
1064 * unnecessary bypass on/off overhead and nasty surprises during
1065 * init. The initial bypass will be finished when the queue is
1066 * registered by blk_register_queue().
1068 q->bypass_depth = 1;
1069 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
1071 init_waitqueue_head(&q->mq_freeze_wq);
1074 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1075 * See blk_register_queue() for details.
1077 if (percpu_ref_init(&q->q_usage_counter,
1078 blk_queue_usage_counter_release,
1079 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
1082 if (blkcg_init_queue(q))
1088 percpu_ref_exit(&q->q_usage_counter);
1090 blk_free_queue_stats(q->stats);
1092 bdi_put(q->backing_dev_info);
1094 bioset_exit(&q->bio_split);
1096 ida_simple_remove(&blk_queue_ida, q->id);
1098 kmem_cache_free(blk_requestq_cachep, q);
1101 EXPORT_SYMBOL(blk_alloc_queue_node);
1104 * blk_init_queue - prepare a request queue for use with a block device
1105 * @rfn: The function to be called to process requests that have been
1106 * placed on the queue.
1107 * @lock: Request queue spin lock
1110 * If a block device wishes to use the standard request handling procedures,
1111 * which sorts requests and coalesces adjacent requests, then it must
1112 * call blk_init_queue(). The function @rfn will be called when there
1113 * are requests on the queue that need to be processed. If the device
1114 * supports plugging, then @rfn may not be called immediately when requests
1115 * are available on the queue, but may be called at some time later instead.
1116 * Plugged queues are generally unplugged when a buffer belonging to one
1117 * of the requests on the queue is needed, or due to memory pressure.
1119 * @rfn is not required, or even expected, to remove all requests off the
1120 * queue, but only as many as it can handle at a time. If it does leave
1121 * requests on the queue, it is responsible for arranging that the requests
1122 * get dealt with eventually.
1124 * The queue spin lock must be held while manipulating the requests on the
1125 * request queue; this lock will be taken also from interrupt context, so irq
1126 * disabling is needed for it.
1128 * Function returns a pointer to the initialized request queue, or %NULL if
1129 * it didn't succeed.
1132 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1133 * when the block device is deactivated (such as at module unload).
1136 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1138 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1140 EXPORT_SYMBOL(blk_init_queue);
1142 struct request_queue *
1143 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1145 struct request_queue *q;
1147 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
1151 q->request_fn = rfn;
1152 if (blk_init_allocated_queue(q) < 0) {
1153 blk_cleanup_queue(q);
1159 EXPORT_SYMBOL(blk_init_queue_node);
1161 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1164 int blk_init_allocated_queue(struct request_queue *q)
1166 WARN_ON_ONCE(q->mq_ops);
1168 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size, GFP_KERNEL);
1172 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1173 goto out_free_flush_queue;
1175 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1176 goto out_exit_flush_rq;
1178 INIT_WORK(&q->timeout_work, blk_timeout_work);
1179 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1182 * This also sets hw/phys segments, boundary and size
1184 blk_queue_make_request(q, blk_queue_bio);
1186 q->sg_reserved_size = INT_MAX;
1188 if (elevator_init(q))
1189 goto out_exit_flush_rq;
1194 q->exit_rq_fn(q, q->fq->flush_rq);
1195 out_free_flush_queue:
1196 blk_free_flush_queue(q->fq);
1200 EXPORT_SYMBOL(blk_init_allocated_queue);
1202 bool blk_get_queue(struct request_queue *q)
1204 if (likely(!blk_queue_dying(q))) {
1211 EXPORT_SYMBOL(blk_get_queue);
1213 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1215 if (rq->rq_flags & RQF_ELVPRIV) {
1216 elv_put_request(rl->q, rq);
1218 put_io_context(rq->elv.icq->ioc);
1221 mempool_free(rq, rl->rq_pool);
1225 * ioc_batching returns true if the ioc is a valid batching request and
1226 * should be given priority access to a request.
1228 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1234 * Make sure the process is able to allocate at least 1 request
1235 * even if the batch times out, otherwise we could theoretically
1238 return ioc->nr_batch_requests == q->nr_batching ||
1239 (ioc->nr_batch_requests > 0
1240 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1244 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1245 * will cause the process to be a "batcher" on all queues in the system. This
1246 * is the behaviour we want though - once it gets a wakeup it should be given
1249 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1251 if (!ioc || ioc_batching(q, ioc))
1254 ioc->nr_batch_requests = q->nr_batching;
1255 ioc->last_waited = jiffies;
1258 static void __freed_request(struct request_list *rl, int sync)
1260 struct request_queue *q = rl->q;
1262 if (rl->count[sync] < queue_congestion_off_threshold(q))
1263 blk_clear_congested(rl, sync);
1265 if (rl->count[sync] + 1 <= q->nr_requests) {
1266 if (waitqueue_active(&rl->wait[sync]))
1267 wake_up(&rl->wait[sync]);
1269 blk_clear_rl_full(rl, sync);
1274 * A request has just been released. Account for it, update the full and
1275 * congestion status, wake up any waiters. Called under q->queue_lock.
1277 static void freed_request(struct request_list *rl, bool sync,
1278 req_flags_t rq_flags)
1280 struct request_queue *q = rl->q;
1284 if (rq_flags & RQF_ELVPRIV)
1285 q->nr_rqs_elvpriv--;
1287 __freed_request(rl, sync);
1289 if (unlikely(rl->starved[sync ^ 1]))
1290 __freed_request(rl, sync ^ 1);
1293 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1295 struct request_list *rl;
1296 int on_thresh, off_thresh;
1298 WARN_ON_ONCE(q->mq_ops);
1300 spin_lock_irq(q->queue_lock);
1301 q->nr_requests = nr;
1302 blk_queue_congestion_threshold(q);
1303 on_thresh = queue_congestion_on_threshold(q);
1304 off_thresh = queue_congestion_off_threshold(q);
1306 blk_queue_for_each_rl(rl, q) {
1307 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1308 blk_set_congested(rl, BLK_RW_SYNC);
1309 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1310 blk_clear_congested(rl, BLK_RW_SYNC);
1312 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1313 blk_set_congested(rl, BLK_RW_ASYNC);
1314 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1315 blk_clear_congested(rl, BLK_RW_ASYNC);
1317 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1318 blk_set_rl_full(rl, BLK_RW_SYNC);
1320 blk_clear_rl_full(rl, BLK_RW_SYNC);
1321 wake_up(&rl->wait[BLK_RW_SYNC]);
1324 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1325 blk_set_rl_full(rl, BLK_RW_ASYNC);
1327 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1328 wake_up(&rl->wait[BLK_RW_ASYNC]);
1332 spin_unlock_irq(q->queue_lock);
1337 * __get_request - get a free request
1338 * @rl: request list to allocate from
1339 * @op: operation and flags
1340 * @bio: bio to allocate request for (can be %NULL)
1341 * @flags: BLQ_MQ_REQ_* flags
1342 * @gfp_mask: allocator flags
1344 * Get a free request from @q. This function may fail under memory
1345 * pressure or if @q is dead.
1347 * Must be called with @q->queue_lock held and,
1348 * Returns ERR_PTR on failure, with @q->queue_lock held.
1349 * Returns request pointer on success, with @q->queue_lock *not held*.
1351 static struct request *__get_request(struct request_list *rl, unsigned int op,
1352 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask)
1354 struct request_queue *q = rl->q;
1356 struct elevator_type *et = q->elevator->type;
1357 struct io_context *ioc = rq_ioc(bio);
1358 struct io_cq *icq = NULL;
1359 const bool is_sync = op_is_sync(op);
1361 req_flags_t rq_flags = RQF_ALLOCED;
1363 lockdep_assert_held(q->queue_lock);
1365 if (unlikely(blk_queue_dying(q)))
1366 return ERR_PTR(-ENODEV);
1368 may_queue = elv_may_queue(q, op);
1369 if (may_queue == ELV_MQUEUE_NO)
1372 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1373 if (rl->count[is_sync]+1 >= q->nr_requests) {
1375 * The queue will fill after this allocation, so set
1376 * it as full, and mark this process as "batching".
1377 * This process will be allowed to complete a batch of
1378 * requests, others will be blocked.
1380 if (!blk_rl_full(rl, is_sync)) {
1381 ioc_set_batching(q, ioc);
1382 blk_set_rl_full(rl, is_sync);
1384 if (may_queue != ELV_MQUEUE_MUST
1385 && !ioc_batching(q, ioc)) {
1387 * The queue is full and the allocating
1388 * process is not a "batcher", and not
1389 * exempted by the IO scheduler
1391 return ERR_PTR(-ENOMEM);
1395 blk_set_congested(rl, is_sync);
1399 * Only allow batching queuers to allocate up to 50% over the defined
1400 * limit of requests, otherwise we could have thousands of requests
1401 * allocated with any setting of ->nr_requests
1403 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1404 return ERR_PTR(-ENOMEM);
1406 q->nr_rqs[is_sync]++;
1407 rl->count[is_sync]++;
1408 rl->starved[is_sync] = 0;
1411 * Decide whether the new request will be managed by elevator. If
1412 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1413 * prevent the current elevator from being destroyed until the new
1414 * request is freed. This guarantees icq's won't be destroyed and
1415 * makes creating new ones safe.
1417 * Flush requests do not use the elevator so skip initialization.
1418 * This allows a request to share the flush and elevator data.
1420 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1421 * it will be created after releasing queue_lock.
1423 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1424 rq_flags |= RQF_ELVPRIV;
1425 q->nr_rqs_elvpriv++;
1426 if (et->icq_cache && ioc)
1427 icq = ioc_lookup_icq(ioc, q);
1430 if (blk_queue_io_stat(q))
1431 rq_flags |= RQF_IO_STAT;
1432 spin_unlock_irq(q->queue_lock);
1434 /* allocate and init request */
1435 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1440 blk_rq_set_rl(rq, rl);
1442 rq->rq_flags = rq_flags;
1443 if (flags & BLK_MQ_REQ_PREEMPT)
1444 rq->rq_flags |= RQF_PREEMPT;
1447 if (rq_flags & RQF_ELVPRIV) {
1448 if (unlikely(et->icq_cache && !icq)) {
1450 icq = ioc_create_icq(ioc, q, gfp_mask);
1456 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1459 /* @rq->elv.icq holds io_context until @rq is freed */
1461 get_io_context(icq->ioc);
1465 * ioc may be NULL here, and ioc_batching will be false. That's
1466 * OK, if the queue is under the request limit then requests need
1467 * not count toward the nr_batch_requests limit. There will always
1468 * be some limit enforced by BLK_BATCH_TIME.
1470 if (ioc_batching(q, ioc))
1471 ioc->nr_batch_requests--;
1473 trace_block_getrq(q, bio, op);
1478 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1479 * and may fail indefinitely under memory pressure and thus
1480 * shouldn't stall IO. Treat this request as !elvpriv. This will
1481 * disturb iosched and blkcg but weird is bettern than dead.
1483 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1484 __func__, dev_name(q->backing_dev_info->dev));
1486 rq->rq_flags &= ~RQF_ELVPRIV;
1489 spin_lock_irq(q->queue_lock);
1490 q->nr_rqs_elvpriv--;
1491 spin_unlock_irq(q->queue_lock);
1496 * Allocation failed presumably due to memory. Undo anything we
1497 * might have messed up.
1499 * Allocating task should really be put onto the front of the wait
1500 * queue, but this is pretty rare.
1502 spin_lock_irq(q->queue_lock);
1503 freed_request(rl, is_sync, rq_flags);
1506 * in the very unlikely event that allocation failed and no
1507 * requests for this direction was pending, mark us starved so that
1508 * freeing of a request in the other direction will notice
1509 * us. another possible fix would be to split the rq mempool into
1513 if (unlikely(rl->count[is_sync] == 0))
1514 rl->starved[is_sync] = 1;
1515 return ERR_PTR(-ENOMEM);
1519 * get_request - get a free request
1520 * @q: request_queue to allocate request from
1521 * @op: operation and flags
1522 * @bio: bio to allocate request for (can be %NULL)
1523 * @flags: BLK_MQ_REQ_* flags.
1524 * @gfp: allocator flags
1526 * Get a free request from @q. If %BLK_MQ_REQ_NOWAIT is set in @flags,
1527 * this function keeps retrying under memory pressure and fails iff @q is dead.
1529 * Must be called with @q->queue_lock held and,
1530 * Returns ERR_PTR on failure, with @q->queue_lock held.
1531 * Returns request pointer on success, with @q->queue_lock *not held*.
1533 static struct request *get_request(struct request_queue *q, unsigned int op,
1534 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp)
1536 const bool is_sync = op_is_sync(op);
1538 struct request_list *rl;
1541 lockdep_assert_held(q->queue_lock);
1542 WARN_ON_ONCE(q->mq_ops);
1544 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1546 rq = __get_request(rl, op, bio, flags, gfp);
1550 if (op & REQ_NOWAIT) {
1552 return ERR_PTR(-EAGAIN);
1555 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1560 /* wait on @rl and retry */
1561 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1562 TASK_UNINTERRUPTIBLE);
1564 trace_block_sleeprq(q, bio, op);
1566 spin_unlock_irq(q->queue_lock);
1570 * After sleeping, we become a "batching" process and will be able
1571 * to allocate at least one request, and up to a big batch of them
1572 * for a small period time. See ioc_batching, ioc_set_batching
1574 ioc_set_batching(q, current->io_context);
1576 spin_lock_irq(q->queue_lock);
1577 finish_wait(&rl->wait[is_sync], &wait);
1582 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1583 static struct request *blk_old_get_request(struct request_queue *q,
1584 unsigned int op, blk_mq_req_flags_t flags)
1587 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO;
1590 WARN_ON_ONCE(q->mq_ops);
1592 /* create ioc upfront */
1593 create_io_context(gfp_mask, q->node);
1595 ret = blk_queue_enter(q, flags);
1597 return ERR_PTR(ret);
1598 spin_lock_irq(q->queue_lock);
1599 rq = get_request(q, op, NULL, flags, gfp_mask);
1601 spin_unlock_irq(q->queue_lock);
1606 /* q->queue_lock is unlocked at this point */
1608 rq->__sector = (sector_t) -1;
1609 rq->bio = rq->biotail = NULL;
1614 * blk_get_request - allocate a request
1615 * @q: request queue to allocate a request for
1616 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1617 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1619 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1620 blk_mq_req_flags_t flags)
1622 struct request *req;
1624 WARN_ON_ONCE(op & REQ_NOWAIT);
1625 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1628 req = blk_mq_alloc_request(q, op, flags);
1629 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1630 q->mq_ops->initialize_rq_fn(req);
1632 req = blk_old_get_request(q, op, flags);
1633 if (!IS_ERR(req) && q->initialize_rq_fn)
1634 q->initialize_rq_fn(req);
1639 EXPORT_SYMBOL(blk_get_request);
1642 * blk_requeue_request - put a request back on queue
1643 * @q: request queue where request should be inserted
1644 * @rq: request to be inserted
1647 * Drivers often keep queueing requests until the hardware cannot accept
1648 * more, when that condition happens we need to put the request back
1649 * on the queue. Must be called with queue lock held.
1651 void blk_requeue_request(struct request_queue *q, struct request *rq)
1653 lockdep_assert_held(q->queue_lock);
1654 WARN_ON_ONCE(q->mq_ops);
1656 blk_delete_timer(rq);
1657 blk_clear_rq_complete(rq);
1658 trace_block_rq_requeue(q, rq);
1659 rq_qos_requeue(q, rq);
1661 if (rq->rq_flags & RQF_QUEUED)
1662 blk_queue_end_tag(q, rq);
1664 BUG_ON(blk_queued_rq(rq));
1666 elv_requeue_request(q, rq);
1668 EXPORT_SYMBOL(blk_requeue_request);
1670 static void add_acct_request(struct request_queue *q, struct request *rq,
1673 blk_account_io_start(rq, true);
1674 __elv_add_request(q, rq, where);
1677 static void part_round_stats_single(struct request_queue *q, int cpu,
1678 struct hd_struct *part, unsigned long now,
1679 unsigned int inflight)
1682 __part_stat_add(cpu, part, time_in_queue,
1683 inflight * (now - part->stamp));
1684 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1690 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1691 * @q: target block queue
1692 * @cpu: cpu number for stats access
1693 * @part: target partition
1695 * The average IO queue length and utilisation statistics are maintained
1696 * by observing the current state of the queue length and the amount of
1697 * time it has been in this state for.
1699 * Normally, that accounting is done on IO completion, but that can result
1700 * in more than a second's worth of IO being accounted for within any one
1701 * second, leading to >100% utilisation. To deal with that, we call this
1702 * function to do a round-off before returning the results when reading
1703 * /proc/diskstats. This accounts immediately for all queue usage up to
1704 * the current jiffies and restarts the counters again.
1706 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1708 struct hd_struct *part2 = NULL;
1709 unsigned long now = jiffies;
1710 unsigned int inflight[2];
1713 if (part->stamp != now)
1717 part2 = &part_to_disk(part)->part0;
1718 if (part2->stamp != now)
1725 part_in_flight(q, part, inflight);
1728 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1730 part_round_stats_single(q, cpu, part, now, inflight[0]);
1732 EXPORT_SYMBOL_GPL(part_round_stats);
1735 static void blk_pm_put_request(struct request *rq)
1737 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1738 pm_runtime_mark_last_busy(rq->q->dev);
1741 static inline void blk_pm_put_request(struct request *rq) {}
1744 void __blk_put_request(struct request_queue *q, struct request *req)
1746 req_flags_t rq_flags = req->rq_flags;
1752 blk_mq_free_request(req);
1756 lockdep_assert_held(q->queue_lock);
1758 blk_req_zone_write_unlock(req);
1759 blk_pm_put_request(req);
1761 elv_completed_request(q, req);
1763 /* this is a bio leak */
1764 WARN_ON(req->bio != NULL);
1766 rq_qos_done(q, req);
1769 * Request may not have originated from ll_rw_blk. if not,
1770 * it didn't come out of our reserved rq pools
1772 if (rq_flags & RQF_ALLOCED) {
1773 struct request_list *rl = blk_rq_rl(req);
1774 bool sync = op_is_sync(req->cmd_flags);
1776 BUG_ON(!list_empty(&req->queuelist));
1777 BUG_ON(ELV_ON_HASH(req));
1779 blk_free_request(rl, req);
1780 freed_request(rl, sync, rq_flags);
1785 EXPORT_SYMBOL_GPL(__blk_put_request);
1787 void blk_put_request(struct request *req)
1789 struct request_queue *q = req->q;
1792 blk_mq_free_request(req);
1794 unsigned long flags;
1796 spin_lock_irqsave(q->queue_lock, flags);
1797 __blk_put_request(q, req);
1798 spin_unlock_irqrestore(q->queue_lock, flags);
1801 EXPORT_SYMBOL(blk_put_request);
1803 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1806 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1808 if (!ll_back_merge_fn(q, req, bio))
1811 trace_block_bio_backmerge(q, req, bio);
1813 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1814 blk_rq_set_mixed_merge(req);
1816 req->biotail->bi_next = bio;
1818 req->__data_len += bio->bi_iter.bi_size;
1819 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1821 blk_account_io_start(req, false);
1825 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1828 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1830 if (!ll_front_merge_fn(q, req, bio))
1833 trace_block_bio_frontmerge(q, req, bio);
1835 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1836 blk_rq_set_mixed_merge(req);
1838 bio->bi_next = req->bio;
1841 req->__sector = bio->bi_iter.bi_sector;
1842 req->__data_len += bio->bi_iter.bi_size;
1843 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1845 blk_account_io_start(req, false);
1849 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1852 unsigned short segments = blk_rq_nr_discard_segments(req);
1854 if (segments >= queue_max_discard_segments(q))
1856 if (blk_rq_sectors(req) + bio_sectors(bio) >
1857 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1860 req->biotail->bi_next = bio;
1862 req->__data_len += bio->bi_iter.bi_size;
1863 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1864 req->nr_phys_segments = segments + 1;
1866 blk_account_io_start(req, false);
1869 req_set_nomerge(q, req);
1874 * blk_attempt_plug_merge - try to merge with %current's plugged list
1875 * @q: request_queue new bio is being queued at
1876 * @bio: new bio being queued
1877 * @request_count: out parameter for number of traversed plugged requests
1878 * @same_queue_rq: pointer to &struct request that gets filled in when
1879 * another request associated with @q is found on the plug list
1880 * (optional, may be %NULL)
1882 * Determine whether @bio being queued on @q can be merged with a request
1883 * on %current's plugged list. Returns %true if merge was successful,
1886 * Plugging coalesces IOs from the same issuer for the same purpose without
1887 * going through @q->queue_lock. As such it's more of an issuing mechanism
1888 * than scheduling, and the request, while may have elvpriv data, is not
1889 * added on the elevator at this point. In addition, we don't have
1890 * reliable access to the elevator outside queue lock. Only check basic
1891 * merging parameters without querying the elevator.
1893 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1895 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1896 unsigned int *request_count,
1897 struct request **same_queue_rq)
1899 struct blk_plug *plug;
1901 struct list_head *plug_list;
1903 plug = current->plug;
1909 plug_list = &plug->mq_list;
1911 plug_list = &plug->list;
1913 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1914 bool merged = false;
1919 * Only blk-mq multiple hardware queues case checks the
1920 * rq in the same queue, there should be only one such
1924 *same_queue_rq = rq;
1927 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1930 switch (blk_try_merge(rq, bio)) {
1931 case ELEVATOR_BACK_MERGE:
1932 merged = bio_attempt_back_merge(q, rq, bio);
1934 case ELEVATOR_FRONT_MERGE:
1935 merged = bio_attempt_front_merge(q, rq, bio);
1937 case ELEVATOR_DISCARD_MERGE:
1938 merged = bio_attempt_discard_merge(q, rq, bio);
1951 unsigned int blk_plug_queued_count(struct request_queue *q)
1953 struct blk_plug *plug;
1955 struct list_head *plug_list;
1956 unsigned int ret = 0;
1958 plug = current->plug;
1963 plug_list = &plug->mq_list;
1965 plug_list = &plug->list;
1967 list_for_each_entry(rq, plug_list, queuelist) {
1975 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1977 struct io_context *ioc = rq_ioc(bio);
1979 if (bio->bi_opf & REQ_RAHEAD)
1980 req->cmd_flags |= REQ_FAILFAST_MASK;
1982 req->__sector = bio->bi_iter.bi_sector;
1983 if (ioprio_valid(bio_prio(bio)))
1984 req->ioprio = bio_prio(bio);
1986 req->ioprio = ioc->ioprio;
1988 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1989 req->write_hint = bio->bi_write_hint;
1990 blk_rq_bio_prep(req->q, req, bio);
1992 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1994 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1996 struct blk_plug *plug;
1997 int where = ELEVATOR_INSERT_SORT;
1998 struct request *req, *free;
1999 unsigned int request_count = 0;
2002 * low level driver can indicate that it wants pages above a
2003 * certain limit bounced to low memory (ie for highmem, or even
2004 * ISA dma in theory)
2006 blk_queue_bounce(q, &bio);
2008 blk_queue_split(q, &bio);
2010 if (!bio_integrity_prep(bio))
2011 return BLK_QC_T_NONE;
2013 if (op_is_flush(bio->bi_opf)) {
2014 spin_lock_irq(q->queue_lock);
2015 where = ELEVATOR_INSERT_FLUSH;
2020 * Check if we can merge with the plugged list before grabbing
2023 if (!blk_queue_nomerges(q)) {
2024 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
2025 return BLK_QC_T_NONE;
2027 request_count = blk_plug_queued_count(q);
2029 spin_lock_irq(q->queue_lock);
2031 switch (elv_merge(q, &req, bio)) {
2032 case ELEVATOR_BACK_MERGE:
2033 if (!bio_attempt_back_merge(q, req, bio))
2035 elv_bio_merged(q, req, bio);
2036 free = attempt_back_merge(q, req);
2038 __blk_put_request(q, free);
2040 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2042 case ELEVATOR_FRONT_MERGE:
2043 if (!bio_attempt_front_merge(q, req, bio))
2045 elv_bio_merged(q, req, bio);
2046 free = attempt_front_merge(q, req);
2048 __blk_put_request(q, free);
2050 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2057 rq_qos_throttle(q, bio, q->queue_lock);
2060 * Grab a free request. This is might sleep but can not fail.
2061 * Returns with the queue unlocked.
2063 blk_queue_enter_live(q);
2064 req = get_request(q, bio->bi_opf, bio, 0, GFP_NOIO);
2067 rq_qos_cleanup(q, bio);
2068 if (PTR_ERR(req) == -ENOMEM)
2069 bio->bi_status = BLK_STS_RESOURCE;
2071 bio->bi_status = BLK_STS_IOERR;
2076 rq_qos_track(q, req, bio);
2079 * After dropping the lock and possibly sleeping here, our request
2080 * may now be mergeable after it had proven unmergeable (above).
2081 * We don't worry about that case for efficiency. It won't happen
2082 * often, and the elevators are able to handle it.
2084 blk_init_request_from_bio(req, bio);
2086 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
2087 req->cpu = raw_smp_processor_id();
2089 plug = current->plug;
2092 * If this is the first request added after a plug, fire
2095 * @request_count may become stale because of schedule
2096 * out, so check plug list again.
2098 if (!request_count || list_empty(&plug->list))
2099 trace_block_plug(q);
2101 struct request *last = list_entry_rq(plug->list.prev);
2102 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2103 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2104 blk_flush_plug_list(plug, false);
2105 trace_block_plug(q);
2108 list_add_tail(&req->queuelist, &plug->list);
2109 blk_account_io_start(req, true);
2111 spin_lock_irq(q->queue_lock);
2112 add_acct_request(q, req, where);
2115 spin_unlock_irq(q->queue_lock);
2118 return BLK_QC_T_NONE;
2121 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
2123 char b[BDEVNAME_SIZE];
2125 printk(KERN_INFO "attempt to access beyond end of device\n");
2126 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2127 bio_devname(bio, b), bio->bi_opf,
2128 (unsigned long long)bio_end_sector(bio),
2129 (long long)maxsector);
2132 #ifdef CONFIG_FAIL_MAKE_REQUEST
2134 static DECLARE_FAULT_ATTR(fail_make_request);
2136 static int __init setup_fail_make_request(char *str)
2138 return setup_fault_attr(&fail_make_request, str);
2140 __setup("fail_make_request=", setup_fail_make_request);
2142 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2144 return part->make_it_fail && should_fail(&fail_make_request, bytes);
2147 static int __init fail_make_request_debugfs(void)
2149 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2150 NULL, &fail_make_request);
2152 return PTR_ERR_OR_ZERO(dir);
2155 late_initcall(fail_make_request_debugfs);
2157 #else /* CONFIG_FAIL_MAKE_REQUEST */
2159 static inline bool should_fail_request(struct hd_struct *part,
2165 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2167 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2169 const int op = bio_op(bio);
2171 if (part->policy && op_is_write(op)) {
2172 char b[BDEVNAME_SIZE];
2174 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
2178 "generic_make_request: Trying to write "
2179 "to read-only block-device %s (partno %d)\n",
2180 bio_devname(bio, b), part->partno);
2181 /* Older lvm-tools actually trigger this */
2188 static noinline int should_fail_bio(struct bio *bio)
2190 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2194 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2197 * Check whether this bio extends beyond the end of the device or partition.
2198 * This may well happen - the kernel calls bread() without checking the size of
2199 * the device, e.g., when mounting a file system.
2201 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2203 unsigned int nr_sectors = bio_sectors(bio);
2205 if (nr_sectors && maxsector &&
2206 (nr_sectors > maxsector ||
2207 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2208 handle_bad_sector(bio, maxsector);
2215 * Remap block n of partition p to block n+start(p) of the disk.
2217 static inline int blk_partition_remap(struct bio *bio)
2219 struct hd_struct *p;
2223 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2226 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2228 if (unlikely(bio_check_ro(bio, p)))
2232 * Zone reset does not include bi_size so bio_sectors() is always 0.
2233 * Include a test for the reset op code and perform the remap if needed.
2235 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2236 if (bio_check_eod(bio, part_nr_sects_read(p)))
2238 bio->bi_iter.bi_sector += p->start_sect;
2239 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2240 bio->bi_iter.bi_sector - p->start_sect);
2249 static noinline_for_stack bool
2250 generic_make_request_checks(struct bio *bio)
2252 struct request_queue *q;
2253 int nr_sectors = bio_sectors(bio);
2254 blk_status_t status = BLK_STS_IOERR;
2255 char b[BDEVNAME_SIZE];
2259 q = bio->bi_disk->queue;
2262 "generic_make_request: Trying to access "
2263 "nonexistent block-device %s (%Lu)\n",
2264 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2269 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2270 * if queue is not a request based queue.
2272 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2275 if (should_fail_bio(bio))
2278 if (bio->bi_partno) {
2279 if (unlikely(blk_partition_remap(bio)))
2282 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2284 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
2289 * Filter flush bio's early so that make_request based
2290 * drivers without flush support don't have to worry
2293 if (op_is_flush(bio->bi_opf) &&
2294 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2295 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2297 status = BLK_STS_OK;
2302 switch (bio_op(bio)) {
2303 case REQ_OP_DISCARD:
2304 if (!blk_queue_discard(q))
2307 case REQ_OP_SECURE_ERASE:
2308 if (!blk_queue_secure_erase(q))
2311 case REQ_OP_WRITE_SAME:
2312 if (!q->limits.max_write_same_sectors)
2315 case REQ_OP_ZONE_REPORT:
2316 case REQ_OP_ZONE_RESET:
2317 if (!blk_queue_is_zoned(q))
2320 case REQ_OP_WRITE_ZEROES:
2321 if (!q->limits.max_write_zeroes_sectors)
2329 * Various block parts want %current->io_context and lazy ioc
2330 * allocation ends up trading a lot of pain for a small amount of
2331 * memory. Just allocate it upfront. This may fail and block
2332 * layer knows how to live with it.
2334 create_io_context(GFP_ATOMIC, q->node);
2336 if (!blkcg_bio_issue_check(q, bio))
2339 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2340 trace_block_bio_queue(q, bio);
2341 /* Now that enqueuing has been traced, we need to trace
2342 * completion as well.
2344 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2349 status = BLK_STS_NOTSUPP;
2351 bio->bi_status = status;
2357 * generic_make_request - hand a buffer to its device driver for I/O
2358 * @bio: The bio describing the location in memory and on the device.
2360 * generic_make_request() is used to make I/O requests of block
2361 * devices. It is passed a &struct bio, which describes the I/O that needs
2364 * generic_make_request() does not return any status. The
2365 * success/failure status of the request, along with notification of
2366 * completion, is delivered asynchronously through the bio->bi_end_io
2367 * function described (one day) else where.
2369 * The caller of generic_make_request must make sure that bi_io_vec
2370 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2371 * set to describe the device address, and the
2372 * bi_end_io and optionally bi_private are set to describe how
2373 * completion notification should be signaled.
2375 * generic_make_request and the drivers it calls may use bi_next if this
2376 * bio happens to be merged with someone else, and may resubmit the bio to
2377 * a lower device by calling into generic_make_request recursively, which
2378 * means the bio should NOT be touched after the call to ->make_request_fn.
2380 blk_qc_t generic_make_request(struct bio *bio)
2383 * bio_list_on_stack[0] contains bios submitted by the current
2385 * bio_list_on_stack[1] contains bios that were submitted before
2386 * the current make_request_fn, but that haven't been processed
2389 struct bio_list bio_list_on_stack[2];
2390 blk_mq_req_flags_t flags = 0;
2391 struct request_queue *q = bio->bi_disk->queue;
2392 blk_qc_t ret = BLK_QC_T_NONE;
2394 if (bio->bi_opf & REQ_NOWAIT)
2395 flags = BLK_MQ_REQ_NOWAIT;
2396 if (bio_flagged(bio, BIO_QUEUE_ENTERED))
2397 blk_queue_enter_live(q);
2398 else if (blk_queue_enter(q, flags) < 0) {
2399 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2400 bio_wouldblock_error(bio);
2406 if (!generic_make_request_checks(bio))
2410 * We only want one ->make_request_fn to be active at a time, else
2411 * stack usage with stacked devices could be a problem. So use
2412 * current->bio_list to keep a list of requests submited by a
2413 * make_request_fn function. current->bio_list is also used as a
2414 * flag to say if generic_make_request is currently active in this
2415 * task or not. If it is NULL, then no make_request is active. If
2416 * it is non-NULL, then a make_request is active, and new requests
2417 * should be added at the tail
2419 if (current->bio_list) {
2420 bio_list_add(¤t->bio_list[0], bio);
2424 /* following loop may be a bit non-obvious, and so deserves some
2426 * Before entering the loop, bio->bi_next is NULL (as all callers
2427 * ensure that) so we have a list with a single bio.
2428 * We pretend that we have just taken it off a longer list, so
2429 * we assign bio_list to a pointer to the bio_list_on_stack,
2430 * thus initialising the bio_list of new bios to be
2431 * added. ->make_request() may indeed add some more bios
2432 * through a recursive call to generic_make_request. If it
2433 * did, we find a non-NULL value in bio_list and re-enter the loop
2434 * from the top. In this case we really did just take the bio
2435 * of the top of the list (no pretending) and so remove it from
2436 * bio_list, and call into ->make_request() again.
2438 BUG_ON(bio->bi_next);
2439 bio_list_init(&bio_list_on_stack[0]);
2440 current->bio_list = bio_list_on_stack;
2442 bool enter_succeeded = true;
2444 if (unlikely(q != bio->bi_disk->queue)) {
2447 q = bio->bi_disk->queue;
2449 if (bio->bi_opf & REQ_NOWAIT)
2450 flags = BLK_MQ_REQ_NOWAIT;
2451 if (blk_queue_enter(q, flags) < 0)
2452 enter_succeeded = false;
2455 if (enter_succeeded) {
2456 struct bio_list lower, same;
2458 /* Create a fresh bio_list for all subordinate requests */
2459 bio_list_on_stack[1] = bio_list_on_stack[0];
2460 bio_list_init(&bio_list_on_stack[0]);
2461 ret = q->make_request_fn(q, bio);
2463 /* sort new bios into those for a lower level
2464 * and those for the same level
2466 bio_list_init(&lower);
2467 bio_list_init(&same);
2468 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2469 if (q == bio->bi_disk->queue)
2470 bio_list_add(&same, bio);
2472 bio_list_add(&lower, bio);
2473 /* now assemble so we handle the lowest level first */
2474 bio_list_merge(&bio_list_on_stack[0], &lower);
2475 bio_list_merge(&bio_list_on_stack[0], &same);
2476 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2478 if (unlikely(!blk_queue_dying(q) &&
2479 (bio->bi_opf & REQ_NOWAIT)))
2480 bio_wouldblock_error(bio);
2485 bio = bio_list_pop(&bio_list_on_stack[0]);
2487 current->bio_list = NULL; /* deactivate */
2494 EXPORT_SYMBOL(generic_make_request);
2497 * direct_make_request - hand a buffer directly to its device driver for I/O
2498 * @bio: The bio describing the location in memory and on the device.
2500 * This function behaves like generic_make_request(), but does not protect
2501 * against recursion. Must only be used if the called driver is known
2502 * to not call generic_make_request (or direct_make_request) again from
2503 * its make_request function. (Calling direct_make_request again from
2504 * a workqueue is perfectly fine as that doesn't recurse).
2506 blk_qc_t direct_make_request(struct bio *bio)
2508 struct request_queue *q = bio->bi_disk->queue;
2509 bool nowait = bio->bi_opf & REQ_NOWAIT;
2512 if (!generic_make_request_checks(bio))
2513 return BLK_QC_T_NONE;
2515 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2516 if (nowait && !blk_queue_dying(q))
2517 bio->bi_status = BLK_STS_AGAIN;
2519 bio->bi_status = BLK_STS_IOERR;
2521 return BLK_QC_T_NONE;
2524 ret = q->make_request_fn(q, bio);
2528 EXPORT_SYMBOL_GPL(direct_make_request);
2531 * submit_bio - submit a bio to the block device layer for I/O
2532 * @bio: The &struct bio which describes the I/O
2534 * submit_bio() is very similar in purpose to generic_make_request(), and
2535 * uses that function to do most of the work. Both are fairly rough
2536 * interfaces; @bio must be presetup and ready for I/O.
2539 blk_qc_t submit_bio(struct bio *bio)
2542 * If it's a regular read/write or a barrier with data attached,
2543 * go through the normal accounting stuff before submission.
2545 if (bio_has_data(bio)) {
2548 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2549 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2551 count = bio_sectors(bio);
2553 if (op_is_write(bio_op(bio))) {
2554 count_vm_events(PGPGOUT, count);
2556 task_io_account_read(bio->bi_iter.bi_size);
2557 count_vm_events(PGPGIN, count);
2560 if (unlikely(block_dump)) {
2561 char b[BDEVNAME_SIZE];
2562 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2563 current->comm, task_pid_nr(current),
2564 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2565 (unsigned long long)bio->bi_iter.bi_sector,
2566 bio_devname(bio, b), count);
2570 return generic_make_request(bio);
2572 EXPORT_SYMBOL(submit_bio);
2574 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2576 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2580 blk_flush_plug_list(current->plug, false);
2581 return q->poll_fn(q, cookie);
2583 EXPORT_SYMBOL_GPL(blk_poll);
2586 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2587 * for new the queue limits
2589 * @rq: the request being checked
2592 * @rq may have been made based on weaker limitations of upper-level queues
2593 * in request stacking drivers, and it may violate the limitation of @q.
2594 * Since the block layer and the underlying device driver trust @rq
2595 * after it is inserted to @q, it should be checked against @q before
2596 * the insertion using this generic function.
2598 * Request stacking drivers like request-based dm may change the queue
2599 * limits when retrying requests on other queues. Those requests need
2600 * to be checked against the new queue limits again during dispatch.
2602 static int blk_cloned_rq_check_limits(struct request_queue *q,
2605 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2606 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2611 * queue's settings related to segment counting like q->bounce_pfn
2612 * may differ from that of other stacking queues.
2613 * Recalculate it to check the request correctly on this queue's
2616 blk_recalc_rq_segments(rq);
2617 if (rq->nr_phys_segments > queue_max_segments(q)) {
2618 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2626 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2627 * @q: the queue to submit the request
2628 * @rq: the request being queued
2630 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2632 unsigned long flags;
2633 int where = ELEVATOR_INSERT_BACK;
2635 if (blk_cloned_rq_check_limits(q, rq))
2636 return BLK_STS_IOERR;
2639 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2640 return BLK_STS_IOERR;
2643 if (blk_queue_io_stat(q))
2644 blk_account_io_start(rq, true);
2646 * Since we have a scheduler attached on the top device,
2647 * bypass a potential scheduler on the bottom device for
2650 return blk_mq_request_issue_directly(rq);
2653 spin_lock_irqsave(q->queue_lock, flags);
2654 if (unlikely(blk_queue_dying(q))) {
2655 spin_unlock_irqrestore(q->queue_lock, flags);
2656 return BLK_STS_IOERR;
2660 * Submitting request must be dequeued before calling this function
2661 * because it will be linked to another request_queue
2663 BUG_ON(blk_queued_rq(rq));
2665 if (op_is_flush(rq->cmd_flags))
2666 where = ELEVATOR_INSERT_FLUSH;
2668 add_acct_request(q, rq, where);
2669 if (where == ELEVATOR_INSERT_FLUSH)
2671 spin_unlock_irqrestore(q->queue_lock, flags);
2675 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2678 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2679 * @rq: request to examine
2682 * A request could be merge of IOs which require different failure
2683 * handling. This function determines the number of bytes which
2684 * can be failed from the beginning of the request without
2685 * crossing into area which need to be retried further.
2688 * The number of bytes to fail.
2690 unsigned int blk_rq_err_bytes(const struct request *rq)
2692 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2693 unsigned int bytes = 0;
2696 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2697 return blk_rq_bytes(rq);
2700 * Currently the only 'mixing' which can happen is between
2701 * different fastfail types. We can safely fail portions
2702 * which have all the failfast bits that the first one has -
2703 * the ones which are at least as eager to fail as the first
2706 for (bio = rq->bio; bio; bio = bio->bi_next) {
2707 if ((bio->bi_opf & ff) != ff)
2709 bytes += bio->bi_iter.bi_size;
2712 /* this could lead to infinite loop */
2713 BUG_ON(blk_rq_bytes(rq) && !bytes);
2716 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2718 void blk_account_io_completion(struct request *req, unsigned int bytes)
2720 if (blk_do_io_stat(req)) {
2721 const int sgrp = op_stat_group(req_op(req));
2722 struct hd_struct *part;
2725 cpu = part_stat_lock();
2727 part_stat_add(cpu, part, sectors[sgrp], bytes >> 9);
2732 void blk_account_io_done(struct request *req, u64 now)
2735 * Account IO completion. flush_rq isn't accounted as a
2736 * normal IO on queueing nor completion. Accounting the
2737 * containing request is enough.
2739 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2740 const int sgrp = op_stat_group(req_op(req));
2741 struct hd_struct *part;
2744 cpu = part_stat_lock();
2747 part_stat_inc(cpu, part, ios[sgrp]);
2748 part_stat_add(cpu, part, nsecs[sgrp], now - req->start_time_ns);
2749 part_round_stats(req->q, cpu, part);
2750 part_dec_in_flight(req->q, part, rq_data_dir(req));
2752 hd_struct_put(part);
2759 * Don't process normal requests when queue is suspended
2760 * or in the process of suspending/resuming
2762 static bool blk_pm_allow_request(struct request *rq)
2764 switch (rq->q->rpm_status) {
2766 case RPM_SUSPENDING:
2767 return rq->rq_flags & RQF_PM;
2775 static bool blk_pm_allow_request(struct request *rq)
2781 void blk_account_io_start(struct request *rq, bool new_io)
2783 struct hd_struct *part;
2784 int rw = rq_data_dir(rq);
2787 if (!blk_do_io_stat(rq))
2790 cpu = part_stat_lock();
2794 part_stat_inc(cpu, part, merges[rw]);
2796 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2797 if (!hd_struct_try_get(part)) {
2799 * The partition is already being removed,
2800 * the request will be accounted on the disk only
2802 * We take a reference on disk->part0 although that
2803 * partition will never be deleted, so we can treat
2804 * it as any other partition.
2806 part = &rq->rq_disk->part0;
2807 hd_struct_get(part);
2809 part_round_stats(rq->q, cpu, part);
2810 part_inc_in_flight(rq->q, part, rw);
2817 static struct request *elv_next_request(struct request_queue *q)
2820 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2822 WARN_ON_ONCE(q->mq_ops);
2825 list_for_each_entry(rq, &q->queue_head, queuelist) {
2826 if (blk_pm_allow_request(rq))
2829 if (rq->rq_flags & RQF_SOFTBARRIER)
2834 * Flush request is running and flush request isn't queueable
2835 * in the drive, we can hold the queue till flush request is
2836 * finished. Even we don't do this, driver can't dispatch next
2837 * requests and will requeue them. And this can improve
2838 * throughput too. For example, we have request flush1, write1,
2839 * flush 2. flush1 is dispatched, then queue is hold, write1
2840 * isn't inserted to queue. After flush1 is finished, flush2
2841 * will be dispatched. Since disk cache is already clean,
2842 * flush2 will be finished very soon, so looks like flush2 is
2844 * Since the queue is hold, a flag is set to indicate the queue
2845 * should be restarted later. Please see flush_end_io() for
2848 if (fq->flush_pending_idx != fq->flush_running_idx &&
2849 !queue_flush_queueable(q)) {
2850 fq->flush_queue_delayed = 1;
2853 if (unlikely(blk_queue_bypass(q)) ||
2854 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2860 * blk_peek_request - peek at the top of a request queue
2861 * @q: request queue to peek at
2864 * Return the request at the top of @q. The returned request
2865 * should be started using blk_start_request() before LLD starts
2869 * Pointer to the request at the top of @q if available. Null
2872 struct request *blk_peek_request(struct request_queue *q)
2877 lockdep_assert_held(q->queue_lock);
2878 WARN_ON_ONCE(q->mq_ops);
2880 while ((rq = elv_next_request(q)) != NULL) {
2881 if (!(rq->rq_flags & RQF_STARTED)) {
2883 * This is the first time the device driver
2884 * sees this request (possibly after
2885 * requeueing). Notify IO scheduler.
2887 if (rq->rq_flags & RQF_SORTED)
2888 elv_activate_rq(q, rq);
2891 * just mark as started even if we don't start
2892 * it, a request that has been delayed should
2893 * not be passed by new incoming requests
2895 rq->rq_flags |= RQF_STARTED;
2896 trace_block_rq_issue(q, rq);
2899 if (!q->boundary_rq || q->boundary_rq == rq) {
2900 q->end_sector = rq_end_sector(rq);
2901 q->boundary_rq = NULL;
2904 if (rq->rq_flags & RQF_DONTPREP)
2907 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2909 * make sure space for the drain appears we
2910 * know we can do this because max_hw_segments
2911 * has been adjusted to be one fewer than the
2914 rq->nr_phys_segments++;
2920 ret = q->prep_rq_fn(q, rq);
2921 if (ret == BLKPREP_OK) {
2923 } else if (ret == BLKPREP_DEFER) {
2925 * the request may have been (partially) prepped.
2926 * we need to keep this request in the front to
2927 * avoid resource deadlock. RQF_STARTED will
2928 * prevent other fs requests from passing this one.
2930 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2931 !(rq->rq_flags & RQF_DONTPREP)) {
2933 * remove the space for the drain we added
2934 * so that we don't add it again
2936 --rq->nr_phys_segments;
2941 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2942 rq->rq_flags |= RQF_QUIET;
2944 * Mark this request as started so we don't trigger
2945 * any debug logic in the end I/O path.
2947 blk_start_request(rq);
2948 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2949 BLK_STS_TARGET : BLK_STS_IOERR);
2951 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2958 EXPORT_SYMBOL(blk_peek_request);
2960 static void blk_dequeue_request(struct request *rq)
2962 struct request_queue *q = rq->q;
2964 BUG_ON(list_empty(&rq->queuelist));
2965 BUG_ON(ELV_ON_HASH(rq));
2967 list_del_init(&rq->queuelist);
2970 * the time frame between a request being removed from the lists
2971 * and to it is freed is accounted as io that is in progress at
2974 if (blk_account_rq(rq))
2975 q->in_flight[rq_is_sync(rq)]++;
2979 * blk_start_request - start request processing on the driver
2980 * @req: request to dequeue
2983 * Dequeue @req and start timeout timer on it. This hands off the
2984 * request to the driver.
2986 void blk_start_request(struct request *req)
2988 lockdep_assert_held(req->q->queue_lock);
2989 WARN_ON_ONCE(req->q->mq_ops);
2991 blk_dequeue_request(req);
2993 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2994 req->io_start_time_ns = ktime_get_ns();
2995 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2996 req->throtl_size = blk_rq_sectors(req);
2998 req->rq_flags |= RQF_STATS;
2999 rq_qos_issue(req->q, req);
3002 BUG_ON(blk_rq_is_complete(req));
3005 EXPORT_SYMBOL(blk_start_request);
3008 * blk_fetch_request - fetch a request from a request queue
3009 * @q: request queue to fetch a request from
3012 * Return the request at the top of @q. The request is started on
3013 * return and LLD can start processing it immediately.
3016 * Pointer to the request at the top of @q if available. Null
3019 struct request *blk_fetch_request(struct request_queue *q)
3023 lockdep_assert_held(q->queue_lock);
3024 WARN_ON_ONCE(q->mq_ops);
3026 rq = blk_peek_request(q);
3028 blk_start_request(rq);
3031 EXPORT_SYMBOL(blk_fetch_request);
3034 * Steal bios from a request and add them to a bio list.
3035 * The request must not have been partially completed before.
3037 void blk_steal_bios(struct bio_list *list, struct request *rq)
3041 list->tail->bi_next = rq->bio;
3043 list->head = rq->bio;
3044 list->tail = rq->biotail;
3052 EXPORT_SYMBOL_GPL(blk_steal_bios);
3055 * blk_update_request - Special helper function for request stacking drivers
3056 * @req: the request being processed
3057 * @error: block status code
3058 * @nr_bytes: number of bytes to complete @req
3061 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3062 * the request structure even if @req doesn't have leftover.
3063 * If @req has leftover, sets it up for the next range of segments.
3065 * This special helper function is only for request stacking drivers
3066 * (e.g. request-based dm) so that they can handle partial completion.
3067 * Actual device drivers should use blk_end_request instead.
3069 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3070 * %false return from this function.
3073 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
3074 * blk_rq_bytes() and in blk_update_request().
3077 * %false - this request doesn't have any more data
3078 * %true - this request has more data
3080 bool blk_update_request(struct request *req, blk_status_t error,
3081 unsigned int nr_bytes)
3085 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
3090 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3091 !(req->rq_flags & RQF_QUIET)))
3092 print_req_error(req, error);
3094 blk_account_io_completion(req, nr_bytes);
3098 struct bio *bio = req->bio;
3099 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3101 if (bio_bytes == bio->bi_iter.bi_size)
3102 req->bio = bio->bi_next;
3104 /* Completion has already been traced */
3105 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
3106 req_bio_endio(req, bio, bio_bytes, error);
3108 total_bytes += bio_bytes;
3109 nr_bytes -= bio_bytes;
3120 * Reset counters so that the request stacking driver
3121 * can find how many bytes remain in the request
3124 req->__data_len = 0;
3128 req->__data_len -= total_bytes;
3130 /* update sector only for requests with clear definition of sector */
3131 if (!blk_rq_is_passthrough(req))
3132 req->__sector += total_bytes >> 9;
3134 /* mixed attributes always follow the first bio */
3135 if (req->rq_flags & RQF_MIXED_MERGE) {
3136 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3137 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3140 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3142 * If total number of sectors is less than the first segment
3143 * size, something has gone terribly wrong.
3145 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3146 blk_dump_rq_flags(req, "request botched");
3147 req->__data_len = blk_rq_cur_bytes(req);
3150 /* recalculate the number of segments */
3151 blk_recalc_rq_segments(req);
3156 EXPORT_SYMBOL_GPL(blk_update_request);
3158 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3159 unsigned int nr_bytes,
3160 unsigned int bidi_bytes)
3162 if (blk_update_request(rq, error, nr_bytes))
3165 /* Bidi request must be completed as a whole */
3166 if (unlikely(blk_bidi_rq(rq)) &&
3167 blk_update_request(rq->next_rq, error, bidi_bytes))
3170 if (blk_queue_add_random(rq->q))
3171 add_disk_randomness(rq->rq_disk);
3177 * blk_unprep_request - unprepare a request
3180 * This function makes a request ready for complete resubmission (or
3181 * completion). It happens only after all error handling is complete,
3182 * so represents the appropriate moment to deallocate any resources
3183 * that were allocated to the request in the prep_rq_fn. The queue
3184 * lock is held when calling this.
3186 void blk_unprep_request(struct request *req)
3188 struct request_queue *q = req->q;
3190 req->rq_flags &= ~RQF_DONTPREP;
3191 if (q->unprep_rq_fn)
3192 q->unprep_rq_fn(q, req);
3194 EXPORT_SYMBOL_GPL(blk_unprep_request);
3196 void blk_finish_request(struct request *req, blk_status_t error)
3198 struct request_queue *q = req->q;
3199 u64 now = ktime_get_ns();
3201 lockdep_assert_held(req->q->queue_lock);
3202 WARN_ON_ONCE(q->mq_ops);
3204 if (req->rq_flags & RQF_STATS)
3205 blk_stat_add(req, now);
3207 if (req->rq_flags & RQF_QUEUED)
3208 blk_queue_end_tag(q, req);
3210 BUG_ON(blk_queued_rq(req));
3212 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3213 laptop_io_completion(req->q->backing_dev_info);
3215 blk_delete_timer(req);
3217 if (req->rq_flags & RQF_DONTPREP)
3218 blk_unprep_request(req);
3220 blk_account_io_done(req, now);
3223 rq_qos_done(q, req);
3224 req->end_io(req, error);
3226 if (blk_bidi_rq(req))
3227 __blk_put_request(req->next_rq->q, req->next_rq);
3229 __blk_put_request(q, req);
3232 EXPORT_SYMBOL(blk_finish_request);
3235 * blk_end_bidi_request - Complete a bidi request
3236 * @rq: the request to complete
3237 * @error: block status code
3238 * @nr_bytes: number of bytes to complete @rq
3239 * @bidi_bytes: number of bytes to complete @rq->next_rq
3242 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3243 * Drivers that supports bidi can safely call this member for any
3244 * type of request, bidi or uni. In the later case @bidi_bytes is
3248 * %false - we are done with this request
3249 * %true - still buffers pending for this request
3251 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3252 unsigned int nr_bytes, unsigned int bidi_bytes)
3254 struct request_queue *q = rq->q;
3255 unsigned long flags;
3257 WARN_ON_ONCE(q->mq_ops);
3259 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3262 spin_lock_irqsave(q->queue_lock, flags);
3263 blk_finish_request(rq, error);
3264 spin_unlock_irqrestore(q->queue_lock, flags);
3270 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3271 * @rq: the request to complete
3272 * @error: block status code
3273 * @nr_bytes: number of bytes to complete @rq
3274 * @bidi_bytes: number of bytes to complete @rq->next_rq
3277 * Identical to blk_end_bidi_request() except that queue lock is
3278 * assumed to be locked on entry and remains so on return.
3281 * %false - we are done with this request
3282 * %true - still buffers pending for this request
3284 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3285 unsigned int nr_bytes, unsigned int bidi_bytes)
3287 lockdep_assert_held(rq->q->queue_lock);
3288 WARN_ON_ONCE(rq->q->mq_ops);
3290 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3293 blk_finish_request(rq, error);
3299 * blk_end_request - Helper function for drivers to complete the request.
3300 * @rq: the request being processed
3301 * @error: block status code
3302 * @nr_bytes: number of bytes to complete
3305 * Ends I/O on a number of bytes attached to @rq.
3306 * If @rq has leftover, sets it up for the next range of segments.
3309 * %false - we are done with this request
3310 * %true - still buffers pending for this request
3312 bool blk_end_request(struct request *rq, blk_status_t error,
3313 unsigned int nr_bytes)
3315 WARN_ON_ONCE(rq->q->mq_ops);
3316 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3318 EXPORT_SYMBOL(blk_end_request);
3321 * blk_end_request_all - Helper function for drives to finish the request.
3322 * @rq: the request to finish
3323 * @error: block status code
3326 * Completely finish @rq.
3328 void blk_end_request_all(struct request *rq, blk_status_t error)
3331 unsigned int bidi_bytes = 0;
3333 if (unlikely(blk_bidi_rq(rq)))
3334 bidi_bytes = blk_rq_bytes(rq->next_rq);
3336 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3339 EXPORT_SYMBOL(blk_end_request_all);
3342 * __blk_end_request - Helper function for drivers to complete the request.
3343 * @rq: the request being processed
3344 * @error: block status code
3345 * @nr_bytes: number of bytes to complete
3348 * Must be called with queue lock held unlike blk_end_request().
3351 * %false - we are done with this request
3352 * %true - still buffers pending for this request
3354 bool __blk_end_request(struct request *rq, blk_status_t error,
3355 unsigned int nr_bytes)
3357 lockdep_assert_held(rq->q->queue_lock);
3358 WARN_ON_ONCE(rq->q->mq_ops);
3360 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3362 EXPORT_SYMBOL(__blk_end_request);
3365 * __blk_end_request_all - Helper function for drives to finish the request.
3366 * @rq: the request to finish
3367 * @error: block status code
3370 * Completely finish @rq. Must be called with queue lock held.
3372 void __blk_end_request_all(struct request *rq, blk_status_t error)
3375 unsigned int bidi_bytes = 0;
3377 lockdep_assert_held(rq->q->queue_lock);
3378 WARN_ON_ONCE(rq->q->mq_ops);
3380 if (unlikely(blk_bidi_rq(rq)))
3381 bidi_bytes = blk_rq_bytes(rq->next_rq);
3383 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3386 EXPORT_SYMBOL(__blk_end_request_all);
3389 * __blk_end_request_cur - Helper function to finish the current request chunk.
3390 * @rq: the request to finish the current chunk for
3391 * @error: block status code
3394 * Complete the current consecutively mapped chunk from @rq. Must
3395 * be called with queue lock held.
3398 * %false - we are done with this request
3399 * %true - still buffers pending for this request
3401 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3403 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3405 EXPORT_SYMBOL(__blk_end_request_cur);
3407 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3410 if (bio_has_data(bio))
3411 rq->nr_phys_segments = bio_phys_segments(q, bio);
3412 else if (bio_op(bio) == REQ_OP_DISCARD)
3413 rq->nr_phys_segments = 1;
3415 rq->__data_len = bio->bi_iter.bi_size;
3416 rq->bio = rq->biotail = bio;
3419 rq->rq_disk = bio->bi_disk;
3422 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3424 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3425 * @rq: the request to be flushed
3428 * Flush all pages in @rq.
3430 void rq_flush_dcache_pages(struct request *rq)
3432 struct req_iterator iter;
3433 struct bio_vec bvec;
3435 rq_for_each_segment(bvec, rq, iter)
3436 flush_dcache_page(bvec.bv_page);
3438 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3442 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3443 * @q : the queue of the device being checked
3446 * Check if underlying low-level drivers of a device are busy.
3447 * If the drivers want to export their busy state, they must set own
3448 * exporting function using blk_queue_lld_busy() first.
3450 * Basically, this function is used only by request stacking drivers
3451 * to stop dispatching requests to underlying devices when underlying
3452 * devices are busy. This behavior helps more I/O merging on the queue
3453 * of the request stacking driver and prevents I/O throughput regression
3454 * on burst I/O load.
3457 * 0 - Not busy (The request stacking driver should dispatch request)
3458 * 1 - Busy (The request stacking driver should stop dispatching request)
3460 int blk_lld_busy(struct request_queue *q)
3463 return q->lld_busy_fn(q);
3467 EXPORT_SYMBOL_GPL(blk_lld_busy);
3470 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3471 * @rq: the clone request to be cleaned up
3474 * Free all bios in @rq for a cloned request.
3476 void blk_rq_unprep_clone(struct request *rq)
3480 while ((bio = rq->bio) != NULL) {
3481 rq->bio = bio->bi_next;
3486 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3489 * Copy attributes of the original request to the clone request.
3490 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3492 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3494 dst->cpu = src->cpu;
3495 dst->__sector = blk_rq_pos(src);
3496 dst->__data_len = blk_rq_bytes(src);
3497 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
3498 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
3499 dst->special_vec = src->special_vec;
3501 dst->nr_phys_segments = src->nr_phys_segments;
3502 dst->ioprio = src->ioprio;
3503 dst->extra_len = src->extra_len;
3507 * blk_rq_prep_clone - Helper function to setup clone request
3508 * @rq: the request to be setup
3509 * @rq_src: original request to be cloned
3510 * @bs: bio_set that bios for clone are allocated from
3511 * @gfp_mask: memory allocation mask for bio
3512 * @bio_ctr: setup function to be called for each clone bio.
3513 * Returns %0 for success, non %0 for failure.
3514 * @data: private data to be passed to @bio_ctr
3517 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3518 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3519 * are not copied, and copying such parts is the caller's responsibility.
3520 * Also, pages which the original bios are pointing to are not copied
3521 * and the cloned bios just point same pages.
3522 * So cloned bios must be completed before original bios, which means
3523 * the caller must complete @rq before @rq_src.
3525 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3526 struct bio_set *bs, gfp_t gfp_mask,
3527 int (*bio_ctr)(struct bio *, struct bio *, void *),
3530 struct bio *bio, *bio_src;
3535 __rq_for_each_bio(bio_src, rq_src) {
3536 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3540 if (bio_ctr && bio_ctr(bio, bio_src, data))
3544 rq->biotail->bi_next = bio;
3547 rq->bio = rq->biotail = bio;
3550 __blk_rq_prep_clone(rq, rq_src);
3557 blk_rq_unprep_clone(rq);
3561 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3563 int kblockd_schedule_work(struct work_struct *work)
3565 return queue_work(kblockd_workqueue, work);
3567 EXPORT_SYMBOL(kblockd_schedule_work);
3569 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3571 return queue_work_on(cpu, kblockd_workqueue, work);
3573 EXPORT_SYMBOL(kblockd_schedule_work_on);
3575 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3576 unsigned long delay)
3578 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3580 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3583 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3584 * @plug: The &struct blk_plug that needs to be initialized
3587 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3588 * pending I/O should the task end up blocking between blk_start_plug() and
3589 * blk_finish_plug(). This is important from a performance perspective, but
3590 * also ensures that we don't deadlock. For instance, if the task is blocking
3591 * for a memory allocation, memory reclaim could end up wanting to free a
3592 * page belonging to that request that is currently residing in our private
3593 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3594 * this kind of deadlock.
3596 void blk_start_plug(struct blk_plug *plug)
3598 struct task_struct *tsk = current;
3601 * If this is a nested plug, don't actually assign it.
3606 INIT_LIST_HEAD(&plug->list);
3607 INIT_LIST_HEAD(&plug->mq_list);
3608 INIT_LIST_HEAD(&plug->cb_list);
3610 * Store ordering should not be needed here, since a potential
3611 * preempt will imply a full memory barrier
3615 EXPORT_SYMBOL(blk_start_plug);
3617 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3619 struct request *rqa = container_of(a, struct request, queuelist);
3620 struct request *rqb = container_of(b, struct request, queuelist);
3622 return !(rqa->q < rqb->q ||
3623 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3627 * If 'from_schedule' is true, then postpone the dispatch of requests
3628 * until a safe kblockd context. We due this to avoid accidental big
3629 * additional stack usage in driver dispatch, in places where the originally
3630 * plugger did not intend it.
3632 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3634 __releases(q->queue_lock)
3636 lockdep_assert_held(q->queue_lock);
3638 trace_block_unplug(q, depth, !from_schedule);
3641 blk_run_queue_async(q);
3644 spin_unlock_irq(q->queue_lock);
3647 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3649 LIST_HEAD(callbacks);
3651 while (!list_empty(&plug->cb_list)) {
3652 list_splice_init(&plug->cb_list, &callbacks);
3654 while (!list_empty(&callbacks)) {
3655 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3658 list_del(&cb->list);
3659 cb->callback(cb, from_schedule);
3664 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3667 struct blk_plug *plug = current->plug;
3668 struct blk_plug_cb *cb;
3673 list_for_each_entry(cb, &plug->cb_list, list)
3674 if (cb->callback == unplug && cb->data == data)
3677 /* Not currently on the callback list */
3678 BUG_ON(size < sizeof(*cb));
3679 cb = kzalloc(size, GFP_ATOMIC);
3682 cb->callback = unplug;
3683 list_add(&cb->list, &plug->cb_list);
3687 EXPORT_SYMBOL(blk_check_plugged);
3689 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3691 struct request_queue *q;
3696 flush_plug_callbacks(plug, from_schedule);
3698 if (!list_empty(&plug->mq_list))
3699 blk_mq_flush_plug_list(plug, from_schedule);
3701 if (list_empty(&plug->list))
3704 list_splice_init(&plug->list, &list);
3706 list_sort(NULL, &list, plug_rq_cmp);
3711 while (!list_empty(&list)) {
3712 rq = list_entry_rq(list.next);
3713 list_del_init(&rq->queuelist);
3717 * This drops the queue lock
3720 queue_unplugged(q, depth, from_schedule);
3723 spin_lock_irq(q->queue_lock);
3727 * Short-circuit if @q is dead
3729 if (unlikely(blk_queue_dying(q))) {
3730 __blk_end_request_all(rq, BLK_STS_IOERR);
3735 * rq is already accounted, so use raw insert
3737 if (op_is_flush(rq->cmd_flags))
3738 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3740 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3746 * This drops the queue lock
3749 queue_unplugged(q, depth, from_schedule);
3752 void blk_finish_plug(struct blk_plug *plug)
3754 if (plug != current->plug)
3756 blk_flush_plug_list(plug, false);
3758 current->plug = NULL;
3760 EXPORT_SYMBOL(blk_finish_plug);
3764 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3765 * @q: the queue of the device
3766 * @dev: the device the queue belongs to
3769 * Initialize runtime-PM-related fields for @q and start auto suspend for
3770 * @dev. Drivers that want to take advantage of request-based runtime PM
3771 * should call this function after @dev has been initialized, and its
3772 * request queue @q has been allocated, and runtime PM for it can not happen
3773 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3774 * cases, driver should call this function before any I/O has taken place.
3776 * This function takes care of setting up using auto suspend for the device,
3777 * the autosuspend delay is set to -1 to make runtime suspend impossible
3778 * until an updated value is either set by user or by driver. Drivers do
3779 * not need to touch other autosuspend settings.
3781 * The block layer runtime PM is request based, so only works for drivers
3782 * that use request as their IO unit instead of those directly use bio's.
3784 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3786 /* Don't enable runtime PM for blk-mq until it is ready */
3788 pm_runtime_disable(dev);
3793 q->rpm_status = RPM_ACTIVE;
3794 pm_runtime_set_autosuspend_delay(q->dev, -1);
3795 pm_runtime_use_autosuspend(q->dev);
3797 EXPORT_SYMBOL(blk_pm_runtime_init);
3800 * blk_pre_runtime_suspend - Pre runtime suspend check
3801 * @q: the queue of the device
3804 * This function will check if runtime suspend is allowed for the device
3805 * by examining if there are any requests pending in the queue. If there
3806 * are requests pending, the device can not be runtime suspended; otherwise,
3807 * the queue's status will be updated to SUSPENDING and the driver can
3808 * proceed to suspend the device.
3810 * For the not allowed case, we mark last busy for the device so that
3811 * runtime PM core will try to autosuspend it some time later.
3813 * This function should be called near the start of the device's
3814 * runtime_suspend callback.
3817 * 0 - OK to runtime suspend the device
3818 * -EBUSY - Device should not be runtime suspended
3820 int blk_pre_runtime_suspend(struct request_queue *q)
3827 spin_lock_irq(q->queue_lock);
3828 if (q->nr_pending) {
3830 pm_runtime_mark_last_busy(q->dev);
3832 q->rpm_status = RPM_SUSPENDING;
3834 spin_unlock_irq(q->queue_lock);
3837 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3840 * blk_post_runtime_suspend - Post runtime suspend processing
3841 * @q: the queue of the device
3842 * @err: return value of the device's runtime_suspend function
3845 * Update the queue's runtime status according to the return value of the
3846 * device's runtime suspend function and mark last busy for the device so
3847 * that PM core will try to auto suspend the device at a later time.
3849 * This function should be called near the end of the device's
3850 * runtime_suspend callback.
3852 void blk_post_runtime_suspend(struct request_queue *q, int err)
3857 spin_lock_irq(q->queue_lock);
3859 q->rpm_status = RPM_SUSPENDED;
3861 q->rpm_status = RPM_ACTIVE;
3862 pm_runtime_mark_last_busy(q->dev);
3864 spin_unlock_irq(q->queue_lock);
3866 EXPORT_SYMBOL(blk_post_runtime_suspend);
3869 * blk_pre_runtime_resume - Pre runtime resume processing
3870 * @q: the queue of the device
3873 * Update the queue's runtime status to RESUMING in preparation for the
3874 * runtime resume of the device.
3876 * This function should be called near the start of the device's
3877 * runtime_resume callback.
3879 void blk_pre_runtime_resume(struct request_queue *q)
3884 spin_lock_irq(q->queue_lock);
3885 q->rpm_status = RPM_RESUMING;
3886 spin_unlock_irq(q->queue_lock);
3888 EXPORT_SYMBOL(blk_pre_runtime_resume);
3891 * blk_post_runtime_resume - Post runtime resume processing
3892 * @q: the queue of the device
3893 * @err: return value of the device's runtime_resume function
3896 * Update the queue's runtime status according to the return value of the
3897 * device's runtime_resume function. If it is successfully resumed, process
3898 * the requests that are queued into the device's queue when it is resuming
3899 * and then mark last busy and initiate autosuspend for it.
3901 * This function should be called near the end of the device's
3902 * runtime_resume callback.
3904 void blk_post_runtime_resume(struct request_queue *q, int err)
3909 spin_lock_irq(q->queue_lock);
3911 q->rpm_status = RPM_ACTIVE;
3913 pm_runtime_mark_last_busy(q->dev);
3914 pm_request_autosuspend(q->dev);
3916 q->rpm_status = RPM_SUSPENDED;
3918 spin_unlock_irq(q->queue_lock);
3920 EXPORT_SYMBOL(blk_post_runtime_resume);
3923 * blk_set_runtime_active - Force runtime status of the queue to be active
3924 * @q: the queue of the device
3926 * If the device is left runtime suspended during system suspend the resume
3927 * hook typically resumes the device and corrects runtime status
3928 * accordingly. However, that does not affect the queue runtime PM status
3929 * which is still "suspended". This prevents processing requests from the
3932 * This function can be used in driver's resume hook to correct queue
3933 * runtime PM status and re-enable peeking requests from the queue. It
3934 * should be called before first request is added to the queue.
3936 void blk_set_runtime_active(struct request_queue *q)
3938 spin_lock_irq(q->queue_lock);
3939 q->rpm_status = RPM_ACTIVE;
3940 pm_runtime_mark_last_busy(q->dev);
3941 pm_request_autosuspend(q->dev);
3942 spin_unlock_irq(q->queue_lock);
3944 EXPORT_SYMBOL(blk_set_runtime_active);
3947 int __init blk_dev_init(void)
3949 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3950 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3951 FIELD_SIZEOF(struct request, cmd_flags));
3952 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3953 FIELD_SIZEOF(struct bio, bi_opf));
3955 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3956 kblockd_workqueue = alloc_workqueue("kblockd",
3957 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3958 if (!kblockd_workqueue)
3959 panic("Failed to create kblockd\n");
3961 request_cachep = kmem_cache_create("blkdev_requests",
3962 sizeof(struct request), 0, SLAB_PANIC, NULL);
3964 blk_requestq_cachep = kmem_cache_create("request_queue",
3965 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3967 #ifdef CONFIG_DEBUG_FS
3968 blk_debugfs_root = debugfs_create_dir("block", NULL);