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();
202 EXPORT_SYMBOL(blk_rq_init);
204 static const struct {
208 [BLK_STS_OK] = { 0, "" },
209 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
210 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
211 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
212 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
213 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
214 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
215 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
216 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
217 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
218 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
219 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
221 /* device mapper special case, should not leak out: */
222 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
224 /* everything else not covered above: */
225 [BLK_STS_IOERR] = { -EIO, "I/O" },
228 blk_status_t errno_to_blk_status(int errno)
232 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
233 if (blk_errors[i].errno == errno)
234 return (__force blk_status_t)i;
237 return BLK_STS_IOERR;
239 EXPORT_SYMBOL_GPL(errno_to_blk_status);
241 int blk_status_to_errno(blk_status_t status)
243 int idx = (__force int)status;
245 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
247 return blk_errors[idx].errno;
249 EXPORT_SYMBOL_GPL(blk_status_to_errno);
251 static void print_req_error(struct request *req, blk_status_t status)
253 int idx = (__force int)status;
255 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
258 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
259 __func__, blk_errors[idx].name, req->rq_disk ?
260 req->rq_disk->disk_name : "?",
261 (unsigned long long)blk_rq_pos(req));
264 static void req_bio_endio(struct request *rq, struct bio *bio,
265 unsigned int nbytes, blk_status_t error)
268 bio->bi_status = error;
270 if (unlikely(rq->rq_flags & RQF_QUIET))
271 bio_set_flag(bio, BIO_QUIET);
273 bio_advance(bio, nbytes);
275 /* don't actually finish bio if it's part of flush sequence */
276 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
280 void blk_dump_rq_flags(struct request *rq, char *msg)
282 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
283 rq->rq_disk ? rq->rq_disk->disk_name : "?",
284 (unsigned long long) rq->cmd_flags);
286 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
287 (unsigned long long)blk_rq_pos(rq),
288 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
289 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
290 rq->bio, rq->biotail, blk_rq_bytes(rq));
292 EXPORT_SYMBOL(blk_dump_rq_flags);
294 static void blk_delay_work(struct work_struct *work)
296 struct request_queue *q;
298 q = container_of(work, struct request_queue, delay_work.work);
299 spin_lock_irq(q->queue_lock);
301 spin_unlock_irq(q->queue_lock);
305 * blk_delay_queue - restart queueing after defined interval
306 * @q: The &struct request_queue in question
307 * @msecs: Delay in msecs
310 * Sometimes queueing needs to be postponed for a little while, to allow
311 * resources to come back. This function will make sure that queueing is
312 * restarted around the specified time.
314 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
316 lockdep_assert_held(q->queue_lock);
317 WARN_ON_ONCE(q->mq_ops);
319 if (likely(!blk_queue_dead(q)))
320 queue_delayed_work(kblockd_workqueue, &q->delay_work,
321 msecs_to_jiffies(msecs));
323 EXPORT_SYMBOL(blk_delay_queue);
326 * blk_start_queue_async - asynchronously restart a previously stopped queue
327 * @q: The &struct request_queue in question
330 * blk_start_queue_async() will clear the stop flag on the queue, and
331 * ensure that the request_fn for the queue is run from an async
334 void blk_start_queue_async(struct request_queue *q)
336 lockdep_assert_held(q->queue_lock);
337 WARN_ON_ONCE(q->mq_ops);
339 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
340 blk_run_queue_async(q);
342 EXPORT_SYMBOL(blk_start_queue_async);
345 * blk_start_queue - restart a previously stopped queue
346 * @q: The &struct request_queue in question
349 * blk_start_queue() will clear the stop flag on the queue, and call
350 * the request_fn for the queue if it was in a stopped state when
351 * entered. Also see blk_stop_queue().
353 void blk_start_queue(struct request_queue *q)
355 lockdep_assert_held(q->queue_lock);
356 WARN_ON_ONCE(q->mq_ops);
358 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
361 EXPORT_SYMBOL(blk_start_queue);
364 * blk_stop_queue - stop a queue
365 * @q: The &struct request_queue in question
368 * The Linux block layer assumes that a block driver will consume all
369 * entries on the request queue when the request_fn strategy is called.
370 * Often this will not happen, because of hardware limitations (queue
371 * depth settings). If a device driver gets a 'queue full' response,
372 * or if it simply chooses not to queue more I/O at one point, it can
373 * call this function to prevent the request_fn from being called until
374 * the driver has signalled it's ready to go again. This happens by calling
375 * blk_start_queue() to restart queue operations.
377 void blk_stop_queue(struct request_queue *q)
379 lockdep_assert_held(q->queue_lock);
380 WARN_ON_ONCE(q->mq_ops);
382 cancel_delayed_work(&q->delay_work);
383 queue_flag_set(QUEUE_FLAG_STOPPED, q);
385 EXPORT_SYMBOL(blk_stop_queue);
388 * blk_sync_queue - cancel any pending callbacks on a queue
392 * The block layer may perform asynchronous callback activity
393 * on a queue, such as calling the unplug function after a timeout.
394 * A block device may call blk_sync_queue to ensure that any
395 * such activity is cancelled, thus allowing it to release resources
396 * that the callbacks might use. The caller must already have made sure
397 * that its ->make_request_fn will not re-add plugging prior to calling
400 * This function does not cancel any asynchronous activity arising
401 * out of elevator or throttling code. That would require elevator_exit()
402 * and blkcg_exit_queue() to be called with queue lock initialized.
405 void blk_sync_queue(struct request_queue *q)
407 del_timer_sync(&q->timeout);
408 cancel_work_sync(&q->timeout_work);
411 struct blk_mq_hw_ctx *hctx;
414 cancel_delayed_work_sync(&q->requeue_work);
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_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
425 * @q: request queue pointer
427 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
428 * set and 1 if the flag was already set.
430 int blk_set_preempt_only(struct request_queue *q)
432 return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
434 EXPORT_SYMBOL_GPL(blk_set_preempt_only);
436 void blk_clear_preempt_only(struct request_queue *q)
438 blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
439 wake_up_all(&q->mq_freeze_wq);
441 EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
444 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
445 * @q: The queue to run
448 * Invoke request handling on a queue if there are any pending requests.
449 * May be used to restart request handling after a request has completed.
450 * This variant runs the queue whether or not the queue has been
451 * stopped. Must be called with the queue lock held and interrupts
452 * disabled. See also @blk_run_queue.
454 inline void __blk_run_queue_uncond(struct request_queue *q)
456 lockdep_assert_held(q->queue_lock);
457 WARN_ON_ONCE(q->mq_ops);
459 if (unlikely(blk_queue_dead(q)))
463 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
464 * the queue lock internally. As a result multiple threads may be
465 * running such a request function concurrently. Keep track of the
466 * number of active request_fn invocations such that blk_drain_queue()
467 * can wait until all these request_fn calls have finished.
469 q->request_fn_active++;
471 q->request_fn_active--;
473 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
476 * __blk_run_queue - run a single device queue
477 * @q: The queue to run
480 * See @blk_run_queue.
482 void __blk_run_queue(struct request_queue *q)
484 lockdep_assert_held(q->queue_lock);
485 WARN_ON_ONCE(q->mq_ops);
487 if (unlikely(blk_queue_stopped(q)))
490 __blk_run_queue_uncond(q);
492 EXPORT_SYMBOL(__blk_run_queue);
495 * blk_run_queue_async - run a single device queue in workqueue context
496 * @q: The queue to run
499 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
503 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
504 * has canceled q->delay_work, callers must hold the queue lock to avoid
505 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
507 void blk_run_queue_async(struct request_queue *q)
509 lockdep_assert_held(q->queue_lock);
510 WARN_ON_ONCE(q->mq_ops);
512 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
513 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
515 EXPORT_SYMBOL(blk_run_queue_async);
518 * blk_run_queue - run a single device queue
519 * @q: The queue to run
522 * Invoke request handling on this queue, if it has pending work to do.
523 * May be used to restart queueing when a request has completed.
525 void blk_run_queue(struct request_queue *q)
529 WARN_ON_ONCE(q->mq_ops);
531 spin_lock_irqsave(q->queue_lock, flags);
533 spin_unlock_irqrestore(q->queue_lock, flags);
535 EXPORT_SYMBOL(blk_run_queue);
537 void blk_put_queue(struct request_queue *q)
539 kobject_put(&q->kobj);
541 EXPORT_SYMBOL(blk_put_queue);
544 * __blk_drain_queue - drain requests from request_queue
546 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
548 * Drain requests from @q. If @drain_all is set, all requests are drained.
549 * If not, only ELVPRIV requests are drained. The caller is responsible
550 * for ensuring that no new requests which need to be drained are queued.
552 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
553 __releases(q->queue_lock)
554 __acquires(q->queue_lock)
558 lockdep_assert_held(q->queue_lock);
559 WARN_ON_ONCE(q->mq_ops);
565 * The caller might be trying to drain @q before its
566 * elevator is initialized.
569 elv_drain_elevator(q);
571 blkcg_drain_queue(q);
574 * This function might be called on a queue which failed
575 * driver init after queue creation or is not yet fully
576 * active yet. Some drivers (e.g. fd and loop) get unhappy
577 * in such cases. Kick queue iff dispatch queue has
578 * something on it and @q has request_fn set.
580 if (!list_empty(&q->queue_head) && q->request_fn)
583 drain |= q->nr_rqs_elvpriv;
584 drain |= q->request_fn_active;
587 * Unfortunately, requests are queued at and tracked from
588 * multiple places and there's no single counter which can
589 * be drained. Check all the queues and counters.
592 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
593 drain |= !list_empty(&q->queue_head);
594 for (i = 0; i < 2; i++) {
595 drain |= q->nr_rqs[i];
596 drain |= q->in_flight[i];
598 drain |= !list_empty(&fq->flush_queue[i]);
605 spin_unlock_irq(q->queue_lock);
609 spin_lock_irq(q->queue_lock);
613 * With queue marked dead, any woken up waiter will fail the
614 * allocation path, so the wakeup chaining is lost and we're
615 * left with hung waiters. We need to wake up those waiters.
618 struct request_list *rl;
620 blk_queue_for_each_rl(rl, q)
621 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
622 wake_up_all(&rl->wait[i]);
626 void blk_drain_queue(struct request_queue *q)
628 spin_lock_irq(q->queue_lock);
629 __blk_drain_queue(q, true);
630 spin_unlock_irq(q->queue_lock);
634 * blk_queue_bypass_start - enter queue bypass mode
635 * @q: queue of interest
637 * In bypass mode, only the dispatch FIFO queue of @q is used. This
638 * function makes @q enter bypass mode and drains all requests which were
639 * throttled or issued before. On return, it's guaranteed that no request
640 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
641 * inside queue or RCU read lock.
643 void blk_queue_bypass_start(struct request_queue *q)
645 WARN_ON_ONCE(q->mq_ops);
647 spin_lock_irq(q->queue_lock);
649 queue_flag_set(QUEUE_FLAG_BYPASS, q);
650 spin_unlock_irq(q->queue_lock);
653 * Queues start drained. Skip actual draining till init is
654 * complete. This avoids lenghty delays during queue init which
655 * can happen many times during boot.
657 if (blk_queue_init_done(q)) {
658 spin_lock_irq(q->queue_lock);
659 __blk_drain_queue(q, false);
660 spin_unlock_irq(q->queue_lock);
662 /* ensure blk_queue_bypass() is %true inside RCU read lock */
666 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
669 * blk_queue_bypass_end - leave queue bypass mode
670 * @q: queue of interest
672 * Leave bypass mode and restore the normal queueing behavior.
674 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
675 * this function is called for both blk-sq and blk-mq queues.
677 void blk_queue_bypass_end(struct request_queue *q)
679 spin_lock_irq(q->queue_lock);
680 if (!--q->bypass_depth)
681 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
682 WARN_ON_ONCE(q->bypass_depth < 0);
683 spin_unlock_irq(q->queue_lock);
685 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
687 void blk_set_queue_dying(struct request_queue *q)
689 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
692 * When queue DYING flag is set, we need to block new req
693 * entering queue, so we call blk_freeze_queue_start() to
694 * prevent I/O from crossing blk_queue_enter().
696 blk_freeze_queue_start(q);
699 blk_mq_wake_waiters(q);
701 struct request_list *rl;
703 spin_lock_irq(q->queue_lock);
704 blk_queue_for_each_rl(rl, q) {
706 wake_up_all(&rl->wait[BLK_RW_SYNC]);
707 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
710 spin_unlock_irq(q->queue_lock);
713 /* Make blk_queue_enter() reexamine the DYING flag. */
714 wake_up_all(&q->mq_freeze_wq);
716 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
718 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
719 void blk_exit_queue(struct request_queue *q)
722 * Since the I/O scheduler exit code may access cgroup information,
723 * perform I/O scheduler exit before disassociating from the block
728 elevator_exit(q, q->elevator);
733 * Remove all references to @q from the block cgroup controller before
734 * restoring @q->queue_lock to avoid that restoring this pointer causes
735 * e.g. blkcg_print_blkgs() to crash.
740 * Since the cgroup code may dereference the @q->backing_dev_info
741 * pointer, only decrease its reference count after having removed the
742 * association with the block cgroup controller.
744 bdi_put(q->backing_dev_info);
748 * blk_cleanup_queue - shutdown a request queue
749 * @q: request queue to shutdown
751 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
752 * put it. All future requests will be failed immediately with -ENODEV.
754 void blk_cleanup_queue(struct request_queue *q)
756 spinlock_t *lock = q->queue_lock;
758 /* mark @q DYING, no new request or merges will be allowed afterwards */
759 mutex_lock(&q->sysfs_lock);
760 blk_set_queue_dying(q);
764 * A dying queue is permanently in bypass mode till released. Note
765 * that, unlike blk_queue_bypass_start(), we aren't performing
766 * synchronize_rcu() after entering bypass mode to avoid the delay
767 * as some drivers create and destroy a lot of queues while
768 * probing. This is still safe because blk_release_queue() will be
769 * called only after the queue refcnt drops to zero and nothing,
770 * RCU or not, would be traversing the queue by then.
773 queue_flag_set(QUEUE_FLAG_BYPASS, q);
775 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
776 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
777 queue_flag_set(QUEUE_FLAG_DYING, q);
778 spin_unlock_irq(lock);
779 mutex_unlock(&q->sysfs_lock);
782 * Drain all requests queued before DYING marking. Set DEAD flag to
783 * prevent that q->request_fn() gets invoked after draining finished.
787 queue_flag_set(QUEUE_FLAG_DEAD, q);
788 spin_unlock_irq(lock);
791 * make sure all in-progress dispatch are completed because
792 * blk_freeze_queue() can only complete all requests, and
793 * dispatch may still be in-progress since we dispatch requests
794 * from more than one contexts.
796 * We rely on driver to deal with the race in case that queue
797 * initialization isn't done.
799 if (q->mq_ops && blk_queue_init_done(q))
800 blk_mq_quiesce_queue(q);
802 /* for synchronous bio-based driver finish in-flight integrity i/o */
803 blk_flush_integrity();
805 /* @q won't process any more request, flush async actions */
806 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
810 * I/O scheduler exit is only safe after the sysfs scheduler attribute
813 WARN_ON_ONCE(q->kobj.state_in_sysfs);
818 blk_mq_free_queue(q);
819 percpu_ref_exit(&q->q_usage_counter);
822 if (q->queue_lock != &q->__queue_lock)
823 q->queue_lock = &q->__queue_lock;
824 spin_unlock_irq(lock);
826 /* @q is and will stay empty, shutdown and put */
829 EXPORT_SYMBOL(blk_cleanup_queue);
831 /* Allocate memory local to the request queue */
832 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
834 struct request_queue *q = data;
836 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
839 static void free_request_simple(void *element, void *data)
841 kmem_cache_free(request_cachep, element);
844 static void *alloc_request_size(gfp_t gfp_mask, void *data)
846 struct request_queue *q = data;
849 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
851 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
858 static void free_request_size(void *element, void *data)
860 struct request_queue *q = data;
863 q->exit_rq_fn(q, element);
867 int blk_init_rl(struct request_list *rl, struct request_queue *q,
870 if (unlikely(rl->rq_pool) || q->mq_ops)
874 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
875 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
876 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
877 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
880 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
881 alloc_request_size, free_request_size,
882 q, gfp_mask, q->node);
884 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
885 alloc_request_simple, free_request_simple,
886 q, gfp_mask, q->node);
891 if (rl != &q->root_rl)
892 WARN_ON_ONCE(!blk_get_queue(q));
897 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
900 mempool_destroy(rl->rq_pool);
901 if (rl != &q->root_rl)
906 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
908 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL);
910 EXPORT_SYMBOL(blk_alloc_queue);
913 * blk_queue_enter() - try to increase q->q_usage_counter
914 * @q: request queue pointer
915 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
917 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
919 const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
922 bool success = false;
925 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
927 * The code that sets the PREEMPT_ONLY flag is
928 * responsible for ensuring that that flag is globally
929 * visible before the queue is unfrozen.
931 if (preempt || !blk_queue_preempt_only(q)) {
934 percpu_ref_put(&q->q_usage_counter);
942 if (flags & BLK_MQ_REQ_NOWAIT)
946 * read pair of barrier in blk_freeze_queue_start(),
947 * we need to order reading __PERCPU_REF_DEAD flag of
948 * .q_usage_counter and reading .mq_freeze_depth or
949 * queue dying flag, otherwise the following wait may
950 * never return if the two reads are reordered.
954 wait_event(q->mq_freeze_wq,
955 (atomic_read(&q->mq_freeze_depth) == 0 &&
956 (preempt || !blk_queue_preempt_only(q))) ||
958 if (blk_queue_dying(q))
963 void blk_queue_exit(struct request_queue *q)
965 percpu_ref_put(&q->q_usage_counter);
968 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
970 struct request_queue *q =
971 container_of(ref, struct request_queue, q_usage_counter);
973 wake_up_all(&q->mq_freeze_wq);
976 static void blk_rq_timed_out_timer(struct timer_list *t)
978 struct request_queue *q = from_timer(q, t, timeout);
980 kblockd_schedule_work(&q->timeout_work);
983 static void blk_timeout_work_dummy(struct work_struct *work)
988 * blk_alloc_queue_node - allocate a request queue
989 * @gfp_mask: memory allocation flags
990 * @node_id: NUMA node to allocate memory from
991 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
992 * serialize calls to the legacy .request_fn() callback. Ignored for
993 * blk-mq request queues.
995 * Note: pass the queue lock as the third argument to this function instead of
996 * setting the queue lock pointer explicitly to avoid triggering a sporadic
997 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
998 * the queue lock pointer must be set before blkcg_init_queue() is called.
1000 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
1003 struct request_queue *q;
1006 q = kmem_cache_alloc_node(blk_requestq_cachep,
1007 gfp_mask | __GFP_ZERO, node_id);
1011 INIT_LIST_HEAD(&q->queue_head);
1012 q->last_merge = NULL;
1014 q->boundary_rq = NULL;
1016 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
1020 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1024 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1025 if (!q->backing_dev_info)
1028 q->stats = blk_alloc_queue_stats();
1032 q->backing_dev_info->ra_pages =
1033 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1034 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1035 q->backing_dev_info->name = "block";
1038 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1039 laptop_mode_timer_fn, 0);
1040 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
1041 INIT_WORK(&q->timeout_work, blk_timeout_work_dummy);
1042 INIT_LIST_HEAD(&q->timeout_list);
1043 INIT_LIST_HEAD(&q->icq_list);
1044 #ifdef CONFIG_BLK_CGROUP
1045 INIT_LIST_HEAD(&q->blkg_list);
1047 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
1049 kobject_init(&q->kobj, &blk_queue_ktype);
1051 #ifdef CONFIG_BLK_DEV_IO_TRACE
1052 mutex_init(&q->blk_trace_mutex);
1054 mutex_init(&q->sysfs_lock);
1055 spin_lock_init(&q->__queue_lock);
1058 q->queue_lock = lock ? : &q->__queue_lock;
1061 * A queue starts its life with bypass turned on to avoid
1062 * unnecessary bypass on/off overhead and nasty surprises during
1063 * init. The initial bypass will be finished when the queue is
1064 * registered by blk_register_queue().
1066 q->bypass_depth = 1;
1067 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
1069 init_waitqueue_head(&q->mq_freeze_wq);
1072 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1073 * See blk_register_queue() for details.
1075 if (percpu_ref_init(&q->q_usage_counter,
1076 blk_queue_usage_counter_release,
1077 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
1080 if (blkcg_init_queue(q))
1086 percpu_ref_exit(&q->q_usage_counter);
1088 blk_free_queue_stats(q->stats);
1090 bdi_put(q->backing_dev_info);
1092 bioset_exit(&q->bio_split);
1094 ida_simple_remove(&blk_queue_ida, q->id);
1096 kmem_cache_free(blk_requestq_cachep, q);
1099 EXPORT_SYMBOL(blk_alloc_queue_node);
1102 * blk_init_queue - prepare a request queue for use with a block device
1103 * @rfn: The function to be called to process requests that have been
1104 * placed on the queue.
1105 * @lock: Request queue spin lock
1108 * If a block device wishes to use the standard request handling procedures,
1109 * which sorts requests and coalesces adjacent requests, then it must
1110 * call blk_init_queue(). The function @rfn will be called when there
1111 * are requests on the queue that need to be processed. If the device
1112 * supports plugging, then @rfn may not be called immediately when requests
1113 * are available on the queue, but may be called at some time later instead.
1114 * Plugged queues are generally unplugged when a buffer belonging to one
1115 * of the requests on the queue is needed, or due to memory pressure.
1117 * @rfn is not required, or even expected, to remove all requests off the
1118 * queue, but only as many as it can handle at a time. If it does leave
1119 * requests on the queue, it is responsible for arranging that the requests
1120 * get dealt with eventually.
1122 * The queue spin lock must be held while manipulating the requests on the
1123 * request queue; this lock will be taken also from interrupt context, so irq
1124 * disabling is needed for it.
1126 * Function returns a pointer to the initialized request queue, or %NULL if
1127 * it didn't succeed.
1130 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1131 * when the block device is deactivated (such as at module unload).
1134 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1136 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1138 EXPORT_SYMBOL(blk_init_queue);
1140 struct request_queue *
1141 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1143 struct request_queue *q;
1145 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
1149 q->request_fn = rfn;
1150 if (blk_init_allocated_queue(q) < 0) {
1151 blk_cleanup_queue(q);
1157 EXPORT_SYMBOL(blk_init_queue_node);
1159 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1162 int blk_init_allocated_queue(struct request_queue *q)
1164 WARN_ON_ONCE(q->mq_ops);
1166 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
1170 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1171 goto out_free_flush_queue;
1173 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1174 goto out_exit_flush_rq;
1176 INIT_WORK(&q->timeout_work, blk_timeout_work);
1177 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1180 * This also sets hw/phys segments, boundary and size
1182 blk_queue_make_request(q, blk_queue_bio);
1184 q->sg_reserved_size = INT_MAX;
1186 if (elevator_init(q))
1187 goto out_exit_flush_rq;
1192 q->exit_rq_fn(q, q->fq->flush_rq);
1193 out_free_flush_queue:
1194 blk_free_flush_queue(q->fq);
1198 EXPORT_SYMBOL(blk_init_allocated_queue);
1200 bool blk_get_queue(struct request_queue *q)
1202 if (likely(!blk_queue_dying(q))) {
1209 EXPORT_SYMBOL(blk_get_queue);
1211 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1213 if (rq->rq_flags & RQF_ELVPRIV) {
1214 elv_put_request(rl->q, rq);
1216 put_io_context(rq->elv.icq->ioc);
1219 mempool_free(rq, rl->rq_pool);
1223 * ioc_batching returns true if the ioc is a valid batching request and
1224 * should be given priority access to a request.
1226 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1232 * Make sure the process is able to allocate at least 1 request
1233 * even if the batch times out, otherwise we could theoretically
1236 return ioc->nr_batch_requests == q->nr_batching ||
1237 (ioc->nr_batch_requests > 0
1238 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1242 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1243 * will cause the process to be a "batcher" on all queues in the system. This
1244 * is the behaviour we want though - once it gets a wakeup it should be given
1247 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1249 if (!ioc || ioc_batching(q, ioc))
1252 ioc->nr_batch_requests = q->nr_batching;
1253 ioc->last_waited = jiffies;
1256 static void __freed_request(struct request_list *rl, int sync)
1258 struct request_queue *q = rl->q;
1260 if (rl->count[sync] < queue_congestion_off_threshold(q))
1261 blk_clear_congested(rl, sync);
1263 if (rl->count[sync] + 1 <= q->nr_requests) {
1264 if (waitqueue_active(&rl->wait[sync]))
1265 wake_up(&rl->wait[sync]);
1267 blk_clear_rl_full(rl, sync);
1272 * A request has just been released. Account for it, update the full and
1273 * congestion status, wake up any waiters. Called under q->queue_lock.
1275 static void freed_request(struct request_list *rl, bool sync,
1276 req_flags_t rq_flags)
1278 struct request_queue *q = rl->q;
1282 if (rq_flags & RQF_ELVPRIV)
1283 q->nr_rqs_elvpriv--;
1285 __freed_request(rl, sync);
1287 if (unlikely(rl->starved[sync ^ 1]))
1288 __freed_request(rl, sync ^ 1);
1291 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1293 struct request_list *rl;
1294 int on_thresh, off_thresh;
1296 WARN_ON_ONCE(q->mq_ops);
1298 spin_lock_irq(q->queue_lock);
1299 q->nr_requests = nr;
1300 blk_queue_congestion_threshold(q);
1301 on_thresh = queue_congestion_on_threshold(q);
1302 off_thresh = queue_congestion_off_threshold(q);
1304 blk_queue_for_each_rl(rl, q) {
1305 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1306 blk_set_congested(rl, BLK_RW_SYNC);
1307 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1308 blk_clear_congested(rl, BLK_RW_SYNC);
1310 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1311 blk_set_congested(rl, BLK_RW_ASYNC);
1312 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1313 blk_clear_congested(rl, BLK_RW_ASYNC);
1315 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1316 blk_set_rl_full(rl, BLK_RW_SYNC);
1318 blk_clear_rl_full(rl, BLK_RW_SYNC);
1319 wake_up(&rl->wait[BLK_RW_SYNC]);
1322 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1323 blk_set_rl_full(rl, BLK_RW_ASYNC);
1325 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1326 wake_up(&rl->wait[BLK_RW_ASYNC]);
1330 spin_unlock_irq(q->queue_lock);
1335 * __get_request - get a free request
1336 * @rl: request list to allocate from
1337 * @op: operation and flags
1338 * @bio: bio to allocate request for (can be %NULL)
1339 * @flags: BLQ_MQ_REQ_* flags
1340 * @gfp_mask: allocator flags
1342 * Get a free request from @q. This function may fail under memory
1343 * pressure or if @q is dead.
1345 * Must be called with @q->queue_lock held and,
1346 * Returns ERR_PTR on failure, with @q->queue_lock held.
1347 * Returns request pointer on success, with @q->queue_lock *not held*.
1349 static struct request *__get_request(struct request_list *rl, unsigned int op,
1350 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask)
1352 struct request_queue *q = rl->q;
1354 struct elevator_type *et = q->elevator->type;
1355 struct io_context *ioc = rq_ioc(bio);
1356 struct io_cq *icq = NULL;
1357 const bool is_sync = op_is_sync(op);
1359 req_flags_t rq_flags = RQF_ALLOCED;
1361 lockdep_assert_held(q->queue_lock);
1363 if (unlikely(blk_queue_dying(q)))
1364 return ERR_PTR(-ENODEV);
1366 may_queue = elv_may_queue(q, op);
1367 if (may_queue == ELV_MQUEUE_NO)
1370 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1371 if (rl->count[is_sync]+1 >= q->nr_requests) {
1373 * The queue will fill after this allocation, so set
1374 * it as full, and mark this process as "batching".
1375 * This process will be allowed to complete a batch of
1376 * requests, others will be blocked.
1378 if (!blk_rl_full(rl, is_sync)) {
1379 ioc_set_batching(q, ioc);
1380 blk_set_rl_full(rl, is_sync);
1382 if (may_queue != ELV_MQUEUE_MUST
1383 && !ioc_batching(q, ioc)) {
1385 * The queue is full and the allocating
1386 * process is not a "batcher", and not
1387 * exempted by the IO scheduler
1389 return ERR_PTR(-ENOMEM);
1393 blk_set_congested(rl, is_sync);
1397 * Only allow batching queuers to allocate up to 50% over the defined
1398 * limit of requests, otherwise we could have thousands of requests
1399 * allocated with any setting of ->nr_requests
1401 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1402 return ERR_PTR(-ENOMEM);
1404 q->nr_rqs[is_sync]++;
1405 rl->count[is_sync]++;
1406 rl->starved[is_sync] = 0;
1409 * Decide whether the new request will be managed by elevator. If
1410 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1411 * prevent the current elevator from being destroyed until the new
1412 * request is freed. This guarantees icq's won't be destroyed and
1413 * makes creating new ones safe.
1415 * Flush requests do not use the elevator so skip initialization.
1416 * This allows a request to share the flush and elevator data.
1418 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1419 * it will be created after releasing queue_lock.
1421 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1422 rq_flags |= RQF_ELVPRIV;
1423 q->nr_rqs_elvpriv++;
1424 if (et->icq_cache && ioc)
1425 icq = ioc_lookup_icq(ioc, q);
1428 if (blk_queue_io_stat(q))
1429 rq_flags |= RQF_IO_STAT;
1430 spin_unlock_irq(q->queue_lock);
1432 /* allocate and init request */
1433 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1438 blk_rq_set_rl(rq, rl);
1440 rq->rq_flags = rq_flags;
1441 if (flags & BLK_MQ_REQ_PREEMPT)
1442 rq->rq_flags |= RQF_PREEMPT;
1445 if (rq_flags & RQF_ELVPRIV) {
1446 if (unlikely(et->icq_cache && !icq)) {
1448 icq = ioc_create_icq(ioc, q, gfp_mask);
1454 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1457 /* @rq->elv.icq holds io_context until @rq is freed */
1459 get_io_context(icq->ioc);
1463 * ioc may be NULL here, and ioc_batching will be false. That's
1464 * OK, if the queue is under the request limit then requests need
1465 * not count toward the nr_batch_requests limit. There will always
1466 * be some limit enforced by BLK_BATCH_TIME.
1468 if (ioc_batching(q, ioc))
1469 ioc->nr_batch_requests--;
1471 trace_block_getrq(q, bio, op);
1476 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1477 * and may fail indefinitely under memory pressure and thus
1478 * shouldn't stall IO. Treat this request as !elvpriv. This will
1479 * disturb iosched and blkcg but weird is bettern than dead.
1481 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1482 __func__, dev_name(q->backing_dev_info->dev));
1484 rq->rq_flags &= ~RQF_ELVPRIV;
1487 spin_lock_irq(q->queue_lock);
1488 q->nr_rqs_elvpriv--;
1489 spin_unlock_irq(q->queue_lock);
1494 * Allocation failed presumably due to memory. Undo anything we
1495 * might have messed up.
1497 * Allocating task should really be put onto the front of the wait
1498 * queue, but this is pretty rare.
1500 spin_lock_irq(q->queue_lock);
1501 freed_request(rl, is_sync, rq_flags);
1504 * in the very unlikely event that allocation failed and no
1505 * requests for this direction was pending, mark us starved so that
1506 * freeing of a request in the other direction will notice
1507 * us. another possible fix would be to split the rq mempool into
1511 if (unlikely(rl->count[is_sync] == 0))
1512 rl->starved[is_sync] = 1;
1513 return ERR_PTR(-ENOMEM);
1517 * get_request - get a free request
1518 * @q: request_queue to allocate request from
1519 * @op: operation and flags
1520 * @bio: bio to allocate request for (can be %NULL)
1521 * @flags: BLK_MQ_REQ_* flags.
1522 * @gfp: allocator flags
1524 * Get a free request from @q. If %BLK_MQ_REQ_NOWAIT is set in @flags,
1525 * this function keeps retrying under memory pressure and fails iff @q is dead.
1527 * Must be called with @q->queue_lock held and,
1528 * Returns ERR_PTR on failure, with @q->queue_lock held.
1529 * Returns request pointer on success, with @q->queue_lock *not held*.
1531 static struct request *get_request(struct request_queue *q, unsigned int op,
1532 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp)
1534 const bool is_sync = op_is_sync(op);
1536 struct request_list *rl;
1539 lockdep_assert_held(q->queue_lock);
1540 WARN_ON_ONCE(q->mq_ops);
1542 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1544 rq = __get_request(rl, op, bio, flags, gfp);
1548 if (op & REQ_NOWAIT) {
1550 return ERR_PTR(-EAGAIN);
1553 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1558 /* wait on @rl and retry */
1559 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1560 TASK_UNINTERRUPTIBLE);
1562 trace_block_sleeprq(q, bio, op);
1564 spin_unlock_irq(q->queue_lock);
1568 * After sleeping, we become a "batching" process and will be able
1569 * to allocate at least one request, and up to a big batch of them
1570 * for a small period time. See ioc_batching, ioc_set_batching
1572 ioc_set_batching(q, current->io_context);
1574 spin_lock_irq(q->queue_lock);
1575 finish_wait(&rl->wait[is_sync], &wait);
1580 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1581 static struct request *blk_old_get_request(struct request_queue *q,
1582 unsigned int op, blk_mq_req_flags_t flags)
1585 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO;
1588 WARN_ON_ONCE(q->mq_ops);
1590 /* create ioc upfront */
1591 create_io_context(gfp_mask, q->node);
1593 ret = blk_queue_enter(q, flags);
1595 return ERR_PTR(ret);
1596 spin_lock_irq(q->queue_lock);
1597 rq = get_request(q, op, NULL, flags, gfp_mask);
1599 spin_unlock_irq(q->queue_lock);
1604 /* q->queue_lock is unlocked at this point */
1606 rq->__sector = (sector_t) -1;
1607 rq->bio = rq->biotail = NULL;
1612 * blk_get_request - allocate a request
1613 * @q: request queue to allocate a request for
1614 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1615 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1617 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1618 blk_mq_req_flags_t flags)
1620 struct request *req;
1622 WARN_ON_ONCE(op & REQ_NOWAIT);
1623 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1626 req = blk_mq_alloc_request(q, op, flags);
1627 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1628 q->mq_ops->initialize_rq_fn(req);
1630 req = blk_old_get_request(q, op, flags);
1631 if (!IS_ERR(req) && q->initialize_rq_fn)
1632 q->initialize_rq_fn(req);
1637 EXPORT_SYMBOL(blk_get_request);
1640 * blk_requeue_request - put a request back on queue
1641 * @q: request queue where request should be inserted
1642 * @rq: request to be inserted
1645 * Drivers often keep queueing requests until the hardware cannot accept
1646 * more, when that condition happens we need to put the request back
1647 * on the queue. Must be called with queue lock held.
1649 void blk_requeue_request(struct request_queue *q, struct request *rq)
1651 lockdep_assert_held(q->queue_lock);
1652 WARN_ON_ONCE(q->mq_ops);
1654 blk_delete_timer(rq);
1655 blk_clear_rq_complete(rq);
1656 trace_block_rq_requeue(q, rq);
1657 rq_qos_requeue(q, rq);
1659 if (rq->rq_flags & RQF_QUEUED)
1660 blk_queue_end_tag(q, rq);
1662 BUG_ON(blk_queued_rq(rq));
1664 elv_requeue_request(q, rq);
1666 EXPORT_SYMBOL(blk_requeue_request);
1668 static void add_acct_request(struct request_queue *q, struct request *rq,
1671 blk_account_io_start(rq, true);
1672 __elv_add_request(q, rq, where);
1675 static void part_round_stats_single(struct request_queue *q, int cpu,
1676 struct hd_struct *part, unsigned long now,
1677 unsigned int inflight)
1680 __part_stat_add(cpu, part, time_in_queue,
1681 inflight * (now - part->stamp));
1682 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1688 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1689 * @q: target block queue
1690 * @cpu: cpu number for stats access
1691 * @part: target partition
1693 * The average IO queue length and utilisation statistics are maintained
1694 * by observing the current state of the queue length and the amount of
1695 * time it has been in this state for.
1697 * Normally, that accounting is done on IO completion, but that can result
1698 * in more than a second's worth of IO being accounted for within any one
1699 * second, leading to >100% utilisation. To deal with that, we call this
1700 * function to do a round-off before returning the results when reading
1701 * /proc/diskstats. This accounts immediately for all queue usage up to
1702 * the current jiffies and restarts the counters again.
1704 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1706 struct hd_struct *part2 = NULL;
1707 unsigned long now = jiffies;
1708 unsigned int inflight[2];
1711 if (part->stamp != now)
1715 part2 = &part_to_disk(part)->part0;
1716 if (part2->stamp != now)
1723 part_in_flight(q, part, inflight);
1726 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1728 part_round_stats_single(q, cpu, part, now, inflight[0]);
1730 EXPORT_SYMBOL_GPL(part_round_stats);
1733 static void blk_pm_put_request(struct request *rq)
1735 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1736 pm_runtime_mark_last_busy(rq->q->dev);
1739 static inline void blk_pm_put_request(struct request *rq) {}
1742 void __blk_put_request(struct request_queue *q, struct request *req)
1744 req_flags_t rq_flags = req->rq_flags;
1750 blk_mq_free_request(req);
1754 lockdep_assert_held(q->queue_lock);
1756 blk_req_zone_write_unlock(req);
1757 blk_pm_put_request(req);
1759 elv_completed_request(q, req);
1761 /* this is a bio leak */
1762 WARN_ON(req->bio != NULL);
1764 rq_qos_done(q, req);
1767 * Request may not have originated from ll_rw_blk. if not,
1768 * it didn't come out of our reserved rq pools
1770 if (rq_flags & RQF_ALLOCED) {
1771 struct request_list *rl = blk_rq_rl(req);
1772 bool sync = op_is_sync(req->cmd_flags);
1774 BUG_ON(!list_empty(&req->queuelist));
1775 BUG_ON(ELV_ON_HASH(req));
1777 blk_free_request(rl, req);
1778 freed_request(rl, sync, rq_flags);
1783 EXPORT_SYMBOL_GPL(__blk_put_request);
1785 void blk_put_request(struct request *req)
1787 struct request_queue *q = req->q;
1790 blk_mq_free_request(req);
1792 unsigned long flags;
1794 spin_lock_irqsave(q->queue_lock, flags);
1795 __blk_put_request(q, req);
1796 spin_unlock_irqrestore(q->queue_lock, flags);
1799 EXPORT_SYMBOL(blk_put_request);
1801 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1804 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1806 if (!ll_back_merge_fn(q, req, bio))
1809 trace_block_bio_backmerge(q, req, bio);
1811 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1812 blk_rq_set_mixed_merge(req);
1814 req->biotail->bi_next = bio;
1816 req->__data_len += bio->bi_iter.bi_size;
1817 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1819 blk_account_io_start(req, false);
1823 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1826 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1828 if (!ll_front_merge_fn(q, req, bio))
1831 trace_block_bio_frontmerge(q, req, bio);
1833 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1834 blk_rq_set_mixed_merge(req);
1836 bio->bi_next = req->bio;
1839 req->__sector = bio->bi_iter.bi_sector;
1840 req->__data_len += bio->bi_iter.bi_size;
1841 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1843 blk_account_io_start(req, false);
1847 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1850 unsigned short segments = blk_rq_nr_discard_segments(req);
1852 if (segments >= queue_max_discard_segments(q))
1854 if (blk_rq_sectors(req) + bio_sectors(bio) >
1855 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1858 req->biotail->bi_next = bio;
1860 req->__data_len += bio->bi_iter.bi_size;
1861 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1862 req->nr_phys_segments = segments + 1;
1864 blk_account_io_start(req, false);
1867 req_set_nomerge(q, req);
1872 * blk_attempt_plug_merge - try to merge with %current's plugged list
1873 * @q: request_queue new bio is being queued at
1874 * @bio: new bio being queued
1875 * @request_count: out parameter for number of traversed plugged requests
1876 * @same_queue_rq: pointer to &struct request that gets filled in when
1877 * another request associated with @q is found on the plug list
1878 * (optional, may be %NULL)
1880 * Determine whether @bio being queued on @q can be merged with a request
1881 * on %current's plugged list. Returns %true if merge was successful,
1884 * Plugging coalesces IOs from the same issuer for the same purpose without
1885 * going through @q->queue_lock. As such it's more of an issuing mechanism
1886 * than scheduling, and the request, while may have elvpriv data, is not
1887 * added on the elevator at this point. In addition, we don't have
1888 * reliable access to the elevator outside queue lock. Only check basic
1889 * merging parameters without querying the elevator.
1891 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1893 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1894 unsigned int *request_count,
1895 struct request **same_queue_rq)
1897 struct blk_plug *plug;
1899 struct list_head *plug_list;
1901 plug = current->plug;
1907 plug_list = &plug->mq_list;
1909 plug_list = &plug->list;
1911 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1912 bool merged = false;
1917 * Only blk-mq multiple hardware queues case checks the
1918 * rq in the same queue, there should be only one such
1922 *same_queue_rq = rq;
1925 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1928 switch (blk_try_merge(rq, bio)) {
1929 case ELEVATOR_BACK_MERGE:
1930 merged = bio_attempt_back_merge(q, rq, bio);
1932 case ELEVATOR_FRONT_MERGE:
1933 merged = bio_attempt_front_merge(q, rq, bio);
1935 case ELEVATOR_DISCARD_MERGE:
1936 merged = bio_attempt_discard_merge(q, rq, bio);
1949 unsigned int blk_plug_queued_count(struct request_queue *q)
1951 struct blk_plug *plug;
1953 struct list_head *plug_list;
1954 unsigned int ret = 0;
1956 plug = current->plug;
1961 plug_list = &plug->mq_list;
1963 plug_list = &plug->list;
1965 list_for_each_entry(rq, plug_list, queuelist) {
1973 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1975 struct io_context *ioc = rq_ioc(bio);
1977 if (bio->bi_opf & REQ_RAHEAD)
1978 req->cmd_flags |= REQ_FAILFAST_MASK;
1980 req->__sector = bio->bi_iter.bi_sector;
1981 if (ioprio_valid(bio_prio(bio)))
1982 req->ioprio = bio_prio(bio);
1984 req->ioprio = ioc->ioprio;
1986 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1987 req->write_hint = bio->bi_write_hint;
1988 blk_rq_bio_prep(req->q, req, bio);
1990 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1992 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1994 struct blk_plug *plug;
1995 int where = ELEVATOR_INSERT_SORT;
1996 struct request *req, *free;
1997 unsigned int request_count = 0;
2000 * low level driver can indicate that it wants pages above a
2001 * certain limit bounced to low memory (ie for highmem, or even
2002 * ISA dma in theory)
2004 blk_queue_bounce(q, &bio);
2006 blk_queue_split(q, &bio);
2008 if (!bio_integrity_prep(bio))
2009 return BLK_QC_T_NONE;
2011 if (op_is_flush(bio->bi_opf)) {
2012 spin_lock_irq(q->queue_lock);
2013 where = ELEVATOR_INSERT_FLUSH;
2018 * Check if we can merge with the plugged list before grabbing
2021 if (!blk_queue_nomerges(q)) {
2022 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
2023 return BLK_QC_T_NONE;
2025 request_count = blk_plug_queued_count(q);
2027 spin_lock_irq(q->queue_lock);
2029 switch (elv_merge(q, &req, bio)) {
2030 case ELEVATOR_BACK_MERGE:
2031 if (!bio_attempt_back_merge(q, req, bio))
2033 elv_bio_merged(q, req, bio);
2034 free = attempt_back_merge(q, req);
2036 __blk_put_request(q, free);
2038 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2040 case ELEVATOR_FRONT_MERGE:
2041 if (!bio_attempt_front_merge(q, req, bio))
2043 elv_bio_merged(q, req, bio);
2044 free = attempt_front_merge(q, req);
2046 __blk_put_request(q, free);
2048 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2055 rq_qos_throttle(q, bio, q->queue_lock);
2058 * Grab a free request. This is might sleep but can not fail.
2059 * Returns with the queue unlocked.
2061 blk_queue_enter_live(q);
2062 req = get_request(q, bio->bi_opf, bio, 0, GFP_NOIO);
2065 rq_qos_cleanup(q, bio);
2066 if (PTR_ERR(req) == -ENOMEM)
2067 bio->bi_status = BLK_STS_RESOURCE;
2069 bio->bi_status = BLK_STS_IOERR;
2074 rq_qos_track(q, req, bio);
2077 * After dropping the lock and possibly sleeping here, our request
2078 * may now be mergeable after it had proven unmergeable (above).
2079 * We don't worry about that case for efficiency. It won't happen
2080 * often, and the elevators are able to handle it.
2082 blk_init_request_from_bio(req, bio);
2084 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
2085 req->cpu = raw_smp_processor_id();
2087 plug = current->plug;
2090 * If this is the first request added after a plug, fire
2093 * @request_count may become stale because of schedule
2094 * out, so check plug list again.
2096 if (!request_count || list_empty(&plug->list))
2097 trace_block_plug(q);
2099 struct request *last = list_entry_rq(plug->list.prev);
2100 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2101 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2102 blk_flush_plug_list(plug, false);
2103 trace_block_plug(q);
2106 list_add_tail(&req->queuelist, &plug->list);
2107 blk_account_io_start(req, true);
2109 spin_lock_irq(q->queue_lock);
2110 add_acct_request(q, req, where);
2113 spin_unlock_irq(q->queue_lock);
2116 return BLK_QC_T_NONE;
2119 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
2121 char b[BDEVNAME_SIZE];
2123 printk(KERN_INFO "attempt to access beyond end of device\n");
2124 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2125 bio_devname(bio, b), bio->bi_opf,
2126 (unsigned long long)bio_end_sector(bio),
2127 (long long)maxsector);
2130 #ifdef CONFIG_FAIL_MAKE_REQUEST
2132 static DECLARE_FAULT_ATTR(fail_make_request);
2134 static int __init setup_fail_make_request(char *str)
2136 return setup_fault_attr(&fail_make_request, str);
2138 __setup("fail_make_request=", setup_fail_make_request);
2140 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2142 return part->make_it_fail && should_fail(&fail_make_request, bytes);
2145 static int __init fail_make_request_debugfs(void)
2147 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2148 NULL, &fail_make_request);
2150 return PTR_ERR_OR_ZERO(dir);
2153 late_initcall(fail_make_request_debugfs);
2155 #else /* CONFIG_FAIL_MAKE_REQUEST */
2157 static inline bool should_fail_request(struct hd_struct *part,
2163 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2165 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2167 const int op = bio_op(bio);
2169 if (part->policy && op_is_write(op)) {
2170 char b[BDEVNAME_SIZE];
2172 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
2176 "generic_make_request: Trying to write "
2177 "to read-only block-device %s (partno %d)\n",
2178 bio_devname(bio, b), part->partno);
2179 /* Older lvm-tools actually trigger this */
2186 static noinline int should_fail_bio(struct bio *bio)
2188 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2192 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2195 * Check whether this bio extends beyond the end of the device or partition.
2196 * This may well happen - the kernel calls bread() without checking the size of
2197 * the device, e.g., when mounting a file system.
2199 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2201 unsigned int nr_sectors = bio_sectors(bio);
2203 if (nr_sectors && maxsector &&
2204 (nr_sectors > maxsector ||
2205 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2206 handle_bad_sector(bio, maxsector);
2213 * Remap block n of partition p to block n+start(p) of the disk.
2215 static inline int blk_partition_remap(struct bio *bio)
2217 struct hd_struct *p;
2221 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2224 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2226 if (unlikely(bio_check_ro(bio, p)))
2230 * Zone reset does not include bi_size so bio_sectors() is always 0.
2231 * Include a test for the reset op code and perform the remap if needed.
2233 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2234 if (bio_check_eod(bio, part_nr_sects_read(p)))
2236 bio->bi_iter.bi_sector += p->start_sect;
2237 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2238 bio->bi_iter.bi_sector - p->start_sect);
2247 static noinline_for_stack bool
2248 generic_make_request_checks(struct bio *bio)
2250 struct request_queue *q;
2251 int nr_sectors = bio_sectors(bio);
2252 blk_status_t status = BLK_STS_IOERR;
2253 char b[BDEVNAME_SIZE];
2257 q = bio->bi_disk->queue;
2260 "generic_make_request: Trying to access "
2261 "nonexistent block-device %s (%Lu)\n",
2262 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2267 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2268 * if queue is not a request based queue.
2270 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2273 if (should_fail_bio(bio))
2276 if (bio->bi_partno) {
2277 if (unlikely(blk_partition_remap(bio)))
2280 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2282 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
2287 * Filter flush bio's early so that make_request based
2288 * drivers without flush support don't have to worry
2291 if (op_is_flush(bio->bi_opf) &&
2292 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2293 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2295 status = BLK_STS_OK;
2300 switch (bio_op(bio)) {
2301 case REQ_OP_DISCARD:
2302 if (!blk_queue_discard(q))
2305 case REQ_OP_SECURE_ERASE:
2306 if (!blk_queue_secure_erase(q))
2309 case REQ_OP_WRITE_SAME:
2310 if (!q->limits.max_write_same_sectors)
2313 case REQ_OP_ZONE_REPORT:
2314 case REQ_OP_ZONE_RESET:
2315 if (!blk_queue_is_zoned(q))
2318 case REQ_OP_WRITE_ZEROES:
2319 if (!q->limits.max_write_zeroes_sectors)
2327 * Various block parts want %current->io_context and lazy ioc
2328 * allocation ends up trading a lot of pain for a small amount of
2329 * memory. Just allocate it upfront. This may fail and block
2330 * layer knows how to live with it.
2332 create_io_context(GFP_ATOMIC, q->node);
2334 if (!blkcg_bio_issue_check(q, bio))
2337 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2338 trace_block_bio_queue(q, bio);
2339 /* Now that enqueuing has been traced, we need to trace
2340 * completion as well.
2342 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2347 status = BLK_STS_NOTSUPP;
2349 bio->bi_status = status;
2355 * generic_make_request - hand a buffer to its device driver for I/O
2356 * @bio: The bio describing the location in memory and on the device.
2358 * generic_make_request() is used to make I/O requests of block
2359 * devices. It is passed a &struct bio, which describes the I/O that needs
2362 * generic_make_request() does not return any status. The
2363 * success/failure status of the request, along with notification of
2364 * completion, is delivered asynchronously through the bio->bi_end_io
2365 * function described (one day) else where.
2367 * The caller of generic_make_request must make sure that bi_io_vec
2368 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2369 * set to describe the device address, and the
2370 * bi_end_io and optionally bi_private are set to describe how
2371 * completion notification should be signaled.
2373 * generic_make_request and the drivers it calls may use bi_next if this
2374 * bio happens to be merged with someone else, and may resubmit the bio to
2375 * a lower device by calling into generic_make_request recursively, which
2376 * means the bio should NOT be touched after the call to ->make_request_fn.
2378 blk_qc_t generic_make_request(struct bio *bio)
2381 * bio_list_on_stack[0] contains bios submitted by the current
2383 * bio_list_on_stack[1] contains bios that were submitted before
2384 * the current make_request_fn, but that haven't been processed
2387 struct bio_list bio_list_on_stack[2];
2388 blk_mq_req_flags_t flags = 0;
2389 struct request_queue *q = bio->bi_disk->queue;
2390 blk_qc_t ret = BLK_QC_T_NONE;
2392 if (bio->bi_opf & REQ_NOWAIT)
2393 flags = BLK_MQ_REQ_NOWAIT;
2394 if (bio_flagged(bio, BIO_QUEUE_ENTERED))
2395 blk_queue_enter_live(q);
2396 else if (blk_queue_enter(q, flags) < 0) {
2397 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2398 bio_wouldblock_error(bio);
2404 if (!generic_make_request_checks(bio))
2408 * We only want one ->make_request_fn to be active at a time, else
2409 * stack usage with stacked devices could be a problem. So use
2410 * current->bio_list to keep a list of requests submited by a
2411 * make_request_fn function. current->bio_list is also used as a
2412 * flag to say if generic_make_request is currently active in this
2413 * task or not. If it is NULL, then no make_request is active. If
2414 * it is non-NULL, then a make_request is active, and new requests
2415 * should be added at the tail
2417 if (current->bio_list) {
2418 bio_list_add(¤t->bio_list[0], bio);
2422 /* following loop may be a bit non-obvious, and so deserves some
2424 * Before entering the loop, bio->bi_next is NULL (as all callers
2425 * ensure that) so we have a list with a single bio.
2426 * We pretend that we have just taken it off a longer list, so
2427 * we assign bio_list to a pointer to the bio_list_on_stack,
2428 * thus initialising the bio_list of new bios to be
2429 * added. ->make_request() may indeed add some more bios
2430 * through a recursive call to generic_make_request. If it
2431 * did, we find a non-NULL value in bio_list and re-enter the loop
2432 * from the top. In this case we really did just take the bio
2433 * of the top of the list (no pretending) and so remove it from
2434 * bio_list, and call into ->make_request() again.
2436 BUG_ON(bio->bi_next);
2437 bio_list_init(&bio_list_on_stack[0]);
2438 current->bio_list = bio_list_on_stack;
2440 bool enter_succeeded = true;
2442 if (unlikely(q != bio->bi_disk->queue)) {
2445 q = bio->bi_disk->queue;
2447 if (bio->bi_opf & REQ_NOWAIT)
2448 flags = BLK_MQ_REQ_NOWAIT;
2449 if (blk_queue_enter(q, flags) < 0) {
2450 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);
2484 bio = bio_list_pop(&bio_list_on_stack[0]);
2486 current->bio_list = NULL; /* deactivate */
2493 EXPORT_SYMBOL(generic_make_request);
2496 * direct_make_request - hand a buffer directly to its device driver for I/O
2497 * @bio: The bio describing the location in memory and on the device.
2499 * This function behaves like generic_make_request(), but does not protect
2500 * against recursion. Must only be used if the called driver is known
2501 * to not call generic_make_request (or direct_make_request) again from
2502 * its make_request function. (Calling direct_make_request again from
2503 * a workqueue is perfectly fine as that doesn't recurse).
2505 blk_qc_t direct_make_request(struct bio *bio)
2507 struct request_queue *q = bio->bi_disk->queue;
2508 bool nowait = bio->bi_opf & REQ_NOWAIT;
2511 if (!generic_make_request_checks(bio))
2512 return BLK_QC_T_NONE;
2514 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2515 if (nowait && !blk_queue_dying(q))
2516 bio->bi_status = BLK_STS_AGAIN;
2518 bio->bi_status = BLK_STS_IOERR;
2520 return BLK_QC_T_NONE;
2523 ret = q->make_request_fn(q, bio);
2527 EXPORT_SYMBOL_GPL(direct_make_request);
2530 * submit_bio - submit a bio to the block device layer for I/O
2531 * @bio: The &struct bio which describes the I/O
2533 * submit_bio() is very similar in purpose to generic_make_request(), and
2534 * uses that function to do most of the work. Both are fairly rough
2535 * interfaces; @bio must be presetup and ready for I/O.
2538 blk_qc_t submit_bio(struct bio *bio)
2541 * If it's a regular read/write or a barrier with data attached,
2542 * go through the normal accounting stuff before submission.
2544 if (bio_has_data(bio)) {
2547 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2548 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2550 count = bio_sectors(bio);
2552 if (op_is_write(bio_op(bio))) {
2553 count_vm_events(PGPGOUT, count);
2555 task_io_account_read(bio->bi_iter.bi_size);
2556 count_vm_events(PGPGIN, count);
2559 if (unlikely(block_dump)) {
2560 char b[BDEVNAME_SIZE];
2561 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2562 current->comm, task_pid_nr(current),
2563 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2564 (unsigned long long)bio->bi_iter.bi_sector,
2565 bio_devname(bio, b), count);
2569 return generic_make_request(bio);
2571 EXPORT_SYMBOL(submit_bio);
2573 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2575 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2579 blk_flush_plug_list(current->plug, false);
2580 return q->poll_fn(q, cookie);
2582 EXPORT_SYMBOL_GPL(blk_poll);
2585 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2586 * for new the queue limits
2588 * @rq: the request being checked
2591 * @rq may have been made based on weaker limitations of upper-level queues
2592 * in request stacking drivers, and it may violate the limitation of @q.
2593 * Since the block layer and the underlying device driver trust @rq
2594 * after it is inserted to @q, it should be checked against @q before
2595 * the insertion using this generic function.
2597 * Request stacking drivers like request-based dm may change the queue
2598 * limits when retrying requests on other queues. Those requests need
2599 * to be checked against the new queue limits again during dispatch.
2601 static int blk_cloned_rq_check_limits(struct request_queue *q,
2604 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2605 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2610 * queue's settings related to segment counting like q->bounce_pfn
2611 * may differ from that of other stacking queues.
2612 * Recalculate it to check the request correctly on this queue's
2615 blk_recalc_rq_segments(rq);
2616 if (rq->nr_phys_segments > queue_max_segments(q)) {
2617 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2625 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2626 * @q: the queue to submit the request
2627 * @rq: the request being queued
2629 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2631 unsigned long flags;
2632 int where = ELEVATOR_INSERT_BACK;
2634 if (blk_cloned_rq_check_limits(q, rq))
2635 return BLK_STS_IOERR;
2638 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2639 return BLK_STS_IOERR;
2642 if (blk_queue_io_stat(q))
2643 blk_account_io_start(rq, true);
2645 * Since we have a scheduler attached on the top device,
2646 * bypass a potential scheduler on the bottom device for
2649 return blk_mq_request_issue_directly(rq);
2652 spin_lock_irqsave(q->queue_lock, flags);
2653 if (unlikely(blk_queue_dying(q))) {
2654 spin_unlock_irqrestore(q->queue_lock, flags);
2655 return BLK_STS_IOERR;
2659 * Submitting request must be dequeued before calling this function
2660 * because it will be linked to another request_queue
2662 BUG_ON(blk_queued_rq(rq));
2664 if (op_is_flush(rq->cmd_flags))
2665 where = ELEVATOR_INSERT_FLUSH;
2667 add_acct_request(q, rq, where);
2668 if (where == ELEVATOR_INSERT_FLUSH)
2670 spin_unlock_irqrestore(q->queue_lock, flags);
2674 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2677 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2678 * @rq: request to examine
2681 * A request could be merge of IOs which require different failure
2682 * handling. This function determines the number of bytes which
2683 * can be failed from the beginning of the request without
2684 * crossing into area which need to be retried further.
2687 * The number of bytes to fail.
2689 unsigned int blk_rq_err_bytes(const struct request *rq)
2691 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2692 unsigned int bytes = 0;
2695 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2696 return blk_rq_bytes(rq);
2699 * Currently the only 'mixing' which can happen is between
2700 * different fastfail types. We can safely fail portions
2701 * which have all the failfast bits that the first one has -
2702 * the ones which are at least as eager to fail as the first
2705 for (bio = rq->bio; bio; bio = bio->bi_next) {
2706 if ((bio->bi_opf & ff) != ff)
2708 bytes += bio->bi_iter.bi_size;
2711 /* this could lead to infinite loop */
2712 BUG_ON(blk_rq_bytes(rq) && !bytes);
2715 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2717 void blk_account_io_completion(struct request *req, unsigned int bytes)
2719 if (blk_do_io_stat(req)) {
2720 const int sgrp = op_stat_group(req_op(req));
2721 struct hd_struct *part;
2724 cpu = part_stat_lock();
2726 part_stat_add(cpu, part, sectors[sgrp], bytes >> 9);
2731 void blk_account_io_done(struct request *req, u64 now)
2734 * Account IO completion. flush_rq isn't accounted as a
2735 * normal IO on queueing nor completion. Accounting the
2736 * containing request is enough.
2738 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2739 const int sgrp = op_stat_group(req_op(req));
2740 struct hd_struct *part;
2743 cpu = part_stat_lock();
2746 part_stat_inc(cpu, part, ios[sgrp]);
2747 part_stat_add(cpu, part, nsecs[sgrp], now - req->start_time_ns);
2748 part_round_stats(req->q, cpu, part);
2749 part_dec_in_flight(req->q, part, rq_data_dir(req));
2751 hd_struct_put(part);
2758 * Don't process normal requests when queue is suspended
2759 * or in the process of suspending/resuming
2761 static bool blk_pm_allow_request(struct request *rq)
2763 switch (rq->q->rpm_status) {
2765 case RPM_SUSPENDING:
2766 return rq->rq_flags & RQF_PM;
2774 static bool blk_pm_allow_request(struct request *rq)
2780 void blk_account_io_start(struct request *rq, bool new_io)
2782 struct hd_struct *part;
2783 int rw = rq_data_dir(rq);
2786 if (!blk_do_io_stat(rq))
2789 cpu = part_stat_lock();
2793 part_stat_inc(cpu, part, merges[rw]);
2795 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2796 if (!hd_struct_try_get(part)) {
2798 * The partition is already being removed,
2799 * the request will be accounted on the disk only
2801 * We take a reference on disk->part0 although that
2802 * partition will never be deleted, so we can treat
2803 * it as any other partition.
2805 part = &rq->rq_disk->part0;
2806 hd_struct_get(part);
2808 part_round_stats(rq->q, cpu, part);
2809 part_inc_in_flight(rq->q, part, rw);
2816 static struct request *elv_next_request(struct request_queue *q)
2819 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2821 WARN_ON_ONCE(q->mq_ops);
2824 list_for_each_entry(rq, &q->queue_head, queuelist) {
2825 if (blk_pm_allow_request(rq))
2828 if (rq->rq_flags & RQF_SOFTBARRIER)
2833 * Flush request is running and flush request isn't queueable
2834 * in the drive, we can hold the queue till flush request is
2835 * finished. Even we don't do this, driver can't dispatch next
2836 * requests and will requeue them. And this can improve
2837 * throughput too. For example, we have request flush1, write1,
2838 * flush 2. flush1 is dispatched, then queue is hold, write1
2839 * isn't inserted to queue. After flush1 is finished, flush2
2840 * will be dispatched. Since disk cache is already clean,
2841 * flush2 will be finished very soon, so looks like flush2 is
2843 * Since the queue is hold, a flag is set to indicate the queue
2844 * should be restarted later. Please see flush_end_io() for
2847 if (fq->flush_pending_idx != fq->flush_running_idx &&
2848 !queue_flush_queueable(q)) {
2849 fq->flush_queue_delayed = 1;
2852 if (unlikely(blk_queue_bypass(q)) ||
2853 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2859 * blk_peek_request - peek at the top of a request queue
2860 * @q: request queue to peek at
2863 * Return the request at the top of @q. The returned request
2864 * should be started using blk_start_request() before LLD starts
2868 * Pointer to the request at the top of @q if available. Null
2871 struct request *blk_peek_request(struct request_queue *q)
2876 lockdep_assert_held(q->queue_lock);
2877 WARN_ON_ONCE(q->mq_ops);
2879 while ((rq = elv_next_request(q)) != NULL) {
2880 if (!(rq->rq_flags & RQF_STARTED)) {
2882 * This is the first time the device driver
2883 * sees this request (possibly after
2884 * requeueing). Notify IO scheduler.
2886 if (rq->rq_flags & RQF_SORTED)
2887 elv_activate_rq(q, rq);
2890 * just mark as started even if we don't start
2891 * it, a request that has been delayed should
2892 * not be passed by new incoming requests
2894 rq->rq_flags |= RQF_STARTED;
2895 trace_block_rq_issue(q, rq);
2898 if (!q->boundary_rq || q->boundary_rq == rq) {
2899 q->end_sector = rq_end_sector(rq);
2900 q->boundary_rq = NULL;
2903 if (rq->rq_flags & RQF_DONTPREP)
2906 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2908 * make sure space for the drain appears we
2909 * know we can do this because max_hw_segments
2910 * has been adjusted to be one fewer than the
2913 rq->nr_phys_segments++;
2919 ret = q->prep_rq_fn(q, rq);
2920 if (ret == BLKPREP_OK) {
2922 } else if (ret == BLKPREP_DEFER) {
2924 * the request may have been (partially) prepped.
2925 * we need to keep this request in the front to
2926 * avoid resource deadlock. RQF_STARTED will
2927 * prevent other fs requests from passing this one.
2929 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2930 !(rq->rq_flags & RQF_DONTPREP)) {
2932 * remove the space for the drain we added
2933 * so that we don't add it again
2935 --rq->nr_phys_segments;
2940 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2941 rq->rq_flags |= RQF_QUIET;
2943 * Mark this request as started so we don't trigger
2944 * any debug logic in the end I/O path.
2946 blk_start_request(rq);
2947 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2948 BLK_STS_TARGET : BLK_STS_IOERR);
2950 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2957 EXPORT_SYMBOL(blk_peek_request);
2959 static void blk_dequeue_request(struct request *rq)
2961 struct request_queue *q = rq->q;
2963 BUG_ON(list_empty(&rq->queuelist));
2964 BUG_ON(ELV_ON_HASH(rq));
2966 list_del_init(&rq->queuelist);
2969 * the time frame between a request being removed from the lists
2970 * and to it is freed is accounted as io that is in progress at
2973 if (blk_account_rq(rq))
2974 q->in_flight[rq_is_sync(rq)]++;
2978 * blk_start_request - start request processing on the driver
2979 * @req: request to dequeue
2982 * Dequeue @req and start timeout timer on it. This hands off the
2983 * request to the driver.
2985 void blk_start_request(struct request *req)
2987 lockdep_assert_held(req->q->queue_lock);
2988 WARN_ON_ONCE(req->q->mq_ops);
2990 blk_dequeue_request(req);
2992 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2993 req->io_start_time_ns = ktime_get_ns();
2994 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2995 req->throtl_size = blk_rq_sectors(req);
2997 req->rq_flags |= RQF_STATS;
2998 rq_qos_issue(req->q, req);
3001 BUG_ON(blk_rq_is_complete(req));
3004 EXPORT_SYMBOL(blk_start_request);
3007 * blk_fetch_request - fetch a request from a request queue
3008 * @q: request queue to fetch a request from
3011 * Return the request at the top of @q. The request is started on
3012 * return and LLD can start processing it immediately.
3015 * Pointer to the request at the top of @q if available. Null
3018 struct request *blk_fetch_request(struct request_queue *q)
3022 lockdep_assert_held(q->queue_lock);
3023 WARN_ON_ONCE(q->mq_ops);
3025 rq = blk_peek_request(q);
3027 blk_start_request(rq);
3030 EXPORT_SYMBOL(blk_fetch_request);
3033 * Steal bios from a request and add them to a bio list.
3034 * The request must not have been partially completed before.
3036 void blk_steal_bios(struct bio_list *list, struct request *rq)
3040 list->tail->bi_next = rq->bio;
3042 list->head = rq->bio;
3043 list->tail = rq->biotail;
3051 EXPORT_SYMBOL_GPL(blk_steal_bios);
3054 * blk_update_request - Special helper function for request stacking drivers
3055 * @req: the request being processed
3056 * @error: block status code
3057 * @nr_bytes: number of bytes to complete @req
3060 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3061 * the request structure even if @req doesn't have leftover.
3062 * If @req has leftover, sets it up for the next range of segments.
3064 * This special helper function is only for request stacking drivers
3065 * (e.g. request-based dm) so that they can handle partial completion.
3066 * Actual device drivers should use blk_end_request instead.
3068 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3069 * %false return from this function.
3072 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
3073 * blk_rq_bytes() and in blk_update_request().
3076 * %false - this request doesn't have any more data
3077 * %true - this request has more data
3079 bool blk_update_request(struct request *req, blk_status_t error,
3080 unsigned int nr_bytes)
3084 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
3089 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3090 !(req->rq_flags & RQF_QUIET)))
3091 print_req_error(req, error);
3093 blk_account_io_completion(req, nr_bytes);
3097 struct bio *bio = req->bio;
3098 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3100 if (bio_bytes == bio->bi_iter.bi_size)
3101 req->bio = bio->bi_next;
3103 /* Completion has already been traced */
3104 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
3105 req_bio_endio(req, bio, bio_bytes, error);
3107 total_bytes += bio_bytes;
3108 nr_bytes -= bio_bytes;
3119 * Reset counters so that the request stacking driver
3120 * can find how many bytes remain in the request
3123 req->__data_len = 0;
3127 req->__data_len -= total_bytes;
3129 /* update sector only for requests with clear definition of sector */
3130 if (!blk_rq_is_passthrough(req))
3131 req->__sector += total_bytes >> 9;
3133 /* mixed attributes always follow the first bio */
3134 if (req->rq_flags & RQF_MIXED_MERGE) {
3135 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3136 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3139 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3141 * If total number of sectors is less than the first segment
3142 * size, something has gone terribly wrong.
3144 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3145 blk_dump_rq_flags(req, "request botched");
3146 req->__data_len = blk_rq_cur_bytes(req);
3149 /* recalculate the number of segments */
3150 blk_recalc_rq_segments(req);
3155 EXPORT_SYMBOL_GPL(blk_update_request);
3157 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3158 unsigned int nr_bytes,
3159 unsigned int bidi_bytes)
3161 if (blk_update_request(rq, error, nr_bytes))
3164 /* Bidi request must be completed as a whole */
3165 if (unlikely(blk_bidi_rq(rq)) &&
3166 blk_update_request(rq->next_rq, error, bidi_bytes))
3169 if (blk_queue_add_random(rq->q))
3170 add_disk_randomness(rq->rq_disk);
3176 * blk_unprep_request - unprepare a request
3179 * This function makes a request ready for complete resubmission (or
3180 * completion). It happens only after all error handling is complete,
3181 * so represents the appropriate moment to deallocate any resources
3182 * that were allocated to the request in the prep_rq_fn. The queue
3183 * lock is held when calling this.
3185 void blk_unprep_request(struct request *req)
3187 struct request_queue *q = req->q;
3189 req->rq_flags &= ~RQF_DONTPREP;
3190 if (q->unprep_rq_fn)
3191 q->unprep_rq_fn(q, req);
3193 EXPORT_SYMBOL_GPL(blk_unprep_request);
3195 void blk_finish_request(struct request *req, blk_status_t error)
3197 struct request_queue *q = req->q;
3198 u64 now = ktime_get_ns();
3200 lockdep_assert_held(req->q->queue_lock);
3201 WARN_ON_ONCE(q->mq_ops);
3203 if (req->rq_flags & RQF_STATS)
3204 blk_stat_add(req, now);
3206 if (req->rq_flags & RQF_QUEUED)
3207 blk_queue_end_tag(q, req);
3209 BUG_ON(blk_queued_rq(req));
3211 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3212 laptop_io_completion(req->q->backing_dev_info);
3214 blk_delete_timer(req);
3216 if (req->rq_flags & RQF_DONTPREP)
3217 blk_unprep_request(req);
3219 blk_account_io_done(req, now);
3222 rq_qos_done(q, req);
3223 req->end_io(req, error);
3225 if (blk_bidi_rq(req))
3226 __blk_put_request(req->next_rq->q, req->next_rq);
3228 __blk_put_request(q, req);
3231 EXPORT_SYMBOL(blk_finish_request);
3234 * blk_end_bidi_request - Complete a bidi request
3235 * @rq: the request to complete
3236 * @error: block status code
3237 * @nr_bytes: number of bytes to complete @rq
3238 * @bidi_bytes: number of bytes to complete @rq->next_rq
3241 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3242 * Drivers that supports bidi can safely call this member for any
3243 * type of request, bidi or uni. In the later case @bidi_bytes is
3247 * %false - we are done with this request
3248 * %true - still buffers pending for this request
3250 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3251 unsigned int nr_bytes, unsigned int bidi_bytes)
3253 struct request_queue *q = rq->q;
3254 unsigned long flags;
3256 WARN_ON_ONCE(q->mq_ops);
3258 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3261 spin_lock_irqsave(q->queue_lock, flags);
3262 blk_finish_request(rq, error);
3263 spin_unlock_irqrestore(q->queue_lock, flags);
3269 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3270 * @rq: the request to complete
3271 * @error: block status code
3272 * @nr_bytes: number of bytes to complete @rq
3273 * @bidi_bytes: number of bytes to complete @rq->next_rq
3276 * Identical to blk_end_bidi_request() except that queue lock is
3277 * assumed to be locked on entry and remains so on return.
3280 * %false - we are done with this request
3281 * %true - still buffers pending for this request
3283 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3284 unsigned int nr_bytes, unsigned int bidi_bytes)
3286 lockdep_assert_held(rq->q->queue_lock);
3287 WARN_ON_ONCE(rq->q->mq_ops);
3289 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3292 blk_finish_request(rq, error);
3298 * blk_end_request - Helper function for drivers to complete the request.
3299 * @rq: the request being processed
3300 * @error: block status code
3301 * @nr_bytes: number of bytes to complete
3304 * Ends I/O on a number of bytes attached to @rq.
3305 * If @rq has leftover, sets it up for the next range of segments.
3308 * %false - we are done with this request
3309 * %true - still buffers pending for this request
3311 bool blk_end_request(struct request *rq, blk_status_t error,
3312 unsigned int nr_bytes)
3314 WARN_ON_ONCE(rq->q->mq_ops);
3315 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3317 EXPORT_SYMBOL(blk_end_request);
3320 * blk_end_request_all - Helper function for drives to finish the request.
3321 * @rq: the request to finish
3322 * @error: block status code
3325 * Completely finish @rq.
3327 void blk_end_request_all(struct request *rq, blk_status_t error)
3330 unsigned int bidi_bytes = 0;
3332 if (unlikely(blk_bidi_rq(rq)))
3333 bidi_bytes = blk_rq_bytes(rq->next_rq);
3335 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3338 EXPORT_SYMBOL(blk_end_request_all);
3341 * __blk_end_request - Helper function for drivers to complete the request.
3342 * @rq: the request being processed
3343 * @error: block status code
3344 * @nr_bytes: number of bytes to complete
3347 * Must be called with queue lock held unlike blk_end_request().
3350 * %false - we are done with this request
3351 * %true - still buffers pending for this request
3353 bool __blk_end_request(struct request *rq, blk_status_t error,
3354 unsigned int nr_bytes)
3356 lockdep_assert_held(rq->q->queue_lock);
3357 WARN_ON_ONCE(rq->q->mq_ops);
3359 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3361 EXPORT_SYMBOL(__blk_end_request);
3364 * __blk_end_request_all - Helper function for drives to finish the request.
3365 * @rq: the request to finish
3366 * @error: block status code
3369 * Completely finish @rq. Must be called with queue lock held.
3371 void __blk_end_request_all(struct request *rq, blk_status_t error)
3374 unsigned int bidi_bytes = 0;
3376 lockdep_assert_held(rq->q->queue_lock);
3377 WARN_ON_ONCE(rq->q->mq_ops);
3379 if (unlikely(blk_bidi_rq(rq)))
3380 bidi_bytes = blk_rq_bytes(rq->next_rq);
3382 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3385 EXPORT_SYMBOL(__blk_end_request_all);
3388 * __blk_end_request_cur - Helper function to finish the current request chunk.
3389 * @rq: the request to finish the current chunk for
3390 * @error: block status code
3393 * Complete the current consecutively mapped chunk from @rq. Must
3394 * be called with queue lock held.
3397 * %false - we are done with this request
3398 * %true - still buffers pending for this request
3400 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3402 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3404 EXPORT_SYMBOL(__blk_end_request_cur);
3406 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3409 if (bio_has_data(bio))
3410 rq->nr_phys_segments = bio_phys_segments(q, bio);
3411 else if (bio_op(bio) == REQ_OP_DISCARD)
3412 rq->nr_phys_segments = 1;
3414 rq->__data_len = bio->bi_iter.bi_size;
3415 rq->bio = rq->biotail = bio;
3418 rq->rq_disk = bio->bi_disk;
3421 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3423 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3424 * @rq: the request to be flushed
3427 * Flush all pages in @rq.
3429 void rq_flush_dcache_pages(struct request *rq)
3431 struct req_iterator iter;
3432 struct bio_vec bvec;
3434 rq_for_each_segment(bvec, rq, iter)
3435 flush_dcache_page(bvec.bv_page);
3437 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3441 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3442 * @q : the queue of the device being checked
3445 * Check if underlying low-level drivers of a device are busy.
3446 * If the drivers want to export their busy state, they must set own
3447 * exporting function using blk_queue_lld_busy() first.
3449 * Basically, this function is used only by request stacking drivers
3450 * to stop dispatching requests to underlying devices when underlying
3451 * devices are busy. This behavior helps more I/O merging on the queue
3452 * of the request stacking driver and prevents I/O throughput regression
3453 * on burst I/O load.
3456 * 0 - Not busy (The request stacking driver should dispatch request)
3457 * 1 - Busy (The request stacking driver should stop dispatching request)
3459 int blk_lld_busy(struct request_queue *q)
3462 return q->lld_busy_fn(q);
3466 EXPORT_SYMBOL_GPL(blk_lld_busy);
3469 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3470 * @rq: the clone request to be cleaned up
3473 * Free all bios in @rq for a cloned request.
3475 void blk_rq_unprep_clone(struct request *rq)
3479 while ((bio = rq->bio) != NULL) {
3480 rq->bio = bio->bi_next;
3485 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3488 * Copy attributes of the original request to the clone request.
3489 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3491 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3493 dst->cpu = src->cpu;
3494 dst->__sector = blk_rq_pos(src);
3495 dst->__data_len = blk_rq_bytes(src);
3496 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
3497 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
3498 dst->special_vec = src->special_vec;
3500 dst->nr_phys_segments = src->nr_phys_segments;
3501 dst->ioprio = src->ioprio;
3502 dst->extra_len = src->extra_len;
3506 * blk_rq_prep_clone - Helper function to setup clone request
3507 * @rq: the request to be setup
3508 * @rq_src: original request to be cloned
3509 * @bs: bio_set that bios for clone are allocated from
3510 * @gfp_mask: memory allocation mask for bio
3511 * @bio_ctr: setup function to be called for each clone bio.
3512 * Returns %0 for success, non %0 for failure.
3513 * @data: private data to be passed to @bio_ctr
3516 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3517 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3518 * are not copied, and copying such parts is the caller's responsibility.
3519 * Also, pages which the original bios are pointing to are not copied
3520 * and the cloned bios just point same pages.
3521 * So cloned bios must be completed before original bios, which means
3522 * the caller must complete @rq before @rq_src.
3524 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3525 struct bio_set *bs, gfp_t gfp_mask,
3526 int (*bio_ctr)(struct bio *, struct bio *, void *),
3529 struct bio *bio, *bio_src;
3534 __rq_for_each_bio(bio_src, rq_src) {
3535 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3539 if (bio_ctr && bio_ctr(bio, bio_src, data))
3543 rq->biotail->bi_next = bio;
3546 rq->bio = rq->biotail = bio;
3549 __blk_rq_prep_clone(rq, rq_src);
3556 blk_rq_unprep_clone(rq);
3560 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3562 int kblockd_schedule_work(struct work_struct *work)
3564 return queue_work(kblockd_workqueue, work);
3566 EXPORT_SYMBOL(kblockd_schedule_work);
3568 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3570 return queue_work_on(cpu, kblockd_workqueue, work);
3572 EXPORT_SYMBOL(kblockd_schedule_work_on);
3574 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3575 unsigned long delay)
3577 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3579 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3582 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3583 * @plug: The &struct blk_plug that needs to be initialized
3586 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3587 * pending I/O should the task end up blocking between blk_start_plug() and
3588 * blk_finish_plug(). This is important from a performance perspective, but
3589 * also ensures that we don't deadlock. For instance, if the task is blocking
3590 * for a memory allocation, memory reclaim could end up wanting to free a
3591 * page belonging to that request that is currently residing in our private
3592 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3593 * this kind of deadlock.
3595 void blk_start_plug(struct blk_plug *plug)
3597 struct task_struct *tsk = current;
3600 * If this is a nested plug, don't actually assign it.
3605 INIT_LIST_HEAD(&plug->list);
3606 INIT_LIST_HEAD(&plug->mq_list);
3607 INIT_LIST_HEAD(&plug->cb_list);
3609 * Store ordering should not be needed here, since a potential
3610 * preempt will imply a full memory barrier
3614 EXPORT_SYMBOL(blk_start_plug);
3616 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3618 struct request *rqa = container_of(a, struct request, queuelist);
3619 struct request *rqb = container_of(b, struct request, queuelist);
3621 return !(rqa->q < rqb->q ||
3622 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3626 * If 'from_schedule' is true, then postpone the dispatch of requests
3627 * until a safe kblockd context. We due this to avoid accidental big
3628 * additional stack usage in driver dispatch, in places where the originally
3629 * plugger did not intend it.
3631 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3633 __releases(q->queue_lock)
3635 lockdep_assert_held(q->queue_lock);
3637 trace_block_unplug(q, depth, !from_schedule);
3640 blk_run_queue_async(q);
3643 spin_unlock_irq(q->queue_lock);
3646 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3648 LIST_HEAD(callbacks);
3650 while (!list_empty(&plug->cb_list)) {
3651 list_splice_init(&plug->cb_list, &callbacks);
3653 while (!list_empty(&callbacks)) {
3654 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3657 list_del(&cb->list);
3658 cb->callback(cb, from_schedule);
3663 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3666 struct blk_plug *plug = current->plug;
3667 struct blk_plug_cb *cb;
3672 list_for_each_entry(cb, &plug->cb_list, list)
3673 if (cb->callback == unplug && cb->data == data)
3676 /* Not currently on the callback list */
3677 BUG_ON(size < sizeof(*cb));
3678 cb = kzalloc(size, GFP_ATOMIC);
3681 cb->callback = unplug;
3682 list_add(&cb->list, &plug->cb_list);
3686 EXPORT_SYMBOL(blk_check_plugged);
3688 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3690 struct request_queue *q;
3695 flush_plug_callbacks(plug, from_schedule);
3697 if (!list_empty(&plug->mq_list))
3698 blk_mq_flush_plug_list(plug, from_schedule);
3700 if (list_empty(&plug->list))
3703 list_splice_init(&plug->list, &list);
3705 list_sort(NULL, &list, plug_rq_cmp);
3710 while (!list_empty(&list)) {
3711 rq = list_entry_rq(list.next);
3712 list_del_init(&rq->queuelist);
3716 * This drops the queue lock
3719 queue_unplugged(q, depth, from_schedule);
3722 spin_lock_irq(q->queue_lock);
3726 * Short-circuit if @q is dead
3728 if (unlikely(blk_queue_dying(q))) {
3729 __blk_end_request_all(rq, BLK_STS_IOERR);
3734 * rq is already accounted, so use raw insert
3736 if (op_is_flush(rq->cmd_flags))
3737 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3739 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3745 * This drops the queue lock
3748 queue_unplugged(q, depth, from_schedule);
3751 void blk_finish_plug(struct blk_plug *plug)
3753 if (plug != current->plug)
3755 blk_flush_plug_list(plug, false);
3757 current->plug = NULL;
3759 EXPORT_SYMBOL(blk_finish_plug);
3763 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3764 * @q: the queue of the device
3765 * @dev: the device the queue belongs to
3768 * Initialize runtime-PM-related fields for @q and start auto suspend for
3769 * @dev. Drivers that want to take advantage of request-based runtime PM
3770 * should call this function after @dev has been initialized, and its
3771 * request queue @q has been allocated, and runtime PM for it can not happen
3772 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3773 * cases, driver should call this function before any I/O has taken place.
3775 * This function takes care of setting up using auto suspend for the device,
3776 * the autosuspend delay is set to -1 to make runtime suspend impossible
3777 * until an updated value is either set by user or by driver. Drivers do
3778 * not need to touch other autosuspend settings.
3780 * The block layer runtime PM is request based, so only works for drivers
3781 * that use request as their IO unit instead of those directly use bio's.
3783 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3785 /* Don't enable runtime PM for blk-mq until it is ready */
3787 pm_runtime_disable(dev);
3792 q->rpm_status = RPM_ACTIVE;
3793 pm_runtime_set_autosuspend_delay(q->dev, -1);
3794 pm_runtime_use_autosuspend(q->dev);
3796 EXPORT_SYMBOL(blk_pm_runtime_init);
3799 * blk_pre_runtime_suspend - Pre runtime suspend check
3800 * @q: the queue of the device
3803 * This function will check if runtime suspend is allowed for the device
3804 * by examining if there are any requests pending in the queue. If there
3805 * are requests pending, the device can not be runtime suspended; otherwise,
3806 * the queue's status will be updated to SUSPENDING and the driver can
3807 * proceed to suspend the device.
3809 * For the not allowed case, we mark last busy for the device so that
3810 * runtime PM core will try to autosuspend it some time later.
3812 * This function should be called near the start of the device's
3813 * runtime_suspend callback.
3816 * 0 - OK to runtime suspend the device
3817 * -EBUSY - Device should not be runtime suspended
3819 int blk_pre_runtime_suspend(struct request_queue *q)
3826 spin_lock_irq(q->queue_lock);
3827 if (q->nr_pending) {
3829 pm_runtime_mark_last_busy(q->dev);
3831 q->rpm_status = RPM_SUSPENDING;
3833 spin_unlock_irq(q->queue_lock);
3836 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3839 * blk_post_runtime_suspend - Post runtime suspend processing
3840 * @q: the queue of the device
3841 * @err: return value of the device's runtime_suspend function
3844 * Update the queue's runtime status according to the return value of the
3845 * device's runtime suspend function and mark last busy for the device so
3846 * that PM core will try to auto suspend the device at a later time.
3848 * This function should be called near the end of the device's
3849 * runtime_suspend callback.
3851 void blk_post_runtime_suspend(struct request_queue *q, int err)
3856 spin_lock_irq(q->queue_lock);
3858 q->rpm_status = RPM_SUSPENDED;
3860 q->rpm_status = RPM_ACTIVE;
3861 pm_runtime_mark_last_busy(q->dev);
3863 spin_unlock_irq(q->queue_lock);
3865 EXPORT_SYMBOL(blk_post_runtime_suspend);
3868 * blk_pre_runtime_resume - Pre runtime resume processing
3869 * @q: the queue of the device
3872 * Update the queue's runtime status to RESUMING in preparation for the
3873 * runtime resume of the device.
3875 * This function should be called near the start of the device's
3876 * runtime_resume callback.
3878 void blk_pre_runtime_resume(struct request_queue *q)
3883 spin_lock_irq(q->queue_lock);
3884 q->rpm_status = RPM_RESUMING;
3885 spin_unlock_irq(q->queue_lock);
3887 EXPORT_SYMBOL(blk_pre_runtime_resume);
3890 * blk_post_runtime_resume - Post runtime resume processing
3891 * @q: the queue of the device
3892 * @err: return value of the device's runtime_resume function
3895 * Update the queue's runtime status according to the return value of the
3896 * device's runtime_resume function. If it is successfully resumed, process
3897 * the requests that are queued into the device's queue when it is resuming
3898 * and then mark last busy and initiate autosuspend for it.
3900 * This function should be called near the end of the device's
3901 * runtime_resume callback.
3903 void blk_post_runtime_resume(struct request_queue *q, int err)
3908 spin_lock_irq(q->queue_lock);
3910 q->rpm_status = RPM_ACTIVE;
3912 pm_runtime_mark_last_busy(q->dev);
3913 pm_request_autosuspend(q->dev);
3915 q->rpm_status = RPM_SUSPENDED;
3917 spin_unlock_irq(q->queue_lock);
3919 EXPORT_SYMBOL(blk_post_runtime_resume);
3922 * blk_set_runtime_active - Force runtime status of the queue to be active
3923 * @q: the queue of the device
3925 * If the device is left runtime suspended during system suspend the resume
3926 * hook typically resumes the device and corrects runtime status
3927 * accordingly. However, that does not affect the queue runtime PM status
3928 * which is still "suspended". This prevents processing requests from the
3931 * This function can be used in driver's resume hook to correct queue
3932 * runtime PM status and re-enable peeking requests from the queue. It
3933 * should be called before first request is added to the queue.
3935 void blk_set_runtime_active(struct request_queue *q)
3937 spin_lock_irq(q->queue_lock);
3938 q->rpm_status = RPM_ACTIVE;
3939 pm_runtime_mark_last_busy(q->dev);
3940 pm_request_autosuspend(q->dev);
3941 spin_unlock_irq(q->queue_lock);
3943 EXPORT_SYMBOL(blk_set_runtime_active);
3946 int __init blk_dev_init(void)
3948 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3949 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3950 FIELD_SIZEOF(struct request, cmd_flags));
3951 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3952 FIELD_SIZEOF(struct bio, bi_opf));
3954 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3955 kblockd_workqueue = alloc_workqueue("kblockd",
3956 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3957 if (!kblockd_workqueue)
3958 panic("Failed to create kblockd\n");
3960 request_cachep = kmem_cache_create("blkdev_requests",
3961 sizeof(struct request), 0, SLAB_PANIC, NULL);
3963 blk_requestq_cachep = kmem_cache_create("request_queue",
3964 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3966 #ifdef CONFIG_DEBUG_FS
3967 blk_debugfs_root = debugfs_create_dir("block", NULL);