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"
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 */
277 * XXX this code looks suspicious - it's not consistent with advancing
280 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
284 void blk_dump_rq_flags(struct request *rq, char *msg)
286 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
287 rq->rq_disk ? rq->rq_disk->disk_name : "?",
288 (unsigned long long) rq->cmd_flags);
290 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
291 (unsigned long long)blk_rq_pos(rq),
292 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
293 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
294 rq->bio, rq->biotail, blk_rq_bytes(rq));
296 EXPORT_SYMBOL(blk_dump_rq_flags);
298 static void blk_delay_work(struct work_struct *work)
300 struct request_queue *q;
302 q = container_of(work, struct request_queue, delay_work.work);
303 spin_lock_irq(q->queue_lock);
305 spin_unlock_irq(q->queue_lock);
309 * blk_delay_queue - restart queueing after defined interval
310 * @q: The &struct request_queue in question
311 * @msecs: Delay in msecs
314 * Sometimes queueing needs to be postponed for a little while, to allow
315 * resources to come back. This function will make sure that queueing is
316 * restarted around the specified time.
318 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
320 lockdep_assert_held(q->queue_lock);
321 WARN_ON_ONCE(q->mq_ops);
323 if (likely(!blk_queue_dead(q)))
324 queue_delayed_work(kblockd_workqueue, &q->delay_work,
325 msecs_to_jiffies(msecs));
327 EXPORT_SYMBOL(blk_delay_queue);
330 * blk_start_queue_async - asynchronously restart a previously stopped queue
331 * @q: The &struct request_queue in question
334 * blk_start_queue_async() will clear the stop flag on the queue, and
335 * ensure that the request_fn for the queue is run from an async
338 void blk_start_queue_async(struct request_queue *q)
340 lockdep_assert_held(q->queue_lock);
341 WARN_ON_ONCE(q->mq_ops);
343 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
344 blk_run_queue_async(q);
346 EXPORT_SYMBOL(blk_start_queue_async);
349 * blk_start_queue - restart a previously stopped queue
350 * @q: The &struct request_queue in question
353 * blk_start_queue() will clear the stop flag on the queue, and call
354 * the request_fn for the queue if it was in a stopped state when
355 * entered. Also see blk_stop_queue().
357 void blk_start_queue(struct request_queue *q)
359 lockdep_assert_held(q->queue_lock);
360 WARN_ON_ONCE(q->mq_ops);
362 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
365 EXPORT_SYMBOL(blk_start_queue);
368 * blk_stop_queue - stop a queue
369 * @q: The &struct request_queue in question
372 * The Linux block layer assumes that a block driver will consume all
373 * entries on the request queue when the request_fn strategy is called.
374 * Often this will not happen, because of hardware limitations (queue
375 * depth settings). If a device driver gets a 'queue full' response,
376 * or if it simply chooses not to queue more I/O at one point, it can
377 * call this function to prevent the request_fn from being called until
378 * the driver has signalled it's ready to go again. This happens by calling
379 * blk_start_queue() to restart queue operations.
381 void blk_stop_queue(struct request_queue *q)
383 lockdep_assert_held(q->queue_lock);
384 WARN_ON_ONCE(q->mq_ops);
386 cancel_delayed_work(&q->delay_work);
387 queue_flag_set(QUEUE_FLAG_STOPPED, q);
389 EXPORT_SYMBOL(blk_stop_queue);
392 * blk_sync_queue - cancel any pending callbacks on a queue
396 * The block layer may perform asynchronous callback activity
397 * on a queue, such as calling the unplug function after a timeout.
398 * A block device may call blk_sync_queue to ensure that any
399 * such activity is cancelled, thus allowing it to release resources
400 * that the callbacks might use. The caller must already have made sure
401 * that its ->make_request_fn will not re-add plugging prior to calling
404 * This function does not cancel any asynchronous activity arising
405 * out of elevator or throttling code. That would require elevator_exit()
406 * and blkcg_exit_queue() to be called with queue lock initialized.
409 void blk_sync_queue(struct request_queue *q)
411 del_timer_sync(&q->timeout);
412 cancel_work_sync(&q->timeout_work);
415 struct blk_mq_hw_ctx *hctx;
418 cancel_delayed_work_sync(&q->requeue_work);
419 queue_for_each_hw_ctx(q, hctx, i)
420 cancel_delayed_work_sync(&hctx->run_work);
422 cancel_delayed_work_sync(&q->delay_work);
425 EXPORT_SYMBOL(blk_sync_queue);
428 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
429 * @q: request queue pointer
431 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
432 * set and 1 if the flag was already set.
434 int blk_set_preempt_only(struct request_queue *q)
436 return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
438 EXPORT_SYMBOL_GPL(blk_set_preempt_only);
440 void blk_clear_preempt_only(struct request_queue *q)
442 blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
443 wake_up_all(&q->mq_freeze_wq);
445 EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
448 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
449 * @q: The queue to run
452 * Invoke request handling on a queue if there are any pending requests.
453 * May be used to restart request handling after a request has completed.
454 * This variant runs the queue whether or not the queue has been
455 * stopped. Must be called with the queue lock held and interrupts
456 * disabled. See also @blk_run_queue.
458 inline void __blk_run_queue_uncond(struct request_queue *q)
460 lockdep_assert_held(q->queue_lock);
461 WARN_ON_ONCE(q->mq_ops);
463 if (unlikely(blk_queue_dead(q)))
467 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
468 * the queue lock internally. As a result multiple threads may be
469 * running such a request function concurrently. Keep track of the
470 * number of active request_fn invocations such that blk_drain_queue()
471 * can wait until all these request_fn calls have finished.
473 q->request_fn_active++;
475 q->request_fn_active--;
477 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
480 * __blk_run_queue - run a single device queue
481 * @q: The queue to run
484 * See @blk_run_queue.
486 void __blk_run_queue(struct request_queue *q)
488 lockdep_assert_held(q->queue_lock);
489 WARN_ON_ONCE(q->mq_ops);
491 if (unlikely(blk_queue_stopped(q)))
494 __blk_run_queue_uncond(q);
496 EXPORT_SYMBOL(__blk_run_queue);
499 * blk_run_queue_async - run a single device queue in workqueue context
500 * @q: The queue to run
503 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
507 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
508 * has canceled q->delay_work, callers must hold the queue lock to avoid
509 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
511 void blk_run_queue_async(struct request_queue *q)
513 lockdep_assert_held(q->queue_lock);
514 WARN_ON_ONCE(q->mq_ops);
516 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
517 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
519 EXPORT_SYMBOL(blk_run_queue_async);
522 * blk_run_queue - run a single device queue
523 * @q: The queue to run
526 * Invoke request handling on this queue, if it has pending work to do.
527 * May be used to restart queueing when a request has completed.
529 void blk_run_queue(struct request_queue *q)
533 WARN_ON_ONCE(q->mq_ops);
535 spin_lock_irqsave(q->queue_lock, flags);
537 spin_unlock_irqrestore(q->queue_lock, flags);
539 EXPORT_SYMBOL(blk_run_queue);
541 void blk_put_queue(struct request_queue *q)
543 kobject_put(&q->kobj);
545 EXPORT_SYMBOL(blk_put_queue);
548 * __blk_drain_queue - drain requests from request_queue
550 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
552 * Drain requests from @q. If @drain_all is set, all requests are drained.
553 * If not, only ELVPRIV requests are drained. The caller is responsible
554 * for ensuring that no new requests which need to be drained are queued.
556 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
557 __releases(q->queue_lock)
558 __acquires(q->queue_lock)
562 lockdep_assert_held(q->queue_lock);
563 WARN_ON_ONCE(q->mq_ops);
569 * The caller might be trying to drain @q before its
570 * elevator is initialized.
573 elv_drain_elevator(q);
575 blkcg_drain_queue(q);
578 * This function might be called on a queue which failed
579 * driver init after queue creation or is not yet fully
580 * active yet. Some drivers (e.g. fd and loop) get unhappy
581 * in such cases. Kick queue iff dispatch queue has
582 * something on it and @q has request_fn set.
584 if (!list_empty(&q->queue_head) && q->request_fn)
587 drain |= q->nr_rqs_elvpriv;
588 drain |= q->request_fn_active;
591 * Unfortunately, requests are queued at and tracked from
592 * multiple places and there's no single counter which can
593 * be drained. Check all the queues and counters.
596 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
597 drain |= !list_empty(&q->queue_head);
598 for (i = 0; i < 2; i++) {
599 drain |= q->nr_rqs[i];
600 drain |= q->in_flight[i];
602 drain |= !list_empty(&fq->flush_queue[i]);
609 spin_unlock_irq(q->queue_lock);
613 spin_lock_irq(q->queue_lock);
617 * With queue marked dead, any woken up waiter will fail the
618 * allocation path, so the wakeup chaining is lost and we're
619 * left with hung waiters. We need to wake up those waiters.
622 struct request_list *rl;
624 blk_queue_for_each_rl(rl, q)
625 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
626 wake_up_all(&rl->wait[i]);
630 void blk_drain_queue(struct request_queue *q)
632 spin_lock_irq(q->queue_lock);
633 __blk_drain_queue(q, true);
634 spin_unlock_irq(q->queue_lock);
638 * blk_queue_bypass_start - enter queue bypass mode
639 * @q: queue of interest
641 * In bypass mode, only the dispatch FIFO queue of @q is used. This
642 * function makes @q enter bypass mode and drains all requests which were
643 * throttled or issued before. On return, it's guaranteed that no request
644 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
645 * inside queue or RCU read lock.
647 void blk_queue_bypass_start(struct request_queue *q)
649 WARN_ON_ONCE(q->mq_ops);
651 spin_lock_irq(q->queue_lock);
653 queue_flag_set(QUEUE_FLAG_BYPASS, q);
654 spin_unlock_irq(q->queue_lock);
657 * Queues start drained. Skip actual draining till init is
658 * complete. This avoids lenghty delays during queue init which
659 * can happen many times during boot.
661 if (blk_queue_init_done(q)) {
662 spin_lock_irq(q->queue_lock);
663 __blk_drain_queue(q, false);
664 spin_unlock_irq(q->queue_lock);
666 /* ensure blk_queue_bypass() is %true inside RCU read lock */
670 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
673 * blk_queue_bypass_end - leave queue bypass mode
674 * @q: queue of interest
676 * Leave bypass mode and restore the normal queueing behavior.
678 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
679 * this function is called for both blk-sq and blk-mq queues.
681 void blk_queue_bypass_end(struct request_queue *q)
683 spin_lock_irq(q->queue_lock);
684 if (!--q->bypass_depth)
685 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
686 WARN_ON_ONCE(q->bypass_depth < 0);
687 spin_unlock_irq(q->queue_lock);
689 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
691 void blk_set_queue_dying(struct request_queue *q)
693 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
696 * When queue DYING flag is set, we need to block new req
697 * entering queue, so we call blk_freeze_queue_start() to
698 * prevent I/O from crossing blk_queue_enter().
700 blk_freeze_queue_start(q);
703 blk_mq_wake_waiters(q);
705 struct request_list *rl;
707 spin_lock_irq(q->queue_lock);
708 blk_queue_for_each_rl(rl, q) {
710 wake_up_all(&rl->wait[BLK_RW_SYNC]);
711 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
714 spin_unlock_irq(q->queue_lock);
717 /* Make blk_queue_enter() reexamine the DYING flag. */
718 wake_up_all(&q->mq_freeze_wq);
720 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
723 * blk_cleanup_queue - shutdown a request queue
724 * @q: request queue to shutdown
726 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
727 * put it. All future requests will be failed immediately with -ENODEV.
729 void blk_cleanup_queue(struct request_queue *q)
731 spinlock_t *lock = q->queue_lock;
733 /* mark @q DYING, no new request or merges will be allowed afterwards */
734 mutex_lock(&q->sysfs_lock);
735 blk_set_queue_dying(q);
739 * A dying queue is permanently in bypass mode till released. Note
740 * that, unlike blk_queue_bypass_start(), we aren't performing
741 * synchronize_rcu() after entering bypass mode to avoid the delay
742 * as some drivers create and destroy a lot of queues while
743 * probing. This is still safe because blk_release_queue() will be
744 * called only after the queue refcnt drops to zero and nothing,
745 * RCU or not, would be traversing the queue by then.
748 queue_flag_set(QUEUE_FLAG_BYPASS, q);
750 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
751 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
752 queue_flag_set(QUEUE_FLAG_DYING, q);
753 spin_unlock_irq(lock);
754 mutex_unlock(&q->sysfs_lock);
757 * Drain all requests queued before DYING marking. Set DEAD flag to
758 * prevent that q->request_fn() gets invoked after draining finished.
762 queue_flag_set(QUEUE_FLAG_DEAD, q);
763 spin_unlock_irq(lock);
766 * make sure all in-progress dispatch are completed because
767 * blk_freeze_queue() can only complete all requests, and
768 * dispatch may still be in-progress since we dispatch requests
769 * from more than one contexts
772 blk_mq_quiesce_queue(q);
774 /* for synchronous bio-based driver finish in-flight integrity i/o */
775 blk_flush_integrity();
777 /* @q won't process any more request, flush async actions */
778 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
782 * I/O scheduler exit is only safe after the sysfs scheduler attribute
785 WARN_ON_ONCE(q->kobj.state_in_sysfs);
788 * Since the I/O scheduler exit code may access cgroup information,
789 * perform I/O scheduler exit before disassociating from the block
794 elevator_exit(q, q->elevator);
799 * Remove all references to @q from the block cgroup controller before
800 * restoring @q->queue_lock to avoid that restoring this pointer causes
801 * e.g. blkcg_print_blkgs() to crash.
806 * Since the cgroup code may dereference the @q->backing_dev_info
807 * pointer, only decrease its reference count after having removed the
808 * association with the block cgroup controller.
810 bdi_put(q->backing_dev_info);
813 blk_mq_free_queue(q);
814 percpu_ref_exit(&q->q_usage_counter);
817 if (q->queue_lock != &q->__queue_lock)
818 q->queue_lock = &q->__queue_lock;
819 spin_unlock_irq(lock);
821 /* @q is and will stay empty, shutdown and put */
824 EXPORT_SYMBOL(blk_cleanup_queue);
826 /* Allocate memory local to the request queue */
827 static void *alloc_request_simple(gfp_t gfp_mask, void *data)
829 struct request_queue *q = data;
831 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
834 static void free_request_simple(void *element, void *data)
836 kmem_cache_free(request_cachep, element);
839 static void *alloc_request_size(gfp_t gfp_mask, void *data)
841 struct request_queue *q = data;
844 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
846 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
853 static void free_request_size(void *element, void *data)
855 struct request_queue *q = data;
858 q->exit_rq_fn(q, element);
862 int blk_init_rl(struct request_list *rl, struct request_queue *q,
865 if (unlikely(rl->rq_pool) || q->mq_ops)
869 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
870 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
871 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
872 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
875 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
876 alloc_request_size, free_request_size,
877 q, gfp_mask, q->node);
879 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
880 alloc_request_simple, free_request_simple,
881 q, gfp_mask, q->node);
886 if (rl != &q->root_rl)
887 WARN_ON_ONCE(!blk_get_queue(q));
892 void blk_exit_rl(struct request_queue *q, struct request_list *rl)
895 mempool_destroy(rl->rq_pool);
896 if (rl != &q->root_rl)
901 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
903 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL);
905 EXPORT_SYMBOL(blk_alloc_queue);
908 * blk_queue_enter() - try to increase q->q_usage_counter
909 * @q: request queue pointer
910 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
912 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
914 const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
917 bool success = false;
920 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
922 * The code that sets the PREEMPT_ONLY flag is
923 * responsible for ensuring that that flag is globally
924 * visible before the queue is unfrozen.
926 if (preempt || !blk_queue_preempt_only(q)) {
929 percpu_ref_put(&q->q_usage_counter);
937 if (flags & BLK_MQ_REQ_NOWAIT)
941 * read pair of barrier in blk_freeze_queue_start(),
942 * we need to order reading __PERCPU_REF_DEAD flag of
943 * .q_usage_counter and reading .mq_freeze_depth or
944 * queue dying flag, otherwise the following wait may
945 * never return if the two reads are reordered.
949 wait_event(q->mq_freeze_wq,
950 (atomic_read(&q->mq_freeze_depth) == 0 &&
951 (preempt || !blk_queue_preempt_only(q))) ||
953 if (blk_queue_dying(q))
958 void blk_queue_exit(struct request_queue *q)
960 percpu_ref_put(&q->q_usage_counter);
963 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
965 struct request_queue *q =
966 container_of(ref, struct request_queue, q_usage_counter);
968 wake_up_all(&q->mq_freeze_wq);
971 static void blk_rq_timed_out_timer(struct timer_list *t)
973 struct request_queue *q = from_timer(q, t, timeout);
975 kblockd_schedule_work(&q->timeout_work);
979 * blk_alloc_queue_node - allocate a request queue
980 * @gfp_mask: memory allocation flags
981 * @node_id: NUMA node to allocate memory from
982 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
983 * serialize calls to the legacy .request_fn() callback. Ignored for
984 * blk-mq request queues.
986 * Note: pass the queue lock as the third argument to this function instead of
987 * setting the queue lock pointer explicitly to avoid triggering a sporadic
988 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
989 * the queue lock pointer must be set before blkcg_init_queue() is called.
991 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
994 struct request_queue *q;
997 q = kmem_cache_alloc_node(blk_requestq_cachep,
998 gfp_mask | __GFP_ZERO, node_id);
1002 INIT_LIST_HEAD(&q->queue_head);
1003 q->last_merge = NULL;
1005 q->boundary_rq = NULL;
1007 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
1011 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1015 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1016 if (!q->backing_dev_info)
1019 q->stats = blk_alloc_queue_stats();
1023 q->backing_dev_info->ra_pages =
1024 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1025 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1026 q->backing_dev_info->name = "block";
1029 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1030 laptop_mode_timer_fn, 0);
1031 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
1032 INIT_WORK(&q->timeout_work, NULL);
1033 INIT_LIST_HEAD(&q->queue_head);
1034 INIT_LIST_HEAD(&q->timeout_list);
1035 INIT_LIST_HEAD(&q->icq_list);
1036 #ifdef CONFIG_BLK_CGROUP
1037 INIT_LIST_HEAD(&q->blkg_list);
1039 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
1041 kobject_init(&q->kobj, &blk_queue_ktype);
1043 #ifdef CONFIG_BLK_DEV_IO_TRACE
1044 mutex_init(&q->blk_trace_mutex);
1046 mutex_init(&q->sysfs_lock);
1047 spin_lock_init(&q->__queue_lock);
1050 q->queue_lock = lock ? : &q->__queue_lock;
1053 * A queue starts its life with bypass turned on to avoid
1054 * unnecessary bypass on/off overhead and nasty surprises during
1055 * init. The initial bypass will be finished when the queue is
1056 * registered by blk_register_queue().
1058 q->bypass_depth = 1;
1059 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
1061 init_waitqueue_head(&q->mq_freeze_wq);
1064 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1065 * See blk_register_queue() for details.
1067 if (percpu_ref_init(&q->q_usage_counter,
1068 blk_queue_usage_counter_release,
1069 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
1072 if (blkcg_init_queue(q))
1078 percpu_ref_exit(&q->q_usage_counter);
1080 blk_free_queue_stats(q->stats);
1082 bdi_put(q->backing_dev_info);
1084 bioset_exit(&q->bio_split);
1086 ida_simple_remove(&blk_queue_ida, q->id);
1088 kmem_cache_free(blk_requestq_cachep, q);
1091 EXPORT_SYMBOL(blk_alloc_queue_node);
1094 * blk_init_queue - prepare a request queue for use with a block device
1095 * @rfn: The function to be called to process requests that have been
1096 * placed on the queue.
1097 * @lock: Request queue spin lock
1100 * If a block device wishes to use the standard request handling procedures,
1101 * which sorts requests and coalesces adjacent requests, then it must
1102 * call blk_init_queue(). The function @rfn will be called when there
1103 * are requests on the queue that need to be processed. If the device
1104 * supports plugging, then @rfn may not be called immediately when requests
1105 * are available on the queue, but may be called at some time later instead.
1106 * Plugged queues are generally unplugged when a buffer belonging to one
1107 * of the requests on the queue is needed, or due to memory pressure.
1109 * @rfn is not required, or even expected, to remove all requests off the
1110 * queue, but only as many as it can handle at a time. If it does leave
1111 * requests on the queue, it is responsible for arranging that the requests
1112 * get dealt with eventually.
1114 * The queue spin lock must be held while manipulating the requests on the
1115 * request queue; this lock will be taken also from interrupt context, so irq
1116 * disabling is needed for it.
1118 * Function returns a pointer to the initialized request queue, or %NULL if
1119 * it didn't succeed.
1122 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1123 * when the block device is deactivated (such as at module unload).
1126 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1128 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1130 EXPORT_SYMBOL(blk_init_queue);
1132 struct request_queue *
1133 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1135 struct request_queue *q;
1137 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
1141 q->request_fn = rfn;
1142 if (blk_init_allocated_queue(q) < 0) {
1143 blk_cleanup_queue(q);
1149 EXPORT_SYMBOL(blk_init_queue_node);
1151 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1154 int blk_init_allocated_queue(struct request_queue *q)
1156 WARN_ON_ONCE(q->mq_ops);
1158 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
1162 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1163 goto out_free_flush_queue;
1165 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1166 goto out_exit_flush_rq;
1168 INIT_WORK(&q->timeout_work, blk_timeout_work);
1169 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1172 * This also sets hw/phys segments, boundary and size
1174 blk_queue_make_request(q, blk_queue_bio);
1176 q->sg_reserved_size = INT_MAX;
1178 /* Protect q->elevator from elevator_change */
1179 mutex_lock(&q->sysfs_lock);
1182 if (elevator_init(q)) {
1183 mutex_unlock(&q->sysfs_lock);
1184 goto out_exit_flush_rq;
1187 mutex_unlock(&q->sysfs_lock);
1192 q->exit_rq_fn(q, q->fq->flush_rq);
1193 out_free_flush_queue:
1194 blk_free_flush_queue(q->fq);
1197 EXPORT_SYMBOL(blk_init_allocated_queue);
1199 bool blk_get_queue(struct request_queue *q)
1201 if (likely(!blk_queue_dying(q))) {
1208 EXPORT_SYMBOL(blk_get_queue);
1210 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1212 if (rq->rq_flags & RQF_ELVPRIV) {
1213 elv_put_request(rl->q, rq);
1215 put_io_context(rq->elv.icq->ioc);
1218 mempool_free(rq, rl->rq_pool);
1222 * ioc_batching returns true if the ioc is a valid batching request and
1223 * should be given priority access to a request.
1225 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1231 * Make sure the process is able to allocate at least 1 request
1232 * even if the batch times out, otherwise we could theoretically
1235 return ioc->nr_batch_requests == q->nr_batching ||
1236 (ioc->nr_batch_requests > 0
1237 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1241 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1242 * will cause the process to be a "batcher" on all queues in the system. This
1243 * is the behaviour we want though - once it gets a wakeup it should be given
1246 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1248 if (!ioc || ioc_batching(q, ioc))
1251 ioc->nr_batch_requests = q->nr_batching;
1252 ioc->last_waited = jiffies;
1255 static void __freed_request(struct request_list *rl, int sync)
1257 struct request_queue *q = rl->q;
1259 if (rl->count[sync] < queue_congestion_off_threshold(q))
1260 blk_clear_congested(rl, sync);
1262 if (rl->count[sync] + 1 <= q->nr_requests) {
1263 if (waitqueue_active(&rl->wait[sync]))
1264 wake_up(&rl->wait[sync]);
1266 blk_clear_rl_full(rl, sync);
1271 * A request has just been released. Account for it, update the full and
1272 * congestion status, wake up any waiters. Called under q->queue_lock.
1274 static void freed_request(struct request_list *rl, bool sync,
1275 req_flags_t rq_flags)
1277 struct request_queue *q = rl->q;
1281 if (rq_flags & RQF_ELVPRIV)
1282 q->nr_rqs_elvpriv--;
1284 __freed_request(rl, sync);
1286 if (unlikely(rl->starved[sync ^ 1]))
1287 __freed_request(rl, sync ^ 1);
1290 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1292 struct request_list *rl;
1293 int on_thresh, off_thresh;
1295 WARN_ON_ONCE(q->mq_ops);
1297 spin_lock_irq(q->queue_lock);
1298 q->nr_requests = nr;
1299 blk_queue_congestion_threshold(q);
1300 on_thresh = queue_congestion_on_threshold(q);
1301 off_thresh = queue_congestion_off_threshold(q);
1303 blk_queue_for_each_rl(rl, q) {
1304 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1305 blk_set_congested(rl, BLK_RW_SYNC);
1306 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1307 blk_clear_congested(rl, BLK_RW_SYNC);
1309 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1310 blk_set_congested(rl, BLK_RW_ASYNC);
1311 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1312 blk_clear_congested(rl, BLK_RW_ASYNC);
1314 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1315 blk_set_rl_full(rl, BLK_RW_SYNC);
1317 blk_clear_rl_full(rl, BLK_RW_SYNC);
1318 wake_up(&rl->wait[BLK_RW_SYNC]);
1321 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1322 blk_set_rl_full(rl, BLK_RW_ASYNC);
1324 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1325 wake_up(&rl->wait[BLK_RW_ASYNC]);
1329 spin_unlock_irq(q->queue_lock);
1334 * __get_request - get a free request
1335 * @rl: request list to allocate from
1336 * @op: operation and flags
1337 * @bio: bio to allocate request for (can be %NULL)
1338 * @flags: BLQ_MQ_REQ_* flags
1339 * @gfp_mask: allocator flags
1341 * Get a free request from @q. This function may fail under memory
1342 * pressure or if @q is dead.
1344 * Must be called with @q->queue_lock held and,
1345 * Returns ERR_PTR on failure, with @q->queue_lock held.
1346 * Returns request pointer on success, with @q->queue_lock *not held*.
1348 static struct request *__get_request(struct request_list *rl, unsigned int op,
1349 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp_mask)
1351 struct request_queue *q = rl->q;
1353 struct elevator_type *et = q->elevator->type;
1354 struct io_context *ioc = rq_ioc(bio);
1355 struct io_cq *icq = NULL;
1356 const bool is_sync = op_is_sync(op);
1358 req_flags_t rq_flags = RQF_ALLOCED;
1360 lockdep_assert_held(q->queue_lock);
1362 if (unlikely(blk_queue_dying(q)))
1363 return ERR_PTR(-ENODEV);
1365 may_queue = elv_may_queue(q, op);
1366 if (may_queue == ELV_MQUEUE_NO)
1369 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1370 if (rl->count[is_sync]+1 >= q->nr_requests) {
1372 * The queue will fill after this allocation, so set
1373 * it as full, and mark this process as "batching".
1374 * This process will be allowed to complete a batch of
1375 * requests, others will be blocked.
1377 if (!blk_rl_full(rl, is_sync)) {
1378 ioc_set_batching(q, ioc);
1379 blk_set_rl_full(rl, is_sync);
1381 if (may_queue != ELV_MQUEUE_MUST
1382 && !ioc_batching(q, ioc)) {
1384 * The queue is full and the allocating
1385 * process is not a "batcher", and not
1386 * exempted by the IO scheduler
1388 return ERR_PTR(-ENOMEM);
1392 blk_set_congested(rl, is_sync);
1396 * Only allow batching queuers to allocate up to 50% over the defined
1397 * limit of requests, otherwise we could have thousands of requests
1398 * allocated with any setting of ->nr_requests
1400 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1401 return ERR_PTR(-ENOMEM);
1403 q->nr_rqs[is_sync]++;
1404 rl->count[is_sync]++;
1405 rl->starved[is_sync] = 0;
1408 * Decide whether the new request will be managed by elevator. If
1409 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1410 * prevent the current elevator from being destroyed until the new
1411 * request is freed. This guarantees icq's won't be destroyed and
1412 * makes creating new ones safe.
1414 * Flush requests do not use the elevator so skip initialization.
1415 * This allows a request to share the flush and elevator data.
1417 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1418 * it will be created after releasing queue_lock.
1420 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1421 rq_flags |= RQF_ELVPRIV;
1422 q->nr_rqs_elvpriv++;
1423 if (et->icq_cache && ioc)
1424 icq = ioc_lookup_icq(ioc, q);
1427 if (blk_queue_io_stat(q))
1428 rq_flags |= RQF_IO_STAT;
1429 spin_unlock_irq(q->queue_lock);
1431 /* allocate and init request */
1432 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1437 blk_rq_set_rl(rq, rl);
1439 rq->rq_flags = rq_flags;
1440 if (flags & BLK_MQ_REQ_PREEMPT)
1441 rq->rq_flags |= RQF_PREEMPT;
1444 if (rq_flags & RQF_ELVPRIV) {
1445 if (unlikely(et->icq_cache && !icq)) {
1447 icq = ioc_create_icq(ioc, q, gfp_mask);
1453 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1456 /* @rq->elv.icq holds io_context until @rq is freed */
1458 get_io_context(icq->ioc);
1462 * ioc may be NULL here, and ioc_batching will be false. That's
1463 * OK, if the queue is under the request limit then requests need
1464 * not count toward the nr_batch_requests limit. There will always
1465 * be some limit enforced by BLK_BATCH_TIME.
1467 if (ioc_batching(q, ioc))
1468 ioc->nr_batch_requests--;
1470 trace_block_getrq(q, bio, op);
1475 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1476 * and may fail indefinitely under memory pressure and thus
1477 * shouldn't stall IO. Treat this request as !elvpriv. This will
1478 * disturb iosched and blkcg but weird is bettern than dead.
1480 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1481 __func__, dev_name(q->backing_dev_info->dev));
1483 rq->rq_flags &= ~RQF_ELVPRIV;
1486 spin_lock_irq(q->queue_lock);
1487 q->nr_rqs_elvpriv--;
1488 spin_unlock_irq(q->queue_lock);
1493 * Allocation failed presumably due to memory. Undo anything we
1494 * might have messed up.
1496 * Allocating task should really be put onto the front of the wait
1497 * queue, but this is pretty rare.
1499 spin_lock_irq(q->queue_lock);
1500 freed_request(rl, is_sync, rq_flags);
1503 * in the very unlikely event that allocation failed and no
1504 * requests for this direction was pending, mark us starved so that
1505 * freeing of a request in the other direction will notice
1506 * us. another possible fix would be to split the rq mempool into
1510 if (unlikely(rl->count[is_sync] == 0))
1511 rl->starved[is_sync] = 1;
1512 return ERR_PTR(-ENOMEM);
1516 * get_request - get a free request
1517 * @q: request_queue to allocate request from
1518 * @op: operation and flags
1519 * @bio: bio to allocate request for (can be %NULL)
1520 * @flags: BLK_MQ_REQ_* flags.
1521 * @gfp: allocator flags
1523 * Get a free request from @q. If %BLK_MQ_REQ_NOWAIT is set in @flags,
1524 * this function keeps retrying under memory pressure and fails iff @q is dead.
1526 * Must be called with @q->queue_lock held and,
1527 * Returns ERR_PTR on failure, with @q->queue_lock held.
1528 * Returns request pointer on success, with @q->queue_lock *not held*.
1530 static struct request *get_request(struct request_queue *q, unsigned int op,
1531 struct bio *bio, blk_mq_req_flags_t flags, gfp_t gfp)
1533 const bool is_sync = op_is_sync(op);
1535 struct request_list *rl;
1538 lockdep_assert_held(q->queue_lock);
1539 WARN_ON_ONCE(q->mq_ops);
1541 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1543 rq = __get_request(rl, op, bio, flags, gfp);
1547 if (op & REQ_NOWAIT) {
1549 return ERR_PTR(-EAGAIN);
1552 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1557 /* wait on @rl and retry */
1558 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1559 TASK_UNINTERRUPTIBLE);
1561 trace_block_sleeprq(q, bio, op);
1563 spin_unlock_irq(q->queue_lock);
1567 * After sleeping, we become a "batching" process and will be able
1568 * to allocate at least one request, and up to a big batch of them
1569 * for a small period time. See ioc_batching, ioc_set_batching
1571 ioc_set_batching(q, current->io_context);
1573 spin_lock_irq(q->queue_lock);
1574 finish_wait(&rl->wait[is_sync], &wait);
1579 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1580 static struct request *blk_old_get_request(struct request_queue *q,
1581 unsigned int op, blk_mq_req_flags_t flags)
1584 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : GFP_NOIO;
1587 WARN_ON_ONCE(q->mq_ops);
1589 /* create ioc upfront */
1590 create_io_context(gfp_mask, q->node);
1592 ret = blk_queue_enter(q, flags);
1594 return ERR_PTR(ret);
1595 spin_lock_irq(q->queue_lock);
1596 rq = get_request(q, op, NULL, flags, gfp_mask);
1598 spin_unlock_irq(q->queue_lock);
1603 /* q->queue_lock is unlocked at this point */
1605 rq->__sector = (sector_t) -1;
1606 rq->bio = rq->biotail = NULL;
1611 * blk_get_request - allocate a request
1612 * @q: request queue to allocate a request for
1613 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1614 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1616 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1617 blk_mq_req_flags_t flags)
1619 struct request *req;
1621 WARN_ON_ONCE(op & REQ_NOWAIT);
1622 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1625 req = blk_mq_alloc_request(q, op, flags);
1626 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1627 q->mq_ops->initialize_rq_fn(req);
1629 req = blk_old_get_request(q, op, flags);
1630 if (!IS_ERR(req) && q->initialize_rq_fn)
1631 q->initialize_rq_fn(req);
1636 EXPORT_SYMBOL(blk_get_request);
1639 * blk_requeue_request - put a request back on queue
1640 * @q: request queue where request should be inserted
1641 * @rq: request to be inserted
1644 * Drivers often keep queueing requests until the hardware cannot accept
1645 * more, when that condition happens we need to put the request back
1646 * on the queue. Must be called with queue lock held.
1648 void blk_requeue_request(struct request_queue *q, struct request *rq)
1650 lockdep_assert_held(q->queue_lock);
1651 WARN_ON_ONCE(q->mq_ops);
1653 blk_delete_timer(rq);
1654 blk_clear_rq_complete(rq);
1655 trace_block_rq_requeue(q, rq);
1656 wbt_requeue(q->rq_wb, rq);
1658 if (rq->rq_flags & RQF_QUEUED)
1659 blk_queue_end_tag(q, rq);
1661 BUG_ON(blk_queued_rq(rq));
1663 elv_requeue_request(q, rq);
1665 EXPORT_SYMBOL(blk_requeue_request);
1667 static void add_acct_request(struct request_queue *q, struct request *rq,
1670 blk_account_io_start(rq, true);
1671 __elv_add_request(q, rq, where);
1674 static void part_round_stats_single(struct request_queue *q, int cpu,
1675 struct hd_struct *part, unsigned long now,
1676 unsigned int inflight)
1679 __part_stat_add(cpu, part, time_in_queue,
1680 inflight * (now - part->stamp));
1681 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1687 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1688 * @q: target block queue
1689 * @cpu: cpu number for stats access
1690 * @part: target partition
1692 * The average IO queue length and utilisation statistics are maintained
1693 * by observing the current state of the queue length and the amount of
1694 * time it has been in this state for.
1696 * Normally, that accounting is done on IO completion, but that can result
1697 * in more than a second's worth of IO being accounted for within any one
1698 * second, leading to >100% utilisation. To deal with that, we call this
1699 * function to do a round-off before returning the results when reading
1700 * /proc/diskstats. This accounts immediately for all queue usage up to
1701 * the current jiffies and restarts the counters again.
1703 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1705 struct hd_struct *part2 = NULL;
1706 unsigned long now = jiffies;
1707 unsigned int inflight[2];
1710 if (part->stamp != now)
1714 part2 = &part_to_disk(part)->part0;
1715 if (part2->stamp != now)
1722 part_in_flight(q, part, inflight);
1725 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1727 part_round_stats_single(q, cpu, part, now, inflight[0]);
1729 EXPORT_SYMBOL_GPL(part_round_stats);
1732 static void blk_pm_put_request(struct request *rq)
1734 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1735 pm_runtime_mark_last_busy(rq->q->dev);
1738 static inline void blk_pm_put_request(struct request *rq) {}
1741 void __blk_put_request(struct request_queue *q, struct request *req)
1743 req_flags_t rq_flags = req->rq_flags;
1749 blk_mq_free_request(req);
1753 lockdep_assert_held(q->queue_lock);
1755 blk_req_zone_write_unlock(req);
1756 blk_pm_put_request(req);
1758 elv_completed_request(q, req);
1760 /* this is a bio leak */
1761 WARN_ON(req->bio != NULL);
1763 wbt_done(q->rq_wb, req);
1766 * Request may not have originated from ll_rw_blk. if not,
1767 * it didn't come out of our reserved rq pools
1769 if (rq_flags & RQF_ALLOCED) {
1770 struct request_list *rl = blk_rq_rl(req);
1771 bool sync = op_is_sync(req->cmd_flags);
1773 BUG_ON(!list_empty(&req->queuelist));
1774 BUG_ON(ELV_ON_HASH(req));
1776 blk_free_request(rl, req);
1777 freed_request(rl, sync, rq_flags);
1782 EXPORT_SYMBOL_GPL(__blk_put_request);
1784 void blk_put_request(struct request *req)
1786 struct request_queue *q = req->q;
1789 blk_mq_free_request(req);
1791 unsigned long flags;
1793 spin_lock_irqsave(q->queue_lock, flags);
1794 __blk_put_request(q, req);
1795 spin_unlock_irqrestore(q->queue_lock, flags);
1798 EXPORT_SYMBOL(blk_put_request);
1800 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1803 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1805 if (!ll_back_merge_fn(q, req, bio))
1808 trace_block_bio_backmerge(q, req, bio);
1810 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1811 blk_rq_set_mixed_merge(req);
1813 req->biotail->bi_next = bio;
1815 req->__data_len += bio->bi_iter.bi_size;
1816 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1818 blk_account_io_start(req, false);
1822 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1825 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1827 if (!ll_front_merge_fn(q, req, bio))
1830 trace_block_bio_frontmerge(q, req, bio);
1832 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1833 blk_rq_set_mixed_merge(req);
1835 bio->bi_next = req->bio;
1838 req->__sector = bio->bi_iter.bi_sector;
1839 req->__data_len += bio->bi_iter.bi_size;
1840 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1842 blk_account_io_start(req, false);
1846 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1849 unsigned short segments = blk_rq_nr_discard_segments(req);
1851 if (segments >= queue_max_discard_segments(q))
1853 if (blk_rq_sectors(req) + bio_sectors(bio) >
1854 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1857 req->biotail->bi_next = bio;
1859 req->__data_len += bio->bi_iter.bi_size;
1860 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1861 req->nr_phys_segments = segments + 1;
1863 blk_account_io_start(req, false);
1866 req_set_nomerge(q, req);
1871 * blk_attempt_plug_merge - try to merge with %current's plugged list
1872 * @q: request_queue new bio is being queued at
1873 * @bio: new bio being queued
1874 * @request_count: out parameter for number of traversed plugged requests
1875 * @same_queue_rq: pointer to &struct request that gets filled in when
1876 * another request associated with @q is found on the plug list
1877 * (optional, may be %NULL)
1879 * Determine whether @bio being queued on @q can be merged with a request
1880 * on %current's plugged list. Returns %true if merge was successful,
1883 * Plugging coalesces IOs from the same issuer for the same purpose without
1884 * going through @q->queue_lock. As such it's more of an issuing mechanism
1885 * than scheduling, and the request, while may have elvpriv data, is not
1886 * added on the elevator at this point. In addition, we don't have
1887 * reliable access to the elevator outside queue lock. Only check basic
1888 * merging parameters without querying the elevator.
1890 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1892 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1893 unsigned int *request_count,
1894 struct request **same_queue_rq)
1896 struct blk_plug *plug;
1898 struct list_head *plug_list;
1900 plug = current->plug;
1906 plug_list = &plug->mq_list;
1908 plug_list = &plug->list;
1910 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1911 bool merged = false;
1916 * Only blk-mq multiple hardware queues case checks the
1917 * rq in the same queue, there should be only one such
1921 *same_queue_rq = rq;
1924 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1927 switch (blk_try_merge(rq, bio)) {
1928 case ELEVATOR_BACK_MERGE:
1929 merged = bio_attempt_back_merge(q, rq, bio);
1931 case ELEVATOR_FRONT_MERGE:
1932 merged = bio_attempt_front_merge(q, rq, bio);
1934 case ELEVATOR_DISCARD_MERGE:
1935 merged = bio_attempt_discard_merge(q, rq, bio);
1948 unsigned int blk_plug_queued_count(struct request_queue *q)
1950 struct blk_plug *plug;
1952 struct list_head *plug_list;
1953 unsigned int ret = 0;
1955 plug = current->plug;
1960 plug_list = &plug->mq_list;
1962 plug_list = &plug->list;
1964 list_for_each_entry(rq, plug_list, queuelist) {
1972 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1974 struct io_context *ioc = rq_ioc(bio);
1976 if (bio->bi_opf & REQ_RAHEAD)
1977 req->cmd_flags |= REQ_FAILFAST_MASK;
1979 req->__sector = bio->bi_iter.bi_sector;
1980 if (ioprio_valid(bio_prio(bio)))
1981 req->ioprio = bio_prio(bio);
1983 req->ioprio = ioc->ioprio;
1985 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1986 req->write_hint = bio->bi_write_hint;
1987 blk_rq_bio_prep(req->q, req, bio);
1989 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1991 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1993 struct blk_plug *plug;
1994 int where = ELEVATOR_INSERT_SORT;
1995 struct request *req, *free;
1996 unsigned int request_count = 0;
1997 unsigned int wb_acct;
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 wb_acct = wbt_wait(q->rq_wb, 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 __wbt_done(q->rq_wb, wb_acct);
2066 if (PTR_ERR(req) == -ENOMEM)
2067 bio->bi_status = BLK_STS_RESOURCE;
2069 bio->bi_status = BLK_STS_IOERR;
2074 wbt_track(req, wb_acct);
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 if (part->policy && op_is_write(bio_op(bio))) {
2168 char b[BDEVNAME_SIZE];
2171 "generic_make_request: Trying to write "
2172 "to read-only block-device %s (partno %d)\n",
2173 bio_devname(bio, b), part->partno);
2180 static noinline int should_fail_bio(struct bio *bio)
2182 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2186 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2189 * Check whether this bio extends beyond the end of the device or partition.
2190 * This may well happen - the kernel calls bread() without checking the size of
2191 * the device, e.g., when mounting a file system.
2193 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2195 unsigned int nr_sectors = bio_sectors(bio);
2197 if (nr_sectors && maxsector &&
2198 (nr_sectors > maxsector ||
2199 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2200 handle_bad_sector(bio, maxsector);
2207 * Remap block n of partition p to block n+start(p) of the disk.
2209 static inline int blk_partition_remap(struct bio *bio)
2211 struct hd_struct *p;
2215 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2218 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2220 if (unlikely(bio_check_ro(bio, p)))
2224 * Zone reset does not include bi_size so bio_sectors() is always 0.
2225 * Include a test for the reset op code and perform the remap if needed.
2227 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2228 if (bio_check_eod(bio, part_nr_sects_read(p)))
2230 bio->bi_iter.bi_sector += p->start_sect;
2232 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2233 bio->bi_iter.bi_sector - p->start_sect);
2241 static noinline_for_stack bool
2242 generic_make_request_checks(struct bio *bio)
2244 struct request_queue *q;
2245 int nr_sectors = bio_sectors(bio);
2246 blk_status_t status = BLK_STS_IOERR;
2247 char b[BDEVNAME_SIZE];
2251 q = bio->bi_disk->queue;
2254 "generic_make_request: Trying to access "
2255 "nonexistent block-device %s (%Lu)\n",
2256 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2261 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2262 * if queue is not a request based queue.
2264 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2267 if (should_fail_bio(bio))
2270 if (bio->bi_partno) {
2271 if (unlikely(blk_partition_remap(bio)))
2274 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2276 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
2281 * Filter flush bio's early so that make_request based
2282 * drivers without flush support don't have to worry
2285 if (op_is_flush(bio->bi_opf) &&
2286 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2287 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2289 status = BLK_STS_OK;
2294 switch (bio_op(bio)) {
2295 case REQ_OP_DISCARD:
2296 if (!blk_queue_discard(q))
2299 case REQ_OP_SECURE_ERASE:
2300 if (!blk_queue_secure_erase(q))
2303 case REQ_OP_WRITE_SAME:
2304 if (!q->limits.max_write_same_sectors)
2307 case REQ_OP_ZONE_REPORT:
2308 case REQ_OP_ZONE_RESET:
2309 if (!blk_queue_is_zoned(q))
2312 case REQ_OP_WRITE_ZEROES:
2313 if (!q->limits.max_write_zeroes_sectors)
2321 * Various block parts want %current->io_context and lazy ioc
2322 * allocation ends up trading a lot of pain for a small amount of
2323 * memory. Just allocate it upfront. This may fail and block
2324 * layer knows how to live with it.
2326 create_io_context(GFP_ATOMIC, q->node);
2328 if (!blkcg_bio_issue_check(q, bio))
2331 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2332 trace_block_bio_queue(q, bio);
2333 /* Now that enqueuing has been traced, we need to trace
2334 * completion as well.
2336 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2341 status = BLK_STS_NOTSUPP;
2343 bio->bi_status = status;
2349 * generic_make_request - hand a buffer to its device driver for I/O
2350 * @bio: The bio describing the location in memory and on the device.
2352 * generic_make_request() is used to make I/O requests of block
2353 * devices. It is passed a &struct bio, which describes the I/O that needs
2356 * generic_make_request() does not return any status. The
2357 * success/failure status of the request, along with notification of
2358 * completion, is delivered asynchronously through the bio->bi_end_io
2359 * function described (one day) else where.
2361 * The caller of generic_make_request must make sure that bi_io_vec
2362 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2363 * set to describe the device address, and the
2364 * bi_end_io and optionally bi_private are set to describe how
2365 * completion notification should be signaled.
2367 * generic_make_request and the drivers it calls may use bi_next if this
2368 * bio happens to be merged with someone else, and may resubmit the bio to
2369 * a lower device by calling into generic_make_request recursively, which
2370 * means the bio should NOT be touched after the call to ->make_request_fn.
2372 blk_qc_t generic_make_request(struct bio *bio)
2375 * bio_list_on_stack[0] contains bios submitted by the current
2377 * bio_list_on_stack[1] contains bios that were submitted before
2378 * the current make_request_fn, but that haven't been processed
2381 struct bio_list bio_list_on_stack[2];
2382 blk_mq_req_flags_t flags = 0;
2383 struct request_queue *q = bio->bi_disk->queue;
2384 blk_qc_t ret = BLK_QC_T_NONE;
2386 if (bio->bi_opf & REQ_NOWAIT)
2387 flags = BLK_MQ_REQ_NOWAIT;
2388 if (blk_queue_enter(q, flags) < 0) {
2389 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2390 bio_wouldblock_error(bio);
2396 if (!generic_make_request_checks(bio))
2400 * We only want one ->make_request_fn to be active at a time, else
2401 * stack usage with stacked devices could be a problem. So use
2402 * current->bio_list to keep a list of requests submited by a
2403 * make_request_fn function. current->bio_list is also used as a
2404 * flag to say if generic_make_request is currently active in this
2405 * task or not. If it is NULL, then no make_request is active. If
2406 * it is non-NULL, then a make_request is active, and new requests
2407 * should be added at the tail
2409 if (current->bio_list) {
2410 bio_list_add(¤t->bio_list[0], bio);
2414 /* following loop may be a bit non-obvious, and so deserves some
2416 * Before entering the loop, bio->bi_next is NULL (as all callers
2417 * ensure that) so we have a list with a single bio.
2418 * We pretend that we have just taken it off a longer list, so
2419 * we assign bio_list to a pointer to the bio_list_on_stack,
2420 * thus initialising the bio_list of new bios to be
2421 * added. ->make_request() may indeed add some more bios
2422 * through a recursive call to generic_make_request. If it
2423 * did, we find a non-NULL value in bio_list and re-enter the loop
2424 * from the top. In this case we really did just take the bio
2425 * of the top of the list (no pretending) and so remove it from
2426 * bio_list, and call into ->make_request() again.
2428 BUG_ON(bio->bi_next);
2429 bio_list_init(&bio_list_on_stack[0]);
2430 current->bio_list = bio_list_on_stack;
2432 bool enter_succeeded = true;
2434 if (unlikely(q != bio->bi_disk->queue)) {
2437 q = bio->bi_disk->queue;
2439 if (bio->bi_opf & REQ_NOWAIT)
2440 flags = BLK_MQ_REQ_NOWAIT;
2441 if (blk_queue_enter(q, flags) < 0) {
2442 enter_succeeded = false;
2447 if (enter_succeeded) {
2448 struct bio_list lower, same;
2450 /* Create a fresh bio_list for all subordinate requests */
2451 bio_list_on_stack[1] = bio_list_on_stack[0];
2452 bio_list_init(&bio_list_on_stack[0]);
2453 ret = q->make_request_fn(q, bio);
2455 /* sort new bios into those for a lower level
2456 * and those for the same level
2458 bio_list_init(&lower);
2459 bio_list_init(&same);
2460 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2461 if (q == bio->bi_disk->queue)
2462 bio_list_add(&same, bio);
2464 bio_list_add(&lower, bio);
2465 /* now assemble so we handle the lowest level first */
2466 bio_list_merge(&bio_list_on_stack[0], &lower);
2467 bio_list_merge(&bio_list_on_stack[0], &same);
2468 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2470 if (unlikely(!blk_queue_dying(q) &&
2471 (bio->bi_opf & REQ_NOWAIT)))
2472 bio_wouldblock_error(bio);
2476 bio = bio_list_pop(&bio_list_on_stack[0]);
2478 current->bio_list = NULL; /* deactivate */
2485 EXPORT_SYMBOL(generic_make_request);
2488 * direct_make_request - hand a buffer directly to its device driver for I/O
2489 * @bio: The bio describing the location in memory and on the device.
2491 * This function behaves like generic_make_request(), but does not protect
2492 * against recursion. Must only be used if the called driver is known
2493 * to not call generic_make_request (or direct_make_request) again from
2494 * its make_request function. (Calling direct_make_request again from
2495 * a workqueue is perfectly fine as that doesn't recurse).
2497 blk_qc_t direct_make_request(struct bio *bio)
2499 struct request_queue *q = bio->bi_disk->queue;
2500 bool nowait = bio->bi_opf & REQ_NOWAIT;
2503 if (!generic_make_request_checks(bio))
2504 return BLK_QC_T_NONE;
2506 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2507 if (nowait && !blk_queue_dying(q))
2508 bio->bi_status = BLK_STS_AGAIN;
2510 bio->bi_status = BLK_STS_IOERR;
2512 return BLK_QC_T_NONE;
2515 ret = q->make_request_fn(q, bio);
2519 EXPORT_SYMBOL_GPL(direct_make_request);
2522 * submit_bio - submit a bio to the block device layer for I/O
2523 * @bio: The &struct bio which describes the I/O
2525 * submit_bio() is very similar in purpose to generic_make_request(), and
2526 * uses that function to do most of the work. Both are fairly rough
2527 * interfaces; @bio must be presetup and ready for I/O.
2530 blk_qc_t submit_bio(struct bio *bio)
2533 * If it's a regular read/write or a barrier with data attached,
2534 * go through the normal accounting stuff before submission.
2536 if (bio_has_data(bio)) {
2539 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2540 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2542 count = bio_sectors(bio);
2544 if (op_is_write(bio_op(bio))) {
2545 count_vm_events(PGPGOUT, count);
2547 task_io_account_read(bio->bi_iter.bi_size);
2548 count_vm_events(PGPGIN, count);
2551 if (unlikely(block_dump)) {
2552 char b[BDEVNAME_SIZE];
2553 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2554 current->comm, task_pid_nr(current),
2555 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2556 (unsigned long long)bio->bi_iter.bi_sector,
2557 bio_devname(bio, b), count);
2561 return generic_make_request(bio);
2563 EXPORT_SYMBOL(submit_bio);
2565 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2567 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2571 blk_flush_plug_list(current->plug, false);
2572 return q->poll_fn(q, cookie);
2574 EXPORT_SYMBOL_GPL(blk_poll);
2577 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2578 * for new the queue limits
2580 * @rq: the request being checked
2583 * @rq may have been made based on weaker limitations of upper-level queues
2584 * in request stacking drivers, and it may violate the limitation of @q.
2585 * Since the block layer and the underlying device driver trust @rq
2586 * after it is inserted to @q, it should be checked against @q before
2587 * the insertion using this generic function.
2589 * Request stacking drivers like request-based dm may change the queue
2590 * limits when retrying requests on other queues. Those requests need
2591 * to be checked against the new queue limits again during dispatch.
2593 static int blk_cloned_rq_check_limits(struct request_queue *q,
2596 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2597 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2602 * queue's settings related to segment counting like q->bounce_pfn
2603 * may differ from that of other stacking queues.
2604 * Recalculate it to check the request correctly on this queue's
2607 blk_recalc_rq_segments(rq);
2608 if (rq->nr_phys_segments > queue_max_segments(q)) {
2609 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2617 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2618 * @q: the queue to submit the request
2619 * @rq: the request being queued
2621 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2623 unsigned long flags;
2624 int where = ELEVATOR_INSERT_BACK;
2626 if (blk_cloned_rq_check_limits(q, rq))
2627 return BLK_STS_IOERR;
2630 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2631 return BLK_STS_IOERR;
2634 if (blk_queue_io_stat(q))
2635 blk_account_io_start(rq, true);
2637 * Since we have a scheduler attached on the top device,
2638 * bypass a potential scheduler on the bottom device for
2641 return blk_mq_request_issue_directly(rq);
2644 spin_lock_irqsave(q->queue_lock, flags);
2645 if (unlikely(blk_queue_dying(q))) {
2646 spin_unlock_irqrestore(q->queue_lock, flags);
2647 return BLK_STS_IOERR;
2651 * Submitting request must be dequeued before calling this function
2652 * because it will be linked to another request_queue
2654 BUG_ON(blk_queued_rq(rq));
2656 if (op_is_flush(rq->cmd_flags))
2657 where = ELEVATOR_INSERT_FLUSH;
2659 add_acct_request(q, rq, where);
2660 if (where == ELEVATOR_INSERT_FLUSH)
2662 spin_unlock_irqrestore(q->queue_lock, flags);
2666 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2669 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2670 * @rq: request to examine
2673 * A request could be merge of IOs which require different failure
2674 * handling. This function determines the number of bytes which
2675 * can be failed from the beginning of the request without
2676 * crossing into area which need to be retried further.
2679 * The number of bytes to fail.
2681 unsigned int blk_rq_err_bytes(const struct request *rq)
2683 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2684 unsigned int bytes = 0;
2687 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2688 return blk_rq_bytes(rq);
2691 * Currently the only 'mixing' which can happen is between
2692 * different fastfail types. We can safely fail portions
2693 * which have all the failfast bits that the first one has -
2694 * the ones which are at least as eager to fail as the first
2697 for (bio = rq->bio; bio; bio = bio->bi_next) {
2698 if ((bio->bi_opf & ff) != ff)
2700 bytes += bio->bi_iter.bi_size;
2703 /* this could lead to infinite loop */
2704 BUG_ON(blk_rq_bytes(rq) && !bytes);
2707 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2709 void blk_account_io_completion(struct request *req, unsigned int bytes)
2711 if (blk_do_io_stat(req)) {
2712 const int rw = rq_data_dir(req);
2713 struct hd_struct *part;
2716 cpu = part_stat_lock();
2718 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2723 void blk_account_io_done(struct request *req, u64 now)
2726 * Account IO completion. flush_rq isn't accounted as a
2727 * normal IO on queueing nor completion. Accounting the
2728 * containing request is enough.
2730 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2731 unsigned long duration;
2732 const int rw = rq_data_dir(req);
2733 struct hd_struct *part;
2736 duration = nsecs_to_jiffies(now - req->start_time_ns);
2737 cpu = part_stat_lock();
2740 part_stat_inc(cpu, part, ios[rw]);
2741 part_stat_add(cpu, part, ticks[rw], duration);
2742 part_round_stats(req->q, cpu, part);
2743 part_dec_in_flight(req->q, part, rw);
2745 hd_struct_put(part);
2752 * Don't process normal requests when queue is suspended
2753 * or in the process of suspending/resuming
2755 static bool blk_pm_allow_request(struct request *rq)
2757 switch (rq->q->rpm_status) {
2759 case RPM_SUSPENDING:
2760 return rq->rq_flags & RQF_PM;
2768 static bool blk_pm_allow_request(struct request *rq)
2774 void blk_account_io_start(struct request *rq, bool new_io)
2776 struct hd_struct *part;
2777 int rw = rq_data_dir(rq);
2780 if (!blk_do_io_stat(rq))
2783 cpu = part_stat_lock();
2787 part_stat_inc(cpu, part, merges[rw]);
2789 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2790 if (!hd_struct_try_get(part)) {
2792 * The partition is already being removed,
2793 * the request will be accounted on the disk only
2795 * We take a reference on disk->part0 although that
2796 * partition will never be deleted, so we can treat
2797 * it as any other partition.
2799 part = &rq->rq_disk->part0;
2800 hd_struct_get(part);
2802 part_round_stats(rq->q, cpu, part);
2803 part_inc_in_flight(rq->q, part, rw);
2810 static struct request *elv_next_request(struct request_queue *q)
2813 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2815 WARN_ON_ONCE(q->mq_ops);
2818 list_for_each_entry(rq, &q->queue_head, queuelist) {
2819 if (blk_pm_allow_request(rq))
2822 if (rq->rq_flags & RQF_SOFTBARRIER)
2827 * Flush request is running and flush request isn't queueable
2828 * in the drive, we can hold the queue till flush request is
2829 * finished. Even we don't do this, driver can't dispatch next
2830 * requests and will requeue them. And this can improve
2831 * throughput too. For example, we have request flush1, write1,
2832 * flush 2. flush1 is dispatched, then queue is hold, write1
2833 * isn't inserted to queue. After flush1 is finished, flush2
2834 * will be dispatched. Since disk cache is already clean,
2835 * flush2 will be finished very soon, so looks like flush2 is
2837 * Since the queue is hold, a flag is set to indicate the queue
2838 * should be restarted later. Please see flush_end_io() for
2841 if (fq->flush_pending_idx != fq->flush_running_idx &&
2842 !queue_flush_queueable(q)) {
2843 fq->flush_queue_delayed = 1;
2846 if (unlikely(blk_queue_bypass(q)) ||
2847 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2853 * blk_peek_request - peek at the top of a request queue
2854 * @q: request queue to peek at
2857 * Return the request at the top of @q. The returned request
2858 * should be started using blk_start_request() before LLD starts
2862 * Pointer to the request at the top of @q if available. Null
2865 struct request *blk_peek_request(struct request_queue *q)
2870 lockdep_assert_held(q->queue_lock);
2871 WARN_ON_ONCE(q->mq_ops);
2873 while ((rq = elv_next_request(q)) != NULL) {
2874 if (!(rq->rq_flags & RQF_STARTED)) {
2876 * This is the first time the device driver
2877 * sees this request (possibly after
2878 * requeueing). Notify IO scheduler.
2880 if (rq->rq_flags & RQF_SORTED)
2881 elv_activate_rq(q, rq);
2884 * just mark as started even if we don't start
2885 * it, a request that has been delayed should
2886 * not be passed by new incoming requests
2888 rq->rq_flags |= RQF_STARTED;
2889 trace_block_rq_issue(q, rq);
2892 if (!q->boundary_rq || q->boundary_rq == rq) {
2893 q->end_sector = rq_end_sector(rq);
2894 q->boundary_rq = NULL;
2897 if (rq->rq_flags & RQF_DONTPREP)
2900 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2902 * make sure space for the drain appears we
2903 * know we can do this because max_hw_segments
2904 * has been adjusted to be one fewer than the
2907 rq->nr_phys_segments++;
2913 ret = q->prep_rq_fn(q, rq);
2914 if (ret == BLKPREP_OK) {
2916 } else if (ret == BLKPREP_DEFER) {
2918 * the request may have been (partially) prepped.
2919 * we need to keep this request in the front to
2920 * avoid resource deadlock. RQF_STARTED will
2921 * prevent other fs requests from passing this one.
2923 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2924 !(rq->rq_flags & RQF_DONTPREP)) {
2926 * remove the space for the drain we added
2927 * so that we don't add it again
2929 --rq->nr_phys_segments;
2934 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2935 rq->rq_flags |= RQF_QUIET;
2937 * Mark this request as started so we don't trigger
2938 * any debug logic in the end I/O path.
2940 blk_start_request(rq);
2941 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2942 BLK_STS_TARGET : BLK_STS_IOERR);
2944 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2951 EXPORT_SYMBOL(blk_peek_request);
2953 static void blk_dequeue_request(struct request *rq)
2955 struct request_queue *q = rq->q;
2957 BUG_ON(list_empty(&rq->queuelist));
2958 BUG_ON(ELV_ON_HASH(rq));
2960 list_del_init(&rq->queuelist);
2963 * the time frame between a request being removed from the lists
2964 * and to it is freed is accounted as io that is in progress at
2967 if (blk_account_rq(rq))
2968 q->in_flight[rq_is_sync(rq)]++;
2972 * blk_start_request - start request processing on the driver
2973 * @req: request to dequeue
2976 * Dequeue @req and start timeout timer on it. This hands off the
2977 * request to the driver.
2979 void blk_start_request(struct request *req)
2981 lockdep_assert_held(req->q->queue_lock);
2982 WARN_ON_ONCE(req->q->mq_ops);
2984 blk_dequeue_request(req);
2986 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2987 req->io_start_time_ns = ktime_get_ns();
2988 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2989 req->throtl_size = blk_rq_sectors(req);
2991 req->rq_flags |= RQF_STATS;
2992 wbt_issue(req->q->rq_wb, req);
2995 BUG_ON(blk_rq_is_complete(req));
2998 EXPORT_SYMBOL(blk_start_request);
3001 * blk_fetch_request - fetch a request from a request queue
3002 * @q: request queue to fetch a request from
3005 * Return the request at the top of @q. The request is started on
3006 * return and LLD can start processing it immediately.
3009 * Pointer to the request at the top of @q if available. Null
3012 struct request *blk_fetch_request(struct request_queue *q)
3016 lockdep_assert_held(q->queue_lock);
3017 WARN_ON_ONCE(q->mq_ops);
3019 rq = blk_peek_request(q);
3021 blk_start_request(rq);
3024 EXPORT_SYMBOL(blk_fetch_request);
3027 * Steal bios from a request and add them to a bio list.
3028 * The request must not have been partially completed before.
3030 void blk_steal_bios(struct bio_list *list, struct request *rq)
3034 list->tail->bi_next = rq->bio;
3036 list->head = rq->bio;
3037 list->tail = rq->biotail;
3045 EXPORT_SYMBOL_GPL(blk_steal_bios);
3048 * blk_update_request - Special helper function for request stacking drivers
3049 * @req: the request being processed
3050 * @error: block status code
3051 * @nr_bytes: number of bytes to complete @req
3054 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3055 * the request structure even if @req doesn't have leftover.
3056 * If @req has leftover, sets it up for the next range of segments.
3058 * This special helper function is only for request stacking drivers
3059 * (e.g. request-based dm) so that they can handle partial completion.
3060 * Actual device drivers should use blk_end_request instead.
3062 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3063 * %false return from this function.
3066 * %false - this request doesn't have any more data
3067 * %true - this request has more data
3069 bool blk_update_request(struct request *req, blk_status_t error,
3070 unsigned int nr_bytes)
3074 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
3079 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3080 !(req->rq_flags & RQF_QUIET)))
3081 print_req_error(req, error);
3083 blk_account_io_completion(req, nr_bytes);
3087 struct bio *bio = req->bio;
3088 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3090 if (bio_bytes == bio->bi_iter.bi_size) {
3091 req->bio = bio->bi_next;
3092 bio->bi_next = NULL;
3095 /* Completion has already been traced */
3096 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
3097 req_bio_endio(req, bio, bio_bytes, error);
3099 total_bytes += bio_bytes;
3100 nr_bytes -= bio_bytes;
3111 * Reset counters so that the request stacking driver
3112 * can find how many bytes remain in the request
3115 req->__data_len = 0;
3119 req->__data_len -= total_bytes;
3121 /* update sector only for requests with clear definition of sector */
3122 if (!blk_rq_is_passthrough(req))
3123 req->__sector += total_bytes >> 9;
3125 /* mixed attributes always follow the first bio */
3126 if (req->rq_flags & RQF_MIXED_MERGE) {
3127 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3128 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3131 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3133 * If total number of sectors is less than the first segment
3134 * size, something has gone terribly wrong.
3136 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3137 blk_dump_rq_flags(req, "request botched");
3138 req->__data_len = blk_rq_cur_bytes(req);
3141 /* recalculate the number of segments */
3142 blk_recalc_rq_segments(req);
3147 EXPORT_SYMBOL_GPL(blk_update_request);
3149 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3150 unsigned int nr_bytes,
3151 unsigned int bidi_bytes)
3153 if (blk_update_request(rq, error, nr_bytes))
3156 /* Bidi request must be completed as a whole */
3157 if (unlikely(blk_bidi_rq(rq)) &&
3158 blk_update_request(rq->next_rq, error, bidi_bytes))
3161 if (blk_queue_add_random(rq->q))
3162 add_disk_randomness(rq->rq_disk);
3168 * blk_unprep_request - unprepare a request
3171 * This function makes a request ready for complete resubmission (or
3172 * completion). It happens only after all error handling is complete,
3173 * so represents the appropriate moment to deallocate any resources
3174 * that were allocated to the request in the prep_rq_fn. The queue
3175 * lock is held when calling this.
3177 void blk_unprep_request(struct request *req)
3179 struct request_queue *q = req->q;
3181 req->rq_flags &= ~RQF_DONTPREP;
3182 if (q->unprep_rq_fn)
3183 q->unprep_rq_fn(q, req);
3185 EXPORT_SYMBOL_GPL(blk_unprep_request);
3187 void blk_finish_request(struct request *req, blk_status_t error)
3189 struct request_queue *q = req->q;
3190 u64 now = ktime_get_ns();
3192 lockdep_assert_held(req->q->queue_lock);
3193 WARN_ON_ONCE(q->mq_ops);
3195 if (req->rq_flags & RQF_STATS)
3196 blk_stat_add(req, now);
3198 if (req->rq_flags & RQF_QUEUED)
3199 blk_queue_end_tag(q, req);
3201 BUG_ON(blk_queued_rq(req));
3203 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3204 laptop_io_completion(req->q->backing_dev_info);
3206 blk_delete_timer(req);
3208 if (req->rq_flags & RQF_DONTPREP)
3209 blk_unprep_request(req);
3211 blk_account_io_done(req, now);
3214 wbt_done(req->q->rq_wb, req);
3215 req->end_io(req, error);
3217 if (blk_bidi_rq(req))
3218 __blk_put_request(req->next_rq->q, req->next_rq);
3220 __blk_put_request(q, req);
3223 EXPORT_SYMBOL(blk_finish_request);
3226 * blk_end_bidi_request - Complete a bidi request
3227 * @rq: the request to complete
3228 * @error: block status code
3229 * @nr_bytes: number of bytes to complete @rq
3230 * @bidi_bytes: number of bytes to complete @rq->next_rq
3233 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3234 * Drivers that supports bidi can safely call this member for any
3235 * type of request, bidi or uni. In the later case @bidi_bytes is
3239 * %false - we are done with this request
3240 * %true - still buffers pending for this request
3242 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3243 unsigned int nr_bytes, unsigned int bidi_bytes)
3245 struct request_queue *q = rq->q;
3246 unsigned long flags;
3248 WARN_ON_ONCE(q->mq_ops);
3250 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3253 spin_lock_irqsave(q->queue_lock, flags);
3254 blk_finish_request(rq, error);
3255 spin_unlock_irqrestore(q->queue_lock, flags);
3261 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3262 * @rq: the request to complete
3263 * @error: block status code
3264 * @nr_bytes: number of bytes to complete @rq
3265 * @bidi_bytes: number of bytes to complete @rq->next_rq
3268 * Identical to blk_end_bidi_request() except that queue lock is
3269 * assumed to be locked on entry and remains so on return.
3272 * %false - we are done with this request
3273 * %true - still buffers pending for this request
3275 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3276 unsigned int nr_bytes, unsigned int bidi_bytes)
3278 lockdep_assert_held(rq->q->queue_lock);
3279 WARN_ON_ONCE(rq->q->mq_ops);
3281 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3284 blk_finish_request(rq, error);
3290 * blk_end_request - Helper function for drivers to complete the request.
3291 * @rq: the request being processed
3292 * @error: block status code
3293 * @nr_bytes: number of bytes to complete
3296 * Ends I/O on a number of bytes attached to @rq.
3297 * If @rq has leftover, sets it up for the next range of segments.
3300 * %false - we are done with this request
3301 * %true - still buffers pending for this request
3303 bool blk_end_request(struct request *rq, blk_status_t error,
3304 unsigned int nr_bytes)
3306 WARN_ON_ONCE(rq->q->mq_ops);
3307 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3309 EXPORT_SYMBOL(blk_end_request);
3312 * blk_end_request_all - Helper function for drives to finish the request.
3313 * @rq: the request to finish
3314 * @error: block status code
3317 * Completely finish @rq.
3319 void blk_end_request_all(struct request *rq, blk_status_t error)
3322 unsigned int bidi_bytes = 0;
3324 if (unlikely(blk_bidi_rq(rq)))
3325 bidi_bytes = blk_rq_bytes(rq->next_rq);
3327 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3330 EXPORT_SYMBOL(blk_end_request_all);
3333 * __blk_end_request - Helper function for drivers to complete the request.
3334 * @rq: the request being processed
3335 * @error: block status code
3336 * @nr_bytes: number of bytes to complete
3339 * Must be called with queue lock held unlike blk_end_request().
3342 * %false - we are done with this request
3343 * %true - still buffers pending for this request
3345 bool __blk_end_request(struct request *rq, blk_status_t error,
3346 unsigned int nr_bytes)
3348 lockdep_assert_held(rq->q->queue_lock);
3349 WARN_ON_ONCE(rq->q->mq_ops);
3351 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3353 EXPORT_SYMBOL(__blk_end_request);
3356 * __blk_end_request_all - Helper function for drives to finish the request.
3357 * @rq: the request to finish
3358 * @error: block status code
3361 * Completely finish @rq. Must be called with queue lock held.
3363 void __blk_end_request_all(struct request *rq, blk_status_t error)
3366 unsigned int bidi_bytes = 0;
3368 lockdep_assert_held(rq->q->queue_lock);
3369 WARN_ON_ONCE(rq->q->mq_ops);
3371 if (unlikely(blk_bidi_rq(rq)))
3372 bidi_bytes = blk_rq_bytes(rq->next_rq);
3374 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3377 EXPORT_SYMBOL(__blk_end_request_all);
3380 * __blk_end_request_cur - Helper function to finish the current request chunk.
3381 * @rq: the request to finish the current chunk for
3382 * @error: block status code
3385 * Complete the current consecutively mapped chunk from @rq. Must
3386 * be called with queue lock held.
3389 * %false - we are done with this request
3390 * %true - still buffers pending for this request
3392 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3394 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3396 EXPORT_SYMBOL(__blk_end_request_cur);
3398 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3401 if (bio_has_data(bio))
3402 rq->nr_phys_segments = bio_phys_segments(q, bio);
3403 else if (bio_op(bio) == REQ_OP_DISCARD)
3404 rq->nr_phys_segments = 1;
3406 rq->__data_len = bio->bi_iter.bi_size;
3407 rq->bio = rq->biotail = bio;
3410 rq->rq_disk = bio->bi_disk;
3413 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3415 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3416 * @rq: the request to be flushed
3419 * Flush all pages in @rq.
3421 void rq_flush_dcache_pages(struct request *rq)
3423 struct req_iterator iter;
3424 struct bio_vec bvec;
3426 rq_for_each_segment(bvec, rq, iter)
3427 flush_dcache_page(bvec.bv_page);
3429 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3433 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3434 * @q : the queue of the device being checked
3437 * Check if underlying low-level drivers of a device are busy.
3438 * If the drivers want to export their busy state, they must set own
3439 * exporting function using blk_queue_lld_busy() first.
3441 * Basically, this function is used only by request stacking drivers
3442 * to stop dispatching requests to underlying devices when underlying
3443 * devices are busy. This behavior helps more I/O merging on the queue
3444 * of the request stacking driver and prevents I/O throughput regression
3445 * on burst I/O load.
3448 * 0 - Not busy (The request stacking driver should dispatch request)
3449 * 1 - Busy (The request stacking driver should stop dispatching request)
3451 int blk_lld_busy(struct request_queue *q)
3454 return q->lld_busy_fn(q);
3458 EXPORT_SYMBOL_GPL(blk_lld_busy);
3461 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3462 * @rq: the clone request to be cleaned up
3465 * Free all bios in @rq for a cloned request.
3467 void blk_rq_unprep_clone(struct request *rq)
3471 while ((bio = rq->bio) != NULL) {
3472 rq->bio = bio->bi_next;
3477 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3480 * Copy attributes of the original request to the clone request.
3481 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3483 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3485 dst->cpu = src->cpu;
3486 dst->__sector = blk_rq_pos(src);
3487 dst->__data_len = blk_rq_bytes(src);
3488 dst->nr_phys_segments = src->nr_phys_segments;
3489 dst->ioprio = src->ioprio;
3490 dst->extra_len = src->extra_len;
3494 * blk_rq_prep_clone - Helper function to setup clone request
3495 * @rq: the request to be setup
3496 * @rq_src: original request to be cloned
3497 * @bs: bio_set that bios for clone are allocated from
3498 * @gfp_mask: memory allocation mask for bio
3499 * @bio_ctr: setup function to be called for each clone bio.
3500 * Returns %0 for success, non %0 for failure.
3501 * @data: private data to be passed to @bio_ctr
3504 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3505 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3506 * are not copied, and copying such parts is the caller's responsibility.
3507 * Also, pages which the original bios are pointing to are not copied
3508 * and the cloned bios just point same pages.
3509 * So cloned bios must be completed before original bios, which means
3510 * the caller must complete @rq before @rq_src.
3512 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3513 struct bio_set *bs, gfp_t gfp_mask,
3514 int (*bio_ctr)(struct bio *, struct bio *, void *),
3517 struct bio *bio, *bio_src;
3522 __rq_for_each_bio(bio_src, rq_src) {
3523 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3527 if (bio_ctr && bio_ctr(bio, bio_src, data))
3531 rq->biotail->bi_next = bio;
3534 rq->bio = rq->biotail = bio;
3537 __blk_rq_prep_clone(rq, rq_src);
3544 blk_rq_unprep_clone(rq);
3548 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3550 int kblockd_schedule_work(struct work_struct *work)
3552 return queue_work(kblockd_workqueue, work);
3554 EXPORT_SYMBOL(kblockd_schedule_work);
3556 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3558 return queue_work_on(cpu, kblockd_workqueue, work);
3560 EXPORT_SYMBOL(kblockd_schedule_work_on);
3562 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3563 unsigned long delay)
3565 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3567 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3570 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3571 * @plug: The &struct blk_plug that needs to be initialized
3574 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3575 * pending I/O should the task end up blocking between blk_start_plug() and
3576 * blk_finish_plug(). This is important from a performance perspective, but
3577 * also ensures that we don't deadlock. For instance, if the task is blocking
3578 * for a memory allocation, memory reclaim could end up wanting to free a
3579 * page belonging to that request that is currently residing in our private
3580 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3581 * this kind of deadlock.
3583 void blk_start_plug(struct blk_plug *plug)
3585 struct task_struct *tsk = current;
3588 * If this is a nested plug, don't actually assign it.
3593 INIT_LIST_HEAD(&plug->list);
3594 INIT_LIST_HEAD(&plug->mq_list);
3595 INIT_LIST_HEAD(&plug->cb_list);
3597 * Store ordering should not be needed here, since a potential
3598 * preempt will imply a full memory barrier
3602 EXPORT_SYMBOL(blk_start_plug);
3604 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3606 struct request *rqa = container_of(a, struct request, queuelist);
3607 struct request *rqb = container_of(b, struct request, queuelist);
3609 return !(rqa->q < rqb->q ||
3610 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3614 * If 'from_schedule' is true, then postpone the dispatch of requests
3615 * until a safe kblockd context. We due this to avoid accidental big
3616 * additional stack usage in driver dispatch, in places where the originally
3617 * plugger did not intend it.
3619 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3621 __releases(q->queue_lock)
3623 lockdep_assert_held(q->queue_lock);
3625 trace_block_unplug(q, depth, !from_schedule);
3628 blk_run_queue_async(q);
3631 spin_unlock_irq(q->queue_lock);
3634 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3636 LIST_HEAD(callbacks);
3638 while (!list_empty(&plug->cb_list)) {
3639 list_splice_init(&plug->cb_list, &callbacks);
3641 while (!list_empty(&callbacks)) {
3642 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3645 list_del(&cb->list);
3646 cb->callback(cb, from_schedule);
3651 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3654 struct blk_plug *plug = current->plug;
3655 struct blk_plug_cb *cb;
3660 list_for_each_entry(cb, &plug->cb_list, list)
3661 if (cb->callback == unplug && cb->data == data)
3664 /* Not currently on the callback list */
3665 BUG_ON(size < sizeof(*cb));
3666 cb = kzalloc(size, GFP_ATOMIC);
3669 cb->callback = unplug;
3670 list_add(&cb->list, &plug->cb_list);
3674 EXPORT_SYMBOL(blk_check_plugged);
3676 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3678 struct request_queue *q;
3683 flush_plug_callbacks(plug, from_schedule);
3685 if (!list_empty(&plug->mq_list))
3686 blk_mq_flush_plug_list(plug, from_schedule);
3688 if (list_empty(&plug->list))
3691 list_splice_init(&plug->list, &list);
3693 list_sort(NULL, &list, plug_rq_cmp);
3698 while (!list_empty(&list)) {
3699 rq = list_entry_rq(list.next);
3700 list_del_init(&rq->queuelist);
3704 * This drops the queue lock
3707 queue_unplugged(q, depth, from_schedule);
3710 spin_lock_irq(q->queue_lock);
3714 * Short-circuit if @q is dead
3716 if (unlikely(blk_queue_dying(q))) {
3717 __blk_end_request_all(rq, BLK_STS_IOERR);
3722 * rq is already accounted, so use raw insert
3724 if (op_is_flush(rq->cmd_flags))
3725 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3727 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3733 * This drops the queue lock
3736 queue_unplugged(q, depth, from_schedule);
3739 void blk_finish_plug(struct blk_plug *plug)
3741 if (plug != current->plug)
3743 blk_flush_plug_list(plug, false);
3745 current->plug = NULL;
3747 EXPORT_SYMBOL(blk_finish_plug);
3751 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3752 * @q: the queue of the device
3753 * @dev: the device the queue belongs to
3756 * Initialize runtime-PM-related fields for @q and start auto suspend for
3757 * @dev. Drivers that want to take advantage of request-based runtime PM
3758 * should call this function after @dev has been initialized, and its
3759 * request queue @q has been allocated, and runtime PM for it can not happen
3760 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3761 * cases, driver should call this function before any I/O has taken place.
3763 * This function takes care of setting up using auto suspend for the device,
3764 * the autosuspend delay is set to -1 to make runtime suspend impossible
3765 * until an updated value is either set by user or by driver. Drivers do
3766 * not need to touch other autosuspend settings.
3768 * The block layer runtime PM is request based, so only works for drivers
3769 * that use request as their IO unit instead of those directly use bio's.
3771 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3773 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3778 q->rpm_status = RPM_ACTIVE;
3779 pm_runtime_set_autosuspend_delay(q->dev, -1);
3780 pm_runtime_use_autosuspend(q->dev);
3782 EXPORT_SYMBOL(blk_pm_runtime_init);
3785 * blk_pre_runtime_suspend - Pre runtime suspend check
3786 * @q: the queue of the device
3789 * This function will check if runtime suspend is allowed for the device
3790 * by examining if there are any requests pending in the queue. If there
3791 * are requests pending, the device can not be runtime suspended; otherwise,
3792 * the queue's status will be updated to SUSPENDING and the driver can
3793 * proceed to suspend the device.
3795 * For the not allowed case, we mark last busy for the device so that
3796 * runtime PM core will try to autosuspend it some time later.
3798 * This function should be called near the start of the device's
3799 * runtime_suspend callback.
3802 * 0 - OK to runtime suspend the device
3803 * -EBUSY - Device should not be runtime suspended
3805 int blk_pre_runtime_suspend(struct request_queue *q)
3812 spin_lock_irq(q->queue_lock);
3813 if (q->nr_pending) {
3815 pm_runtime_mark_last_busy(q->dev);
3817 q->rpm_status = RPM_SUSPENDING;
3819 spin_unlock_irq(q->queue_lock);
3822 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3825 * blk_post_runtime_suspend - Post runtime suspend processing
3826 * @q: the queue of the device
3827 * @err: return value of the device's runtime_suspend function
3830 * Update the queue's runtime status according to the return value of the
3831 * device's runtime suspend function and mark last busy for the device so
3832 * that PM core will try to auto suspend the device at a later time.
3834 * This function should be called near the end of the device's
3835 * runtime_suspend callback.
3837 void blk_post_runtime_suspend(struct request_queue *q, int err)
3842 spin_lock_irq(q->queue_lock);
3844 q->rpm_status = RPM_SUSPENDED;
3846 q->rpm_status = RPM_ACTIVE;
3847 pm_runtime_mark_last_busy(q->dev);
3849 spin_unlock_irq(q->queue_lock);
3851 EXPORT_SYMBOL(blk_post_runtime_suspend);
3854 * blk_pre_runtime_resume - Pre runtime resume processing
3855 * @q: the queue of the device
3858 * Update the queue's runtime status to RESUMING in preparation for the
3859 * runtime resume of the device.
3861 * This function should be called near the start of the device's
3862 * runtime_resume callback.
3864 void blk_pre_runtime_resume(struct request_queue *q)
3869 spin_lock_irq(q->queue_lock);
3870 q->rpm_status = RPM_RESUMING;
3871 spin_unlock_irq(q->queue_lock);
3873 EXPORT_SYMBOL(blk_pre_runtime_resume);
3876 * blk_post_runtime_resume - Post runtime resume processing
3877 * @q: the queue of the device
3878 * @err: return value of the device's runtime_resume function
3881 * Update the queue's runtime status according to the return value of the
3882 * device's runtime_resume function. If it is successfully resumed, process
3883 * the requests that are queued into the device's queue when it is resuming
3884 * and then mark last busy and initiate autosuspend for it.
3886 * This function should be called near the end of the device's
3887 * runtime_resume callback.
3889 void blk_post_runtime_resume(struct request_queue *q, int err)
3894 spin_lock_irq(q->queue_lock);
3896 q->rpm_status = RPM_ACTIVE;
3898 pm_runtime_mark_last_busy(q->dev);
3899 pm_request_autosuspend(q->dev);
3901 q->rpm_status = RPM_SUSPENDED;
3903 spin_unlock_irq(q->queue_lock);
3905 EXPORT_SYMBOL(blk_post_runtime_resume);
3908 * blk_set_runtime_active - Force runtime status of the queue to be active
3909 * @q: the queue of the device
3911 * If the device is left runtime suspended during system suspend the resume
3912 * hook typically resumes the device and corrects runtime status
3913 * accordingly. However, that does not affect the queue runtime PM status
3914 * which is still "suspended". This prevents processing requests from the
3917 * This function can be used in driver's resume hook to correct queue
3918 * runtime PM status and re-enable peeking requests from the queue. It
3919 * should be called before first request is added to the queue.
3921 void blk_set_runtime_active(struct request_queue *q)
3923 spin_lock_irq(q->queue_lock);
3924 q->rpm_status = RPM_ACTIVE;
3925 pm_runtime_mark_last_busy(q->dev);
3926 pm_request_autosuspend(q->dev);
3927 spin_unlock_irq(q->queue_lock);
3929 EXPORT_SYMBOL(blk_set_runtime_active);
3932 int __init blk_dev_init(void)
3934 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3935 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3936 FIELD_SIZEOF(struct request, cmd_flags));
3937 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3938 FIELD_SIZEOF(struct bio, bi_opf));
3940 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3941 kblockd_workqueue = alloc_workqueue("kblockd",
3942 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3943 if (!kblockd_workqueue)
3944 panic("Failed to create kblockd\n");
3946 request_cachep = kmem_cache_create("blkdev_requests",
3947 sizeof(struct request), 0, SLAB_PANIC, NULL);
3949 blk_requestq_cachep = kmem_cache_create("request_queue",
3950 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3952 #ifdef CONFIG_DEBUG_FS
3953 blk_debugfs_root = debugfs_create_dir("block", NULL);