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 = jiffies;
200 set_start_time_ns(rq);
202 seqcount_init(&rq->gstate_seq);
203 u64_stats_init(&rq->aborted_gstate_sync);
205 EXPORT_SYMBOL(blk_rq_init);
207 static const struct {
211 [BLK_STS_OK] = { 0, "" },
212 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
213 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
214 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
215 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
216 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
217 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
218 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
219 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
220 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
221 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
222 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
224 /* device mapper special case, should not leak out: */
225 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
227 /* everything else not covered above: */
228 [BLK_STS_IOERR] = { -EIO, "I/O" },
231 blk_status_t errno_to_blk_status(int errno)
235 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
236 if (blk_errors[i].errno == errno)
237 return (__force blk_status_t)i;
240 return BLK_STS_IOERR;
242 EXPORT_SYMBOL_GPL(errno_to_blk_status);
244 int blk_status_to_errno(blk_status_t status)
246 int idx = (__force int)status;
248 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
250 return blk_errors[idx].errno;
252 EXPORT_SYMBOL_GPL(blk_status_to_errno);
254 static void print_req_error(struct request *req, blk_status_t status)
256 int idx = (__force int)status;
258 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
261 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
262 __func__, blk_errors[idx].name, req->rq_disk ?
263 req->rq_disk->disk_name : "?",
264 (unsigned long long)blk_rq_pos(req));
267 static void req_bio_endio(struct request *rq, struct bio *bio,
268 unsigned int nbytes, blk_status_t error)
271 bio->bi_status = error;
273 if (unlikely(rq->rq_flags & RQF_QUIET))
274 bio_set_flag(bio, BIO_QUIET);
276 bio_advance(bio, nbytes);
278 /* don't actually finish bio if it's part of flush sequence */
279 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
283 void blk_dump_rq_flags(struct request *rq, char *msg)
285 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
286 rq->rq_disk ? rq->rq_disk->disk_name : "?",
287 (unsigned long long) rq->cmd_flags);
289 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
290 (unsigned long long)blk_rq_pos(rq),
291 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
292 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
293 rq->bio, rq->biotail, blk_rq_bytes(rq));
295 EXPORT_SYMBOL(blk_dump_rq_flags);
297 static void blk_delay_work(struct work_struct *work)
299 struct request_queue *q;
301 q = container_of(work, struct request_queue, delay_work.work);
302 spin_lock_irq(q->queue_lock);
304 spin_unlock_irq(q->queue_lock);
308 * blk_delay_queue - restart queueing after defined interval
309 * @q: The &struct request_queue in question
310 * @msecs: Delay in msecs
313 * Sometimes queueing needs to be postponed for a little while, to allow
314 * resources to come back. This function will make sure that queueing is
315 * restarted around the specified time.
317 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
319 lockdep_assert_held(q->queue_lock);
320 WARN_ON_ONCE(q->mq_ops);
322 if (likely(!blk_queue_dead(q)))
323 queue_delayed_work(kblockd_workqueue, &q->delay_work,
324 msecs_to_jiffies(msecs));
326 EXPORT_SYMBOL(blk_delay_queue);
329 * blk_start_queue_async - asynchronously restart a previously stopped queue
330 * @q: The &struct request_queue in question
333 * blk_start_queue_async() will clear the stop flag on the queue, and
334 * ensure that the request_fn for the queue is run from an async
337 void blk_start_queue_async(struct request_queue *q)
339 lockdep_assert_held(q->queue_lock);
340 WARN_ON_ONCE(q->mq_ops);
342 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
343 blk_run_queue_async(q);
345 EXPORT_SYMBOL(blk_start_queue_async);
348 * blk_start_queue - restart a previously stopped queue
349 * @q: The &struct request_queue in question
352 * blk_start_queue() will clear the stop flag on the queue, and call
353 * the request_fn for the queue if it was in a stopped state when
354 * entered. Also see blk_stop_queue().
356 void blk_start_queue(struct request_queue *q)
358 lockdep_assert_held(q->queue_lock);
359 WARN_ON(!in_interrupt() && !irqs_disabled());
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;
921 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
923 * The code that sets the PREEMPT_ONLY flag is
924 * responsible for ensuring that that flag is globally
925 * visible before the queue is unfrozen.
927 if (preempt || !blk_queue_preempt_only(q)) {
930 percpu_ref_put(&q->q_usage_counter);
938 if (flags & BLK_MQ_REQ_NOWAIT)
942 * read pair of barrier in blk_freeze_queue_start(),
943 * we need to order reading __PERCPU_REF_DEAD flag of
944 * .q_usage_counter and reading .mq_freeze_depth or
945 * queue dying flag, otherwise the following wait may
946 * never return if the two reads are reordered.
950 ret = wait_event_interruptible(q->mq_freeze_wq,
951 (atomic_read(&q->mq_freeze_depth) == 0 &&
952 (preempt || !blk_queue_preempt_only(q))) ||
954 if (blk_queue_dying(q))
961 void blk_queue_exit(struct request_queue *q)
963 percpu_ref_put(&q->q_usage_counter);
966 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
968 struct request_queue *q =
969 container_of(ref, struct request_queue, q_usage_counter);
971 wake_up_all(&q->mq_freeze_wq);
974 static void blk_rq_timed_out_timer(struct timer_list *t)
976 struct request_queue *q = from_timer(q, t, timeout);
978 kblockd_schedule_work(&q->timeout_work);
982 * blk_alloc_queue_node - allocate a request queue
983 * @gfp_mask: memory allocation flags
984 * @node_id: NUMA node to allocate memory from
985 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
986 * serialize calls to the legacy .request_fn() callback. Ignored for
987 * blk-mq request queues.
989 * Note: pass the queue lock as the third argument to this function instead of
990 * setting the queue lock pointer explicitly to avoid triggering a sporadic
991 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
992 * the queue lock pointer must be set before blkcg_init_queue() is called.
994 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
997 struct request_queue *q;
999 q = kmem_cache_alloc_node(blk_requestq_cachep,
1000 gfp_mask | __GFP_ZERO, node_id);
1004 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
1008 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1012 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1013 if (!q->backing_dev_info)
1016 q->stats = blk_alloc_queue_stats();
1020 q->backing_dev_info->ra_pages =
1021 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1022 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1023 q->backing_dev_info->name = "block";
1026 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1027 laptop_mode_timer_fn, 0);
1028 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
1029 INIT_WORK(&q->timeout_work, NULL);
1030 INIT_LIST_HEAD(&q->queue_head);
1031 INIT_LIST_HEAD(&q->timeout_list);
1032 INIT_LIST_HEAD(&q->icq_list);
1033 #ifdef CONFIG_BLK_CGROUP
1034 INIT_LIST_HEAD(&q->blkg_list);
1036 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
1038 kobject_init(&q->kobj, &blk_queue_ktype);
1040 #ifdef CONFIG_BLK_DEV_IO_TRACE
1041 mutex_init(&q->blk_trace_mutex);
1043 mutex_init(&q->sysfs_lock);
1044 spin_lock_init(&q->__queue_lock);
1047 q->queue_lock = lock ? : &q->__queue_lock;
1050 * A queue starts its life with bypass turned on to avoid
1051 * unnecessary bypass on/off overhead and nasty surprises during
1052 * init. The initial bypass will be finished when the queue is
1053 * registered by blk_register_queue().
1055 q->bypass_depth = 1;
1056 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
1058 init_waitqueue_head(&q->mq_freeze_wq);
1061 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1062 * See blk_register_queue() for details.
1064 if (percpu_ref_init(&q->q_usage_counter,
1065 blk_queue_usage_counter_release,
1066 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
1069 if (blkcg_init_queue(q))
1075 percpu_ref_exit(&q->q_usage_counter);
1077 blk_free_queue_stats(q->stats);
1079 bdi_put(q->backing_dev_info);
1081 bioset_free(q->bio_split);
1083 ida_simple_remove(&blk_queue_ida, q->id);
1085 kmem_cache_free(blk_requestq_cachep, q);
1088 EXPORT_SYMBOL(blk_alloc_queue_node);
1091 * blk_init_queue - prepare a request queue for use with a block device
1092 * @rfn: The function to be called to process requests that have been
1093 * placed on the queue.
1094 * @lock: Request queue spin lock
1097 * If a block device wishes to use the standard request handling procedures,
1098 * which sorts requests and coalesces adjacent requests, then it must
1099 * call blk_init_queue(). The function @rfn will be called when there
1100 * are requests on the queue that need to be processed. If the device
1101 * supports plugging, then @rfn may not be called immediately when requests
1102 * are available on the queue, but may be called at some time later instead.
1103 * Plugged queues are generally unplugged when a buffer belonging to one
1104 * of the requests on the queue is needed, or due to memory pressure.
1106 * @rfn is not required, or even expected, to remove all requests off the
1107 * queue, but only as many as it can handle at a time. If it does leave
1108 * requests on the queue, it is responsible for arranging that the requests
1109 * get dealt with eventually.
1111 * The queue spin lock must be held while manipulating the requests on the
1112 * request queue; this lock will be taken also from interrupt context, so irq
1113 * disabling is needed for it.
1115 * Function returns a pointer to the initialized request queue, or %NULL if
1116 * it didn't succeed.
1119 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1120 * when the block device is deactivated (such as at module unload).
1123 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1125 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1127 EXPORT_SYMBOL(blk_init_queue);
1129 struct request_queue *
1130 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1132 struct request_queue *q;
1134 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
1138 q->request_fn = rfn;
1139 if (blk_init_allocated_queue(q) < 0) {
1140 blk_cleanup_queue(q);
1146 EXPORT_SYMBOL(blk_init_queue_node);
1148 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1151 int blk_init_allocated_queue(struct request_queue *q)
1153 WARN_ON_ONCE(q->mq_ops);
1155 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
1159 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1160 goto out_free_flush_queue;
1162 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1163 goto out_exit_flush_rq;
1165 INIT_WORK(&q->timeout_work, blk_timeout_work);
1166 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1169 * This also sets hw/phys segments, boundary and size
1171 blk_queue_make_request(q, blk_queue_bio);
1173 q->sg_reserved_size = INT_MAX;
1175 /* Protect q->elevator from elevator_change */
1176 mutex_lock(&q->sysfs_lock);
1179 if (elevator_init(q, NULL)) {
1180 mutex_unlock(&q->sysfs_lock);
1181 goto out_exit_flush_rq;
1184 mutex_unlock(&q->sysfs_lock);
1189 q->exit_rq_fn(q, q->fq->flush_rq);
1190 out_free_flush_queue:
1191 blk_free_flush_queue(q->fq);
1194 EXPORT_SYMBOL(blk_init_allocated_queue);
1196 bool blk_get_queue(struct request_queue *q)
1198 if (likely(!blk_queue_dying(q))) {
1205 EXPORT_SYMBOL(blk_get_queue);
1207 static inline void blk_free_request(struct request_list *rl, struct request *rq)
1209 if (rq->rq_flags & RQF_ELVPRIV) {
1210 elv_put_request(rl->q, rq);
1212 put_io_context(rq->elv.icq->ioc);
1215 mempool_free(rq, rl->rq_pool);
1219 * ioc_batching returns true if the ioc is a valid batching request and
1220 * should be given priority access to a request.
1222 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1228 * Make sure the process is able to allocate at least 1 request
1229 * even if the batch times out, otherwise we could theoretically
1232 return ioc->nr_batch_requests == q->nr_batching ||
1233 (ioc->nr_batch_requests > 0
1234 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1238 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1239 * will cause the process to be a "batcher" on all queues in the system. This
1240 * is the behaviour we want though - once it gets a wakeup it should be given
1243 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1245 if (!ioc || ioc_batching(q, ioc))
1248 ioc->nr_batch_requests = q->nr_batching;
1249 ioc->last_waited = jiffies;
1252 static void __freed_request(struct request_list *rl, int sync)
1254 struct request_queue *q = rl->q;
1256 if (rl->count[sync] < queue_congestion_off_threshold(q))
1257 blk_clear_congested(rl, sync);
1259 if (rl->count[sync] + 1 <= q->nr_requests) {
1260 if (waitqueue_active(&rl->wait[sync]))
1261 wake_up(&rl->wait[sync]);
1263 blk_clear_rl_full(rl, sync);
1268 * A request has just been released. Account for it, update the full and
1269 * congestion status, wake up any waiters. Called under q->queue_lock.
1271 static void freed_request(struct request_list *rl, bool sync,
1272 req_flags_t rq_flags)
1274 struct request_queue *q = rl->q;
1278 if (rq_flags & RQF_ELVPRIV)
1279 q->nr_rqs_elvpriv--;
1281 __freed_request(rl, sync);
1283 if (unlikely(rl->starved[sync ^ 1]))
1284 __freed_request(rl, sync ^ 1);
1287 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1289 struct request_list *rl;
1290 int on_thresh, off_thresh;
1292 WARN_ON_ONCE(q->mq_ops);
1294 spin_lock_irq(q->queue_lock);
1295 q->nr_requests = nr;
1296 blk_queue_congestion_threshold(q);
1297 on_thresh = queue_congestion_on_threshold(q);
1298 off_thresh = queue_congestion_off_threshold(q);
1300 blk_queue_for_each_rl(rl, q) {
1301 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1302 blk_set_congested(rl, BLK_RW_SYNC);
1303 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1304 blk_clear_congested(rl, BLK_RW_SYNC);
1306 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1307 blk_set_congested(rl, BLK_RW_ASYNC);
1308 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1309 blk_clear_congested(rl, BLK_RW_ASYNC);
1311 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1312 blk_set_rl_full(rl, BLK_RW_SYNC);
1314 blk_clear_rl_full(rl, BLK_RW_SYNC);
1315 wake_up(&rl->wait[BLK_RW_SYNC]);
1318 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1319 blk_set_rl_full(rl, BLK_RW_ASYNC);
1321 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1322 wake_up(&rl->wait[BLK_RW_ASYNC]);
1326 spin_unlock_irq(q->queue_lock);
1331 * __get_request - get a free request
1332 * @rl: request list to allocate from
1333 * @op: operation and flags
1334 * @bio: bio to allocate request for (can be %NULL)
1335 * @flags: BLQ_MQ_REQ_* flags
1337 * Get a free request from @q. This function may fail under memory
1338 * pressure or if @q is dead.
1340 * Must be called with @q->queue_lock held and,
1341 * Returns ERR_PTR on failure, with @q->queue_lock held.
1342 * Returns request pointer on success, with @q->queue_lock *not held*.
1344 static struct request *__get_request(struct request_list *rl, unsigned int op,
1345 struct bio *bio, blk_mq_req_flags_t flags)
1347 struct request_queue *q = rl->q;
1349 struct elevator_type *et = q->elevator->type;
1350 struct io_context *ioc = rq_ioc(bio);
1351 struct io_cq *icq = NULL;
1352 const bool is_sync = op_is_sync(op);
1354 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1355 __GFP_DIRECT_RECLAIM;
1356 req_flags_t rq_flags = RQF_ALLOCED;
1358 lockdep_assert_held(q->queue_lock);
1360 if (unlikely(blk_queue_dying(q)))
1361 return ERR_PTR(-ENODEV);
1363 may_queue = elv_may_queue(q, op);
1364 if (may_queue == ELV_MQUEUE_NO)
1367 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1368 if (rl->count[is_sync]+1 >= q->nr_requests) {
1370 * The queue will fill after this allocation, so set
1371 * it as full, and mark this process as "batching".
1372 * This process will be allowed to complete a batch of
1373 * requests, others will be blocked.
1375 if (!blk_rl_full(rl, is_sync)) {
1376 ioc_set_batching(q, ioc);
1377 blk_set_rl_full(rl, is_sync);
1379 if (may_queue != ELV_MQUEUE_MUST
1380 && !ioc_batching(q, ioc)) {
1382 * The queue is full and the allocating
1383 * process is not a "batcher", and not
1384 * exempted by the IO scheduler
1386 return ERR_PTR(-ENOMEM);
1390 blk_set_congested(rl, is_sync);
1394 * Only allow batching queuers to allocate up to 50% over the defined
1395 * limit of requests, otherwise we could have thousands of requests
1396 * allocated with any setting of ->nr_requests
1398 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1399 return ERR_PTR(-ENOMEM);
1401 q->nr_rqs[is_sync]++;
1402 rl->count[is_sync]++;
1403 rl->starved[is_sync] = 0;
1406 * Decide whether the new request will be managed by elevator. If
1407 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1408 * prevent the current elevator from being destroyed until the new
1409 * request is freed. This guarantees icq's won't be destroyed and
1410 * makes creating new ones safe.
1412 * Flush requests do not use the elevator so skip initialization.
1413 * This allows a request to share the flush and elevator data.
1415 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1416 * it will be created after releasing queue_lock.
1418 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1419 rq_flags |= RQF_ELVPRIV;
1420 q->nr_rqs_elvpriv++;
1421 if (et->icq_cache && ioc)
1422 icq = ioc_lookup_icq(ioc, q);
1425 if (blk_queue_io_stat(q))
1426 rq_flags |= RQF_IO_STAT;
1427 spin_unlock_irq(q->queue_lock);
1429 /* allocate and init request */
1430 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1435 blk_rq_set_rl(rq, rl);
1437 rq->rq_flags = rq_flags;
1438 if (flags & BLK_MQ_REQ_PREEMPT)
1439 rq->rq_flags |= RQF_PREEMPT;
1442 if (rq_flags & RQF_ELVPRIV) {
1443 if (unlikely(et->icq_cache && !icq)) {
1445 icq = ioc_create_icq(ioc, q, gfp_mask);
1451 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1454 /* @rq->elv.icq holds io_context until @rq is freed */
1456 get_io_context(icq->ioc);
1460 * ioc may be NULL here, and ioc_batching will be false. That's
1461 * OK, if the queue is under the request limit then requests need
1462 * not count toward the nr_batch_requests limit. There will always
1463 * be some limit enforced by BLK_BATCH_TIME.
1465 if (ioc_batching(q, ioc))
1466 ioc->nr_batch_requests--;
1468 trace_block_getrq(q, bio, op);
1473 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1474 * and may fail indefinitely under memory pressure and thus
1475 * shouldn't stall IO. Treat this request as !elvpriv. This will
1476 * disturb iosched and blkcg but weird is bettern than dead.
1478 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1479 __func__, dev_name(q->backing_dev_info->dev));
1481 rq->rq_flags &= ~RQF_ELVPRIV;
1484 spin_lock_irq(q->queue_lock);
1485 q->nr_rqs_elvpriv--;
1486 spin_unlock_irq(q->queue_lock);
1491 * Allocation failed presumably due to memory. Undo anything we
1492 * might have messed up.
1494 * Allocating task should really be put onto the front of the wait
1495 * queue, but this is pretty rare.
1497 spin_lock_irq(q->queue_lock);
1498 freed_request(rl, is_sync, rq_flags);
1501 * in the very unlikely event that allocation failed and no
1502 * requests for this direction was pending, mark us starved so that
1503 * freeing of a request in the other direction will notice
1504 * us. another possible fix would be to split the rq mempool into
1508 if (unlikely(rl->count[is_sync] == 0))
1509 rl->starved[is_sync] = 1;
1510 return ERR_PTR(-ENOMEM);
1514 * get_request - get a free request
1515 * @q: request_queue to allocate request from
1516 * @op: operation and flags
1517 * @bio: bio to allocate request for (can be %NULL)
1518 * @flags: BLK_MQ_REQ_* flags.
1520 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1521 * this function keeps retrying under memory pressure and fails iff @q is dead.
1523 * Must be called with @q->queue_lock held and,
1524 * Returns ERR_PTR on failure, with @q->queue_lock held.
1525 * Returns request pointer on success, with @q->queue_lock *not held*.
1527 static struct request *get_request(struct request_queue *q, unsigned int op,
1528 struct bio *bio, blk_mq_req_flags_t flags)
1530 const bool is_sync = op_is_sync(op);
1532 struct request_list *rl;
1535 lockdep_assert_held(q->queue_lock);
1536 WARN_ON_ONCE(q->mq_ops);
1538 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1540 rq = __get_request(rl, op, bio, flags);
1544 if (op & REQ_NOWAIT) {
1546 return ERR_PTR(-EAGAIN);
1549 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1554 /* wait on @rl and retry */
1555 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1556 TASK_UNINTERRUPTIBLE);
1558 trace_block_sleeprq(q, bio, op);
1560 spin_unlock_irq(q->queue_lock);
1564 * After sleeping, we become a "batching" process and will be able
1565 * to allocate at least one request, and up to a big batch of them
1566 * for a small period time. See ioc_batching, ioc_set_batching
1568 ioc_set_batching(q, current->io_context);
1570 spin_lock_irq(q->queue_lock);
1571 finish_wait(&rl->wait[is_sync], &wait);
1576 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1577 static struct request *blk_old_get_request(struct request_queue *q,
1578 unsigned int op, blk_mq_req_flags_t flags)
1581 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1582 __GFP_DIRECT_RECLAIM;
1585 WARN_ON_ONCE(q->mq_ops);
1587 /* create ioc upfront */
1588 create_io_context(gfp_mask, q->node);
1590 ret = blk_queue_enter(q, flags);
1592 return ERR_PTR(ret);
1593 spin_lock_irq(q->queue_lock);
1594 rq = get_request(q, op, NULL, flags);
1596 spin_unlock_irq(q->queue_lock);
1601 /* q->queue_lock is unlocked at this point */
1603 rq->__sector = (sector_t) -1;
1604 rq->bio = rq->biotail = NULL;
1609 * blk_get_request_flags - allocate a request
1610 * @q: request queue to allocate a request for
1611 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1612 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1614 struct request *blk_get_request_flags(struct request_queue *q, unsigned int op,
1615 blk_mq_req_flags_t flags)
1617 struct request *req;
1619 WARN_ON_ONCE(op & REQ_NOWAIT);
1620 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1623 req = blk_mq_alloc_request(q, op, flags);
1624 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1625 q->mq_ops->initialize_rq_fn(req);
1627 req = blk_old_get_request(q, op, flags);
1628 if (!IS_ERR(req) && q->initialize_rq_fn)
1629 q->initialize_rq_fn(req);
1634 EXPORT_SYMBOL(blk_get_request_flags);
1636 struct request *blk_get_request(struct request_queue *q, unsigned int op,
1639 return blk_get_request_flags(q, op, gfp_mask & __GFP_DIRECT_RECLAIM ?
1640 0 : BLK_MQ_REQ_NOWAIT);
1642 EXPORT_SYMBOL(blk_get_request);
1645 * blk_requeue_request - put a request back on queue
1646 * @q: request queue where request should be inserted
1647 * @rq: request to be inserted
1650 * Drivers often keep queueing requests until the hardware cannot accept
1651 * more, when that condition happens we need to put the request back
1652 * on the queue. Must be called with queue lock held.
1654 void blk_requeue_request(struct request_queue *q, struct request *rq)
1656 lockdep_assert_held(q->queue_lock);
1657 WARN_ON_ONCE(q->mq_ops);
1659 blk_delete_timer(rq);
1660 blk_clear_rq_complete(rq);
1661 trace_block_rq_requeue(q, rq);
1662 wbt_requeue(q->rq_wb, &rq->issue_stat);
1664 if (rq->rq_flags & RQF_QUEUED)
1665 blk_queue_end_tag(q, rq);
1667 BUG_ON(blk_queued_rq(rq));
1669 elv_requeue_request(q, rq);
1671 EXPORT_SYMBOL(blk_requeue_request);
1673 static void add_acct_request(struct request_queue *q, struct request *rq,
1676 blk_account_io_start(rq, true);
1677 __elv_add_request(q, rq, where);
1680 static void part_round_stats_single(struct request_queue *q, int cpu,
1681 struct hd_struct *part, unsigned long now,
1682 unsigned int inflight)
1685 __part_stat_add(cpu, part, time_in_queue,
1686 inflight * (now - part->stamp));
1687 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1693 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1694 * @q: target block queue
1695 * @cpu: cpu number for stats access
1696 * @part: target partition
1698 * The average IO queue length and utilisation statistics are maintained
1699 * by observing the current state of the queue length and the amount of
1700 * time it has been in this state for.
1702 * Normally, that accounting is done on IO completion, but that can result
1703 * in more than a second's worth of IO being accounted for within any one
1704 * second, leading to >100% utilisation. To deal with that, we call this
1705 * function to do a round-off before returning the results when reading
1706 * /proc/diskstats. This accounts immediately for all queue usage up to
1707 * the current jiffies and restarts the counters again.
1709 void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1711 struct hd_struct *part2 = NULL;
1712 unsigned long now = jiffies;
1713 unsigned int inflight[2];
1716 if (part->stamp != now)
1720 part2 = &part_to_disk(part)->part0;
1721 if (part2->stamp != now)
1728 part_in_flight(q, part, inflight);
1731 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1733 part_round_stats_single(q, cpu, part, now, inflight[0]);
1735 EXPORT_SYMBOL_GPL(part_round_stats);
1738 static void blk_pm_put_request(struct request *rq)
1740 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1741 pm_runtime_mark_last_busy(rq->q->dev);
1744 static inline void blk_pm_put_request(struct request *rq) {}
1747 void __blk_put_request(struct request_queue *q, struct request *req)
1749 req_flags_t rq_flags = req->rq_flags;
1755 blk_mq_free_request(req);
1759 lockdep_assert_held(q->queue_lock);
1761 blk_req_zone_write_unlock(req);
1762 blk_pm_put_request(req);
1764 elv_completed_request(q, req);
1766 /* this is a bio leak */
1767 WARN_ON(req->bio != NULL);
1769 wbt_done(q->rq_wb, &req->issue_stat);
1772 * Request may not have originated from ll_rw_blk. if not,
1773 * it didn't come out of our reserved rq pools
1775 if (rq_flags & RQF_ALLOCED) {
1776 struct request_list *rl = blk_rq_rl(req);
1777 bool sync = op_is_sync(req->cmd_flags);
1779 BUG_ON(!list_empty(&req->queuelist));
1780 BUG_ON(ELV_ON_HASH(req));
1782 blk_free_request(rl, req);
1783 freed_request(rl, sync, rq_flags);
1788 EXPORT_SYMBOL_GPL(__blk_put_request);
1790 void blk_put_request(struct request *req)
1792 struct request_queue *q = req->q;
1795 blk_mq_free_request(req);
1797 unsigned long flags;
1799 spin_lock_irqsave(q->queue_lock, flags);
1800 __blk_put_request(q, req);
1801 spin_unlock_irqrestore(q->queue_lock, flags);
1804 EXPORT_SYMBOL(blk_put_request);
1806 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1809 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1811 if (!ll_back_merge_fn(q, req, bio))
1814 trace_block_bio_backmerge(q, req, bio);
1816 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1817 blk_rq_set_mixed_merge(req);
1819 req->biotail->bi_next = bio;
1821 req->__data_len += bio->bi_iter.bi_size;
1822 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1824 blk_account_io_start(req, false);
1828 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1831 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1833 if (!ll_front_merge_fn(q, req, bio))
1836 trace_block_bio_frontmerge(q, req, bio);
1838 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1839 blk_rq_set_mixed_merge(req);
1841 bio->bi_next = req->bio;
1844 req->__sector = bio->bi_iter.bi_sector;
1845 req->__data_len += bio->bi_iter.bi_size;
1846 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1848 blk_account_io_start(req, false);
1852 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1855 unsigned short segments = blk_rq_nr_discard_segments(req);
1857 if (segments >= queue_max_discard_segments(q))
1859 if (blk_rq_sectors(req) + bio_sectors(bio) >
1860 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1863 req->biotail->bi_next = bio;
1865 req->__data_len += bio->bi_iter.bi_size;
1866 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1867 req->nr_phys_segments = segments + 1;
1869 blk_account_io_start(req, false);
1872 req_set_nomerge(q, req);
1877 * blk_attempt_plug_merge - try to merge with %current's plugged list
1878 * @q: request_queue new bio is being queued at
1879 * @bio: new bio being queued
1880 * @request_count: out parameter for number of traversed plugged requests
1881 * @same_queue_rq: pointer to &struct request that gets filled in when
1882 * another request associated with @q is found on the plug list
1883 * (optional, may be %NULL)
1885 * Determine whether @bio being queued on @q can be merged with a request
1886 * on %current's plugged list. Returns %true if merge was successful,
1889 * Plugging coalesces IOs from the same issuer for the same purpose without
1890 * going through @q->queue_lock. As such it's more of an issuing mechanism
1891 * than scheduling, and the request, while may have elvpriv data, is not
1892 * added on the elevator at this point. In addition, we don't have
1893 * reliable access to the elevator outside queue lock. Only check basic
1894 * merging parameters without querying the elevator.
1896 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1898 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1899 unsigned int *request_count,
1900 struct request **same_queue_rq)
1902 struct blk_plug *plug;
1904 struct list_head *plug_list;
1906 plug = current->plug;
1912 plug_list = &plug->mq_list;
1914 plug_list = &plug->list;
1916 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1917 bool merged = false;
1922 * Only blk-mq multiple hardware queues case checks the
1923 * rq in the same queue, there should be only one such
1927 *same_queue_rq = rq;
1930 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1933 switch (blk_try_merge(rq, bio)) {
1934 case ELEVATOR_BACK_MERGE:
1935 merged = bio_attempt_back_merge(q, rq, bio);
1937 case ELEVATOR_FRONT_MERGE:
1938 merged = bio_attempt_front_merge(q, rq, bio);
1940 case ELEVATOR_DISCARD_MERGE:
1941 merged = bio_attempt_discard_merge(q, rq, bio);
1954 unsigned int blk_plug_queued_count(struct request_queue *q)
1956 struct blk_plug *plug;
1958 struct list_head *plug_list;
1959 unsigned int ret = 0;
1961 plug = current->plug;
1966 plug_list = &plug->mq_list;
1968 plug_list = &plug->list;
1970 list_for_each_entry(rq, plug_list, queuelist) {
1978 void blk_init_request_from_bio(struct request *req, struct bio *bio)
1980 struct io_context *ioc = rq_ioc(bio);
1982 if (bio->bi_opf & REQ_RAHEAD)
1983 req->cmd_flags |= REQ_FAILFAST_MASK;
1985 req->__sector = bio->bi_iter.bi_sector;
1986 if (ioprio_valid(bio_prio(bio)))
1987 req->ioprio = bio_prio(bio);
1989 req->ioprio = ioc->ioprio;
1991 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1992 req->write_hint = bio->bi_write_hint;
1993 blk_rq_bio_prep(req->q, req, bio);
1995 EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1997 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1999 struct blk_plug *plug;
2000 int where = ELEVATOR_INSERT_SORT;
2001 struct request *req, *free;
2002 unsigned int request_count = 0;
2003 unsigned int wb_acct;
2006 * low level driver can indicate that it wants pages above a
2007 * certain limit bounced to low memory (ie for highmem, or even
2008 * ISA dma in theory)
2010 blk_queue_bounce(q, &bio);
2012 blk_queue_split(q, &bio);
2014 if (!bio_integrity_prep(bio))
2015 return BLK_QC_T_NONE;
2017 if (op_is_flush(bio->bi_opf)) {
2018 spin_lock_irq(q->queue_lock);
2019 where = ELEVATOR_INSERT_FLUSH;
2024 * Check if we can merge with the plugged list before grabbing
2027 if (!blk_queue_nomerges(q)) {
2028 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
2029 return BLK_QC_T_NONE;
2031 request_count = blk_plug_queued_count(q);
2033 spin_lock_irq(q->queue_lock);
2035 switch (elv_merge(q, &req, bio)) {
2036 case ELEVATOR_BACK_MERGE:
2037 if (!bio_attempt_back_merge(q, req, bio))
2039 elv_bio_merged(q, req, bio);
2040 free = attempt_back_merge(q, req);
2042 __blk_put_request(q, free);
2044 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2046 case ELEVATOR_FRONT_MERGE:
2047 if (!bio_attempt_front_merge(q, req, bio))
2049 elv_bio_merged(q, req, bio);
2050 free = attempt_front_merge(q, req);
2052 __blk_put_request(q, free);
2054 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2061 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
2064 * Grab a free request. This is might sleep but can not fail.
2065 * Returns with the queue unlocked.
2067 blk_queue_enter_live(q);
2068 req = get_request(q, bio->bi_opf, bio, 0);
2071 __wbt_done(q->rq_wb, wb_acct);
2072 if (PTR_ERR(req) == -ENOMEM)
2073 bio->bi_status = BLK_STS_RESOURCE;
2075 bio->bi_status = BLK_STS_IOERR;
2080 wbt_track(&req->issue_stat, wb_acct);
2083 * After dropping the lock and possibly sleeping here, our request
2084 * may now be mergeable after it had proven unmergeable (above).
2085 * We don't worry about that case for efficiency. It won't happen
2086 * often, and the elevators are able to handle it.
2088 blk_init_request_from_bio(req, bio);
2090 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
2091 req->cpu = raw_smp_processor_id();
2093 plug = current->plug;
2096 * If this is the first request added after a plug, fire
2099 * @request_count may become stale because of schedule
2100 * out, so check plug list again.
2102 if (!request_count || list_empty(&plug->list))
2103 trace_block_plug(q);
2105 struct request *last = list_entry_rq(plug->list.prev);
2106 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2107 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2108 blk_flush_plug_list(plug, false);
2109 trace_block_plug(q);
2112 list_add_tail(&req->queuelist, &plug->list);
2113 blk_account_io_start(req, true);
2115 spin_lock_irq(q->queue_lock);
2116 add_acct_request(q, req, where);
2119 spin_unlock_irq(q->queue_lock);
2122 return BLK_QC_T_NONE;
2125 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
2127 char b[BDEVNAME_SIZE];
2129 printk(KERN_INFO "attempt to access beyond end of device\n");
2130 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2131 bio_devname(bio, b), bio->bi_opf,
2132 (unsigned long long)bio_end_sector(bio),
2133 (long long)maxsector);
2136 #ifdef CONFIG_FAIL_MAKE_REQUEST
2138 static DECLARE_FAULT_ATTR(fail_make_request);
2140 static int __init setup_fail_make_request(char *str)
2142 return setup_fault_attr(&fail_make_request, str);
2144 __setup("fail_make_request=", setup_fail_make_request);
2146 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2148 return part->make_it_fail && should_fail(&fail_make_request, bytes);
2151 static int __init fail_make_request_debugfs(void)
2153 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2154 NULL, &fail_make_request);
2156 return PTR_ERR_OR_ZERO(dir);
2159 late_initcall(fail_make_request_debugfs);
2161 #else /* CONFIG_FAIL_MAKE_REQUEST */
2163 static inline bool should_fail_request(struct hd_struct *part,
2169 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2171 static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2173 if (part->policy && op_is_write(bio_op(bio))) {
2174 char b[BDEVNAME_SIZE];
2177 "generic_make_request: Trying to write "
2178 "to read-only block-device %s (partno %d)\n",
2179 bio_devname(bio, b), part->partno);
2186 static noinline int should_fail_bio(struct bio *bio)
2188 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2192 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2195 * Check whether this bio extends beyond the end of the device or partition.
2196 * This may well happen - the kernel calls bread() without checking the size of
2197 * the device, e.g., when mounting a file system.
2199 static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2201 unsigned int nr_sectors = bio_sectors(bio);
2203 if (nr_sectors && maxsector &&
2204 (nr_sectors > maxsector ||
2205 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2206 handle_bad_sector(bio, maxsector);
2213 * Remap block n of partition p to block n+start(p) of the disk.
2215 static inline int blk_partition_remap(struct bio *bio)
2217 struct hd_struct *p;
2221 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2224 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2226 if (unlikely(bio_check_ro(bio, p)))
2230 * Zone reset does not include bi_size so bio_sectors() is always 0.
2231 * Include a test for the reset op code and perform the remap if needed.
2233 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2234 if (bio_check_eod(bio, part_nr_sects_read(p)))
2236 bio->bi_iter.bi_sector += p->start_sect;
2238 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2239 bio->bi_iter.bi_sector - p->start_sect);
2247 static noinline_for_stack bool
2248 generic_make_request_checks(struct bio *bio)
2250 struct request_queue *q;
2251 int nr_sectors = bio_sectors(bio);
2252 blk_status_t status = BLK_STS_IOERR;
2253 char b[BDEVNAME_SIZE];
2257 q = bio->bi_disk->queue;
2260 "generic_make_request: Trying to access "
2261 "nonexistent block-device %s (%Lu)\n",
2262 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2267 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2268 * if queue is not a request based queue.
2270 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2273 if (should_fail_bio(bio))
2276 if (bio->bi_partno) {
2277 if (unlikely(blk_partition_remap(bio)))
2280 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2282 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
2287 * Filter flush bio's early so that make_request based
2288 * drivers without flush support don't have to worry
2291 if (op_is_flush(bio->bi_opf) &&
2292 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2293 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2295 status = BLK_STS_OK;
2300 switch (bio_op(bio)) {
2301 case REQ_OP_DISCARD:
2302 if (!blk_queue_discard(q))
2305 case REQ_OP_SECURE_ERASE:
2306 if (!blk_queue_secure_erase(q))
2309 case REQ_OP_WRITE_SAME:
2310 if (!q->limits.max_write_same_sectors)
2313 case REQ_OP_ZONE_REPORT:
2314 case REQ_OP_ZONE_RESET:
2315 if (!blk_queue_is_zoned(q))
2318 case REQ_OP_WRITE_ZEROES:
2319 if (!q->limits.max_write_zeroes_sectors)
2327 * Various block parts want %current->io_context and lazy ioc
2328 * allocation ends up trading a lot of pain for a small amount of
2329 * memory. Just allocate it upfront. This may fail and block
2330 * layer knows how to live with it.
2332 create_io_context(GFP_ATOMIC, q->node);
2334 if (!blkcg_bio_issue_check(q, bio))
2337 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2338 trace_block_bio_queue(q, bio);
2339 /* Now that enqueuing has been traced, we need to trace
2340 * completion as well.
2342 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2347 status = BLK_STS_NOTSUPP;
2349 bio->bi_status = status;
2355 * generic_make_request - hand a buffer to its device driver for I/O
2356 * @bio: The bio describing the location in memory and on the device.
2358 * generic_make_request() is used to make I/O requests of block
2359 * devices. It is passed a &struct bio, which describes the I/O that needs
2362 * generic_make_request() does not return any status. The
2363 * success/failure status of the request, along with notification of
2364 * completion, is delivered asynchronously through the bio->bi_end_io
2365 * function described (one day) else where.
2367 * The caller of generic_make_request must make sure that bi_io_vec
2368 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2369 * set to describe the device address, and the
2370 * bi_end_io and optionally bi_private are set to describe how
2371 * completion notification should be signaled.
2373 * generic_make_request and the drivers it calls may use bi_next if this
2374 * bio happens to be merged with someone else, and may resubmit the bio to
2375 * a lower device by calling into generic_make_request recursively, which
2376 * means the bio should NOT be touched after the call to ->make_request_fn.
2378 blk_qc_t generic_make_request(struct bio *bio)
2381 * bio_list_on_stack[0] contains bios submitted by the current
2383 * bio_list_on_stack[1] contains bios that were submitted before
2384 * the current make_request_fn, but that haven't been processed
2387 struct bio_list bio_list_on_stack[2];
2388 blk_qc_t ret = BLK_QC_T_NONE;
2390 if (!generic_make_request_checks(bio))
2394 * We only want one ->make_request_fn to be active at a time, else
2395 * stack usage with stacked devices could be a problem. So use
2396 * current->bio_list to keep a list of requests submited by a
2397 * make_request_fn function. current->bio_list is also used as a
2398 * flag to say if generic_make_request is currently active in this
2399 * task or not. If it is NULL, then no make_request is active. If
2400 * it is non-NULL, then a make_request is active, and new requests
2401 * should be added at the tail
2403 if (current->bio_list) {
2404 bio_list_add(¤t->bio_list[0], bio);
2408 /* following loop may be a bit non-obvious, and so deserves some
2410 * Before entering the loop, bio->bi_next is NULL (as all callers
2411 * ensure that) so we have a list with a single bio.
2412 * We pretend that we have just taken it off a longer list, so
2413 * we assign bio_list to a pointer to the bio_list_on_stack,
2414 * thus initialising the bio_list of new bios to be
2415 * added. ->make_request() may indeed add some more bios
2416 * through a recursive call to generic_make_request. If it
2417 * did, we find a non-NULL value in bio_list and re-enter the loop
2418 * from the top. In this case we really did just take the bio
2419 * of the top of the list (no pretending) and so remove it from
2420 * bio_list, and call into ->make_request() again.
2422 BUG_ON(bio->bi_next);
2423 bio_list_init(&bio_list_on_stack[0]);
2424 current->bio_list = bio_list_on_stack;
2426 struct request_queue *q = bio->bi_disk->queue;
2427 blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
2428 BLK_MQ_REQ_NOWAIT : 0;
2430 if (likely(blk_queue_enter(q, flags) == 0)) {
2431 struct bio_list lower, same;
2433 /* Create a fresh bio_list for all subordinate requests */
2434 bio_list_on_stack[1] = bio_list_on_stack[0];
2435 bio_list_init(&bio_list_on_stack[0]);
2436 ret = q->make_request_fn(q, bio);
2440 /* sort new bios into those for a lower level
2441 * and those for the same level
2443 bio_list_init(&lower);
2444 bio_list_init(&same);
2445 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2446 if (q == bio->bi_disk->queue)
2447 bio_list_add(&same, bio);
2449 bio_list_add(&lower, bio);
2450 /* now assemble so we handle the lowest level first */
2451 bio_list_merge(&bio_list_on_stack[0], &lower);
2452 bio_list_merge(&bio_list_on_stack[0], &same);
2453 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2455 if (unlikely(!blk_queue_dying(q) &&
2456 (bio->bi_opf & REQ_NOWAIT)))
2457 bio_wouldblock_error(bio);
2461 bio = bio_list_pop(&bio_list_on_stack[0]);
2463 current->bio_list = NULL; /* deactivate */
2468 EXPORT_SYMBOL(generic_make_request);
2471 * direct_make_request - hand a buffer directly to its device driver for I/O
2472 * @bio: The bio describing the location in memory and on the device.
2474 * This function behaves like generic_make_request(), but does not protect
2475 * against recursion. Must only be used if the called driver is known
2476 * to not call generic_make_request (or direct_make_request) again from
2477 * its make_request function. (Calling direct_make_request again from
2478 * a workqueue is perfectly fine as that doesn't recurse).
2480 blk_qc_t direct_make_request(struct bio *bio)
2482 struct request_queue *q = bio->bi_disk->queue;
2483 bool nowait = bio->bi_opf & REQ_NOWAIT;
2486 if (!generic_make_request_checks(bio))
2487 return BLK_QC_T_NONE;
2489 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2490 if (nowait && !blk_queue_dying(q))
2491 bio->bi_status = BLK_STS_AGAIN;
2493 bio->bi_status = BLK_STS_IOERR;
2495 return BLK_QC_T_NONE;
2498 ret = q->make_request_fn(q, bio);
2502 EXPORT_SYMBOL_GPL(direct_make_request);
2505 * submit_bio - submit a bio to the block device layer for I/O
2506 * @bio: The &struct bio which describes the I/O
2508 * submit_bio() is very similar in purpose to generic_make_request(), and
2509 * uses that function to do most of the work. Both are fairly rough
2510 * interfaces; @bio must be presetup and ready for I/O.
2513 blk_qc_t submit_bio(struct bio *bio)
2516 * If it's a regular read/write or a barrier with data attached,
2517 * go through the normal accounting stuff before submission.
2519 if (bio_has_data(bio)) {
2522 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2523 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2525 count = bio_sectors(bio);
2527 if (op_is_write(bio_op(bio))) {
2528 count_vm_events(PGPGOUT, count);
2530 task_io_account_read(bio->bi_iter.bi_size);
2531 count_vm_events(PGPGIN, count);
2534 if (unlikely(block_dump)) {
2535 char b[BDEVNAME_SIZE];
2536 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2537 current->comm, task_pid_nr(current),
2538 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2539 (unsigned long long)bio->bi_iter.bi_sector,
2540 bio_devname(bio, b), count);
2544 return generic_make_request(bio);
2546 EXPORT_SYMBOL(submit_bio);
2548 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2550 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2554 blk_flush_plug_list(current->plug, false);
2555 return q->poll_fn(q, cookie);
2557 EXPORT_SYMBOL_GPL(blk_poll);
2560 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2561 * for new the queue limits
2563 * @rq: the request being checked
2566 * @rq may have been made based on weaker limitations of upper-level queues
2567 * in request stacking drivers, and it may violate the limitation of @q.
2568 * Since the block layer and the underlying device driver trust @rq
2569 * after it is inserted to @q, it should be checked against @q before
2570 * the insertion using this generic function.
2572 * Request stacking drivers like request-based dm may change the queue
2573 * limits when retrying requests on other queues. Those requests need
2574 * to be checked against the new queue limits again during dispatch.
2576 static int blk_cloned_rq_check_limits(struct request_queue *q,
2579 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2580 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2585 * queue's settings related to segment counting like q->bounce_pfn
2586 * may differ from that of other stacking queues.
2587 * Recalculate it to check the request correctly on this queue's
2590 blk_recalc_rq_segments(rq);
2591 if (rq->nr_phys_segments > queue_max_segments(q)) {
2592 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2600 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2601 * @q: the queue to submit the request
2602 * @rq: the request being queued
2604 blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2606 unsigned long flags;
2607 int where = ELEVATOR_INSERT_BACK;
2609 if (blk_cloned_rq_check_limits(q, rq))
2610 return BLK_STS_IOERR;
2613 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2614 return BLK_STS_IOERR;
2617 if (blk_queue_io_stat(q))
2618 blk_account_io_start(rq, true);
2620 * Since we have a scheduler attached on the top device,
2621 * bypass a potential scheduler on the bottom device for
2624 return blk_mq_request_issue_directly(rq);
2627 spin_lock_irqsave(q->queue_lock, flags);
2628 if (unlikely(blk_queue_dying(q))) {
2629 spin_unlock_irqrestore(q->queue_lock, flags);
2630 return BLK_STS_IOERR;
2634 * Submitting request must be dequeued before calling this function
2635 * because it will be linked to another request_queue
2637 BUG_ON(blk_queued_rq(rq));
2639 if (op_is_flush(rq->cmd_flags))
2640 where = ELEVATOR_INSERT_FLUSH;
2642 add_acct_request(q, rq, where);
2643 if (where == ELEVATOR_INSERT_FLUSH)
2645 spin_unlock_irqrestore(q->queue_lock, flags);
2649 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2652 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2653 * @rq: request to examine
2656 * A request could be merge of IOs which require different failure
2657 * handling. This function determines the number of bytes which
2658 * can be failed from the beginning of the request without
2659 * crossing into area which need to be retried further.
2662 * The number of bytes to fail.
2664 unsigned int blk_rq_err_bytes(const struct request *rq)
2666 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2667 unsigned int bytes = 0;
2670 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2671 return blk_rq_bytes(rq);
2674 * Currently the only 'mixing' which can happen is between
2675 * different fastfail types. We can safely fail portions
2676 * which have all the failfast bits that the first one has -
2677 * the ones which are at least as eager to fail as the first
2680 for (bio = rq->bio; bio; bio = bio->bi_next) {
2681 if ((bio->bi_opf & ff) != ff)
2683 bytes += bio->bi_iter.bi_size;
2686 /* this could lead to infinite loop */
2687 BUG_ON(blk_rq_bytes(rq) && !bytes);
2690 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2692 void blk_account_io_completion(struct request *req, unsigned int bytes)
2694 if (blk_do_io_stat(req)) {
2695 const int rw = rq_data_dir(req);
2696 struct hd_struct *part;
2699 cpu = part_stat_lock();
2701 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2706 void blk_account_io_done(struct request *req)
2709 * Account IO completion. flush_rq isn't accounted as a
2710 * normal IO on queueing nor completion. Accounting the
2711 * containing request is enough.
2713 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2714 unsigned long duration = jiffies - req->start_time;
2715 const int rw = rq_data_dir(req);
2716 struct hd_struct *part;
2719 cpu = part_stat_lock();
2722 part_stat_inc(cpu, part, ios[rw]);
2723 part_stat_add(cpu, part, ticks[rw], duration);
2724 part_round_stats(req->q, cpu, part);
2725 part_dec_in_flight(req->q, part, rw);
2727 hd_struct_put(part);
2734 * Don't process normal requests when queue is suspended
2735 * or in the process of suspending/resuming
2737 static bool blk_pm_allow_request(struct request *rq)
2739 switch (rq->q->rpm_status) {
2741 case RPM_SUSPENDING:
2742 return rq->rq_flags & RQF_PM;
2750 static bool blk_pm_allow_request(struct request *rq)
2756 void blk_account_io_start(struct request *rq, bool new_io)
2758 struct hd_struct *part;
2759 int rw = rq_data_dir(rq);
2762 if (!blk_do_io_stat(rq))
2765 cpu = part_stat_lock();
2769 part_stat_inc(cpu, part, merges[rw]);
2771 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2772 if (!hd_struct_try_get(part)) {
2774 * The partition is already being removed,
2775 * the request will be accounted on the disk only
2777 * We take a reference on disk->part0 although that
2778 * partition will never be deleted, so we can treat
2779 * it as any other partition.
2781 part = &rq->rq_disk->part0;
2782 hd_struct_get(part);
2784 part_round_stats(rq->q, cpu, part);
2785 part_inc_in_flight(rq->q, part, rw);
2792 static struct request *elv_next_request(struct request_queue *q)
2795 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2797 WARN_ON_ONCE(q->mq_ops);
2800 list_for_each_entry(rq, &q->queue_head, queuelist) {
2801 if (blk_pm_allow_request(rq))
2804 if (rq->rq_flags & RQF_SOFTBARRIER)
2809 * Flush request is running and flush request isn't queueable
2810 * in the drive, we can hold the queue till flush request is
2811 * finished. Even we don't do this, driver can't dispatch next
2812 * requests and will requeue them. And this can improve
2813 * throughput too. For example, we have request flush1, write1,
2814 * flush 2. flush1 is dispatched, then queue is hold, write1
2815 * isn't inserted to queue. After flush1 is finished, flush2
2816 * will be dispatched. Since disk cache is already clean,
2817 * flush2 will be finished very soon, so looks like flush2 is
2819 * Since the queue is hold, a flag is set to indicate the queue
2820 * should be restarted later. Please see flush_end_io() for
2823 if (fq->flush_pending_idx != fq->flush_running_idx &&
2824 !queue_flush_queueable(q)) {
2825 fq->flush_queue_delayed = 1;
2828 if (unlikely(blk_queue_bypass(q)) ||
2829 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2835 * blk_peek_request - peek at the top of a request queue
2836 * @q: request queue to peek at
2839 * Return the request at the top of @q. The returned request
2840 * should be started using blk_start_request() before LLD starts
2844 * Pointer to the request at the top of @q if available. Null
2847 struct request *blk_peek_request(struct request_queue *q)
2852 lockdep_assert_held(q->queue_lock);
2853 WARN_ON_ONCE(q->mq_ops);
2855 while ((rq = elv_next_request(q)) != NULL) {
2856 if (!(rq->rq_flags & RQF_STARTED)) {
2858 * This is the first time the device driver
2859 * sees this request (possibly after
2860 * requeueing). Notify IO scheduler.
2862 if (rq->rq_flags & RQF_SORTED)
2863 elv_activate_rq(q, rq);
2866 * just mark as started even if we don't start
2867 * it, a request that has been delayed should
2868 * not be passed by new incoming requests
2870 rq->rq_flags |= RQF_STARTED;
2871 trace_block_rq_issue(q, rq);
2874 if (!q->boundary_rq || q->boundary_rq == rq) {
2875 q->end_sector = rq_end_sector(rq);
2876 q->boundary_rq = NULL;
2879 if (rq->rq_flags & RQF_DONTPREP)
2882 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2884 * make sure space for the drain appears we
2885 * know we can do this because max_hw_segments
2886 * has been adjusted to be one fewer than the
2889 rq->nr_phys_segments++;
2895 ret = q->prep_rq_fn(q, rq);
2896 if (ret == BLKPREP_OK) {
2898 } else if (ret == BLKPREP_DEFER) {
2900 * the request may have been (partially) prepped.
2901 * we need to keep this request in the front to
2902 * avoid resource deadlock. RQF_STARTED will
2903 * prevent other fs requests from passing this one.
2905 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2906 !(rq->rq_flags & RQF_DONTPREP)) {
2908 * remove the space for the drain we added
2909 * so that we don't add it again
2911 --rq->nr_phys_segments;
2916 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2917 rq->rq_flags |= RQF_QUIET;
2919 * Mark this request as started so we don't trigger
2920 * any debug logic in the end I/O path.
2922 blk_start_request(rq);
2923 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2924 BLK_STS_TARGET : BLK_STS_IOERR);
2926 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2933 EXPORT_SYMBOL(blk_peek_request);
2935 static void blk_dequeue_request(struct request *rq)
2937 struct request_queue *q = rq->q;
2939 BUG_ON(list_empty(&rq->queuelist));
2940 BUG_ON(ELV_ON_HASH(rq));
2942 list_del_init(&rq->queuelist);
2945 * the time frame between a request being removed from the lists
2946 * and to it is freed is accounted as io that is in progress at
2949 if (blk_account_rq(rq)) {
2950 q->in_flight[rq_is_sync(rq)]++;
2951 set_io_start_time_ns(rq);
2956 * blk_start_request - start request processing on the driver
2957 * @req: request to dequeue
2960 * Dequeue @req and start timeout timer on it. This hands off the
2961 * request to the driver.
2963 void blk_start_request(struct request *req)
2965 lockdep_assert_held(req->q->queue_lock);
2966 WARN_ON_ONCE(req->q->mq_ops);
2968 blk_dequeue_request(req);
2970 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2971 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2972 req->rq_flags |= RQF_STATS;
2973 wbt_issue(req->q->rq_wb, &req->issue_stat);
2976 BUG_ON(blk_rq_is_complete(req));
2979 EXPORT_SYMBOL(blk_start_request);
2982 * blk_fetch_request - fetch a request from a request queue
2983 * @q: request queue to fetch a request from
2986 * Return the request at the top of @q. The request is started on
2987 * return and LLD can start processing it immediately.
2990 * Pointer to the request at the top of @q if available. Null
2993 struct request *blk_fetch_request(struct request_queue *q)
2997 lockdep_assert_held(q->queue_lock);
2998 WARN_ON_ONCE(q->mq_ops);
3000 rq = blk_peek_request(q);
3002 blk_start_request(rq);
3005 EXPORT_SYMBOL(blk_fetch_request);
3008 * Steal bios from a request and add them to a bio list.
3009 * The request must not have been partially completed before.
3011 void blk_steal_bios(struct bio_list *list, struct request *rq)
3015 list->tail->bi_next = rq->bio;
3017 list->head = rq->bio;
3018 list->tail = rq->biotail;
3026 EXPORT_SYMBOL_GPL(blk_steal_bios);
3029 * blk_update_request - Special helper function for request stacking drivers
3030 * @req: the request being processed
3031 * @error: block status code
3032 * @nr_bytes: number of bytes to complete @req
3035 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3036 * the request structure even if @req doesn't have leftover.
3037 * If @req has leftover, sets it up for the next range of segments.
3039 * This special helper function is only for request stacking drivers
3040 * (e.g. request-based dm) so that they can handle partial completion.
3041 * Actual device drivers should use blk_end_request instead.
3043 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3044 * %false return from this function.
3047 * %false - this request doesn't have any more data
3048 * %true - this request has more data
3050 bool blk_update_request(struct request *req, blk_status_t error,
3051 unsigned int nr_bytes)
3055 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
3060 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3061 !(req->rq_flags & RQF_QUIET)))
3062 print_req_error(req, error);
3064 blk_account_io_completion(req, nr_bytes);
3068 struct bio *bio = req->bio;
3069 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3071 if (bio_bytes == bio->bi_iter.bi_size)
3072 req->bio = bio->bi_next;
3074 /* Completion has already been traced */
3075 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
3076 req_bio_endio(req, bio, bio_bytes, error);
3078 total_bytes += bio_bytes;
3079 nr_bytes -= bio_bytes;
3090 * Reset counters so that the request stacking driver
3091 * can find how many bytes remain in the request
3094 req->__data_len = 0;
3098 req->__data_len -= total_bytes;
3100 /* update sector only for requests with clear definition of sector */
3101 if (!blk_rq_is_passthrough(req))
3102 req->__sector += total_bytes >> 9;
3104 /* mixed attributes always follow the first bio */
3105 if (req->rq_flags & RQF_MIXED_MERGE) {
3106 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3107 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3110 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3112 * If total number of sectors is less than the first segment
3113 * size, something has gone terribly wrong.
3115 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3116 blk_dump_rq_flags(req, "request botched");
3117 req->__data_len = blk_rq_cur_bytes(req);
3120 /* recalculate the number of segments */
3121 blk_recalc_rq_segments(req);
3126 EXPORT_SYMBOL_GPL(blk_update_request);
3128 static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3129 unsigned int nr_bytes,
3130 unsigned int bidi_bytes)
3132 if (blk_update_request(rq, error, nr_bytes))
3135 /* Bidi request must be completed as a whole */
3136 if (unlikely(blk_bidi_rq(rq)) &&
3137 blk_update_request(rq->next_rq, error, bidi_bytes))
3140 if (blk_queue_add_random(rq->q))
3141 add_disk_randomness(rq->rq_disk);
3147 * blk_unprep_request - unprepare a request
3150 * This function makes a request ready for complete resubmission (or
3151 * completion). It happens only after all error handling is complete,
3152 * so represents the appropriate moment to deallocate any resources
3153 * that were allocated to the request in the prep_rq_fn. The queue
3154 * lock is held when calling this.
3156 void blk_unprep_request(struct request *req)
3158 struct request_queue *q = req->q;
3160 req->rq_flags &= ~RQF_DONTPREP;
3161 if (q->unprep_rq_fn)
3162 q->unprep_rq_fn(q, req);
3164 EXPORT_SYMBOL_GPL(blk_unprep_request);
3166 void blk_finish_request(struct request *req, blk_status_t error)
3168 struct request_queue *q = req->q;
3170 lockdep_assert_held(req->q->queue_lock);
3171 WARN_ON_ONCE(q->mq_ops);
3173 if (req->rq_flags & RQF_STATS)
3176 if (req->rq_flags & RQF_QUEUED)
3177 blk_queue_end_tag(q, req);
3179 BUG_ON(blk_queued_rq(req));
3181 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3182 laptop_io_completion(req->q->backing_dev_info);
3184 blk_delete_timer(req);
3186 if (req->rq_flags & RQF_DONTPREP)
3187 blk_unprep_request(req);
3189 blk_account_io_done(req);
3192 wbt_done(req->q->rq_wb, &req->issue_stat);
3193 req->end_io(req, error);
3195 if (blk_bidi_rq(req))
3196 __blk_put_request(req->next_rq->q, req->next_rq);
3198 __blk_put_request(q, req);
3201 EXPORT_SYMBOL(blk_finish_request);
3204 * blk_end_bidi_request - Complete a bidi request
3205 * @rq: the request to complete
3206 * @error: block status code
3207 * @nr_bytes: number of bytes to complete @rq
3208 * @bidi_bytes: number of bytes to complete @rq->next_rq
3211 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3212 * Drivers that supports bidi can safely call this member for any
3213 * type of request, bidi or uni. In the later case @bidi_bytes is
3217 * %false - we are done with this request
3218 * %true - still buffers pending for this request
3220 static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3221 unsigned int nr_bytes, unsigned int bidi_bytes)
3223 struct request_queue *q = rq->q;
3224 unsigned long flags;
3226 WARN_ON_ONCE(q->mq_ops);
3228 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3231 spin_lock_irqsave(q->queue_lock, flags);
3232 blk_finish_request(rq, error);
3233 spin_unlock_irqrestore(q->queue_lock, flags);
3239 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3240 * @rq: the request to complete
3241 * @error: block status code
3242 * @nr_bytes: number of bytes to complete @rq
3243 * @bidi_bytes: number of bytes to complete @rq->next_rq
3246 * Identical to blk_end_bidi_request() except that queue lock is
3247 * assumed to be locked on entry and remains so on return.
3250 * %false - we are done with this request
3251 * %true - still buffers pending for this request
3253 static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3254 unsigned int nr_bytes, unsigned int bidi_bytes)
3256 lockdep_assert_held(rq->q->queue_lock);
3257 WARN_ON_ONCE(rq->q->mq_ops);
3259 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3262 blk_finish_request(rq, error);
3268 * blk_end_request - Helper function for drivers to complete the request.
3269 * @rq: the request being processed
3270 * @error: block status code
3271 * @nr_bytes: number of bytes to complete
3274 * Ends I/O on a number of bytes attached to @rq.
3275 * If @rq has leftover, sets it up for the next range of segments.
3278 * %false - we are done with this request
3279 * %true - still buffers pending for this request
3281 bool blk_end_request(struct request *rq, blk_status_t error,
3282 unsigned int nr_bytes)
3284 WARN_ON_ONCE(rq->q->mq_ops);
3285 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3287 EXPORT_SYMBOL(blk_end_request);
3290 * blk_end_request_all - Helper function for drives to finish the request.
3291 * @rq: the request to finish
3292 * @error: block status code
3295 * Completely finish @rq.
3297 void blk_end_request_all(struct request *rq, blk_status_t error)
3300 unsigned int bidi_bytes = 0;
3302 if (unlikely(blk_bidi_rq(rq)))
3303 bidi_bytes = blk_rq_bytes(rq->next_rq);
3305 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3308 EXPORT_SYMBOL(blk_end_request_all);
3311 * __blk_end_request - Helper function for drivers to complete the request.
3312 * @rq: the request being processed
3313 * @error: block status code
3314 * @nr_bytes: number of bytes to complete
3317 * Must be called with queue lock held unlike blk_end_request().
3320 * %false - we are done with this request
3321 * %true - still buffers pending for this request
3323 bool __blk_end_request(struct request *rq, blk_status_t error,
3324 unsigned int nr_bytes)
3326 lockdep_assert_held(rq->q->queue_lock);
3327 WARN_ON_ONCE(rq->q->mq_ops);
3329 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3331 EXPORT_SYMBOL(__blk_end_request);
3334 * __blk_end_request_all - Helper function for drives to finish the request.
3335 * @rq: the request to finish
3336 * @error: block status code
3339 * Completely finish @rq. Must be called with queue lock held.
3341 void __blk_end_request_all(struct request *rq, blk_status_t error)
3344 unsigned int bidi_bytes = 0;
3346 lockdep_assert_held(rq->q->queue_lock);
3347 WARN_ON_ONCE(rq->q->mq_ops);
3349 if (unlikely(blk_bidi_rq(rq)))
3350 bidi_bytes = blk_rq_bytes(rq->next_rq);
3352 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3355 EXPORT_SYMBOL(__blk_end_request_all);
3358 * __blk_end_request_cur - Helper function to finish the current request chunk.
3359 * @rq: the request to finish the current chunk for
3360 * @error: block status code
3363 * Complete the current consecutively mapped chunk from @rq. Must
3364 * be called with queue lock held.
3367 * %false - we are done with this request
3368 * %true - still buffers pending for this request
3370 bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3372 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3374 EXPORT_SYMBOL(__blk_end_request_cur);
3376 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3379 if (bio_has_data(bio))
3380 rq->nr_phys_segments = bio_phys_segments(q, bio);
3381 else if (bio_op(bio) == REQ_OP_DISCARD)
3382 rq->nr_phys_segments = 1;
3384 rq->__data_len = bio->bi_iter.bi_size;
3385 rq->bio = rq->biotail = bio;
3388 rq->rq_disk = bio->bi_disk;
3391 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3393 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3394 * @rq: the request to be flushed
3397 * Flush all pages in @rq.
3399 void rq_flush_dcache_pages(struct request *rq)
3401 struct req_iterator iter;
3402 struct bio_vec bvec;
3404 rq_for_each_segment(bvec, rq, iter)
3405 flush_dcache_page(bvec.bv_page);
3407 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3411 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3412 * @q : the queue of the device being checked
3415 * Check if underlying low-level drivers of a device are busy.
3416 * If the drivers want to export their busy state, they must set own
3417 * exporting function using blk_queue_lld_busy() first.
3419 * Basically, this function is used only by request stacking drivers
3420 * to stop dispatching requests to underlying devices when underlying
3421 * devices are busy. This behavior helps more I/O merging on the queue
3422 * of the request stacking driver and prevents I/O throughput regression
3423 * on burst I/O load.
3426 * 0 - Not busy (The request stacking driver should dispatch request)
3427 * 1 - Busy (The request stacking driver should stop dispatching request)
3429 int blk_lld_busy(struct request_queue *q)
3432 return q->lld_busy_fn(q);
3436 EXPORT_SYMBOL_GPL(blk_lld_busy);
3439 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3440 * @rq: the clone request to be cleaned up
3443 * Free all bios in @rq for a cloned request.
3445 void blk_rq_unprep_clone(struct request *rq)
3449 while ((bio = rq->bio) != NULL) {
3450 rq->bio = bio->bi_next;
3455 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3458 * Copy attributes of the original request to the clone request.
3459 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3461 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3463 dst->cpu = src->cpu;
3464 dst->__sector = blk_rq_pos(src);
3465 dst->__data_len = blk_rq_bytes(src);
3466 dst->nr_phys_segments = src->nr_phys_segments;
3467 dst->ioprio = src->ioprio;
3468 dst->extra_len = src->extra_len;
3472 * blk_rq_prep_clone - Helper function to setup clone request
3473 * @rq: the request to be setup
3474 * @rq_src: original request to be cloned
3475 * @bs: bio_set that bios for clone are allocated from
3476 * @gfp_mask: memory allocation mask for bio
3477 * @bio_ctr: setup function to be called for each clone bio.
3478 * Returns %0 for success, non %0 for failure.
3479 * @data: private data to be passed to @bio_ctr
3482 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3483 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3484 * are not copied, and copying such parts is the caller's responsibility.
3485 * Also, pages which the original bios are pointing to are not copied
3486 * and the cloned bios just point same pages.
3487 * So cloned bios must be completed before original bios, which means
3488 * the caller must complete @rq before @rq_src.
3490 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3491 struct bio_set *bs, gfp_t gfp_mask,
3492 int (*bio_ctr)(struct bio *, struct bio *, void *),
3495 struct bio *bio, *bio_src;
3500 __rq_for_each_bio(bio_src, rq_src) {
3501 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3505 if (bio_ctr && bio_ctr(bio, bio_src, data))
3509 rq->biotail->bi_next = bio;
3512 rq->bio = rq->biotail = bio;
3515 __blk_rq_prep_clone(rq, rq_src);
3522 blk_rq_unprep_clone(rq);
3526 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3528 int kblockd_schedule_work(struct work_struct *work)
3530 return queue_work(kblockd_workqueue, work);
3532 EXPORT_SYMBOL(kblockd_schedule_work);
3534 int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3536 return queue_work_on(cpu, kblockd_workqueue, work);
3538 EXPORT_SYMBOL(kblockd_schedule_work_on);
3540 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3541 unsigned long delay)
3543 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3545 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3548 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3549 * @plug: The &struct blk_plug that needs to be initialized
3552 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3553 * pending I/O should the task end up blocking between blk_start_plug() and
3554 * blk_finish_plug(). This is important from a performance perspective, but
3555 * also ensures that we don't deadlock. For instance, if the task is blocking
3556 * for a memory allocation, memory reclaim could end up wanting to free a
3557 * page belonging to that request that is currently residing in our private
3558 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3559 * this kind of deadlock.
3561 void blk_start_plug(struct blk_plug *plug)
3563 struct task_struct *tsk = current;
3566 * If this is a nested plug, don't actually assign it.
3571 INIT_LIST_HEAD(&plug->list);
3572 INIT_LIST_HEAD(&plug->mq_list);
3573 INIT_LIST_HEAD(&plug->cb_list);
3575 * Store ordering should not be needed here, since a potential
3576 * preempt will imply a full memory barrier
3580 EXPORT_SYMBOL(blk_start_plug);
3582 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3584 struct request *rqa = container_of(a, struct request, queuelist);
3585 struct request *rqb = container_of(b, struct request, queuelist);
3587 return !(rqa->q < rqb->q ||
3588 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3592 * If 'from_schedule' is true, then postpone the dispatch of requests
3593 * until a safe kblockd context. We due this to avoid accidental big
3594 * additional stack usage in driver dispatch, in places where the originally
3595 * plugger did not intend it.
3597 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3599 __releases(q->queue_lock)
3601 lockdep_assert_held(q->queue_lock);
3603 trace_block_unplug(q, depth, !from_schedule);
3606 blk_run_queue_async(q);
3609 spin_unlock(q->queue_lock);
3612 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3614 LIST_HEAD(callbacks);
3616 while (!list_empty(&plug->cb_list)) {
3617 list_splice_init(&plug->cb_list, &callbacks);
3619 while (!list_empty(&callbacks)) {
3620 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3623 list_del(&cb->list);
3624 cb->callback(cb, from_schedule);
3629 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3632 struct blk_plug *plug = current->plug;
3633 struct blk_plug_cb *cb;
3638 list_for_each_entry(cb, &plug->cb_list, list)
3639 if (cb->callback == unplug && cb->data == data)
3642 /* Not currently on the callback list */
3643 BUG_ON(size < sizeof(*cb));
3644 cb = kzalloc(size, GFP_ATOMIC);
3647 cb->callback = unplug;
3648 list_add(&cb->list, &plug->cb_list);
3652 EXPORT_SYMBOL(blk_check_plugged);
3654 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3656 struct request_queue *q;
3657 unsigned long flags;
3662 flush_plug_callbacks(plug, from_schedule);
3664 if (!list_empty(&plug->mq_list))
3665 blk_mq_flush_plug_list(plug, from_schedule);
3667 if (list_empty(&plug->list))
3670 list_splice_init(&plug->list, &list);
3672 list_sort(NULL, &list, plug_rq_cmp);
3678 * Save and disable interrupts here, to avoid doing it for every
3679 * queue lock we have to take.
3681 local_irq_save(flags);
3682 while (!list_empty(&list)) {
3683 rq = list_entry_rq(list.next);
3684 list_del_init(&rq->queuelist);
3688 * This drops the queue lock
3691 queue_unplugged(q, depth, from_schedule);
3694 spin_lock(q->queue_lock);
3698 * Short-circuit if @q is dead
3700 if (unlikely(blk_queue_dying(q))) {
3701 __blk_end_request_all(rq, BLK_STS_IOERR);
3706 * rq is already accounted, so use raw insert
3708 if (op_is_flush(rq->cmd_flags))
3709 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3711 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3717 * This drops the queue lock
3720 queue_unplugged(q, depth, from_schedule);
3722 local_irq_restore(flags);
3725 void blk_finish_plug(struct blk_plug *plug)
3727 if (plug != current->plug)
3729 blk_flush_plug_list(plug, false);
3731 current->plug = NULL;
3733 EXPORT_SYMBOL(blk_finish_plug);
3737 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3738 * @q: the queue of the device
3739 * @dev: the device the queue belongs to
3742 * Initialize runtime-PM-related fields for @q and start auto suspend for
3743 * @dev. Drivers that want to take advantage of request-based runtime PM
3744 * should call this function after @dev has been initialized, and its
3745 * request queue @q has been allocated, and runtime PM for it can not happen
3746 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3747 * cases, driver should call this function before any I/O has taken place.
3749 * This function takes care of setting up using auto suspend for the device,
3750 * the autosuspend delay is set to -1 to make runtime suspend impossible
3751 * until an updated value is either set by user or by driver. Drivers do
3752 * not need to touch other autosuspend settings.
3754 * The block layer runtime PM is request based, so only works for drivers
3755 * that use request as their IO unit instead of those directly use bio's.
3757 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3759 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3764 q->rpm_status = RPM_ACTIVE;
3765 pm_runtime_set_autosuspend_delay(q->dev, -1);
3766 pm_runtime_use_autosuspend(q->dev);
3768 EXPORT_SYMBOL(blk_pm_runtime_init);
3771 * blk_pre_runtime_suspend - Pre runtime suspend check
3772 * @q: the queue of the device
3775 * This function will check if runtime suspend is allowed for the device
3776 * by examining if there are any requests pending in the queue. If there
3777 * are requests pending, the device can not be runtime suspended; otherwise,
3778 * the queue's status will be updated to SUSPENDING and the driver can
3779 * proceed to suspend the device.
3781 * For the not allowed case, we mark last busy for the device so that
3782 * runtime PM core will try to autosuspend it some time later.
3784 * This function should be called near the start of the device's
3785 * runtime_suspend callback.
3788 * 0 - OK to runtime suspend the device
3789 * -EBUSY - Device should not be runtime suspended
3791 int blk_pre_runtime_suspend(struct request_queue *q)
3798 spin_lock_irq(q->queue_lock);
3799 if (q->nr_pending) {
3801 pm_runtime_mark_last_busy(q->dev);
3803 q->rpm_status = RPM_SUSPENDING;
3805 spin_unlock_irq(q->queue_lock);
3808 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3811 * blk_post_runtime_suspend - Post runtime suspend processing
3812 * @q: the queue of the device
3813 * @err: return value of the device's runtime_suspend function
3816 * Update the queue's runtime status according to the return value of the
3817 * device's runtime suspend function and mark last busy for the device so
3818 * that PM core will try to auto suspend the device at a later time.
3820 * This function should be called near the end of the device's
3821 * runtime_suspend callback.
3823 void blk_post_runtime_suspend(struct request_queue *q, int err)
3828 spin_lock_irq(q->queue_lock);
3830 q->rpm_status = RPM_SUSPENDED;
3832 q->rpm_status = RPM_ACTIVE;
3833 pm_runtime_mark_last_busy(q->dev);
3835 spin_unlock_irq(q->queue_lock);
3837 EXPORT_SYMBOL(blk_post_runtime_suspend);
3840 * blk_pre_runtime_resume - Pre runtime resume processing
3841 * @q: the queue of the device
3844 * Update the queue's runtime status to RESUMING in preparation for the
3845 * runtime resume of the device.
3847 * This function should be called near the start of the device's
3848 * runtime_resume callback.
3850 void blk_pre_runtime_resume(struct request_queue *q)
3855 spin_lock_irq(q->queue_lock);
3856 q->rpm_status = RPM_RESUMING;
3857 spin_unlock_irq(q->queue_lock);
3859 EXPORT_SYMBOL(blk_pre_runtime_resume);
3862 * blk_post_runtime_resume - Post runtime resume processing
3863 * @q: the queue of the device
3864 * @err: return value of the device's runtime_resume function
3867 * Update the queue's runtime status according to the return value of the
3868 * device's runtime_resume function. If it is successfully resumed, process
3869 * the requests that are queued into the device's queue when it is resuming
3870 * and then mark last busy and initiate autosuspend for it.
3872 * This function should be called near the end of the device's
3873 * runtime_resume callback.
3875 void blk_post_runtime_resume(struct request_queue *q, int err)
3880 spin_lock_irq(q->queue_lock);
3882 q->rpm_status = RPM_ACTIVE;
3884 pm_runtime_mark_last_busy(q->dev);
3885 pm_request_autosuspend(q->dev);
3887 q->rpm_status = RPM_SUSPENDED;
3889 spin_unlock_irq(q->queue_lock);
3891 EXPORT_SYMBOL(blk_post_runtime_resume);
3894 * blk_set_runtime_active - Force runtime status of the queue to be active
3895 * @q: the queue of the device
3897 * If the device is left runtime suspended during system suspend the resume
3898 * hook typically resumes the device and corrects runtime status
3899 * accordingly. However, that does not affect the queue runtime PM status
3900 * which is still "suspended". This prevents processing requests from the
3903 * This function can be used in driver's resume hook to correct queue
3904 * runtime PM status and re-enable peeking requests from the queue. It
3905 * should be called before first request is added to the queue.
3907 void blk_set_runtime_active(struct request_queue *q)
3909 spin_lock_irq(q->queue_lock);
3910 q->rpm_status = RPM_ACTIVE;
3911 pm_runtime_mark_last_busy(q->dev);
3912 pm_request_autosuspend(q->dev);
3913 spin_unlock_irq(q->queue_lock);
3915 EXPORT_SYMBOL(blk_set_runtime_active);
3918 int __init blk_dev_init(void)
3920 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3921 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3922 FIELD_SIZEOF(struct request, cmd_flags));
3923 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3924 FIELD_SIZEOF(struct bio, bi_opf));
3926 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3927 kblockd_workqueue = alloc_workqueue("kblockd",
3928 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3929 if (!kblockd_workqueue)
3930 panic("Failed to create kblockd\n");
3932 request_cachep = kmem_cache_create("blkdev_requests",
3933 sizeof(struct request), 0, SLAB_PANIC, NULL);
3935 blk_requestq_cachep = kmem_cache_create("request_queue",
3936 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3938 #ifdef CONFIG_DEBUG_FS
3939 blk_debugfs_root = debugfs_create_dir("block", NULL);