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>
36 #define CREATE_TRACE_POINTS
37 #include <trace/events/block.h>
40 #include "blk-cgroup.h"
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
49 DEFINE_IDA(blk_queue_ida);
52 * For the allocated request tables
54 struct kmem_cache *request_cachep = NULL;
57 * For queue allocation
59 struct kmem_cache *blk_requestq_cachep;
62 * Controlling structure to kblockd
64 static struct workqueue_struct *kblockd_workqueue;
66 void blk_queue_congestion_threshold(struct request_queue *q)
70 nr = q->nr_requests - (q->nr_requests / 8) + 1;
71 if (nr > q->nr_requests)
73 q->nr_congestion_on = nr;
75 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
78 q->nr_congestion_off = nr;
82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
85 * Locates the passed device's request queue and returns the address of its
86 * backing_dev_info. This function can only be called if @bdev is opened
87 * and the return value is never NULL.
89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
91 struct request_queue *q = bdev_get_queue(bdev);
93 return &q->backing_dev_info;
95 EXPORT_SYMBOL(blk_get_backing_dev_info);
97 void blk_rq_init(struct request_queue *q, struct request *rq)
99 memset(rq, 0, sizeof(*rq));
101 INIT_LIST_HEAD(&rq->queuelist);
102 INIT_LIST_HEAD(&rq->timeout_list);
105 rq->__sector = (sector_t) -1;
106 INIT_HLIST_NODE(&rq->hash);
107 RB_CLEAR_NODE(&rq->rb_node);
109 rq->cmd_len = BLK_MAX_CDB;
111 rq->start_time = jiffies;
112 set_start_time_ns(rq);
115 EXPORT_SYMBOL(blk_rq_init);
117 static void req_bio_endio(struct request *rq, struct bio *bio,
118 unsigned int nbytes, int error)
120 if (error && !(rq->cmd_flags & REQ_CLONE))
121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
122 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
125 if (unlikely(rq->cmd_flags & REQ_QUIET))
126 set_bit(BIO_QUIET, &bio->bi_flags);
128 bio_advance(bio, nbytes);
130 /* don't actually finish bio if it's part of flush sequence */
131 if (bio->bi_iter.bi_size == 0 &&
132 !(rq->cmd_flags & (REQ_FLUSH_SEQ|REQ_CLONE)))
133 bio_endio(bio, error);
136 void blk_dump_rq_flags(struct request *rq, char *msg)
140 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
141 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
142 (unsigned long long) rq->cmd_flags);
144 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
145 (unsigned long long)blk_rq_pos(rq),
146 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
147 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
148 rq->bio, rq->biotail, blk_rq_bytes(rq));
150 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
151 printk(KERN_INFO " cdb: ");
152 for (bit = 0; bit < BLK_MAX_CDB; bit++)
153 printk("%02x ", rq->cmd[bit]);
157 EXPORT_SYMBOL(blk_dump_rq_flags);
159 static void blk_delay_work(struct work_struct *work)
161 struct request_queue *q;
163 q = container_of(work, struct request_queue, delay_work.work);
164 spin_lock_irq(q->queue_lock);
166 spin_unlock_irq(q->queue_lock);
170 * blk_delay_queue - restart queueing after defined interval
171 * @q: The &struct request_queue in question
172 * @msecs: Delay in msecs
175 * Sometimes queueing needs to be postponed for a little while, to allow
176 * resources to come back. This function will make sure that queueing is
177 * restarted around the specified time. Queue lock must be held.
179 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
181 if (likely(!blk_queue_dead(q)))
182 queue_delayed_work(kblockd_workqueue, &q->delay_work,
183 msecs_to_jiffies(msecs));
185 EXPORT_SYMBOL(blk_delay_queue);
188 * blk_start_queue - restart a previously stopped queue
189 * @q: The &struct request_queue in question
192 * blk_start_queue() will clear the stop flag on the queue, and call
193 * the request_fn for the queue if it was in a stopped state when
194 * entered. Also see blk_stop_queue(). Queue lock must be held.
196 void blk_start_queue(struct request_queue *q)
198 WARN_ON(!irqs_disabled());
200 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
203 EXPORT_SYMBOL(blk_start_queue);
206 * blk_stop_queue - stop a queue
207 * @q: The &struct request_queue in question
210 * The Linux block layer assumes that a block driver will consume all
211 * entries on the request queue when the request_fn strategy is called.
212 * Often this will not happen, because of hardware limitations (queue
213 * depth settings). If a device driver gets a 'queue full' response,
214 * or if it simply chooses not to queue more I/O at one point, it can
215 * call this function to prevent the request_fn from being called until
216 * the driver has signalled it's ready to go again. This happens by calling
217 * blk_start_queue() to restart queue operations. Queue lock must be held.
219 void blk_stop_queue(struct request_queue *q)
221 cancel_delayed_work(&q->delay_work);
222 queue_flag_set(QUEUE_FLAG_STOPPED, q);
224 EXPORT_SYMBOL(blk_stop_queue);
227 * blk_sync_queue - cancel any pending callbacks on a queue
231 * The block layer may perform asynchronous callback activity
232 * on a queue, such as calling the unplug function after a timeout.
233 * A block device may call blk_sync_queue to ensure that any
234 * such activity is cancelled, thus allowing it to release resources
235 * that the callbacks might use. The caller must already have made sure
236 * that its ->make_request_fn will not re-add plugging prior to calling
239 * This function does not cancel any asynchronous activity arising
240 * out of elevator or throttling code. That would require elevator_exit()
241 * and blkcg_exit_queue() to be called with queue lock initialized.
244 void blk_sync_queue(struct request_queue *q)
246 del_timer_sync(&q->timeout);
249 struct blk_mq_hw_ctx *hctx;
252 queue_for_each_hw_ctx(q, hctx, i) {
253 cancel_delayed_work_sync(&hctx->run_work);
254 cancel_delayed_work_sync(&hctx->delay_work);
257 cancel_delayed_work_sync(&q->delay_work);
260 EXPORT_SYMBOL(blk_sync_queue);
263 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
264 * @q: The queue to run
267 * Invoke request handling on a queue if there are any pending requests.
268 * May be used to restart request handling after a request has completed.
269 * This variant runs the queue whether or not the queue has been
270 * stopped. Must be called with the queue lock held and interrupts
271 * disabled. See also @blk_run_queue.
273 inline void __blk_run_queue_uncond(struct request_queue *q)
275 if (unlikely(blk_queue_dead(q)))
279 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
280 * the queue lock internally. As a result multiple threads may be
281 * running such a request function concurrently. Keep track of the
282 * number of active request_fn invocations such that blk_drain_queue()
283 * can wait until all these request_fn calls have finished.
285 q->request_fn_active++;
287 q->request_fn_active--;
289 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
292 * __blk_run_queue - run a single device queue
293 * @q: The queue to run
296 * See @blk_run_queue. This variant must be called with the queue lock
297 * held and interrupts disabled.
299 void __blk_run_queue(struct request_queue *q)
301 if (unlikely(blk_queue_stopped(q)))
304 __blk_run_queue_uncond(q);
306 EXPORT_SYMBOL(__blk_run_queue);
309 * blk_run_queue_async - run a single device queue in workqueue context
310 * @q: The queue to run
313 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
314 * of us. The caller must hold the queue lock.
316 void blk_run_queue_async(struct request_queue *q)
318 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
319 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
321 EXPORT_SYMBOL(blk_run_queue_async);
324 * blk_run_queue - run a single device queue
325 * @q: The queue to run
328 * Invoke request handling on this queue, if it has pending work to do.
329 * May be used to restart queueing when a request has completed.
331 void blk_run_queue(struct request_queue *q)
335 spin_lock_irqsave(q->queue_lock, flags);
337 spin_unlock_irqrestore(q->queue_lock, flags);
339 EXPORT_SYMBOL(blk_run_queue);
341 void blk_put_queue(struct request_queue *q)
343 kobject_put(&q->kobj);
345 EXPORT_SYMBOL(blk_put_queue);
348 * __blk_drain_queue - drain requests from request_queue
350 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
352 * Drain requests from @q. If @drain_all is set, all requests are drained.
353 * If not, only ELVPRIV requests are drained. The caller is responsible
354 * for ensuring that no new requests which need to be drained are queued.
356 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
357 __releases(q->queue_lock)
358 __acquires(q->queue_lock)
362 lockdep_assert_held(q->queue_lock);
368 * The caller might be trying to drain @q before its
369 * elevator is initialized.
372 elv_drain_elevator(q);
374 blkcg_drain_queue(q);
377 * This function might be called on a queue which failed
378 * driver init after queue creation or is not yet fully
379 * active yet. Some drivers (e.g. fd and loop) get unhappy
380 * in such cases. Kick queue iff dispatch queue has
381 * something on it and @q has request_fn set.
383 if (!list_empty(&q->queue_head) && q->request_fn)
386 drain |= q->nr_rqs_elvpriv;
387 drain |= q->request_fn_active;
390 * Unfortunately, requests are queued at and tracked from
391 * multiple places and there's no single counter which can
392 * be drained. Check all the queues and counters.
395 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
396 drain |= !list_empty(&q->queue_head);
397 for (i = 0; i < 2; i++) {
398 drain |= q->nr_rqs[i];
399 drain |= q->in_flight[i];
401 drain |= !list_empty(&fq->flush_queue[i]);
408 spin_unlock_irq(q->queue_lock);
412 spin_lock_irq(q->queue_lock);
416 * With queue marked dead, any woken up waiter will fail the
417 * allocation path, so the wakeup chaining is lost and we're
418 * left with hung waiters. We need to wake up those waiters.
421 struct request_list *rl;
423 blk_queue_for_each_rl(rl, q)
424 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
425 wake_up_all(&rl->wait[i]);
430 * blk_queue_bypass_start - enter queue bypass mode
431 * @q: queue of interest
433 * In bypass mode, only the dispatch FIFO queue of @q is used. This
434 * function makes @q enter bypass mode and drains all requests which were
435 * throttled or issued before. On return, it's guaranteed that no request
436 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
437 * inside queue or RCU read lock.
439 void blk_queue_bypass_start(struct request_queue *q)
441 spin_lock_irq(q->queue_lock);
443 queue_flag_set(QUEUE_FLAG_BYPASS, q);
444 spin_unlock_irq(q->queue_lock);
447 * Queues start drained. Skip actual draining till init is
448 * complete. This avoids lenghty delays during queue init which
449 * can happen many times during boot.
451 if (blk_queue_init_done(q)) {
452 spin_lock_irq(q->queue_lock);
453 __blk_drain_queue(q, false);
454 spin_unlock_irq(q->queue_lock);
456 /* ensure blk_queue_bypass() is %true inside RCU read lock */
460 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
463 * blk_queue_bypass_end - leave queue bypass mode
464 * @q: queue of interest
466 * Leave bypass mode and restore the normal queueing behavior.
468 void blk_queue_bypass_end(struct request_queue *q)
470 spin_lock_irq(q->queue_lock);
471 if (!--q->bypass_depth)
472 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
473 WARN_ON_ONCE(q->bypass_depth < 0);
474 spin_unlock_irq(q->queue_lock);
476 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
478 void blk_set_queue_dying(struct request_queue *q)
480 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
483 blk_mq_wake_waiters(q);
485 struct request_list *rl;
487 blk_queue_for_each_rl(rl, q) {
489 wake_up(&rl->wait[BLK_RW_SYNC]);
490 wake_up(&rl->wait[BLK_RW_ASYNC]);
495 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
498 * blk_cleanup_queue - shutdown a request queue
499 * @q: request queue to shutdown
501 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
502 * put it. All future requests will be failed immediately with -ENODEV.
504 void blk_cleanup_queue(struct request_queue *q)
506 spinlock_t *lock = q->queue_lock;
508 /* mark @q DYING, no new request or merges will be allowed afterwards */
509 mutex_lock(&q->sysfs_lock);
510 blk_set_queue_dying(q);
514 * A dying queue is permanently in bypass mode till released. Note
515 * that, unlike blk_queue_bypass_start(), we aren't performing
516 * synchronize_rcu() after entering bypass mode to avoid the delay
517 * as some drivers create and destroy a lot of queues while
518 * probing. This is still safe because blk_release_queue() will be
519 * called only after the queue refcnt drops to zero and nothing,
520 * RCU or not, would be traversing the queue by then.
523 queue_flag_set(QUEUE_FLAG_BYPASS, q);
525 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
526 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
527 queue_flag_set(QUEUE_FLAG_DYING, q);
528 spin_unlock_irq(lock);
529 mutex_unlock(&q->sysfs_lock);
532 * Drain all requests queued before DYING marking. Set DEAD flag to
533 * prevent that q->request_fn() gets invoked after draining finished.
536 blk_mq_freeze_queue(q);
540 __blk_drain_queue(q, true);
542 queue_flag_set(QUEUE_FLAG_DEAD, q);
543 spin_unlock_irq(lock);
545 /* @q won't process any more request, flush async actions */
546 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
550 blk_mq_free_queue(q);
553 if (q->queue_lock != &q->__queue_lock)
554 q->queue_lock = &q->__queue_lock;
555 spin_unlock_irq(lock);
557 bdi_destroy(&q->backing_dev_info);
559 /* @q is and will stay empty, shutdown and put */
562 EXPORT_SYMBOL(blk_cleanup_queue);
564 /* Allocate memory local to the request queue */
565 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
567 int nid = (int)(long)data;
568 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
571 static void free_request_struct(void *element, void *unused)
573 kmem_cache_free(request_cachep, element);
576 int blk_init_rl(struct request_list *rl, struct request_queue *q,
579 if (unlikely(rl->rq_pool))
583 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
584 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
585 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
586 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
588 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
590 (void *)(long)q->node, gfp_mask,
598 void blk_exit_rl(struct request_list *rl)
601 mempool_destroy(rl->rq_pool);
604 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
606 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
608 EXPORT_SYMBOL(blk_alloc_queue);
610 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
612 struct request_queue *q;
615 q = kmem_cache_alloc_node(blk_requestq_cachep,
616 gfp_mask | __GFP_ZERO, node_id);
620 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
624 q->backing_dev_info.ra_pages =
625 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
626 q->backing_dev_info.state = 0;
627 q->backing_dev_info.capabilities = 0;
628 q->backing_dev_info.name = "block";
631 err = bdi_init(&q->backing_dev_info);
635 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
636 laptop_mode_timer_fn, (unsigned long) q);
637 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
638 INIT_LIST_HEAD(&q->queue_head);
639 INIT_LIST_HEAD(&q->timeout_list);
640 INIT_LIST_HEAD(&q->icq_list);
641 #ifdef CONFIG_BLK_CGROUP
642 INIT_LIST_HEAD(&q->blkg_list);
644 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
646 kobject_init(&q->kobj, &blk_queue_ktype);
648 mutex_init(&q->sysfs_lock);
649 spin_lock_init(&q->__queue_lock);
652 * By default initialize queue_lock to internal lock and driver can
653 * override it later if need be.
655 q->queue_lock = &q->__queue_lock;
658 * A queue starts its life with bypass turned on to avoid
659 * unnecessary bypass on/off overhead and nasty surprises during
660 * init. The initial bypass will be finished when the queue is
661 * registered by blk_register_queue().
664 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
666 init_waitqueue_head(&q->mq_freeze_wq);
668 if (blkcg_init_queue(q))
674 bdi_destroy(&q->backing_dev_info);
676 ida_simple_remove(&blk_queue_ida, q->id);
678 kmem_cache_free(blk_requestq_cachep, q);
681 EXPORT_SYMBOL(blk_alloc_queue_node);
684 * blk_init_queue - prepare a request queue for use with a block device
685 * @rfn: The function to be called to process requests that have been
686 * placed on the queue.
687 * @lock: Request queue spin lock
690 * If a block device wishes to use the standard request handling procedures,
691 * which sorts requests and coalesces adjacent requests, then it must
692 * call blk_init_queue(). The function @rfn will be called when there
693 * are requests on the queue that need to be processed. If the device
694 * supports plugging, then @rfn may not be called immediately when requests
695 * are available on the queue, but may be called at some time later instead.
696 * Plugged queues are generally unplugged when a buffer belonging to one
697 * of the requests on the queue is needed, or due to memory pressure.
699 * @rfn is not required, or even expected, to remove all requests off the
700 * queue, but only as many as it can handle at a time. If it does leave
701 * requests on the queue, it is responsible for arranging that the requests
702 * get dealt with eventually.
704 * The queue spin lock must be held while manipulating the requests on the
705 * request queue; this lock will be taken also from interrupt context, so irq
706 * disabling is needed for it.
708 * Function returns a pointer to the initialized request queue, or %NULL if
712 * blk_init_queue() must be paired with a blk_cleanup_queue() call
713 * when the block device is deactivated (such as at module unload).
716 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
718 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
720 EXPORT_SYMBOL(blk_init_queue);
722 struct request_queue *
723 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
725 struct request_queue *uninit_q, *q;
727 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
731 q = blk_init_allocated_queue(uninit_q, rfn, lock);
733 blk_cleanup_queue(uninit_q);
737 EXPORT_SYMBOL(blk_init_queue_node);
739 static void blk_queue_bio(struct request_queue *q, struct bio *bio);
741 struct request_queue *
742 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
748 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
752 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
756 q->prep_rq_fn = NULL;
757 q->unprep_rq_fn = NULL;
758 q->queue_flags |= QUEUE_FLAG_DEFAULT;
760 /* Override internal queue lock with supplied lock pointer */
762 q->queue_lock = lock;
765 * This also sets hw/phys segments, boundary and size
767 blk_queue_make_request(q, blk_queue_bio);
769 q->sg_reserved_size = INT_MAX;
771 /* Protect q->elevator from elevator_change */
772 mutex_lock(&q->sysfs_lock);
775 if (elevator_init(q, NULL)) {
776 mutex_unlock(&q->sysfs_lock);
780 mutex_unlock(&q->sysfs_lock);
785 blk_free_flush_queue(q->fq);
788 EXPORT_SYMBOL(blk_init_allocated_queue);
790 bool blk_get_queue(struct request_queue *q)
792 if (likely(!blk_queue_dying(q))) {
799 EXPORT_SYMBOL(blk_get_queue);
801 static inline void blk_free_request(struct request_list *rl, struct request *rq)
803 if (rq->cmd_flags & REQ_ELVPRIV) {
804 elv_put_request(rl->q, rq);
806 put_io_context(rq->elv.icq->ioc);
809 mempool_free(rq, rl->rq_pool);
813 * ioc_batching returns true if the ioc is a valid batching request and
814 * should be given priority access to a request.
816 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
822 * Make sure the process is able to allocate at least 1 request
823 * even if the batch times out, otherwise we could theoretically
826 return ioc->nr_batch_requests == q->nr_batching ||
827 (ioc->nr_batch_requests > 0
828 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
832 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
833 * will cause the process to be a "batcher" on all queues in the system. This
834 * is the behaviour we want though - once it gets a wakeup it should be given
837 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
839 if (!ioc || ioc_batching(q, ioc))
842 ioc->nr_batch_requests = q->nr_batching;
843 ioc->last_waited = jiffies;
846 static void __freed_request(struct request_list *rl, int sync)
848 struct request_queue *q = rl->q;
851 * bdi isn't aware of blkcg yet. As all async IOs end up root
852 * blkcg anyway, just use root blkcg state.
854 if (rl == &q->root_rl &&
855 rl->count[sync] < queue_congestion_off_threshold(q))
856 blk_clear_queue_congested(q, sync);
858 if (rl->count[sync] + 1 <= q->nr_requests) {
859 if (waitqueue_active(&rl->wait[sync]))
860 wake_up(&rl->wait[sync]);
862 blk_clear_rl_full(rl, sync);
867 * A request has just been released. Account for it, update the full and
868 * congestion status, wake up any waiters. Called under q->queue_lock.
870 static void freed_request(struct request_list *rl, unsigned int flags)
872 struct request_queue *q = rl->q;
873 int sync = rw_is_sync(flags);
877 if (flags & REQ_ELVPRIV)
880 __freed_request(rl, sync);
882 if (unlikely(rl->starved[sync ^ 1]))
883 __freed_request(rl, sync ^ 1);
886 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
888 struct request_list *rl;
890 spin_lock_irq(q->queue_lock);
892 blk_queue_congestion_threshold(q);
894 /* congestion isn't cgroup aware and follows root blkcg for now */
897 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
898 blk_set_queue_congested(q, BLK_RW_SYNC);
899 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
900 blk_clear_queue_congested(q, BLK_RW_SYNC);
902 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
903 blk_set_queue_congested(q, BLK_RW_ASYNC);
904 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
905 blk_clear_queue_congested(q, BLK_RW_ASYNC);
907 blk_queue_for_each_rl(rl, q) {
908 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
909 blk_set_rl_full(rl, BLK_RW_SYNC);
911 blk_clear_rl_full(rl, BLK_RW_SYNC);
912 wake_up(&rl->wait[BLK_RW_SYNC]);
915 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
916 blk_set_rl_full(rl, BLK_RW_ASYNC);
918 blk_clear_rl_full(rl, BLK_RW_ASYNC);
919 wake_up(&rl->wait[BLK_RW_ASYNC]);
923 spin_unlock_irq(q->queue_lock);
928 * Determine if elevator data should be initialized when allocating the
929 * request associated with @bio.
931 static bool blk_rq_should_init_elevator(struct bio *bio)
937 * Flush requests do not use the elevator so skip initialization.
938 * This allows a request to share the flush and elevator data.
940 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
947 * rq_ioc - determine io_context for request allocation
948 * @bio: request being allocated is for this bio (can be %NULL)
950 * Determine io_context to use for request allocation for @bio. May return
951 * %NULL if %current->io_context doesn't exist.
953 static struct io_context *rq_ioc(struct bio *bio)
955 #ifdef CONFIG_BLK_CGROUP
956 if (bio && bio->bi_ioc)
959 return current->io_context;
963 * __get_request - get a free request
964 * @rl: request list to allocate from
965 * @rw_flags: RW and SYNC flags
966 * @bio: bio to allocate request for (can be %NULL)
967 * @gfp_mask: allocation mask
969 * Get a free request from @q. This function may fail under memory
970 * pressure or if @q is dead.
972 * Must be called with @q->queue_lock held and,
973 * Returns ERR_PTR on failure, with @q->queue_lock held.
974 * Returns request pointer on success, with @q->queue_lock *not held*.
976 static struct request *__get_request(struct request_list *rl, int rw_flags,
977 struct bio *bio, gfp_t gfp_mask)
979 struct request_queue *q = rl->q;
981 struct elevator_type *et = q->elevator->type;
982 struct io_context *ioc = rq_ioc(bio);
983 struct io_cq *icq = NULL;
984 const bool is_sync = rw_is_sync(rw_flags) != 0;
987 if (unlikely(blk_queue_dying(q)))
988 return ERR_PTR(-ENODEV);
990 may_queue = elv_may_queue(q, rw_flags);
991 if (may_queue == ELV_MQUEUE_NO)
994 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
995 if (rl->count[is_sync]+1 >= q->nr_requests) {
997 * The queue will fill after this allocation, so set
998 * it as full, and mark this process as "batching".
999 * This process will be allowed to complete a batch of
1000 * requests, others will be blocked.
1002 if (!blk_rl_full(rl, is_sync)) {
1003 ioc_set_batching(q, ioc);
1004 blk_set_rl_full(rl, is_sync);
1006 if (may_queue != ELV_MQUEUE_MUST
1007 && !ioc_batching(q, ioc)) {
1009 * The queue is full and the allocating
1010 * process is not a "batcher", and not
1011 * exempted by the IO scheduler
1013 return ERR_PTR(-ENOMEM);
1018 * bdi isn't aware of blkcg yet. As all async IOs end up
1019 * root blkcg anyway, just use root blkcg state.
1021 if (rl == &q->root_rl)
1022 blk_set_queue_congested(q, is_sync);
1026 * Only allow batching queuers to allocate up to 50% over the defined
1027 * limit of requests, otherwise we could have thousands of requests
1028 * allocated with any setting of ->nr_requests
1030 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1031 return ERR_PTR(-ENOMEM);
1033 q->nr_rqs[is_sync]++;
1034 rl->count[is_sync]++;
1035 rl->starved[is_sync] = 0;
1038 * Decide whether the new request will be managed by elevator. If
1039 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1040 * prevent the current elevator from being destroyed until the new
1041 * request is freed. This guarantees icq's won't be destroyed and
1042 * makes creating new ones safe.
1044 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1045 * it will be created after releasing queue_lock.
1047 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1048 rw_flags |= REQ_ELVPRIV;
1049 q->nr_rqs_elvpriv++;
1050 if (et->icq_cache && ioc)
1051 icq = ioc_lookup_icq(ioc, q);
1054 if (blk_queue_io_stat(q))
1055 rw_flags |= REQ_IO_STAT;
1056 spin_unlock_irq(q->queue_lock);
1058 /* allocate and init request */
1059 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1064 blk_rq_set_rl(rq, rl);
1065 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1068 if (rw_flags & REQ_ELVPRIV) {
1069 if (unlikely(et->icq_cache && !icq)) {
1071 icq = ioc_create_icq(ioc, q, gfp_mask);
1077 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1080 /* @rq->elv.icq holds io_context until @rq is freed */
1082 get_io_context(icq->ioc);
1086 * ioc may be NULL here, and ioc_batching will be false. That's
1087 * OK, if the queue is under the request limit then requests need
1088 * not count toward the nr_batch_requests limit. There will always
1089 * be some limit enforced by BLK_BATCH_TIME.
1091 if (ioc_batching(q, ioc))
1092 ioc->nr_batch_requests--;
1094 trace_block_getrq(q, bio, rw_flags & 1);
1099 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1100 * and may fail indefinitely under memory pressure and thus
1101 * shouldn't stall IO. Treat this request as !elvpriv. This will
1102 * disturb iosched and blkcg but weird is bettern than dead.
1104 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1105 __func__, dev_name(q->backing_dev_info.dev));
1107 rq->cmd_flags &= ~REQ_ELVPRIV;
1110 spin_lock_irq(q->queue_lock);
1111 q->nr_rqs_elvpriv--;
1112 spin_unlock_irq(q->queue_lock);
1117 * Allocation failed presumably due to memory. Undo anything we
1118 * might have messed up.
1120 * Allocating task should really be put onto the front of the wait
1121 * queue, but this is pretty rare.
1123 spin_lock_irq(q->queue_lock);
1124 freed_request(rl, rw_flags);
1127 * in the very unlikely event that allocation failed and no
1128 * requests for this direction was pending, mark us starved so that
1129 * freeing of a request in the other direction will notice
1130 * us. another possible fix would be to split the rq mempool into
1134 if (unlikely(rl->count[is_sync] == 0))
1135 rl->starved[is_sync] = 1;
1136 return ERR_PTR(-ENOMEM);
1140 * get_request - get a free request
1141 * @q: request_queue to allocate request from
1142 * @rw_flags: RW and SYNC flags
1143 * @bio: bio to allocate request for (can be %NULL)
1144 * @gfp_mask: allocation mask
1146 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1147 * function keeps retrying under memory pressure and fails iff @q is dead.
1149 * Must be called with @q->queue_lock held and,
1150 * Returns ERR_PTR on failure, with @q->queue_lock held.
1151 * Returns request pointer on success, with @q->queue_lock *not held*.
1153 static struct request *get_request(struct request_queue *q, int rw_flags,
1154 struct bio *bio, gfp_t gfp_mask)
1156 const bool is_sync = rw_is_sync(rw_flags) != 0;
1158 struct request_list *rl;
1161 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1163 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1167 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1172 /* wait on @rl and retry */
1173 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1174 TASK_UNINTERRUPTIBLE);
1176 trace_block_sleeprq(q, bio, rw_flags & 1);
1178 spin_unlock_irq(q->queue_lock);
1182 * After sleeping, we become a "batching" process and will be able
1183 * to allocate at least one request, and up to a big batch of them
1184 * for a small period time. See ioc_batching, ioc_set_batching
1186 ioc_set_batching(q, current->io_context);
1188 spin_lock_irq(q->queue_lock);
1189 finish_wait(&rl->wait[is_sync], &wait);
1194 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1199 BUG_ON(rw != READ && rw != WRITE);
1201 /* create ioc upfront */
1202 create_io_context(gfp_mask, q->node);
1204 spin_lock_irq(q->queue_lock);
1205 rq = get_request(q, rw, NULL, gfp_mask);
1207 spin_unlock_irq(q->queue_lock);
1208 /* q->queue_lock is unlocked at this point */
1213 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1216 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1218 return blk_old_get_request(q, rw, gfp_mask);
1220 EXPORT_SYMBOL(blk_get_request);
1223 * blk_make_request - given a bio, allocate a corresponding struct request.
1224 * @q: target request queue
1225 * @bio: The bio describing the memory mappings that will be submitted for IO.
1226 * It may be a chained-bio properly constructed by block/bio layer.
1227 * @gfp_mask: gfp flags to be used for memory allocation
1229 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1230 * type commands. Where the struct request needs to be farther initialized by
1231 * the caller. It is passed a &struct bio, which describes the memory info of
1234 * The caller of blk_make_request must make sure that bi_io_vec
1235 * are set to describe the memory buffers. That bio_data_dir() will return
1236 * the needed direction of the request. (And all bio's in the passed bio-chain
1237 * are properly set accordingly)
1239 * If called under none-sleepable conditions, mapped bio buffers must not
1240 * need bouncing, by calling the appropriate masked or flagged allocator,
1241 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1244 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1245 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1246 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1247 * completion of a bio that hasn't been submitted yet, thus resulting in a
1248 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1249 * of bio_alloc(), as that avoids the mempool deadlock.
1250 * If possible a big IO should be split into smaller parts when allocation
1251 * fails. Partial allocation should not be an error, or you risk a live-lock.
1253 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1256 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1261 blk_rq_set_block_pc(rq);
1264 struct bio *bounce_bio = bio;
1267 blk_queue_bounce(q, &bounce_bio);
1268 ret = blk_rq_append_bio(q, rq, bounce_bio);
1269 if (unlikely(ret)) {
1270 blk_put_request(rq);
1271 return ERR_PTR(ret);
1277 EXPORT_SYMBOL(blk_make_request);
1280 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1281 * @rq: request to be initialized
1284 void blk_rq_set_block_pc(struct request *rq)
1286 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1288 rq->__sector = (sector_t) -1;
1289 rq->bio = rq->biotail = NULL;
1290 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1292 EXPORT_SYMBOL(blk_rq_set_block_pc);
1295 * blk_requeue_request - put a request back on queue
1296 * @q: request queue where request should be inserted
1297 * @rq: request to be inserted
1300 * Drivers often keep queueing requests until the hardware cannot accept
1301 * more, when that condition happens we need to put the request back
1302 * on the queue. Must be called with queue lock held.
1304 void blk_requeue_request(struct request_queue *q, struct request *rq)
1306 blk_delete_timer(rq);
1307 blk_clear_rq_complete(rq);
1308 trace_block_rq_requeue(q, rq);
1310 if (rq->cmd_flags & REQ_QUEUED)
1311 blk_queue_end_tag(q, rq);
1313 BUG_ON(blk_queued_rq(rq));
1315 elv_requeue_request(q, rq);
1317 EXPORT_SYMBOL(blk_requeue_request);
1319 static void add_acct_request(struct request_queue *q, struct request *rq,
1322 blk_account_io_start(rq, true);
1323 __elv_add_request(q, rq, where);
1326 static void part_round_stats_single(int cpu, struct hd_struct *part,
1331 if (now == part->stamp)
1334 inflight = part_in_flight(part);
1336 __part_stat_add(cpu, part, time_in_queue,
1337 inflight * (now - part->stamp));
1338 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1344 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1345 * @cpu: cpu number for stats access
1346 * @part: target partition
1348 * The average IO queue length and utilisation statistics are maintained
1349 * by observing the current state of the queue length and the amount of
1350 * time it has been in this state for.
1352 * Normally, that accounting is done on IO completion, but that can result
1353 * in more than a second's worth of IO being accounted for within any one
1354 * second, leading to >100% utilisation. To deal with that, we call this
1355 * function to do a round-off before returning the results when reading
1356 * /proc/diskstats. This accounts immediately for all queue usage up to
1357 * the current jiffies and restarts the counters again.
1359 void part_round_stats(int cpu, struct hd_struct *part)
1361 unsigned long now = jiffies;
1364 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1365 part_round_stats_single(cpu, part, now);
1367 EXPORT_SYMBOL_GPL(part_round_stats);
1370 static void blk_pm_put_request(struct request *rq)
1372 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1373 pm_runtime_mark_last_busy(rq->q->dev);
1376 static inline void blk_pm_put_request(struct request *rq) {}
1380 * queue lock must be held
1382 void __blk_put_request(struct request_queue *q, struct request *req)
1388 blk_mq_free_request(req);
1392 blk_pm_put_request(req);
1394 elv_completed_request(q, req);
1396 /* this is a bio leak */
1397 WARN_ON(req->bio != NULL);
1400 * Request may not have originated from ll_rw_blk. if not,
1401 * it didn't come out of our reserved rq pools
1403 if (req->cmd_flags & REQ_ALLOCED) {
1404 unsigned int flags = req->cmd_flags;
1405 struct request_list *rl = blk_rq_rl(req);
1407 BUG_ON(!list_empty(&req->queuelist));
1408 BUG_ON(ELV_ON_HASH(req));
1410 blk_free_request(rl, req);
1411 freed_request(rl, flags);
1415 EXPORT_SYMBOL_GPL(__blk_put_request);
1417 void blk_put_request(struct request *req)
1419 struct request_queue *q = req->q;
1422 blk_mq_free_request(req);
1424 unsigned long flags;
1426 spin_lock_irqsave(q->queue_lock, flags);
1427 __blk_put_request(q, req);
1428 spin_unlock_irqrestore(q->queue_lock, flags);
1431 EXPORT_SYMBOL(blk_put_request);
1434 * blk_add_request_payload - add a payload to a request
1435 * @rq: request to update
1436 * @page: page backing the payload
1437 * @len: length of the payload.
1439 * This allows to later add a payload to an already submitted request by
1440 * a block driver. The driver needs to take care of freeing the payload
1443 * Note that this is a quite horrible hack and nothing but handling of
1444 * discard requests should ever use it.
1446 void blk_add_request_payload(struct request *rq, struct page *page,
1449 struct bio *bio = rq->bio;
1451 bio->bi_io_vec->bv_page = page;
1452 bio->bi_io_vec->bv_offset = 0;
1453 bio->bi_io_vec->bv_len = len;
1455 bio->bi_iter.bi_size = len;
1457 bio->bi_phys_segments = 1;
1459 rq->__data_len = rq->resid_len = len;
1460 rq->nr_phys_segments = 1;
1462 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1464 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1467 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1469 if (!ll_back_merge_fn(q, req, bio))
1472 trace_block_bio_backmerge(q, req, bio);
1474 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1475 blk_rq_set_mixed_merge(req);
1477 req->biotail->bi_next = bio;
1479 req->__data_len += bio->bi_iter.bi_size;
1480 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1482 blk_account_io_start(req, false);
1486 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1489 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1491 if (!ll_front_merge_fn(q, req, bio))
1494 trace_block_bio_frontmerge(q, req, bio);
1496 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1497 blk_rq_set_mixed_merge(req);
1499 bio->bi_next = req->bio;
1502 req->__sector = bio->bi_iter.bi_sector;
1503 req->__data_len += bio->bi_iter.bi_size;
1504 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1506 blk_account_io_start(req, false);
1511 * blk_attempt_plug_merge - try to merge with %current's plugged list
1512 * @q: request_queue new bio is being queued at
1513 * @bio: new bio being queued
1514 * @request_count: out parameter for number of traversed plugged requests
1516 * Determine whether @bio being queued on @q can be merged with a request
1517 * on %current's plugged list. Returns %true if merge was successful,
1520 * Plugging coalesces IOs from the same issuer for the same purpose without
1521 * going through @q->queue_lock. As such it's more of an issuing mechanism
1522 * than scheduling, and the request, while may have elvpriv data, is not
1523 * added on the elevator at this point. In addition, we don't have
1524 * reliable access to the elevator outside queue lock. Only check basic
1525 * merging parameters without querying the elevator.
1527 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1529 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1530 unsigned int *request_count,
1531 struct request **same_queue_rq)
1533 struct blk_plug *plug;
1536 struct list_head *plug_list;
1538 plug = current->plug;
1544 plug_list = &plug->mq_list;
1546 plug_list = &plug->list;
1548 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1554 * Only blk-mq multiple hardware queues case checks the
1555 * rq in the same queue, there should be only one such
1559 *same_queue_rq = rq;
1562 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1565 el_ret = blk_try_merge(rq, bio);
1566 if (el_ret == ELEVATOR_BACK_MERGE) {
1567 ret = bio_attempt_back_merge(q, rq, bio);
1570 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1571 ret = bio_attempt_front_merge(q, rq, bio);
1580 void init_request_from_bio(struct request *req, struct bio *bio)
1582 req->cmd_type = REQ_TYPE_FS;
1584 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1585 if (bio->bi_rw & REQ_RAHEAD)
1586 req->cmd_flags |= REQ_FAILFAST_MASK;
1589 req->__sector = bio->bi_iter.bi_sector;
1590 req->ioprio = bio_prio(bio);
1591 blk_rq_bio_prep(req->q, req, bio);
1594 static void blk_queue_bio(struct request_queue *q, struct bio *bio)
1596 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1597 struct blk_plug *plug;
1598 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1599 struct request *req;
1600 unsigned int request_count = 0;
1603 * low level driver can indicate that it wants pages above a
1604 * certain limit bounced to low memory (ie for highmem, or even
1605 * ISA dma in theory)
1607 blk_queue_bounce(q, &bio);
1609 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1610 bio_endio(bio, -EIO);
1614 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1615 spin_lock_irq(q->queue_lock);
1616 where = ELEVATOR_INSERT_FLUSH;
1621 * Check if we can merge with the plugged list before grabbing
1624 if (!blk_queue_nomerges(q) &&
1625 blk_attempt_plug_merge(q, bio, &request_count, NULL))
1628 spin_lock_irq(q->queue_lock);
1630 el_ret = elv_merge(q, &req, bio);
1631 if (el_ret == ELEVATOR_BACK_MERGE) {
1632 if (bio_attempt_back_merge(q, req, bio)) {
1633 elv_bio_merged(q, req, bio);
1634 if (!attempt_back_merge(q, req))
1635 elv_merged_request(q, req, el_ret);
1638 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1639 if (bio_attempt_front_merge(q, req, bio)) {
1640 elv_bio_merged(q, req, bio);
1641 if (!attempt_front_merge(q, req))
1642 elv_merged_request(q, req, el_ret);
1649 * This sync check and mask will be re-done in init_request_from_bio(),
1650 * but we need to set it earlier to expose the sync flag to the
1651 * rq allocator and io schedulers.
1653 rw_flags = bio_data_dir(bio);
1655 rw_flags |= REQ_SYNC;
1658 * Grab a free request. This is might sleep but can not fail.
1659 * Returns with the queue unlocked.
1661 req = get_request(q, rw_flags, bio, GFP_NOIO);
1663 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
1668 * After dropping the lock and possibly sleeping here, our request
1669 * may now be mergeable after it had proven unmergeable (above).
1670 * We don't worry about that case for efficiency. It won't happen
1671 * often, and the elevators are able to handle it.
1673 init_request_from_bio(req, bio);
1675 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1676 req->cpu = raw_smp_processor_id();
1678 plug = current->plug;
1681 * If this is the first request added after a plug, fire
1685 trace_block_plug(q);
1687 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1688 blk_flush_plug_list(plug, false);
1689 trace_block_plug(q);
1692 list_add_tail(&req->queuelist, &plug->list);
1693 blk_account_io_start(req, true);
1695 spin_lock_irq(q->queue_lock);
1696 add_acct_request(q, req, where);
1699 spin_unlock_irq(q->queue_lock);
1704 * If bio->bi_dev is a partition, remap the location
1706 static inline void blk_partition_remap(struct bio *bio)
1708 struct block_device *bdev = bio->bi_bdev;
1710 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1711 struct hd_struct *p = bdev->bd_part;
1713 bio->bi_iter.bi_sector += p->start_sect;
1714 bio->bi_bdev = bdev->bd_contains;
1716 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1718 bio->bi_iter.bi_sector - p->start_sect);
1722 static void handle_bad_sector(struct bio *bio)
1724 char b[BDEVNAME_SIZE];
1726 printk(KERN_INFO "attempt to access beyond end of device\n");
1727 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1728 bdevname(bio->bi_bdev, b),
1730 (unsigned long long)bio_end_sector(bio),
1731 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1734 #ifdef CONFIG_FAIL_MAKE_REQUEST
1736 static DECLARE_FAULT_ATTR(fail_make_request);
1738 static int __init setup_fail_make_request(char *str)
1740 return setup_fault_attr(&fail_make_request, str);
1742 __setup("fail_make_request=", setup_fail_make_request);
1744 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1746 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1749 static int __init fail_make_request_debugfs(void)
1751 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1752 NULL, &fail_make_request);
1754 return PTR_ERR_OR_ZERO(dir);
1757 late_initcall(fail_make_request_debugfs);
1759 #else /* CONFIG_FAIL_MAKE_REQUEST */
1761 static inline bool should_fail_request(struct hd_struct *part,
1767 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1770 * Check whether this bio extends beyond the end of the device.
1772 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1779 /* Test device or partition size, when known. */
1780 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1782 sector_t sector = bio->bi_iter.bi_sector;
1784 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1786 * This may well happen - the kernel calls bread()
1787 * without checking the size of the device, e.g., when
1788 * mounting a device.
1790 handle_bad_sector(bio);
1798 static noinline_for_stack bool
1799 generic_make_request_checks(struct bio *bio)
1801 struct request_queue *q;
1802 int nr_sectors = bio_sectors(bio);
1804 char b[BDEVNAME_SIZE];
1805 struct hd_struct *part;
1809 if (bio_check_eod(bio, nr_sectors))
1812 q = bdev_get_queue(bio->bi_bdev);
1815 "generic_make_request: Trying to access "
1816 "nonexistent block-device %s (%Lu)\n",
1817 bdevname(bio->bi_bdev, b),
1818 (long long) bio->bi_iter.bi_sector);
1822 if (likely(bio_is_rw(bio) &&
1823 nr_sectors > queue_max_hw_sectors(q))) {
1824 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1825 bdevname(bio->bi_bdev, b),
1827 queue_max_hw_sectors(q));
1831 part = bio->bi_bdev->bd_part;
1832 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1833 should_fail_request(&part_to_disk(part)->part0,
1834 bio->bi_iter.bi_size))
1838 * If this device has partitions, remap block n
1839 * of partition p to block n+start(p) of the disk.
1841 blk_partition_remap(bio);
1843 if (bio_check_eod(bio, nr_sectors))
1847 * Filter flush bio's early so that make_request based
1848 * drivers without flush support don't have to worry
1851 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1852 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1859 if ((bio->bi_rw & REQ_DISCARD) &&
1860 (!blk_queue_discard(q) ||
1861 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1866 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1872 * Various block parts want %current->io_context and lazy ioc
1873 * allocation ends up trading a lot of pain for a small amount of
1874 * memory. Just allocate it upfront. This may fail and block
1875 * layer knows how to live with it.
1877 create_io_context(GFP_ATOMIC, q->node);
1879 if (blk_throtl_bio(q, bio))
1880 return false; /* throttled, will be resubmitted later */
1882 trace_block_bio_queue(q, bio);
1886 bio_endio(bio, err);
1891 * generic_make_request - hand a buffer to its device driver for I/O
1892 * @bio: The bio describing the location in memory and on the device.
1894 * generic_make_request() is used to make I/O requests of block
1895 * devices. It is passed a &struct bio, which describes the I/O that needs
1898 * generic_make_request() does not return any status. The
1899 * success/failure status of the request, along with notification of
1900 * completion, is delivered asynchronously through the bio->bi_end_io
1901 * function described (one day) else where.
1903 * The caller of generic_make_request must make sure that bi_io_vec
1904 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1905 * set to describe the device address, and the
1906 * bi_end_io and optionally bi_private are set to describe how
1907 * completion notification should be signaled.
1909 * generic_make_request and the drivers it calls may use bi_next if this
1910 * bio happens to be merged with someone else, and may resubmit the bio to
1911 * a lower device by calling into generic_make_request recursively, which
1912 * means the bio should NOT be touched after the call to ->make_request_fn.
1914 void generic_make_request(struct bio *bio)
1916 struct bio_list bio_list_on_stack;
1918 if (!generic_make_request_checks(bio))
1922 * We only want one ->make_request_fn to be active at a time, else
1923 * stack usage with stacked devices could be a problem. So use
1924 * current->bio_list to keep a list of requests submited by a
1925 * make_request_fn function. current->bio_list is also used as a
1926 * flag to say if generic_make_request is currently active in this
1927 * task or not. If it is NULL, then no make_request is active. If
1928 * it is non-NULL, then a make_request is active, and new requests
1929 * should be added at the tail
1931 if (current->bio_list) {
1932 bio_list_add(current->bio_list, bio);
1936 /* following loop may be a bit non-obvious, and so deserves some
1938 * Before entering the loop, bio->bi_next is NULL (as all callers
1939 * ensure that) so we have a list with a single bio.
1940 * We pretend that we have just taken it off a longer list, so
1941 * we assign bio_list to a pointer to the bio_list_on_stack,
1942 * thus initialising the bio_list of new bios to be
1943 * added. ->make_request() may indeed add some more bios
1944 * through a recursive call to generic_make_request. If it
1945 * did, we find a non-NULL value in bio_list and re-enter the loop
1946 * from the top. In this case we really did just take the bio
1947 * of the top of the list (no pretending) and so remove it from
1948 * bio_list, and call into ->make_request() again.
1950 BUG_ON(bio->bi_next);
1951 bio_list_init(&bio_list_on_stack);
1952 current->bio_list = &bio_list_on_stack;
1954 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1956 q->make_request_fn(q, bio);
1958 bio = bio_list_pop(current->bio_list);
1960 current->bio_list = NULL; /* deactivate */
1962 EXPORT_SYMBOL(generic_make_request);
1965 * submit_bio - submit a bio to the block device layer for I/O
1966 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1967 * @bio: The &struct bio which describes the I/O
1969 * submit_bio() is very similar in purpose to generic_make_request(), and
1970 * uses that function to do most of the work. Both are fairly rough
1971 * interfaces; @bio must be presetup and ready for I/O.
1974 void submit_bio(int rw, struct bio *bio)
1979 * If it's a regular read/write or a barrier with data attached,
1980 * go through the normal accounting stuff before submission.
1982 if (bio_has_data(bio)) {
1985 if (unlikely(rw & REQ_WRITE_SAME))
1986 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1988 count = bio_sectors(bio);
1991 count_vm_events(PGPGOUT, count);
1993 task_io_account_read(bio->bi_iter.bi_size);
1994 count_vm_events(PGPGIN, count);
1997 if (unlikely(block_dump)) {
1998 char b[BDEVNAME_SIZE];
1999 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2000 current->comm, task_pid_nr(current),
2001 (rw & WRITE) ? "WRITE" : "READ",
2002 (unsigned long long)bio->bi_iter.bi_sector,
2003 bdevname(bio->bi_bdev, b),
2008 generic_make_request(bio);
2010 EXPORT_SYMBOL(submit_bio);
2013 * blk_rq_check_limits - Helper function to check a request for the queue limit
2015 * @rq: the request being checked
2018 * @rq may have been made based on weaker limitations of upper-level queues
2019 * in request stacking drivers, and it may violate the limitation of @q.
2020 * Since the block layer and the underlying device driver trust @rq
2021 * after it is inserted to @q, it should be checked against @q before
2022 * the insertion using this generic function.
2024 * This function should also be useful for request stacking drivers
2025 * in some cases below, so export this function.
2026 * Request stacking drivers like request-based dm may change the queue
2027 * limits while requests are in the queue (e.g. dm's table swapping).
2028 * Such request stacking drivers should check those requests against
2029 * the new queue limits again when they dispatch those requests,
2030 * although such checkings are also done against the old queue limits
2031 * when submitting requests.
2033 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2035 if (!rq_mergeable(rq))
2038 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2039 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2044 * queue's settings related to segment counting like q->bounce_pfn
2045 * may differ from that of other stacking queues.
2046 * Recalculate it to check the request correctly on this queue's
2049 blk_recalc_rq_segments(rq);
2050 if (rq->nr_phys_segments > queue_max_segments(q)) {
2051 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2057 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2060 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2061 * @q: the queue to submit the request
2062 * @rq: the request being queued
2064 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2066 unsigned long flags;
2067 int where = ELEVATOR_INSERT_BACK;
2069 if (blk_rq_check_limits(q, rq))
2073 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2077 if (blk_queue_io_stat(q))
2078 blk_account_io_start(rq, true);
2079 blk_mq_insert_request(rq, false, true, true);
2083 spin_lock_irqsave(q->queue_lock, flags);
2084 if (unlikely(blk_queue_dying(q))) {
2085 spin_unlock_irqrestore(q->queue_lock, flags);
2090 * Submitting request must be dequeued before calling this function
2091 * because it will be linked to another request_queue
2093 BUG_ON(blk_queued_rq(rq));
2095 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2096 where = ELEVATOR_INSERT_FLUSH;
2098 add_acct_request(q, rq, where);
2099 if (where == ELEVATOR_INSERT_FLUSH)
2101 spin_unlock_irqrestore(q->queue_lock, flags);
2105 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2108 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2109 * @rq: request to examine
2112 * A request could be merge of IOs which require different failure
2113 * handling. This function determines the number of bytes which
2114 * can be failed from the beginning of the request without
2115 * crossing into area which need to be retried further.
2118 * The number of bytes to fail.
2121 * queue_lock must be held.
2123 unsigned int blk_rq_err_bytes(const struct request *rq)
2125 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2126 unsigned int bytes = 0;
2129 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2130 return blk_rq_bytes(rq);
2133 * Currently the only 'mixing' which can happen is between
2134 * different fastfail types. We can safely fail portions
2135 * which have all the failfast bits that the first one has -
2136 * the ones which are at least as eager to fail as the first
2139 for (bio = rq->bio; bio; bio = bio->bi_next) {
2140 if ((bio->bi_rw & ff) != ff)
2142 bytes += bio->bi_iter.bi_size;
2145 /* this could lead to infinite loop */
2146 BUG_ON(blk_rq_bytes(rq) && !bytes);
2149 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2151 void blk_account_io_completion(struct request *req, unsigned int bytes)
2153 if (blk_do_io_stat(req)) {
2154 const int rw = rq_data_dir(req);
2155 struct hd_struct *part;
2158 cpu = part_stat_lock();
2160 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2165 void blk_account_io_done(struct request *req)
2168 * Account IO completion. flush_rq isn't accounted as a
2169 * normal IO on queueing nor completion. Accounting the
2170 * containing request is enough.
2172 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2173 unsigned long duration = jiffies - req->start_time;
2174 const int rw = rq_data_dir(req);
2175 struct hd_struct *part;
2178 cpu = part_stat_lock();
2181 part_stat_inc(cpu, part, ios[rw]);
2182 part_stat_add(cpu, part, ticks[rw], duration);
2183 part_round_stats(cpu, part);
2184 part_dec_in_flight(part, rw);
2186 hd_struct_put(part);
2193 * Don't process normal requests when queue is suspended
2194 * or in the process of suspending/resuming
2196 static struct request *blk_pm_peek_request(struct request_queue *q,
2199 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2200 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2206 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2213 void blk_account_io_start(struct request *rq, bool new_io)
2215 struct hd_struct *part;
2216 int rw = rq_data_dir(rq);
2219 if (!blk_do_io_stat(rq))
2222 cpu = part_stat_lock();
2226 part_stat_inc(cpu, part, merges[rw]);
2228 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2229 if (!hd_struct_try_get(part)) {
2231 * The partition is already being removed,
2232 * the request will be accounted on the disk only
2234 * We take a reference on disk->part0 although that
2235 * partition will never be deleted, so we can treat
2236 * it as any other partition.
2238 part = &rq->rq_disk->part0;
2239 hd_struct_get(part);
2241 part_round_stats(cpu, part);
2242 part_inc_in_flight(part, rw);
2250 * blk_peek_request - peek at the top of a request queue
2251 * @q: request queue to peek at
2254 * Return the request at the top of @q. The returned request
2255 * should be started using blk_start_request() before LLD starts
2259 * Pointer to the request at the top of @q if available. Null
2263 * queue_lock must be held.
2265 struct request *blk_peek_request(struct request_queue *q)
2270 while ((rq = __elv_next_request(q)) != NULL) {
2272 rq = blk_pm_peek_request(q, rq);
2276 if (!(rq->cmd_flags & REQ_STARTED)) {
2278 * This is the first time the device driver
2279 * sees this request (possibly after
2280 * requeueing). Notify IO scheduler.
2282 if (rq->cmd_flags & REQ_SORTED)
2283 elv_activate_rq(q, rq);
2286 * just mark as started even if we don't start
2287 * it, a request that has been delayed should
2288 * not be passed by new incoming requests
2290 rq->cmd_flags |= REQ_STARTED;
2291 trace_block_rq_issue(q, rq);
2294 if (!q->boundary_rq || q->boundary_rq == rq) {
2295 q->end_sector = rq_end_sector(rq);
2296 q->boundary_rq = NULL;
2299 if (rq->cmd_flags & REQ_DONTPREP)
2302 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2304 * make sure space for the drain appears we
2305 * know we can do this because max_hw_segments
2306 * has been adjusted to be one fewer than the
2309 rq->nr_phys_segments++;
2315 ret = q->prep_rq_fn(q, rq);
2316 if (ret == BLKPREP_OK) {
2318 } else if (ret == BLKPREP_DEFER) {
2320 * the request may have been (partially) prepped.
2321 * we need to keep this request in the front to
2322 * avoid resource deadlock. REQ_STARTED will
2323 * prevent other fs requests from passing this one.
2325 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2326 !(rq->cmd_flags & REQ_DONTPREP)) {
2328 * remove the space for the drain we added
2329 * so that we don't add it again
2331 --rq->nr_phys_segments;
2336 } else if (ret == BLKPREP_KILL) {
2337 rq->cmd_flags |= REQ_QUIET;
2339 * Mark this request as started so we don't trigger
2340 * any debug logic in the end I/O path.
2342 blk_start_request(rq);
2343 __blk_end_request_all(rq, -EIO);
2345 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2352 EXPORT_SYMBOL(blk_peek_request);
2354 void blk_dequeue_request(struct request *rq)
2356 struct request_queue *q = rq->q;
2358 BUG_ON(list_empty(&rq->queuelist));
2359 BUG_ON(ELV_ON_HASH(rq));
2361 list_del_init(&rq->queuelist);
2364 * the time frame between a request being removed from the lists
2365 * and to it is freed is accounted as io that is in progress at
2368 if (blk_account_rq(rq)) {
2369 q->in_flight[rq_is_sync(rq)]++;
2370 set_io_start_time_ns(rq);
2375 * blk_start_request - start request processing on the driver
2376 * @req: request to dequeue
2379 * Dequeue @req and start timeout timer on it. This hands off the
2380 * request to the driver.
2382 * Block internal functions which don't want to start timer should
2383 * call blk_dequeue_request().
2386 * queue_lock must be held.
2388 void blk_start_request(struct request *req)
2390 blk_dequeue_request(req);
2393 * We are now handing the request to the hardware, initialize
2394 * resid_len to full count and add the timeout handler.
2396 req->resid_len = blk_rq_bytes(req);
2397 if (unlikely(blk_bidi_rq(req)))
2398 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2400 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2403 EXPORT_SYMBOL(blk_start_request);
2406 * blk_fetch_request - fetch a request from a request queue
2407 * @q: request queue to fetch a request from
2410 * Return the request at the top of @q. The request is started on
2411 * return and LLD can start processing it immediately.
2414 * Pointer to the request at the top of @q if available. Null
2418 * queue_lock must be held.
2420 struct request *blk_fetch_request(struct request_queue *q)
2424 rq = blk_peek_request(q);
2426 blk_start_request(rq);
2429 EXPORT_SYMBOL(blk_fetch_request);
2432 * blk_update_request - Special helper function for request stacking drivers
2433 * @req: the request being processed
2434 * @error: %0 for success, < %0 for error
2435 * @nr_bytes: number of bytes to complete @req
2438 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2439 * the request structure even if @req doesn't have leftover.
2440 * If @req has leftover, sets it up for the next range of segments.
2442 * This special helper function is only for request stacking drivers
2443 * (e.g. request-based dm) so that they can handle partial completion.
2444 * Actual device drivers should use blk_end_request instead.
2446 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2447 * %false return from this function.
2450 * %false - this request doesn't have any more data
2451 * %true - this request has more data
2453 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2457 trace_block_rq_complete(req->q, req, nr_bytes);
2463 * For fs requests, rq is just carrier of independent bio's
2464 * and each partial completion should be handled separately.
2465 * Reset per-request error on each partial completion.
2467 * TODO: tj: This is too subtle. It would be better to let
2468 * low level drivers do what they see fit.
2470 if (req->cmd_type == REQ_TYPE_FS)
2473 if (error && req->cmd_type == REQ_TYPE_FS &&
2474 !(req->cmd_flags & REQ_QUIET)) {
2479 error_type = "recoverable transport";
2482 error_type = "critical target";
2485 error_type = "critical nexus";
2488 error_type = "timeout";
2491 error_type = "critical space allocation";
2494 error_type = "critical medium";
2501 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2502 __func__, error_type, req->rq_disk ?
2503 req->rq_disk->disk_name : "?",
2504 (unsigned long long)blk_rq_pos(req));
2508 blk_account_io_completion(req, nr_bytes);
2512 struct bio *bio = req->bio;
2513 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2515 if (bio_bytes == bio->bi_iter.bi_size)
2516 req->bio = bio->bi_next;
2518 req_bio_endio(req, bio, bio_bytes, error);
2520 total_bytes += bio_bytes;
2521 nr_bytes -= bio_bytes;
2532 * Reset counters so that the request stacking driver
2533 * can find how many bytes remain in the request
2536 req->__data_len = 0;
2540 req->__data_len -= total_bytes;
2542 /* update sector only for requests with clear definition of sector */
2543 if (req->cmd_type == REQ_TYPE_FS)
2544 req->__sector += total_bytes >> 9;
2546 /* mixed attributes always follow the first bio */
2547 if (req->cmd_flags & REQ_MIXED_MERGE) {
2548 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2549 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2553 * If total number of sectors is less than the first segment
2554 * size, something has gone terribly wrong.
2556 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2557 blk_dump_rq_flags(req, "request botched");
2558 req->__data_len = blk_rq_cur_bytes(req);
2561 /* recalculate the number of segments */
2562 blk_recalc_rq_segments(req);
2566 EXPORT_SYMBOL_GPL(blk_update_request);
2568 static bool blk_update_bidi_request(struct request *rq, int error,
2569 unsigned int nr_bytes,
2570 unsigned int bidi_bytes)
2572 if (blk_update_request(rq, error, nr_bytes))
2575 /* Bidi request must be completed as a whole */
2576 if (unlikely(blk_bidi_rq(rq)) &&
2577 blk_update_request(rq->next_rq, error, bidi_bytes))
2580 if (blk_queue_add_random(rq->q))
2581 add_disk_randomness(rq->rq_disk);
2587 * blk_unprep_request - unprepare a request
2590 * This function makes a request ready for complete resubmission (or
2591 * completion). It happens only after all error handling is complete,
2592 * so represents the appropriate moment to deallocate any resources
2593 * that were allocated to the request in the prep_rq_fn. The queue
2594 * lock is held when calling this.
2596 void blk_unprep_request(struct request *req)
2598 struct request_queue *q = req->q;
2600 req->cmd_flags &= ~REQ_DONTPREP;
2601 if (q->unprep_rq_fn)
2602 q->unprep_rq_fn(q, req);
2604 EXPORT_SYMBOL_GPL(blk_unprep_request);
2607 * queue lock must be held
2609 void blk_finish_request(struct request *req, int error)
2611 if (req->cmd_flags & REQ_QUEUED)
2612 blk_queue_end_tag(req->q, req);
2614 BUG_ON(blk_queued_rq(req));
2616 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2617 laptop_io_completion(&req->q->backing_dev_info);
2619 blk_delete_timer(req);
2621 if (req->cmd_flags & REQ_DONTPREP)
2622 blk_unprep_request(req);
2624 blk_account_io_done(req);
2627 req->end_io(req, error);
2629 if (blk_bidi_rq(req))
2630 __blk_put_request(req->next_rq->q, req->next_rq);
2632 __blk_put_request(req->q, req);
2635 EXPORT_SYMBOL(blk_finish_request);
2638 * blk_end_bidi_request - Complete a bidi request
2639 * @rq: the request to complete
2640 * @error: %0 for success, < %0 for error
2641 * @nr_bytes: number of bytes to complete @rq
2642 * @bidi_bytes: number of bytes to complete @rq->next_rq
2645 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2646 * Drivers that supports bidi can safely call this member for any
2647 * type of request, bidi or uni. In the later case @bidi_bytes is
2651 * %false - we are done with this request
2652 * %true - still buffers pending for this request
2654 static bool blk_end_bidi_request(struct request *rq, int error,
2655 unsigned int nr_bytes, unsigned int bidi_bytes)
2657 struct request_queue *q = rq->q;
2658 unsigned long flags;
2660 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2663 spin_lock_irqsave(q->queue_lock, flags);
2664 blk_finish_request(rq, error);
2665 spin_unlock_irqrestore(q->queue_lock, flags);
2671 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2672 * @rq: the request to complete
2673 * @error: %0 for success, < %0 for error
2674 * @nr_bytes: number of bytes to complete @rq
2675 * @bidi_bytes: number of bytes to complete @rq->next_rq
2678 * Identical to blk_end_bidi_request() except that queue lock is
2679 * assumed to be locked on entry and remains so on return.
2682 * %false - we are done with this request
2683 * %true - still buffers pending for this request
2685 bool __blk_end_bidi_request(struct request *rq, int error,
2686 unsigned int nr_bytes, unsigned int bidi_bytes)
2688 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2691 blk_finish_request(rq, error);
2697 * blk_end_request - Helper function for drivers to complete the request.
2698 * @rq: the request being processed
2699 * @error: %0 for success, < %0 for error
2700 * @nr_bytes: number of bytes to complete
2703 * Ends I/O on a number of bytes attached to @rq.
2704 * If @rq has leftover, sets it up for the next range of segments.
2707 * %false - we are done with this request
2708 * %true - still buffers pending for this request
2710 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2712 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2714 EXPORT_SYMBOL(blk_end_request);
2717 * blk_end_request_all - Helper function for drives to finish the request.
2718 * @rq: the request to finish
2719 * @error: %0 for success, < %0 for error
2722 * Completely finish @rq.
2724 void blk_end_request_all(struct request *rq, int error)
2727 unsigned int bidi_bytes = 0;
2729 if (unlikely(blk_bidi_rq(rq)))
2730 bidi_bytes = blk_rq_bytes(rq->next_rq);
2732 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2735 EXPORT_SYMBOL(blk_end_request_all);
2738 * blk_end_request_cur - Helper function to finish the current request chunk.
2739 * @rq: the request to finish the current chunk for
2740 * @error: %0 for success, < %0 for error
2743 * Complete the current consecutively mapped chunk from @rq.
2746 * %false - we are done with this request
2747 * %true - still buffers pending for this request
2749 bool blk_end_request_cur(struct request *rq, int error)
2751 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2753 EXPORT_SYMBOL(blk_end_request_cur);
2756 * blk_end_request_err - Finish a request till the next failure boundary.
2757 * @rq: the request to finish till the next failure boundary for
2758 * @error: must be negative errno
2761 * Complete @rq till the next failure boundary.
2764 * %false - we are done with this request
2765 * %true - still buffers pending for this request
2767 bool blk_end_request_err(struct request *rq, int error)
2769 WARN_ON(error >= 0);
2770 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2772 EXPORT_SYMBOL_GPL(blk_end_request_err);
2775 * __blk_end_request - Helper function for drivers to complete the request.
2776 * @rq: the request being processed
2777 * @error: %0 for success, < %0 for error
2778 * @nr_bytes: number of bytes to complete
2781 * Must be called with queue lock held unlike blk_end_request().
2784 * %false - we are done with this request
2785 * %true - still buffers pending for this request
2787 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2789 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2791 EXPORT_SYMBOL(__blk_end_request);
2794 * __blk_end_request_all - Helper function for drives to finish the request.
2795 * @rq: the request to finish
2796 * @error: %0 for success, < %0 for error
2799 * Completely finish @rq. Must be called with queue lock held.
2801 void __blk_end_request_all(struct request *rq, int error)
2804 unsigned int bidi_bytes = 0;
2806 if (unlikely(blk_bidi_rq(rq)))
2807 bidi_bytes = blk_rq_bytes(rq->next_rq);
2809 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2812 EXPORT_SYMBOL(__blk_end_request_all);
2815 * __blk_end_request_cur - Helper function to finish the current request chunk.
2816 * @rq: the request to finish the current chunk for
2817 * @error: %0 for success, < %0 for error
2820 * Complete the current consecutively mapped chunk from @rq. Must
2821 * be called with queue lock held.
2824 * %false - we are done with this request
2825 * %true - still buffers pending for this request
2827 bool __blk_end_request_cur(struct request *rq, int error)
2829 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2831 EXPORT_SYMBOL(__blk_end_request_cur);
2834 * __blk_end_request_err - Finish a request till the next failure boundary.
2835 * @rq: the request to finish till the next failure boundary for
2836 * @error: must be negative errno
2839 * Complete @rq till the next failure boundary. Must be called
2840 * with queue lock held.
2843 * %false - we are done with this request
2844 * %true - still buffers pending for this request
2846 bool __blk_end_request_err(struct request *rq, int error)
2848 WARN_ON(error >= 0);
2849 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2851 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2853 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2856 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2857 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2859 if (bio_has_data(bio))
2860 rq->nr_phys_segments = bio_phys_segments(q, bio);
2862 rq->__data_len = bio->bi_iter.bi_size;
2863 rq->bio = rq->biotail = bio;
2866 rq->rq_disk = bio->bi_bdev->bd_disk;
2869 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2871 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2872 * @rq: the request to be flushed
2875 * Flush all pages in @rq.
2877 void rq_flush_dcache_pages(struct request *rq)
2879 struct req_iterator iter;
2880 struct bio_vec bvec;
2882 rq_for_each_segment(bvec, rq, iter)
2883 flush_dcache_page(bvec.bv_page);
2885 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2889 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2890 * @q : the queue of the device being checked
2893 * Check if underlying low-level drivers of a device are busy.
2894 * If the drivers want to export their busy state, they must set own
2895 * exporting function using blk_queue_lld_busy() first.
2897 * Basically, this function is used only by request stacking drivers
2898 * to stop dispatching requests to underlying devices when underlying
2899 * devices are busy. This behavior helps more I/O merging on the queue
2900 * of the request stacking driver and prevents I/O throughput regression
2901 * on burst I/O load.
2904 * 0 - Not busy (The request stacking driver should dispatch request)
2905 * 1 - Busy (The request stacking driver should stop dispatching request)
2907 int blk_lld_busy(struct request_queue *q)
2910 return q->lld_busy_fn(q);
2914 EXPORT_SYMBOL_GPL(blk_lld_busy);
2916 void blk_rq_prep_clone(struct request *dst, struct request *src)
2918 dst->cpu = src->cpu;
2919 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK);
2920 dst->cmd_flags |= REQ_NOMERGE | REQ_CLONE;
2921 dst->cmd_type = src->cmd_type;
2922 dst->__sector = blk_rq_pos(src);
2923 dst->__data_len = blk_rq_bytes(src);
2924 dst->nr_phys_segments = src->nr_phys_segments;
2925 dst->ioprio = src->ioprio;
2926 dst->extra_len = src->extra_len;
2927 dst->bio = src->bio;
2928 dst->biotail = src->biotail;
2929 dst->cmd = src->cmd;
2930 dst->cmd_len = src->cmd_len;
2931 dst->sense = src->sense;
2933 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2935 int kblockd_schedule_work(struct work_struct *work)
2937 return queue_work(kblockd_workqueue, work);
2939 EXPORT_SYMBOL(kblockd_schedule_work);
2941 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2942 unsigned long delay)
2944 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2946 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2948 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2949 unsigned long delay)
2951 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2953 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2956 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2957 * @plug: The &struct blk_plug that needs to be initialized
2960 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2961 * pending I/O should the task end up blocking between blk_start_plug() and
2962 * blk_finish_plug(). This is important from a performance perspective, but
2963 * also ensures that we don't deadlock. For instance, if the task is blocking
2964 * for a memory allocation, memory reclaim could end up wanting to free a
2965 * page belonging to that request that is currently residing in our private
2966 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2967 * this kind of deadlock.
2969 void blk_start_plug(struct blk_plug *plug)
2971 struct task_struct *tsk = current;
2974 * If this is a nested plug, don't actually assign it.
2979 INIT_LIST_HEAD(&plug->list);
2980 INIT_LIST_HEAD(&plug->mq_list);
2981 INIT_LIST_HEAD(&plug->cb_list);
2983 * Store ordering should not be needed here, since a potential
2984 * preempt will imply a full memory barrier
2988 EXPORT_SYMBOL(blk_start_plug);
2990 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2992 struct request *rqa = container_of(a, struct request, queuelist);
2993 struct request *rqb = container_of(b, struct request, queuelist);
2995 return !(rqa->q < rqb->q ||
2996 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3000 * If 'from_schedule' is true, then postpone the dispatch of requests
3001 * until a safe kblockd context. We due this to avoid accidental big
3002 * additional stack usage in driver dispatch, in places where the originally
3003 * plugger did not intend it.
3005 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3007 __releases(q->queue_lock)
3009 trace_block_unplug(q, depth, !from_schedule);
3012 blk_run_queue_async(q);
3015 spin_unlock(q->queue_lock);
3018 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3020 LIST_HEAD(callbacks);
3022 while (!list_empty(&plug->cb_list)) {
3023 list_splice_init(&plug->cb_list, &callbacks);
3025 while (!list_empty(&callbacks)) {
3026 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3029 list_del(&cb->list);
3030 cb->callback(cb, from_schedule);
3035 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3038 struct blk_plug *plug = current->plug;
3039 struct blk_plug_cb *cb;
3044 list_for_each_entry(cb, &plug->cb_list, list)
3045 if (cb->callback == unplug && cb->data == data)
3048 /* Not currently on the callback list */
3049 BUG_ON(size < sizeof(*cb));
3050 cb = kzalloc(size, GFP_ATOMIC);
3053 cb->callback = unplug;
3054 list_add(&cb->list, &plug->cb_list);
3058 EXPORT_SYMBOL(blk_check_plugged);
3060 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3062 struct request_queue *q;
3063 unsigned long flags;
3068 flush_plug_callbacks(plug, from_schedule);
3070 if (!list_empty(&plug->mq_list))
3071 blk_mq_flush_plug_list(plug, from_schedule);
3073 if (list_empty(&plug->list))
3076 list_splice_init(&plug->list, &list);
3078 list_sort(NULL, &list, plug_rq_cmp);
3084 * Save and disable interrupts here, to avoid doing it for every
3085 * queue lock we have to take.
3087 local_irq_save(flags);
3088 while (!list_empty(&list)) {
3089 rq = list_entry_rq(list.next);
3090 list_del_init(&rq->queuelist);
3094 * This drops the queue lock
3097 queue_unplugged(q, depth, from_schedule);
3100 spin_lock(q->queue_lock);
3104 * Short-circuit if @q is dead
3106 if (unlikely(blk_queue_dying(q))) {
3107 __blk_end_request_all(rq, -ENODEV);
3112 * rq is already accounted, so use raw insert
3114 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3115 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3117 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3123 * This drops the queue lock
3126 queue_unplugged(q, depth, from_schedule);
3128 local_irq_restore(flags);
3131 void blk_finish_plug(struct blk_plug *plug)
3133 if (plug != current->plug)
3135 blk_flush_plug_list(plug, false);
3137 current->plug = NULL;
3139 EXPORT_SYMBOL(blk_finish_plug);
3143 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3144 * @q: the queue of the device
3145 * @dev: the device the queue belongs to
3148 * Initialize runtime-PM-related fields for @q and start auto suspend for
3149 * @dev. Drivers that want to take advantage of request-based runtime PM
3150 * should call this function after @dev has been initialized, and its
3151 * request queue @q has been allocated, and runtime PM for it can not happen
3152 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3153 * cases, driver should call this function before any I/O has taken place.
3155 * This function takes care of setting up using auto suspend for the device,
3156 * the autosuspend delay is set to -1 to make runtime suspend impossible
3157 * until an updated value is either set by user or by driver. Drivers do
3158 * not need to touch other autosuspend settings.
3160 * The block layer runtime PM is request based, so only works for drivers
3161 * that use request as their IO unit instead of those directly use bio's.
3163 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3166 q->rpm_status = RPM_ACTIVE;
3167 pm_runtime_set_autosuspend_delay(q->dev, -1);
3168 pm_runtime_use_autosuspend(q->dev);
3170 EXPORT_SYMBOL(blk_pm_runtime_init);
3173 * blk_pre_runtime_suspend - Pre runtime suspend check
3174 * @q: the queue of the device
3177 * This function will check if runtime suspend is allowed for the device
3178 * by examining if there are any requests pending in the queue. If there
3179 * are requests pending, the device can not be runtime suspended; otherwise,
3180 * the queue's status will be updated to SUSPENDING and the driver can
3181 * proceed to suspend the device.
3183 * For the not allowed case, we mark last busy for the device so that
3184 * runtime PM core will try to autosuspend it some time later.
3186 * This function should be called near the start of the device's
3187 * runtime_suspend callback.
3190 * 0 - OK to runtime suspend the device
3191 * -EBUSY - Device should not be runtime suspended
3193 int blk_pre_runtime_suspend(struct request_queue *q)
3197 spin_lock_irq(q->queue_lock);
3198 if (q->nr_pending) {
3200 pm_runtime_mark_last_busy(q->dev);
3202 q->rpm_status = RPM_SUSPENDING;
3204 spin_unlock_irq(q->queue_lock);
3207 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3210 * blk_post_runtime_suspend - Post runtime suspend processing
3211 * @q: the queue of the device
3212 * @err: return value of the device's runtime_suspend function
3215 * Update the queue's runtime status according to the return value of the
3216 * device's runtime suspend function and mark last busy for the device so
3217 * that PM core will try to auto suspend the device at a later time.
3219 * This function should be called near the end of the device's
3220 * runtime_suspend callback.
3222 void blk_post_runtime_suspend(struct request_queue *q, int err)
3224 spin_lock_irq(q->queue_lock);
3226 q->rpm_status = RPM_SUSPENDED;
3228 q->rpm_status = RPM_ACTIVE;
3229 pm_runtime_mark_last_busy(q->dev);
3231 spin_unlock_irq(q->queue_lock);
3233 EXPORT_SYMBOL(blk_post_runtime_suspend);
3236 * blk_pre_runtime_resume - Pre runtime resume processing
3237 * @q: the queue of the device
3240 * Update the queue's runtime status to RESUMING in preparation for the
3241 * runtime resume of the device.
3243 * This function should be called near the start of the device's
3244 * runtime_resume callback.
3246 void blk_pre_runtime_resume(struct request_queue *q)
3248 spin_lock_irq(q->queue_lock);
3249 q->rpm_status = RPM_RESUMING;
3250 spin_unlock_irq(q->queue_lock);
3252 EXPORT_SYMBOL(blk_pre_runtime_resume);
3255 * blk_post_runtime_resume - Post runtime resume processing
3256 * @q: the queue of the device
3257 * @err: return value of the device's runtime_resume function
3260 * Update the queue's runtime status according to the return value of the
3261 * device's runtime_resume function. If it is successfully resumed, process
3262 * the requests that are queued into the device's queue when it is resuming
3263 * and then mark last busy and initiate autosuspend for it.
3265 * This function should be called near the end of the device's
3266 * runtime_resume callback.
3268 void blk_post_runtime_resume(struct request_queue *q, int err)
3270 spin_lock_irq(q->queue_lock);
3272 q->rpm_status = RPM_ACTIVE;
3274 pm_runtime_mark_last_busy(q->dev);
3275 pm_request_autosuspend(q->dev);
3277 q->rpm_status = RPM_SUSPENDED;
3279 spin_unlock_irq(q->queue_lock);
3281 EXPORT_SYMBOL(blk_post_runtime_resume);
3284 int __init blk_dev_init(void)
3286 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3287 sizeof(((struct request *)0)->cmd_flags));
3289 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3290 kblockd_workqueue = alloc_workqueue("kblockd",
3291 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3292 if (!kblockd_workqueue)
3293 panic("Failed to create kblockd\n");
3295 request_cachep = kmem_cache_create("blkdev_requests",
3296 sizeof(struct request), 0, SLAB_PANIC, NULL);
3298 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3299 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);