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_unplug);
48 DEFINE_IDA(blk_queue_ida);
51 * For the allocated request tables
53 struct kmem_cache *request_cachep = NULL;
56 * For queue allocation
58 struct kmem_cache *blk_requestq_cachep;
61 * Controlling structure to kblockd
63 static struct workqueue_struct *kblockd_workqueue;
65 void blk_queue_congestion_threshold(struct request_queue *q)
69 nr = q->nr_requests - (q->nr_requests / 8) + 1;
70 if (nr > q->nr_requests)
72 q->nr_congestion_on = nr;
74 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
77 q->nr_congestion_off = nr;
81 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
84 * Locates the passed device's request queue and returns the address of its
87 * Will return NULL if the request queue cannot be located.
89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
91 struct backing_dev_info *ret = NULL;
92 struct request_queue *q = bdev_get_queue(bdev);
95 ret = &q->backing_dev_info;
98 EXPORT_SYMBOL(blk_get_backing_dev_info);
100 void blk_rq_init(struct request_queue *q, struct request *rq)
102 memset(rq, 0, sizeof(*rq));
104 INIT_LIST_HEAD(&rq->queuelist);
105 INIT_LIST_HEAD(&rq->timeout_list);
108 rq->__sector = (sector_t) -1;
109 INIT_HLIST_NODE(&rq->hash);
110 RB_CLEAR_NODE(&rq->rb_node);
112 rq->cmd_len = BLK_MAX_CDB;
114 rq->start_time = jiffies;
115 set_start_time_ns(rq);
118 EXPORT_SYMBOL(blk_rq_init);
120 static void req_bio_endio(struct request *rq, struct bio *bio,
121 unsigned int nbytes, int error)
124 clear_bit(BIO_UPTODATE, &bio->bi_flags);
125 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
128 if (unlikely(rq->cmd_flags & REQ_QUIET))
129 set_bit(BIO_QUIET, &bio->bi_flags);
131 bio_advance(bio, nbytes);
133 /* don't actually finish bio if it's part of flush sequence */
134 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
135 bio_endio(bio, error);
138 void blk_dump_rq_flags(struct request *rq, char *msg)
142 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
143 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
144 (unsigned long long) rq->cmd_flags);
146 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
147 (unsigned long long)blk_rq_pos(rq),
148 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
149 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
150 rq->bio, rq->biotail, blk_rq_bytes(rq));
152 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
153 printk(KERN_INFO " cdb: ");
154 for (bit = 0; bit < BLK_MAX_CDB; bit++)
155 printk("%02x ", rq->cmd[bit]);
159 EXPORT_SYMBOL(blk_dump_rq_flags);
161 static void blk_delay_work(struct work_struct *work)
163 struct request_queue *q;
165 q = container_of(work, struct request_queue, delay_work.work);
166 spin_lock_irq(q->queue_lock);
168 spin_unlock_irq(q->queue_lock);
172 * blk_delay_queue - restart queueing after defined interval
173 * @q: The &struct request_queue in question
174 * @msecs: Delay in msecs
177 * Sometimes queueing needs to be postponed for a little while, to allow
178 * resources to come back. This function will make sure that queueing is
179 * restarted around the specified time. Queue lock must be held.
181 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
183 if (likely(!blk_queue_dead(q)))
184 queue_delayed_work(kblockd_workqueue, &q->delay_work,
185 msecs_to_jiffies(msecs));
187 EXPORT_SYMBOL(blk_delay_queue);
190 * blk_start_queue - restart a previously stopped queue
191 * @q: The &struct request_queue in question
194 * blk_start_queue() will clear the stop flag on the queue, and call
195 * the request_fn for the queue if it was in a stopped state when
196 * entered. Also see blk_stop_queue(). Queue lock must be held.
198 void blk_start_queue(struct request_queue *q)
200 WARN_ON(!irqs_disabled());
202 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
205 EXPORT_SYMBOL(blk_start_queue);
208 * blk_stop_queue - stop a queue
209 * @q: The &struct request_queue in question
212 * The Linux block layer assumes that a block driver will consume all
213 * entries on the request queue when the request_fn strategy is called.
214 * Often this will not happen, because of hardware limitations (queue
215 * depth settings). If a device driver gets a 'queue full' response,
216 * or if it simply chooses not to queue more I/O at one point, it can
217 * call this function to prevent the request_fn from being called until
218 * the driver has signalled it's ready to go again. This happens by calling
219 * blk_start_queue() to restart queue operations. Queue lock must be held.
221 void blk_stop_queue(struct request_queue *q)
223 cancel_delayed_work(&q->delay_work);
224 queue_flag_set(QUEUE_FLAG_STOPPED, q);
226 EXPORT_SYMBOL(blk_stop_queue);
229 * blk_sync_queue - cancel any pending callbacks on a queue
233 * The block layer may perform asynchronous callback activity
234 * on a queue, such as calling the unplug function after a timeout.
235 * A block device may call blk_sync_queue to ensure that any
236 * such activity is cancelled, thus allowing it to release resources
237 * that the callbacks might use. The caller must already have made sure
238 * that its ->make_request_fn will not re-add plugging prior to calling
241 * This function does not cancel any asynchronous activity arising
242 * out of elevator or throttling code. That would require elevaotor_exit()
243 * and blkcg_exit_queue() to be called with queue lock initialized.
246 void blk_sync_queue(struct request_queue *q)
248 del_timer_sync(&q->timeout);
251 struct blk_mq_hw_ctx *hctx;
254 queue_for_each_hw_ctx(q, hctx, i) {
255 cancel_delayed_work_sync(&hctx->run_work);
256 cancel_delayed_work_sync(&hctx->delay_work);
259 cancel_delayed_work_sync(&q->delay_work);
262 EXPORT_SYMBOL(blk_sync_queue);
265 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
266 * @q: The queue to run
269 * Invoke request handling on a queue if there are any pending requests.
270 * May be used to restart request handling after a request has completed.
271 * This variant runs the queue whether or not the queue has been
272 * stopped. Must be called with the queue lock held and interrupts
273 * disabled. See also @blk_run_queue.
275 inline void __blk_run_queue_uncond(struct request_queue *q)
277 if (unlikely(blk_queue_dead(q)))
281 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
282 * the queue lock internally. As a result multiple threads may be
283 * running such a request function concurrently. Keep track of the
284 * number of active request_fn invocations such that blk_drain_queue()
285 * can wait until all these request_fn calls have finished.
287 q->request_fn_active++;
289 q->request_fn_active--;
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue *q)
302 if (unlikely(blk_queue_stopped(q)))
305 __blk_run_queue_uncond(q);
307 EXPORT_SYMBOL(__blk_run_queue);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 * of us. The caller must hold the queue lock.
317 void blk_run_queue_async(struct request_queue *q)
319 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
320 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
322 EXPORT_SYMBOL(blk_run_queue_async);
325 * blk_run_queue - run a single device queue
326 * @q: The queue to run
329 * Invoke request handling on this queue, if it has pending work to do.
330 * May be used to restart queueing when a request has completed.
332 void blk_run_queue(struct request_queue *q)
336 spin_lock_irqsave(q->queue_lock, flags);
338 spin_unlock_irqrestore(q->queue_lock, flags);
340 EXPORT_SYMBOL(blk_run_queue);
342 void blk_put_queue(struct request_queue *q)
344 kobject_put(&q->kobj);
346 EXPORT_SYMBOL(blk_put_queue);
349 * __blk_drain_queue - drain requests from request_queue
351 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
353 * Drain requests from @q. If @drain_all is set, all requests are drained.
354 * If not, only ELVPRIV requests are drained. The caller is responsible
355 * for ensuring that no new requests which need to be drained are queued.
357 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
358 __releases(q->queue_lock)
359 __acquires(q->queue_lock)
363 lockdep_assert_held(q->queue_lock);
369 * The caller might be trying to drain @q before its
370 * elevator is initialized.
373 elv_drain_elevator(q);
375 blkcg_drain_queue(q);
378 * This function might be called on a queue which failed
379 * driver init after queue creation or is not yet fully
380 * active yet. Some drivers (e.g. fd and loop) get unhappy
381 * in such cases. Kick queue iff dispatch queue has
382 * something on it and @q has request_fn set.
384 if (!list_empty(&q->queue_head) && q->request_fn)
387 drain |= q->nr_rqs_elvpriv;
388 drain |= q->request_fn_active;
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
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];
400 drain |= !list_empty(&q->flush_queue[i]);
407 spin_unlock_irq(q->queue_lock);
411 spin_lock_irq(q->queue_lock);
415 * With queue marked dead, any woken up waiter will fail the
416 * allocation path, so the wakeup chaining is lost and we're
417 * left with hung waiters. We need to wake up those waiters.
420 struct request_list *rl;
422 blk_queue_for_each_rl(rl, q)
423 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
424 wake_up_all(&rl->wait[i]);
429 * blk_queue_bypass_start - enter queue bypass mode
430 * @q: queue of interest
432 * In bypass mode, only the dispatch FIFO queue of @q is used. This
433 * function makes @q enter bypass mode and drains all requests which were
434 * throttled or issued before. On return, it's guaranteed that no request
435 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
436 * inside queue or RCU read lock.
438 void blk_queue_bypass_start(struct request_queue *q)
442 spin_lock_irq(q->queue_lock);
443 drain = !q->bypass_depth++;
444 queue_flag_set(QUEUE_FLAG_BYPASS, q);
445 spin_unlock_irq(q->queue_lock);
448 spin_lock_irq(q->queue_lock);
449 __blk_drain_queue(q, false);
450 spin_unlock_irq(q->queue_lock);
452 /* ensure blk_queue_bypass() is %true inside RCU read lock */
456 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
459 * blk_queue_bypass_end - leave queue bypass mode
460 * @q: queue of interest
462 * Leave bypass mode and restore the normal queueing behavior.
464 void blk_queue_bypass_end(struct request_queue *q)
466 spin_lock_irq(q->queue_lock);
467 if (!--q->bypass_depth)
468 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
469 WARN_ON_ONCE(q->bypass_depth < 0);
470 spin_unlock_irq(q->queue_lock);
472 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
475 * blk_cleanup_queue - shutdown a request queue
476 * @q: request queue to shutdown
478 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
479 * put it. All future requests will be failed immediately with -ENODEV.
481 void blk_cleanup_queue(struct request_queue *q)
483 spinlock_t *lock = q->queue_lock;
485 /* mark @q DYING, no new request or merges will be allowed afterwards */
486 mutex_lock(&q->sysfs_lock);
487 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
491 * A dying queue is permanently in bypass mode till released. Note
492 * that, unlike blk_queue_bypass_start(), we aren't performing
493 * synchronize_rcu() after entering bypass mode to avoid the delay
494 * as some drivers create and destroy a lot of queues while
495 * probing. This is still safe because blk_release_queue() will be
496 * called only after the queue refcnt drops to zero and nothing,
497 * RCU or not, would be traversing the queue by then.
500 queue_flag_set(QUEUE_FLAG_BYPASS, q);
502 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
503 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
504 queue_flag_set(QUEUE_FLAG_DYING, q);
505 spin_unlock_irq(lock);
506 mutex_unlock(&q->sysfs_lock);
509 * Drain all requests queued before DYING marking. Set DEAD flag to
510 * prevent that q->request_fn() gets invoked after draining finished.
513 blk_mq_drain_queue(q);
517 __blk_drain_queue(q, true);
519 queue_flag_set(QUEUE_FLAG_DEAD, q);
520 spin_unlock_irq(lock);
522 /* @q won't process any more request, flush async actions */
523 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
527 if (q->queue_lock != &q->__queue_lock)
528 q->queue_lock = &q->__queue_lock;
529 spin_unlock_irq(lock);
531 /* @q is and will stay empty, shutdown and put */
534 EXPORT_SYMBOL(blk_cleanup_queue);
536 int blk_init_rl(struct request_list *rl, struct request_queue *q,
539 if (unlikely(rl->rq_pool))
543 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
544 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
545 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
546 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
548 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
549 mempool_free_slab, request_cachep,
557 void blk_exit_rl(struct request_list *rl)
560 mempool_destroy(rl->rq_pool);
563 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
565 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
567 EXPORT_SYMBOL(blk_alloc_queue);
569 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
571 struct request_queue *q;
574 q = kmem_cache_alloc_node(blk_requestq_cachep,
575 gfp_mask | __GFP_ZERO, node_id);
579 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
583 q->backing_dev_info.ra_pages =
584 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
585 q->backing_dev_info.state = 0;
586 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
587 q->backing_dev_info.name = "block";
590 err = bdi_init(&q->backing_dev_info);
594 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
595 laptop_mode_timer_fn, (unsigned long) q);
596 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
597 INIT_LIST_HEAD(&q->queue_head);
598 INIT_LIST_HEAD(&q->timeout_list);
599 INIT_LIST_HEAD(&q->icq_list);
600 #ifdef CONFIG_BLK_CGROUP
601 INIT_LIST_HEAD(&q->blkg_list);
603 INIT_LIST_HEAD(&q->flush_queue[0]);
604 INIT_LIST_HEAD(&q->flush_queue[1]);
605 INIT_LIST_HEAD(&q->flush_data_in_flight);
606 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
608 kobject_init(&q->kobj, &blk_queue_ktype);
610 mutex_init(&q->sysfs_lock);
611 spin_lock_init(&q->__queue_lock);
614 * By default initialize queue_lock to internal lock and driver can
615 * override it later if need be.
617 q->queue_lock = &q->__queue_lock;
620 * A queue starts its life with bypass turned on to avoid
621 * unnecessary bypass on/off overhead and nasty surprises during
622 * init. The initial bypass will be finished when the queue is
623 * registered by blk_register_queue().
626 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
628 init_waitqueue_head(&q->mq_freeze_wq);
630 if (blkcg_init_queue(q))
636 bdi_destroy(&q->backing_dev_info);
638 ida_simple_remove(&blk_queue_ida, q->id);
640 kmem_cache_free(blk_requestq_cachep, q);
643 EXPORT_SYMBOL(blk_alloc_queue_node);
646 * blk_init_queue - prepare a request queue for use with a block device
647 * @rfn: The function to be called to process requests that have been
648 * placed on the queue.
649 * @lock: Request queue spin lock
652 * If a block device wishes to use the standard request handling procedures,
653 * which sorts requests and coalesces adjacent requests, then it must
654 * call blk_init_queue(). The function @rfn will be called when there
655 * are requests on the queue that need to be processed. If the device
656 * supports plugging, then @rfn may not be called immediately when requests
657 * are available on the queue, but may be called at some time later instead.
658 * Plugged queues are generally unplugged when a buffer belonging to one
659 * of the requests on the queue is needed, or due to memory pressure.
661 * @rfn is not required, or even expected, to remove all requests off the
662 * queue, but only as many as it can handle at a time. If it does leave
663 * requests on the queue, it is responsible for arranging that the requests
664 * get dealt with eventually.
666 * The queue spin lock must be held while manipulating the requests on the
667 * request queue; this lock will be taken also from interrupt context, so irq
668 * disabling is needed for it.
670 * Function returns a pointer to the initialized request queue, or %NULL if
674 * blk_init_queue() must be paired with a blk_cleanup_queue() call
675 * when the block device is deactivated (such as at module unload).
678 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
680 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
682 EXPORT_SYMBOL(blk_init_queue);
684 struct request_queue *
685 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
687 struct request_queue *uninit_q, *q;
689 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
693 q = blk_init_allocated_queue(uninit_q, rfn, lock);
695 blk_cleanup_queue(uninit_q);
699 EXPORT_SYMBOL(blk_init_queue_node);
701 struct request_queue *
702 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
708 q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
712 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
716 q->prep_rq_fn = NULL;
717 q->unprep_rq_fn = NULL;
718 q->queue_flags |= QUEUE_FLAG_DEFAULT;
720 /* Override internal queue lock with supplied lock pointer */
722 q->queue_lock = lock;
725 * This also sets hw/phys segments, boundary and size
727 blk_queue_make_request(q, blk_queue_bio);
729 q->sg_reserved_size = INT_MAX;
731 /* Protect q->elevator from elevator_change */
732 mutex_lock(&q->sysfs_lock);
735 if (elevator_init(q, NULL)) {
736 mutex_unlock(&q->sysfs_lock);
740 mutex_unlock(&q->sysfs_lock);
748 EXPORT_SYMBOL(blk_init_allocated_queue);
750 bool blk_get_queue(struct request_queue *q)
752 if (likely(!blk_queue_dying(q))) {
759 EXPORT_SYMBOL(blk_get_queue);
761 static inline void blk_free_request(struct request_list *rl, struct request *rq)
763 if (rq->cmd_flags & REQ_ELVPRIV) {
764 elv_put_request(rl->q, rq);
766 put_io_context(rq->elv.icq->ioc);
769 mempool_free(rq, rl->rq_pool);
773 * ioc_batching returns true if the ioc is a valid batching request and
774 * should be given priority access to a request.
776 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
782 * Make sure the process is able to allocate at least 1 request
783 * even if the batch times out, otherwise we could theoretically
786 return ioc->nr_batch_requests == q->nr_batching ||
787 (ioc->nr_batch_requests > 0
788 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
792 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
793 * will cause the process to be a "batcher" on all queues in the system. This
794 * is the behaviour we want though - once it gets a wakeup it should be given
797 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
799 if (!ioc || ioc_batching(q, ioc))
802 ioc->nr_batch_requests = q->nr_batching;
803 ioc->last_waited = jiffies;
806 static void __freed_request(struct request_list *rl, int sync)
808 struct request_queue *q = rl->q;
811 * bdi isn't aware of blkcg yet. As all async IOs end up root
812 * blkcg anyway, just use root blkcg state.
814 if (rl == &q->root_rl &&
815 rl->count[sync] < queue_congestion_off_threshold(q))
816 blk_clear_queue_congested(q, sync);
818 if (rl->count[sync] + 1 <= q->nr_requests) {
819 if (waitqueue_active(&rl->wait[sync]))
820 wake_up(&rl->wait[sync]);
822 blk_clear_rl_full(rl, sync);
827 * A request has just been released. Account for it, update the full and
828 * congestion status, wake up any waiters. Called under q->queue_lock.
830 static void freed_request(struct request_list *rl, unsigned int flags)
832 struct request_queue *q = rl->q;
833 int sync = rw_is_sync(flags);
837 if (flags & REQ_ELVPRIV)
840 __freed_request(rl, sync);
842 if (unlikely(rl->starved[sync ^ 1]))
843 __freed_request(rl, sync ^ 1);
846 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
848 struct request_list *rl;
850 spin_lock_irq(q->queue_lock);
852 blk_queue_congestion_threshold(q);
854 /* congestion isn't cgroup aware and follows root blkcg for now */
857 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
858 blk_set_queue_congested(q, BLK_RW_SYNC);
859 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
860 blk_clear_queue_congested(q, BLK_RW_SYNC);
862 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
863 blk_set_queue_congested(q, BLK_RW_ASYNC);
864 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
865 blk_clear_queue_congested(q, BLK_RW_ASYNC);
867 blk_queue_for_each_rl(rl, q) {
868 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
869 blk_set_rl_full(rl, BLK_RW_SYNC);
871 blk_clear_rl_full(rl, BLK_RW_SYNC);
872 wake_up(&rl->wait[BLK_RW_SYNC]);
875 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
876 blk_set_rl_full(rl, BLK_RW_ASYNC);
878 blk_clear_rl_full(rl, BLK_RW_ASYNC);
879 wake_up(&rl->wait[BLK_RW_ASYNC]);
883 spin_unlock_irq(q->queue_lock);
888 * Determine if elevator data should be initialized when allocating the
889 * request associated with @bio.
891 static bool blk_rq_should_init_elevator(struct bio *bio)
897 * Flush requests do not use the elevator so skip initialization.
898 * This allows a request to share the flush and elevator data.
900 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
907 * rq_ioc - determine io_context for request allocation
908 * @bio: request being allocated is for this bio (can be %NULL)
910 * Determine io_context to use for request allocation for @bio. May return
911 * %NULL if %current->io_context doesn't exist.
913 static struct io_context *rq_ioc(struct bio *bio)
915 #ifdef CONFIG_BLK_CGROUP
916 if (bio && bio->bi_ioc)
919 return current->io_context;
923 * __get_request - get a free request
924 * @rl: request list to allocate from
925 * @rw_flags: RW and SYNC flags
926 * @bio: bio to allocate request for (can be %NULL)
927 * @gfp_mask: allocation mask
929 * Get a free request from @q. This function may fail under memory
930 * pressure or if @q is dead.
932 * Must be callled with @q->queue_lock held and,
933 * Returns %NULL on failure, with @q->queue_lock held.
934 * Returns !%NULL on success, with @q->queue_lock *not held*.
936 static struct request *__get_request(struct request_list *rl, int rw_flags,
937 struct bio *bio, gfp_t gfp_mask)
939 struct request_queue *q = rl->q;
941 struct elevator_type *et = q->elevator->type;
942 struct io_context *ioc = rq_ioc(bio);
943 struct io_cq *icq = NULL;
944 const bool is_sync = rw_is_sync(rw_flags) != 0;
947 if (unlikely(blk_queue_dying(q)))
950 may_queue = elv_may_queue(q, rw_flags);
951 if (may_queue == ELV_MQUEUE_NO)
954 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
955 if (rl->count[is_sync]+1 >= q->nr_requests) {
957 * The queue will fill after this allocation, so set
958 * it as full, and mark this process as "batching".
959 * This process will be allowed to complete a batch of
960 * requests, others will be blocked.
962 if (!blk_rl_full(rl, is_sync)) {
963 ioc_set_batching(q, ioc);
964 blk_set_rl_full(rl, is_sync);
966 if (may_queue != ELV_MQUEUE_MUST
967 && !ioc_batching(q, ioc)) {
969 * The queue is full and the allocating
970 * process is not a "batcher", and not
971 * exempted by the IO scheduler
978 * bdi isn't aware of blkcg yet. As all async IOs end up
979 * root blkcg anyway, just use root blkcg state.
981 if (rl == &q->root_rl)
982 blk_set_queue_congested(q, is_sync);
986 * Only allow batching queuers to allocate up to 50% over the defined
987 * limit of requests, otherwise we could have thousands of requests
988 * allocated with any setting of ->nr_requests
990 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
993 q->nr_rqs[is_sync]++;
994 rl->count[is_sync]++;
995 rl->starved[is_sync] = 0;
998 * Decide whether the new request will be managed by elevator. If
999 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1000 * prevent the current elevator from being destroyed until the new
1001 * request is freed. This guarantees icq's won't be destroyed and
1002 * makes creating new ones safe.
1004 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1005 * it will be created after releasing queue_lock.
1007 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1008 rw_flags |= REQ_ELVPRIV;
1009 q->nr_rqs_elvpriv++;
1010 if (et->icq_cache && ioc)
1011 icq = ioc_lookup_icq(ioc, q);
1014 if (blk_queue_io_stat(q))
1015 rw_flags |= REQ_IO_STAT;
1016 spin_unlock_irq(q->queue_lock);
1018 /* allocate and init request */
1019 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1024 blk_rq_set_rl(rq, rl);
1025 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1028 if (rw_flags & REQ_ELVPRIV) {
1029 if (unlikely(et->icq_cache && !icq)) {
1031 icq = ioc_create_icq(ioc, q, gfp_mask);
1037 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1040 /* @rq->elv.icq holds io_context until @rq is freed */
1042 get_io_context(icq->ioc);
1046 * ioc may be NULL here, and ioc_batching will be false. That's
1047 * OK, if the queue is under the request limit then requests need
1048 * not count toward the nr_batch_requests limit. There will always
1049 * be some limit enforced by BLK_BATCH_TIME.
1051 if (ioc_batching(q, ioc))
1052 ioc->nr_batch_requests--;
1054 trace_block_getrq(q, bio, rw_flags & 1);
1059 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1060 * and may fail indefinitely under memory pressure and thus
1061 * shouldn't stall IO. Treat this request as !elvpriv. This will
1062 * disturb iosched and blkcg but weird is bettern than dead.
1064 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1065 dev_name(q->backing_dev_info.dev));
1067 rq->cmd_flags &= ~REQ_ELVPRIV;
1070 spin_lock_irq(q->queue_lock);
1071 q->nr_rqs_elvpriv--;
1072 spin_unlock_irq(q->queue_lock);
1077 * Allocation failed presumably due to memory. Undo anything we
1078 * might have messed up.
1080 * Allocating task should really be put onto the front of the wait
1081 * queue, but this is pretty rare.
1083 spin_lock_irq(q->queue_lock);
1084 freed_request(rl, rw_flags);
1087 * in the very unlikely event that allocation failed and no
1088 * requests for this direction was pending, mark us starved so that
1089 * freeing of a request in the other direction will notice
1090 * us. another possible fix would be to split the rq mempool into
1094 if (unlikely(rl->count[is_sync] == 0))
1095 rl->starved[is_sync] = 1;
1100 * get_request - get a free request
1101 * @q: request_queue to allocate request from
1102 * @rw_flags: RW and SYNC flags
1103 * @bio: bio to allocate request for (can be %NULL)
1104 * @gfp_mask: allocation mask
1106 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1107 * function keeps retrying under memory pressure and fails iff @q is dead.
1109 * Must be callled with @q->queue_lock held and,
1110 * Returns %NULL on failure, with @q->queue_lock held.
1111 * Returns !%NULL on success, with @q->queue_lock *not held*.
1113 static struct request *get_request(struct request_queue *q, int rw_flags,
1114 struct bio *bio, gfp_t gfp_mask)
1116 const bool is_sync = rw_is_sync(rw_flags) != 0;
1118 struct request_list *rl;
1121 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1123 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1127 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1132 /* wait on @rl and retry */
1133 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1134 TASK_UNINTERRUPTIBLE);
1136 trace_block_sleeprq(q, bio, rw_flags & 1);
1138 spin_unlock_irq(q->queue_lock);
1142 * After sleeping, we become a "batching" process and will be able
1143 * to allocate at least one request, and up to a big batch of them
1144 * for a small period time. See ioc_batching, ioc_set_batching
1146 ioc_set_batching(q, current->io_context);
1148 spin_lock_irq(q->queue_lock);
1149 finish_wait(&rl->wait[is_sync], &wait);
1154 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1159 BUG_ON(rw != READ && rw != WRITE);
1161 /* create ioc upfront */
1162 create_io_context(gfp_mask, q->node);
1164 spin_lock_irq(q->queue_lock);
1165 rq = get_request(q, rw, NULL, gfp_mask);
1167 spin_unlock_irq(q->queue_lock);
1168 /* q->queue_lock is unlocked at this point */
1173 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1176 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1178 return blk_old_get_request(q, rw, gfp_mask);
1180 EXPORT_SYMBOL(blk_get_request);
1183 * blk_make_request - given a bio, allocate a corresponding struct request.
1184 * @q: target request queue
1185 * @bio: The bio describing the memory mappings that will be submitted for IO.
1186 * It may be a chained-bio properly constructed by block/bio layer.
1187 * @gfp_mask: gfp flags to be used for memory allocation
1189 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1190 * type commands. Where the struct request needs to be farther initialized by
1191 * the caller. It is passed a &struct bio, which describes the memory info of
1194 * The caller of blk_make_request must make sure that bi_io_vec
1195 * are set to describe the memory buffers. That bio_data_dir() will return
1196 * the needed direction of the request. (And all bio's in the passed bio-chain
1197 * are properly set accordingly)
1199 * If called under none-sleepable conditions, mapped bio buffers must not
1200 * need bouncing, by calling the appropriate masked or flagged allocator,
1201 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1204 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1205 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1206 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1207 * completion of a bio that hasn't been submitted yet, thus resulting in a
1208 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1209 * of bio_alloc(), as that avoids the mempool deadlock.
1210 * If possible a big IO should be split into smaller parts when allocation
1211 * fails. Partial allocation should not be an error, or you risk a live-lock.
1213 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1216 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1219 return ERR_PTR(-ENOMEM);
1221 blk_rq_set_block_pc(rq);
1224 struct bio *bounce_bio = bio;
1227 blk_queue_bounce(q, &bounce_bio);
1228 ret = blk_rq_append_bio(q, rq, bounce_bio);
1229 if (unlikely(ret)) {
1230 blk_put_request(rq);
1231 return ERR_PTR(ret);
1237 EXPORT_SYMBOL(blk_make_request);
1240 * blk_rq_set_block_pc - initialize a requeest to type BLOCK_PC
1241 * @rq: request to be initialized
1244 void blk_rq_set_block_pc(struct request *rq)
1246 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1248 rq->__sector = (sector_t) -1;
1249 rq->bio = rq->biotail = NULL;
1250 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1251 rq->cmd = rq->__cmd;
1253 EXPORT_SYMBOL(blk_rq_set_block_pc);
1256 * blk_requeue_request - put a request back on queue
1257 * @q: request queue where request should be inserted
1258 * @rq: request to be inserted
1261 * Drivers often keep queueing requests until the hardware cannot accept
1262 * more, when that condition happens we need to put the request back
1263 * on the queue. Must be called with queue lock held.
1265 void blk_requeue_request(struct request_queue *q, struct request *rq)
1267 blk_delete_timer(rq);
1268 blk_clear_rq_complete(rq);
1269 trace_block_rq_requeue(q, rq);
1271 if (blk_rq_tagged(rq))
1272 blk_queue_end_tag(q, rq);
1274 BUG_ON(blk_queued_rq(rq));
1276 elv_requeue_request(q, rq);
1278 EXPORT_SYMBOL(blk_requeue_request);
1280 static void add_acct_request(struct request_queue *q, struct request *rq,
1283 blk_account_io_start(rq, true);
1284 __elv_add_request(q, rq, where);
1287 static void part_round_stats_single(int cpu, struct hd_struct *part,
1292 if (now == part->stamp)
1295 inflight = part_in_flight(part);
1297 __part_stat_add(cpu, part, time_in_queue,
1298 inflight * (now - part->stamp));
1299 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1305 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1306 * @cpu: cpu number for stats access
1307 * @part: target partition
1309 * The average IO queue length and utilisation statistics are maintained
1310 * by observing the current state of the queue length and the amount of
1311 * time it has been in this state for.
1313 * Normally, that accounting is done on IO completion, but that can result
1314 * in more than a second's worth of IO being accounted for within any one
1315 * second, leading to >100% utilisation. To deal with that, we call this
1316 * function to do a round-off before returning the results when reading
1317 * /proc/diskstats. This accounts immediately for all queue usage up to
1318 * the current jiffies and restarts the counters again.
1320 void part_round_stats(int cpu, struct hd_struct *part)
1322 unsigned long now = jiffies;
1325 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1326 part_round_stats_single(cpu, part, now);
1328 EXPORT_SYMBOL_GPL(part_round_stats);
1330 #ifdef CONFIG_PM_RUNTIME
1331 static void blk_pm_put_request(struct request *rq)
1333 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1334 pm_runtime_mark_last_busy(rq->q->dev);
1337 static inline void blk_pm_put_request(struct request *rq) {}
1341 * queue lock must be held
1343 void __blk_put_request(struct request_queue *q, struct request *req)
1349 blk_mq_free_request(req);
1353 blk_pm_put_request(req);
1355 elv_completed_request(q, req);
1357 /* this is a bio leak */
1358 WARN_ON(req->bio != NULL);
1361 * Request may not have originated from ll_rw_blk. if not,
1362 * it didn't come out of our reserved rq pools
1364 if (req->cmd_flags & REQ_ALLOCED) {
1365 unsigned int flags = req->cmd_flags;
1366 struct request_list *rl = blk_rq_rl(req);
1368 BUG_ON(!list_empty(&req->queuelist));
1369 BUG_ON(ELV_ON_HASH(req));
1371 blk_free_request(rl, req);
1372 freed_request(rl, flags);
1376 EXPORT_SYMBOL_GPL(__blk_put_request);
1378 void blk_put_request(struct request *req)
1380 struct request_queue *q = req->q;
1383 blk_mq_free_request(req);
1385 unsigned long flags;
1387 spin_lock_irqsave(q->queue_lock, flags);
1388 __blk_put_request(q, req);
1389 spin_unlock_irqrestore(q->queue_lock, flags);
1392 EXPORT_SYMBOL(blk_put_request);
1395 * blk_add_request_payload - add a payload to a request
1396 * @rq: request to update
1397 * @page: page backing the payload
1398 * @len: length of the payload.
1400 * This allows to later add a payload to an already submitted request by
1401 * a block driver. The driver needs to take care of freeing the payload
1404 * Note that this is a quite horrible hack and nothing but handling of
1405 * discard requests should ever use it.
1407 void blk_add_request_payload(struct request *rq, struct page *page,
1410 struct bio *bio = rq->bio;
1412 bio->bi_io_vec->bv_page = page;
1413 bio->bi_io_vec->bv_offset = 0;
1414 bio->bi_io_vec->bv_len = len;
1416 bio->bi_iter.bi_size = len;
1418 bio->bi_phys_segments = 1;
1420 rq->__data_len = rq->resid_len = len;
1421 rq->nr_phys_segments = 1;
1423 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1425 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1428 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1430 if (!ll_back_merge_fn(q, req, bio))
1433 trace_block_bio_backmerge(q, req, bio);
1435 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1436 blk_rq_set_mixed_merge(req);
1438 req->biotail->bi_next = bio;
1440 req->__data_len += bio->bi_iter.bi_size;
1441 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1443 blk_account_io_start(req, false);
1447 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1450 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1452 if (!ll_front_merge_fn(q, req, bio))
1455 trace_block_bio_frontmerge(q, req, bio);
1457 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1458 blk_rq_set_mixed_merge(req);
1460 bio->bi_next = req->bio;
1463 req->__sector = bio->bi_iter.bi_sector;
1464 req->__data_len += bio->bi_iter.bi_size;
1465 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1467 blk_account_io_start(req, false);
1472 * blk_attempt_plug_merge - try to merge with %current's plugged list
1473 * @q: request_queue new bio is being queued at
1474 * @bio: new bio being queued
1475 * @request_count: out parameter for number of traversed plugged requests
1477 * Determine whether @bio being queued on @q can be merged with a request
1478 * on %current's plugged list. Returns %true if merge was successful,
1481 * Plugging coalesces IOs from the same issuer for the same purpose without
1482 * going through @q->queue_lock. As such it's more of an issuing mechanism
1483 * than scheduling, and the request, while may have elvpriv data, is not
1484 * added on the elevator at this point. In addition, we don't have
1485 * reliable access to the elevator outside queue lock. Only check basic
1486 * merging parameters without querying the elevator.
1488 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1490 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1491 unsigned int *request_count)
1493 struct blk_plug *plug;
1496 struct list_head *plug_list;
1498 plug = current->plug;
1504 plug_list = &plug->mq_list;
1506 plug_list = &plug->list;
1508 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1514 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1517 el_ret = blk_try_merge(rq, bio);
1518 if (el_ret == ELEVATOR_BACK_MERGE) {
1519 ret = bio_attempt_back_merge(q, rq, bio);
1522 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1523 ret = bio_attempt_front_merge(q, rq, bio);
1532 void init_request_from_bio(struct request *req, struct bio *bio)
1534 req->cmd_type = REQ_TYPE_FS;
1536 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1537 if (bio->bi_rw & REQ_RAHEAD)
1538 req->cmd_flags |= REQ_FAILFAST_MASK;
1541 req->__sector = bio->bi_iter.bi_sector;
1542 req->ioprio = bio_prio(bio);
1543 blk_rq_bio_prep(req->q, req, bio);
1546 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1548 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1549 struct blk_plug *plug;
1550 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1551 struct request *req;
1552 unsigned int request_count = 0;
1555 * low level driver can indicate that it wants pages above a
1556 * certain limit bounced to low memory (ie for highmem, or even
1557 * ISA dma in theory)
1559 blk_queue_bounce(q, &bio);
1561 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1562 bio_endio(bio, -EIO);
1566 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1567 spin_lock_irq(q->queue_lock);
1568 where = ELEVATOR_INSERT_FLUSH;
1573 * Check if we can merge with the plugged list before grabbing
1576 if (!blk_queue_nomerges(q) &&
1577 blk_attempt_plug_merge(q, bio, &request_count))
1580 spin_lock_irq(q->queue_lock);
1582 el_ret = elv_merge(q, &req, bio);
1583 if (el_ret == ELEVATOR_BACK_MERGE) {
1584 if (bio_attempt_back_merge(q, req, bio)) {
1585 elv_bio_merged(q, req, bio);
1586 if (!attempt_back_merge(q, req))
1587 elv_merged_request(q, req, el_ret);
1590 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1591 if (bio_attempt_front_merge(q, req, bio)) {
1592 elv_bio_merged(q, req, bio);
1593 if (!attempt_front_merge(q, req))
1594 elv_merged_request(q, req, el_ret);
1601 * This sync check and mask will be re-done in init_request_from_bio(),
1602 * but we need to set it earlier to expose the sync flag to the
1603 * rq allocator and io schedulers.
1605 rw_flags = bio_data_dir(bio);
1607 rw_flags |= REQ_SYNC;
1610 * Grab a free request. This is might sleep but can not fail.
1611 * Returns with the queue unlocked.
1613 req = get_request(q, rw_flags, bio, GFP_NOIO);
1614 if (unlikely(!req)) {
1615 bio_endio(bio, -ENODEV); /* @q is dead */
1620 * After dropping the lock and possibly sleeping here, our request
1621 * may now be mergeable after it had proven unmergeable (above).
1622 * We don't worry about that case for efficiency. It won't happen
1623 * often, and the elevators are able to handle it.
1625 init_request_from_bio(req, bio);
1627 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1628 req->cpu = raw_smp_processor_id();
1630 plug = current->plug;
1633 * If this is the first request added after a plug, fire
1637 trace_block_plug(q);
1639 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1640 blk_flush_plug_list(plug, false);
1641 trace_block_plug(q);
1644 list_add_tail(&req->queuelist, &plug->list);
1645 blk_account_io_start(req, true);
1647 spin_lock_irq(q->queue_lock);
1648 add_acct_request(q, req, where);
1651 spin_unlock_irq(q->queue_lock);
1654 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1657 * If bio->bi_dev is a partition, remap the location
1659 static inline void blk_partition_remap(struct bio *bio)
1661 struct block_device *bdev = bio->bi_bdev;
1663 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1664 struct hd_struct *p = bdev->bd_part;
1666 bio->bi_iter.bi_sector += p->start_sect;
1667 bio->bi_bdev = bdev->bd_contains;
1669 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1671 bio->bi_iter.bi_sector - p->start_sect);
1675 static void handle_bad_sector(struct bio *bio)
1677 char b[BDEVNAME_SIZE];
1679 printk(KERN_INFO "attempt to access beyond end of device\n");
1680 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1681 bdevname(bio->bi_bdev, b),
1683 (unsigned long long)bio_end_sector(bio),
1684 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1686 set_bit(BIO_EOF, &bio->bi_flags);
1689 #ifdef CONFIG_FAIL_MAKE_REQUEST
1691 static DECLARE_FAULT_ATTR(fail_make_request);
1693 static int __init setup_fail_make_request(char *str)
1695 return setup_fault_attr(&fail_make_request, str);
1697 __setup("fail_make_request=", setup_fail_make_request);
1699 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1701 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1704 static int __init fail_make_request_debugfs(void)
1706 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1707 NULL, &fail_make_request);
1709 return PTR_ERR_OR_ZERO(dir);
1712 late_initcall(fail_make_request_debugfs);
1714 #else /* CONFIG_FAIL_MAKE_REQUEST */
1716 static inline bool should_fail_request(struct hd_struct *part,
1722 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1725 * Check whether this bio extends beyond the end of the device.
1727 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1734 /* Test device or partition size, when known. */
1735 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1737 sector_t sector = bio->bi_iter.bi_sector;
1739 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1741 * This may well happen - the kernel calls bread()
1742 * without checking the size of the device, e.g., when
1743 * mounting a device.
1745 handle_bad_sector(bio);
1753 static noinline_for_stack bool
1754 generic_make_request_checks(struct bio *bio)
1756 struct request_queue *q;
1757 int nr_sectors = bio_sectors(bio);
1759 char b[BDEVNAME_SIZE];
1760 struct hd_struct *part;
1764 if (bio_check_eod(bio, nr_sectors))
1767 q = bdev_get_queue(bio->bi_bdev);
1770 "generic_make_request: Trying to access "
1771 "nonexistent block-device %s (%Lu)\n",
1772 bdevname(bio->bi_bdev, b),
1773 (long long) bio->bi_iter.bi_sector);
1777 if (likely(bio_is_rw(bio) &&
1778 nr_sectors > queue_max_hw_sectors(q))) {
1779 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1780 bdevname(bio->bi_bdev, b),
1782 queue_max_hw_sectors(q));
1786 part = bio->bi_bdev->bd_part;
1787 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1788 should_fail_request(&part_to_disk(part)->part0,
1789 bio->bi_iter.bi_size))
1793 * If this device has partitions, remap block n
1794 * of partition p to block n+start(p) of the disk.
1796 blk_partition_remap(bio);
1798 if (bio_check_eod(bio, nr_sectors))
1802 * Filter flush bio's early so that make_request based
1803 * drivers without flush support don't have to worry
1806 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1807 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1814 if ((bio->bi_rw & REQ_DISCARD) &&
1815 (!blk_queue_discard(q) ||
1816 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1821 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1827 * Various block parts want %current->io_context and lazy ioc
1828 * allocation ends up trading a lot of pain for a small amount of
1829 * memory. Just allocate it upfront. This may fail and block
1830 * layer knows how to live with it.
1832 create_io_context(GFP_ATOMIC, q->node);
1834 if (blk_throtl_bio(q, bio))
1835 return false; /* throttled, will be resubmitted later */
1837 trace_block_bio_queue(q, bio);
1841 bio_endio(bio, err);
1846 * generic_make_request - hand a buffer to its device driver for I/O
1847 * @bio: The bio describing the location in memory and on the device.
1849 * generic_make_request() is used to make I/O requests of block
1850 * devices. It is passed a &struct bio, which describes the I/O that needs
1853 * generic_make_request() does not return any status. The
1854 * success/failure status of the request, along with notification of
1855 * completion, is delivered asynchronously through the bio->bi_end_io
1856 * function described (one day) else where.
1858 * The caller of generic_make_request must make sure that bi_io_vec
1859 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1860 * set to describe the device address, and the
1861 * bi_end_io and optionally bi_private are set to describe how
1862 * completion notification should be signaled.
1864 * generic_make_request and the drivers it calls may use bi_next if this
1865 * bio happens to be merged with someone else, and may resubmit the bio to
1866 * a lower device by calling into generic_make_request recursively, which
1867 * means the bio should NOT be touched after the call to ->make_request_fn.
1869 void generic_make_request(struct bio *bio)
1871 struct bio_list bio_list_on_stack;
1873 if (!generic_make_request_checks(bio))
1877 * We only want one ->make_request_fn to be active at a time, else
1878 * stack usage with stacked devices could be a problem. So use
1879 * current->bio_list to keep a list of requests submited by a
1880 * make_request_fn function. current->bio_list is also used as a
1881 * flag to say if generic_make_request is currently active in this
1882 * task or not. If it is NULL, then no make_request is active. If
1883 * it is non-NULL, then a make_request is active, and new requests
1884 * should be added at the tail
1886 if (current->bio_list) {
1887 bio_list_add(current->bio_list, bio);
1891 /* following loop may be a bit non-obvious, and so deserves some
1893 * Before entering the loop, bio->bi_next is NULL (as all callers
1894 * ensure that) so we have a list with a single bio.
1895 * We pretend that we have just taken it off a longer list, so
1896 * we assign bio_list to a pointer to the bio_list_on_stack,
1897 * thus initialising the bio_list of new bios to be
1898 * added. ->make_request() may indeed add some more bios
1899 * through a recursive call to generic_make_request. If it
1900 * did, we find a non-NULL value in bio_list and re-enter the loop
1901 * from the top. In this case we really did just take the bio
1902 * of the top of the list (no pretending) and so remove it from
1903 * bio_list, and call into ->make_request() again.
1905 BUG_ON(bio->bi_next);
1906 bio_list_init(&bio_list_on_stack);
1907 current->bio_list = &bio_list_on_stack;
1909 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1911 q->make_request_fn(q, bio);
1913 bio = bio_list_pop(current->bio_list);
1915 current->bio_list = NULL; /* deactivate */
1917 EXPORT_SYMBOL(generic_make_request);
1920 * submit_bio - submit a bio to the block device layer for I/O
1921 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1922 * @bio: The &struct bio which describes the I/O
1924 * submit_bio() is very similar in purpose to generic_make_request(), and
1925 * uses that function to do most of the work. Both are fairly rough
1926 * interfaces; @bio must be presetup and ready for I/O.
1929 void submit_bio(int rw, struct bio *bio)
1934 * If it's a regular read/write or a barrier with data attached,
1935 * go through the normal accounting stuff before submission.
1937 if (bio_has_data(bio)) {
1940 if (unlikely(rw & REQ_WRITE_SAME))
1941 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1943 count = bio_sectors(bio);
1946 count_vm_events(PGPGOUT, count);
1948 task_io_account_read(bio->bi_iter.bi_size);
1949 count_vm_events(PGPGIN, count);
1952 if (unlikely(block_dump)) {
1953 char b[BDEVNAME_SIZE];
1954 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1955 current->comm, task_pid_nr(current),
1956 (rw & WRITE) ? "WRITE" : "READ",
1957 (unsigned long long)bio->bi_iter.bi_sector,
1958 bdevname(bio->bi_bdev, b),
1963 generic_make_request(bio);
1965 EXPORT_SYMBOL(submit_bio);
1968 * blk_rq_check_limits - Helper function to check a request for the queue limit
1970 * @rq: the request being checked
1973 * @rq may have been made based on weaker limitations of upper-level queues
1974 * in request stacking drivers, and it may violate the limitation of @q.
1975 * Since the block layer and the underlying device driver trust @rq
1976 * after it is inserted to @q, it should be checked against @q before
1977 * the insertion using this generic function.
1979 * This function should also be useful for request stacking drivers
1980 * in some cases below, so export this function.
1981 * Request stacking drivers like request-based dm may change the queue
1982 * limits while requests are in the queue (e.g. dm's table swapping).
1983 * Such request stacking drivers should check those requests against
1984 * the new queue limits again when they dispatch those requests,
1985 * although such checkings are also done against the old queue limits
1986 * when submitting requests.
1988 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1990 if (!rq_mergeable(rq))
1993 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1994 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1999 * queue's settings related to segment counting like q->bounce_pfn
2000 * may differ from that of other stacking queues.
2001 * Recalculate it to check the request correctly on this queue's
2004 blk_recalc_rq_segments(rq);
2005 if (rq->nr_phys_segments > queue_max_segments(q)) {
2006 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2012 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2015 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2016 * @q: the queue to submit the request
2017 * @rq: the request being queued
2019 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2021 unsigned long flags;
2022 int where = ELEVATOR_INSERT_BACK;
2024 if (blk_rq_check_limits(q, rq))
2028 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2031 spin_lock_irqsave(q->queue_lock, flags);
2032 if (unlikely(blk_queue_dying(q))) {
2033 spin_unlock_irqrestore(q->queue_lock, flags);
2038 * Submitting request must be dequeued before calling this function
2039 * because it will be linked to another request_queue
2041 BUG_ON(blk_queued_rq(rq));
2043 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2044 where = ELEVATOR_INSERT_FLUSH;
2046 add_acct_request(q, rq, where);
2047 if (where == ELEVATOR_INSERT_FLUSH)
2049 spin_unlock_irqrestore(q->queue_lock, flags);
2053 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2056 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2057 * @rq: request to examine
2060 * A request could be merge of IOs which require different failure
2061 * handling. This function determines the number of bytes which
2062 * can be failed from the beginning of the request without
2063 * crossing into area which need to be retried further.
2066 * The number of bytes to fail.
2069 * queue_lock must be held.
2071 unsigned int blk_rq_err_bytes(const struct request *rq)
2073 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2074 unsigned int bytes = 0;
2077 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2078 return blk_rq_bytes(rq);
2081 * Currently the only 'mixing' which can happen is between
2082 * different fastfail types. We can safely fail portions
2083 * which have all the failfast bits that the first one has -
2084 * the ones which are at least as eager to fail as the first
2087 for (bio = rq->bio; bio; bio = bio->bi_next) {
2088 if ((bio->bi_rw & ff) != ff)
2090 bytes += bio->bi_iter.bi_size;
2093 /* this could lead to infinite loop */
2094 BUG_ON(blk_rq_bytes(rq) && !bytes);
2097 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2099 void blk_account_io_completion(struct request *req, unsigned int bytes)
2101 if (blk_do_io_stat(req)) {
2102 const int rw = rq_data_dir(req);
2103 struct hd_struct *part;
2106 cpu = part_stat_lock();
2108 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2113 void blk_account_io_done(struct request *req)
2116 * Account IO completion. flush_rq isn't accounted as a
2117 * normal IO on queueing nor completion. Accounting the
2118 * containing request is enough.
2120 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2121 unsigned long duration = jiffies - req->start_time;
2122 const int rw = rq_data_dir(req);
2123 struct hd_struct *part;
2126 cpu = part_stat_lock();
2129 part_stat_inc(cpu, part, ios[rw]);
2130 part_stat_add(cpu, part, ticks[rw], duration);
2131 part_round_stats(cpu, part);
2132 part_dec_in_flight(part, rw);
2134 hd_struct_put(part);
2139 #ifdef CONFIG_PM_RUNTIME
2141 * Don't process normal requests when queue is suspended
2142 * or in the process of suspending/resuming
2144 static struct request *blk_pm_peek_request(struct request_queue *q,
2147 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2148 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2154 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2161 void blk_account_io_start(struct request *rq, bool new_io)
2163 struct hd_struct *part;
2164 int rw = rq_data_dir(rq);
2167 if (!blk_do_io_stat(rq))
2170 cpu = part_stat_lock();
2174 part_stat_inc(cpu, part, merges[rw]);
2176 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2177 if (!hd_struct_try_get(part)) {
2179 * The partition is already being removed,
2180 * the request will be accounted on the disk only
2182 * We take a reference on disk->part0 although that
2183 * partition will never be deleted, so we can treat
2184 * it as any other partition.
2186 part = &rq->rq_disk->part0;
2187 hd_struct_get(part);
2189 part_round_stats(cpu, part);
2190 part_inc_in_flight(part, rw);
2198 * blk_peek_request - peek at the top of a request queue
2199 * @q: request queue to peek at
2202 * Return the request at the top of @q. The returned request
2203 * should be started using blk_start_request() before LLD starts
2207 * Pointer to the request at the top of @q if available. Null
2211 * queue_lock must be held.
2213 struct request *blk_peek_request(struct request_queue *q)
2218 while ((rq = __elv_next_request(q)) != NULL) {
2220 rq = blk_pm_peek_request(q, rq);
2224 if (!(rq->cmd_flags & REQ_STARTED)) {
2226 * This is the first time the device driver
2227 * sees this request (possibly after
2228 * requeueing). Notify IO scheduler.
2230 if (rq->cmd_flags & REQ_SORTED)
2231 elv_activate_rq(q, rq);
2234 * just mark as started even if we don't start
2235 * it, a request that has been delayed should
2236 * not be passed by new incoming requests
2238 rq->cmd_flags |= REQ_STARTED;
2239 trace_block_rq_issue(q, rq);
2242 if (!q->boundary_rq || q->boundary_rq == rq) {
2243 q->end_sector = rq_end_sector(rq);
2244 q->boundary_rq = NULL;
2247 if (rq->cmd_flags & REQ_DONTPREP)
2250 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2252 * make sure space for the drain appears we
2253 * know we can do this because max_hw_segments
2254 * has been adjusted to be one fewer than the
2257 rq->nr_phys_segments++;
2263 ret = q->prep_rq_fn(q, rq);
2264 if (ret == BLKPREP_OK) {
2266 } else if (ret == BLKPREP_DEFER) {
2268 * the request may have been (partially) prepped.
2269 * we need to keep this request in the front to
2270 * avoid resource deadlock. REQ_STARTED will
2271 * prevent other fs requests from passing this one.
2273 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2274 !(rq->cmd_flags & REQ_DONTPREP)) {
2276 * remove the space for the drain we added
2277 * so that we don't add it again
2279 --rq->nr_phys_segments;
2284 } else if (ret == BLKPREP_KILL) {
2285 rq->cmd_flags |= REQ_QUIET;
2287 * Mark this request as started so we don't trigger
2288 * any debug logic in the end I/O path.
2290 blk_start_request(rq);
2291 __blk_end_request_all(rq, -EIO);
2293 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2300 EXPORT_SYMBOL(blk_peek_request);
2302 void blk_dequeue_request(struct request *rq)
2304 struct request_queue *q = rq->q;
2306 BUG_ON(list_empty(&rq->queuelist));
2307 BUG_ON(ELV_ON_HASH(rq));
2309 list_del_init(&rq->queuelist);
2312 * the time frame between a request being removed from the lists
2313 * and to it is freed is accounted as io that is in progress at
2316 if (blk_account_rq(rq)) {
2317 q->in_flight[rq_is_sync(rq)]++;
2318 set_io_start_time_ns(rq);
2323 * blk_start_request - start request processing on the driver
2324 * @req: request to dequeue
2327 * Dequeue @req and start timeout timer on it. This hands off the
2328 * request to the driver.
2330 * Block internal functions which don't want to start timer should
2331 * call blk_dequeue_request().
2334 * queue_lock must be held.
2336 void blk_start_request(struct request *req)
2338 blk_dequeue_request(req);
2341 * We are now handing the request to the hardware, initialize
2342 * resid_len to full count and add the timeout handler.
2344 req->resid_len = blk_rq_bytes(req);
2345 if (unlikely(blk_bidi_rq(req)))
2346 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2348 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2351 EXPORT_SYMBOL(blk_start_request);
2354 * blk_fetch_request - fetch a request from a request queue
2355 * @q: request queue to fetch a request from
2358 * Return the request at the top of @q. The request is started on
2359 * return and LLD can start processing it immediately.
2362 * Pointer to the request at the top of @q if available. Null
2366 * queue_lock must be held.
2368 struct request *blk_fetch_request(struct request_queue *q)
2372 rq = blk_peek_request(q);
2374 blk_start_request(rq);
2377 EXPORT_SYMBOL(blk_fetch_request);
2380 * blk_update_request - Special helper function for request stacking drivers
2381 * @req: the request being processed
2382 * @error: %0 for success, < %0 for error
2383 * @nr_bytes: number of bytes to complete @req
2386 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2387 * the request structure even if @req doesn't have leftover.
2388 * If @req has leftover, sets it up for the next range of segments.
2390 * This special helper function is only for request stacking drivers
2391 * (e.g. request-based dm) so that they can handle partial completion.
2392 * Actual device drivers should use blk_end_request instead.
2394 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2395 * %false return from this function.
2398 * %false - this request doesn't have any more data
2399 * %true - this request has more data
2401 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2408 trace_block_rq_complete(req->q, req, nr_bytes);
2411 * For fs requests, rq is just carrier of independent bio's
2412 * and each partial completion should be handled separately.
2413 * Reset per-request error on each partial completion.
2415 * TODO: tj: This is too subtle. It would be better to let
2416 * low level drivers do what they see fit.
2418 if (req->cmd_type == REQ_TYPE_FS)
2421 if (error && req->cmd_type == REQ_TYPE_FS &&
2422 !(req->cmd_flags & REQ_QUIET)) {
2427 error_type = "recoverable transport";
2430 error_type = "critical target";
2433 error_type = "critical nexus";
2436 error_type = "timeout";
2439 error_type = "critical space allocation";
2442 error_type = "critical medium";
2449 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2450 error_type, req->rq_disk ?
2451 req->rq_disk->disk_name : "?",
2452 (unsigned long long)blk_rq_pos(req));
2456 blk_account_io_completion(req, nr_bytes);
2460 struct bio *bio = req->bio;
2461 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2463 if (bio_bytes == bio->bi_iter.bi_size)
2464 req->bio = bio->bi_next;
2466 req_bio_endio(req, bio, bio_bytes, error);
2468 total_bytes += bio_bytes;
2469 nr_bytes -= bio_bytes;
2480 * Reset counters so that the request stacking driver
2481 * can find how many bytes remain in the request
2484 req->__data_len = 0;
2488 req->__data_len -= total_bytes;
2490 /* update sector only for requests with clear definition of sector */
2491 if (req->cmd_type == REQ_TYPE_FS)
2492 req->__sector += total_bytes >> 9;
2494 /* mixed attributes always follow the first bio */
2495 if (req->cmd_flags & REQ_MIXED_MERGE) {
2496 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2497 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2501 * If total number of sectors is less than the first segment
2502 * size, something has gone terribly wrong.
2504 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2505 blk_dump_rq_flags(req, "request botched");
2506 req->__data_len = blk_rq_cur_bytes(req);
2509 /* recalculate the number of segments */
2510 blk_recalc_rq_segments(req);
2514 EXPORT_SYMBOL_GPL(blk_update_request);
2516 static bool blk_update_bidi_request(struct request *rq, int error,
2517 unsigned int nr_bytes,
2518 unsigned int bidi_bytes)
2520 if (blk_update_request(rq, error, nr_bytes))
2523 /* Bidi request must be completed as a whole */
2524 if (unlikely(blk_bidi_rq(rq)) &&
2525 blk_update_request(rq->next_rq, error, bidi_bytes))
2528 if (blk_queue_add_random(rq->q))
2529 add_disk_randomness(rq->rq_disk);
2535 * blk_unprep_request - unprepare a request
2538 * This function makes a request ready for complete resubmission (or
2539 * completion). It happens only after all error handling is complete,
2540 * so represents the appropriate moment to deallocate any resources
2541 * that were allocated to the request in the prep_rq_fn. The queue
2542 * lock is held when calling this.
2544 void blk_unprep_request(struct request *req)
2546 struct request_queue *q = req->q;
2548 req->cmd_flags &= ~REQ_DONTPREP;
2549 if (q->unprep_rq_fn)
2550 q->unprep_rq_fn(q, req);
2552 EXPORT_SYMBOL_GPL(blk_unprep_request);
2555 * queue lock must be held
2557 void blk_finish_request(struct request *req, int error)
2559 if (blk_rq_tagged(req))
2560 blk_queue_end_tag(req->q, req);
2562 BUG_ON(blk_queued_rq(req));
2564 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2565 laptop_io_completion(&req->q->backing_dev_info);
2567 blk_delete_timer(req);
2569 if (req->cmd_flags & REQ_DONTPREP)
2570 blk_unprep_request(req);
2572 blk_account_io_done(req);
2575 req->end_io(req, error);
2577 if (blk_bidi_rq(req))
2578 __blk_put_request(req->next_rq->q, req->next_rq);
2580 __blk_put_request(req->q, req);
2583 EXPORT_SYMBOL(blk_finish_request);
2586 * blk_end_bidi_request - Complete a bidi request
2587 * @rq: the request to complete
2588 * @error: %0 for success, < %0 for error
2589 * @nr_bytes: number of bytes to complete @rq
2590 * @bidi_bytes: number of bytes to complete @rq->next_rq
2593 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2594 * Drivers that supports bidi can safely call this member for any
2595 * type of request, bidi or uni. In the later case @bidi_bytes is
2599 * %false - we are done with this request
2600 * %true - still buffers pending for this request
2602 static bool blk_end_bidi_request(struct request *rq, int error,
2603 unsigned int nr_bytes, unsigned int bidi_bytes)
2605 struct request_queue *q = rq->q;
2606 unsigned long flags;
2608 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2611 spin_lock_irqsave(q->queue_lock, flags);
2612 blk_finish_request(rq, error);
2613 spin_unlock_irqrestore(q->queue_lock, flags);
2619 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2620 * @rq: the request to complete
2621 * @error: %0 for success, < %0 for error
2622 * @nr_bytes: number of bytes to complete @rq
2623 * @bidi_bytes: number of bytes to complete @rq->next_rq
2626 * Identical to blk_end_bidi_request() except that queue lock is
2627 * assumed to be locked on entry and remains so on return.
2630 * %false - we are done with this request
2631 * %true - still buffers pending for this request
2633 bool __blk_end_bidi_request(struct request *rq, int error,
2634 unsigned int nr_bytes, unsigned int bidi_bytes)
2636 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2639 blk_finish_request(rq, error);
2645 * blk_end_request - Helper function for drivers to complete the request.
2646 * @rq: the request being processed
2647 * @error: %0 for success, < %0 for error
2648 * @nr_bytes: number of bytes to complete
2651 * Ends I/O on a number of bytes attached to @rq.
2652 * If @rq has leftover, sets it up for the next range of segments.
2655 * %false - we are done with this request
2656 * %true - still buffers pending for this request
2658 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2660 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2662 EXPORT_SYMBOL(blk_end_request);
2665 * blk_end_request_all - Helper function for drives to finish the request.
2666 * @rq: the request to finish
2667 * @error: %0 for success, < %0 for error
2670 * Completely finish @rq.
2672 void blk_end_request_all(struct request *rq, int error)
2675 unsigned int bidi_bytes = 0;
2677 if (unlikely(blk_bidi_rq(rq)))
2678 bidi_bytes = blk_rq_bytes(rq->next_rq);
2680 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2683 EXPORT_SYMBOL(blk_end_request_all);
2686 * blk_end_request_cur - Helper function to finish the current request chunk.
2687 * @rq: the request to finish the current chunk for
2688 * @error: %0 for success, < %0 for error
2691 * Complete the current consecutively mapped chunk from @rq.
2694 * %false - we are done with this request
2695 * %true - still buffers pending for this request
2697 bool blk_end_request_cur(struct request *rq, int error)
2699 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2701 EXPORT_SYMBOL(blk_end_request_cur);
2704 * blk_end_request_err - Finish a request till the next failure boundary.
2705 * @rq: the request to finish till the next failure boundary for
2706 * @error: must be negative errno
2709 * Complete @rq till the next failure boundary.
2712 * %false - we are done with this request
2713 * %true - still buffers pending for this request
2715 bool blk_end_request_err(struct request *rq, int error)
2717 WARN_ON(error >= 0);
2718 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2720 EXPORT_SYMBOL_GPL(blk_end_request_err);
2723 * __blk_end_request - Helper function for drivers to complete the request.
2724 * @rq: the request being processed
2725 * @error: %0 for success, < %0 for error
2726 * @nr_bytes: number of bytes to complete
2729 * Must be called with queue lock held unlike blk_end_request().
2732 * %false - we are done with this request
2733 * %true - still buffers pending for this request
2735 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2737 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2739 EXPORT_SYMBOL(__blk_end_request);
2742 * __blk_end_request_all - Helper function for drives to finish the request.
2743 * @rq: the request to finish
2744 * @error: %0 for success, < %0 for error
2747 * Completely finish @rq. Must be called with queue lock held.
2749 void __blk_end_request_all(struct request *rq, int error)
2752 unsigned int bidi_bytes = 0;
2754 if (unlikely(blk_bidi_rq(rq)))
2755 bidi_bytes = blk_rq_bytes(rq->next_rq);
2757 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2760 EXPORT_SYMBOL(__blk_end_request_all);
2763 * __blk_end_request_cur - Helper function to finish the current request chunk.
2764 * @rq: the request to finish the current chunk for
2765 * @error: %0 for success, < %0 for error
2768 * Complete the current consecutively mapped chunk from @rq. Must
2769 * be called with queue lock held.
2772 * %false - we are done with this request
2773 * %true - still buffers pending for this request
2775 bool __blk_end_request_cur(struct request *rq, int error)
2777 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2779 EXPORT_SYMBOL(__blk_end_request_cur);
2782 * __blk_end_request_err - Finish a request till the next failure boundary.
2783 * @rq: the request to finish till the next failure boundary for
2784 * @error: must be negative errno
2787 * Complete @rq till the next failure boundary. Must be called
2788 * with queue lock held.
2791 * %false - we are done with this request
2792 * %true - still buffers pending for this request
2794 bool __blk_end_request_err(struct request *rq, int error)
2796 WARN_ON(error >= 0);
2797 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2799 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2801 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2804 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2805 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2807 if (bio_has_data(bio))
2808 rq->nr_phys_segments = bio_phys_segments(q, bio);
2810 rq->__data_len = bio->bi_iter.bi_size;
2811 rq->bio = rq->biotail = bio;
2814 rq->rq_disk = bio->bi_bdev->bd_disk;
2817 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2819 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2820 * @rq: the request to be flushed
2823 * Flush all pages in @rq.
2825 void rq_flush_dcache_pages(struct request *rq)
2827 struct req_iterator iter;
2828 struct bio_vec bvec;
2830 rq_for_each_segment(bvec, rq, iter)
2831 flush_dcache_page(bvec.bv_page);
2833 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2837 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2838 * @q : the queue of the device being checked
2841 * Check if underlying low-level drivers of a device are busy.
2842 * If the drivers want to export their busy state, they must set own
2843 * exporting function using blk_queue_lld_busy() first.
2845 * Basically, this function is used only by request stacking drivers
2846 * to stop dispatching requests to underlying devices when underlying
2847 * devices are busy. This behavior helps more I/O merging on the queue
2848 * of the request stacking driver and prevents I/O throughput regression
2849 * on burst I/O load.
2852 * 0 - Not busy (The request stacking driver should dispatch request)
2853 * 1 - Busy (The request stacking driver should stop dispatching request)
2855 int blk_lld_busy(struct request_queue *q)
2858 return q->lld_busy_fn(q);
2862 EXPORT_SYMBOL_GPL(blk_lld_busy);
2865 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2866 * @rq: the clone request to be cleaned up
2869 * Free all bios in @rq for a cloned request.
2871 void blk_rq_unprep_clone(struct request *rq)
2875 while ((bio = rq->bio) != NULL) {
2876 rq->bio = bio->bi_next;
2881 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2884 * Copy attributes of the original request to the clone request.
2885 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2887 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2889 dst->cpu = src->cpu;
2890 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2891 dst->cmd_type = src->cmd_type;
2892 dst->__sector = blk_rq_pos(src);
2893 dst->__data_len = blk_rq_bytes(src);
2894 dst->nr_phys_segments = src->nr_phys_segments;
2895 dst->ioprio = src->ioprio;
2896 dst->extra_len = src->extra_len;
2900 * blk_rq_prep_clone - Helper function to setup clone request
2901 * @rq: the request to be setup
2902 * @rq_src: original request to be cloned
2903 * @bs: bio_set that bios for clone are allocated from
2904 * @gfp_mask: memory allocation mask for bio
2905 * @bio_ctr: setup function to be called for each clone bio.
2906 * Returns %0 for success, non %0 for failure.
2907 * @data: private data to be passed to @bio_ctr
2910 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2911 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2912 * are not copied, and copying such parts is the caller's responsibility.
2913 * Also, pages which the original bios are pointing to are not copied
2914 * and the cloned bios just point same pages.
2915 * So cloned bios must be completed before original bios, which means
2916 * the caller must complete @rq before @rq_src.
2918 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2919 struct bio_set *bs, gfp_t gfp_mask,
2920 int (*bio_ctr)(struct bio *, struct bio *, void *),
2923 struct bio *bio, *bio_src;
2928 blk_rq_init(NULL, rq);
2930 __rq_for_each_bio(bio_src, rq_src) {
2931 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2935 if (bio_ctr && bio_ctr(bio, bio_src, data))
2939 rq->biotail->bi_next = bio;
2942 rq->bio = rq->biotail = bio;
2945 __blk_rq_prep_clone(rq, rq_src);
2952 blk_rq_unprep_clone(rq);
2956 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2958 int kblockd_schedule_work(struct work_struct *work)
2960 return queue_work(kblockd_workqueue, work);
2962 EXPORT_SYMBOL(kblockd_schedule_work);
2964 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2965 unsigned long delay)
2967 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2969 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2971 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2972 unsigned long delay)
2974 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2976 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2979 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2980 * @plug: The &struct blk_plug that needs to be initialized
2983 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2984 * pending I/O should the task end up blocking between blk_start_plug() and
2985 * blk_finish_plug(). This is important from a performance perspective, but
2986 * also ensures that we don't deadlock. For instance, if the task is blocking
2987 * for a memory allocation, memory reclaim could end up wanting to free a
2988 * page belonging to that request that is currently residing in our private
2989 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2990 * this kind of deadlock.
2992 void blk_start_plug(struct blk_plug *plug)
2994 struct task_struct *tsk = current;
2996 INIT_LIST_HEAD(&plug->list);
2997 INIT_LIST_HEAD(&plug->mq_list);
2998 INIT_LIST_HEAD(&plug->cb_list);
3001 * If this is a nested plug, don't actually assign it. It will be
3002 * flushed on its own.
3006 * Store ordering should not be needed here, since a potential
3007 * preempt will imply a full memory barrier
3012 EXPORT_SYMBOL(blk_start_plug);
3014 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3016 struct request *rqa = container_of(a, struct request, queuelist);
3017 struct request *rqb = container_of(b, struct request, queuelist);
3019 return !(rqa->q < rqb->q ||
3020 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3024 * If 'from_schedule' is true, then postpone the dispatch of requests
3025 * until a safe kblockd context. We due this to avoid accidental big
3026 * additional stack usage in driver dispatch, in places where the originally
3027 * plugger did not intend it.
3029 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3031 __releases(q->queue_lock)
3033 trace_block_unplug(q, depth, !from_schedule);
3036 blk_run_queue_async(q);
3039 spin_unlock(q->queue_lock);
3042 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3044 LIST_HEAD(callbacks);
3046 while (!list_empty(&plug->cb_list)) {
3047 list_splice_init(&plug->cb_list, &callbacks);
3049 while (!list_empty(&callbacks)) {
3050 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3053 list_del(&cb->list);
3054 cb->callback(cb, from_schedule);
3059 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3062 struct blk_plug *plug = current->plug;
3063 struct blk_plug_cb *cb;
3068 list_for_each_entry(cb, &plug->cb_list, list)
3069 if (cb->callback == unplug && cb->data == data)
3072 /* Not currently on the callback list */
3073 BUG_ON(size < sizeof(*cb));
3074 cb = kzalloc(size, GFP_ATOMIC);
3077 cb->callback = unplug;
3078 list_add(&cb->list, &plug->cb_list);
3082 EXPORT_SYMBOL(blk_check_plugged);
3084 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3086 struct request_queue *q;
3087 unsigned long flags;
3092 flush_plug_callbacks(plug, from_schedule);
3094 if (!list_empty(&plug->mq_list))
3095 blk_mq_flush_plug_list(plug, from_schedule);
3097 if (list_empty(&plug->list))
3100 list_splice_init(&plug->list, &list);
3102 list_sort(NULL, &list, plug_rq_cmp);
3108 * Save and disable interrupts here, to avoid doing it for every
3109 * queue lock we have to take.
3111 local_irq_save(flags);
3112 while (!list_empty(&list)) {
3113 rq = list_entry_rq(list.next);
3114 list_del_init(&rq->queuelist);
3118 * This drops the queue lock
3121 queue_unplugged(q, depth, from_schedule);
3124 spin_lock(q->queue_lock);
3128 * Short-circuit if @q is dead
3130 if (unlikely(blk_queue_dying(q))) {
3131 __blk_end_request_all(rq, -ENODEV);
3136 * rq is already accounted, so use raw insert
3138 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3139 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3141 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3147 * This drops the queue lock
3150 queue_unplugged(q, depth, from_schedule);
3152 local_irq_restore(flags);
3155 void blk_finish_plug(struct blk_plug *plug)
3157 blk_flush_plug_list(plug, false);
3159 if (plug == current->plug)
3160 current->plug = NULL;
3162 EXPORT_SYMBOL(blk_finish_plug);
3164 #ifdef CONFIG_PM_RUNTIME
3166 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3167 * @q: the queue of the device
3168 * @dev: the device the queue belongs to
3171 * Initialize runtime-PM-related fields for @q and start auto suspend for
3172 * @dev. Drivers that want to take advantage of request-based runtime PM
3173 * should call this function after @dev has been initialized, and its
3174 * request queue @q has been allocated, and runtime PM for it can not happen
3175 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3176 * cases, driver should call this function before any I/O has taken place.
3178 * This function takes care of setting up using auto suspend for the device,
3179 * the autosuspend delay is set to -1 to make runtime suspend impossible
3180 * until an updated value is either set by user or by driver. Drivers do
3181 * not need to touch other autosuspend settings.
3183 * The block layer runtime PM is request based, so only works for drivers
3184 * that use request as their IO unit instead of those directly use bio's.
3186 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3189 q->rpm_status = RPM_ACTIVE;
3190 pm_runtime_set_autosuspend_delay(q->dev, -1);
3191 pm_runtime_use_autosuspend(q->dev);
3193 EXPORT_SYMBOL(blk_pm_runtime_init);
3196 * blk_pre_runtime_suspend - Pre runtime suspend check
3197 * @q: the queue of the device
3200 * This function will check if runtime suspend is allowed for the device
3201 * by examining if there are any requests pending in the queue. If there
3202 * are requests pending, the device can not be runtime suspended; otherwise,
3203 * the queue's status will be updated to SUSPENDING and the driver can
3204 * proceed to suspend the device.
3206 * For the not allowed case, we mark last busy for the device so that
3207 * runtime PM core will try to autosuspend it some time later.
3209 * This function should be called near the start of the device's
3210 * runtime_suspend callback.
3213 * 0 - OK to runtime suspend the device
3214 * -EBUSY - Device should not be runtime suspended
3216 int blk_pre_runtime_suspend(struct request_queue *q)
3220 spin_lock_irq(q->queue_lock);
3221 if (q->nr_pending) {
3223 pm_runtime_mark_last_busy(q->dev);
3225 q->rpm_status = RPM_SUSPENDING;
3227 spin_unlock_irq(q->queue_lock);
3230 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3233 * blk_post_runtime_suspend - Post runtime suspend processing
3234 * @q: the queue of the device
3235 * @err: return value of the device's runtime_suspend function
3238 * Update the queue's runtime status according to the return value of the
3239 * device's runtime suspend function and mark last busy for the device so
3240 * that PM core will try to auto suspend the device at a later time.
3242 * This function should be called near the end of the device's
3243 * runtime_suspend callback.
3245 void blk_post_runtime_suspend(struct request_queue *q, int err)
3247 spin_lock_irq(q->queue_lock);
3249 q->rpm_status = RPM_SUSPENDED;
3251 q->rpm_status = RPM_ACTIVE;
3252 pm_runtime_mark_last_busy(q->dev);
3254 spin_unlock_irq(q->queue_lock);
3256 EXPORT_SYMBOL(blk_post_runtime_suspend);
3259 * blk_pre_runtime_resume - Pre runtime resume processing
3260 * @q: the queue of the device
3263 * Update the queue's runtime status to RESUMING in preparation for the
3264 * runtime resume of the device.
3266 * This function should be called near the start of the device's
3267 * runtime_resume callback.
3269 void blk_pre_runtime_resume(struct request_queue *q)
3271 spin_lock_irq(q->queue_lock);
3272 q->rpm_status = RPM_RESUMING;
3273 spin_unlock_irq(q->queue_lock);
3275 EXPORT_SYMBOL(blk_pre_runtime_resume);
3278 * blk_post_runtime_resume - Post runtime resume processing
3279 * @q: the queue of the device
3280 * @err: return value of the device's runtime_resume function
3283 * Update the queue's runtime status according to the return value of the
3284 * device's runtime_resume function. If it is successfully resumed, process
3285 * the requests that are queued into the device's queue when it is resuming
3286 * and then mark last busy and initiate autosuspend for it.
3288 * This function should be called near the end of the device's
3289 * runtime_resume callback.
3291 void blk_post_runtime_resume(struct request_queue *q, int err)
3293 spin_lock_irq(q->queue_lock);
3295 q->rpm_status = RPM_ACTIVE;
3297 pm_runtime_mark_last_busy(q->dev);
3298 pm_request_autosuspend(q->dev);
3300 q->rpm_status = RPM_SUSPENDED;
3302 spin_unlock_irq(q->queue_lock);
3304 EXPORT_SYMBOL(blk_post_runtime_resume);
3307 int __init blk_dev_init(void)
3309 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3310 sizeof(((struct request *)0)->cmd_flags));
3312 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3313 kblockd_workqueue = alloc_workqueue("kblockd",
3314 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3315 WQ_POWER_EFFICIENT, 0);
3316 if (!kblockd_workqueue)
3317 panic("Failed to create kblockd\n");
3319 request_cachep = kmem_cache_create("blkdev_requests",
3320 sizeof(struct request), 0, SLAB_PANIC, NULL);
3322 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3323 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);