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);
103 #ifdef CONFIG_PREEMPT_RT_FULL
104 INIT_WORK(&rq->work, __blk_mq_complete_request_remote_work);
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_NONRT(!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 elevator_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 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
397 drain |= !list_empty(&q->queue_head);
398 for (i = 0; i < 2; i++) {
399 drain |= q->nr_rqs[i];
400 drain |= q->in_flight[i];
402 drain |= !list_empty(&fq->flush_queue[i]);
409 spin_unlock_irq(q->queue_lock);
413 spin_lock_irq(q->queue_lock);
417 * With queue marked dead, any woken up waiter will fail the
418 * allocation path, so the wakeup chaining is lost and we're
419 * left with hung waiters. We need to wake up those waiters.
422 struct request_list *rl;
424 blk_queue_for_each_rl(rl, q)
425 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
426 wake_up_all(&rl->wait[i]);
431 * blk_queue_bypass_start - enter queue bypass mode
432 * @q: queue of interest
434 * In bypass mode, only the dispatch FIFO queue of @q is used. This
435 * function makes @q enter bypass mode and drains all requests which were
436 * throttled or issued before. On return, it's guaranteed that no request
437 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
438 * inside queue or RCU read lock.
440 void blk_queue_bypass_start(struct request_queue *q)
442 spin_lock_irq(q->queue_lock);
444 queue_flag_set(QUEUE_FLAG_BYPASS, q);
445 spin_unlock_irq(q->queue_lock);
448 * Queues start drained. Skip actual draining till init is
449 * complete. This avoids lenghty delays during queue init which
450 * can happen many times during boot.
452 if (blk_queue_init_done(q)) {
453 spin_lock_irq(q->queue_lock);
454 __blk_drain_queue(q, false);
455 spin_unlock_irq(q->queue_lock);
457 /* ensure blk_queue_bypass() is %true inside RCU read lock */
461 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
464 * blk_queue_bypass_end - leave queue bypass mode
465 * @q: queue of interest
467 * Leave bypass mode and restore the normal queueing behavior.
469 void blk_queue_bypass_end(struct request_queue *q)
471 spin_lock_irq(q->queue_lock);
472 if (!--q->bypass_depth)
473 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
474 WARN_ON_ONCE(q->bypass_depth < 0);
475 spin_unlock_irq(q->queue_lock);
477 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
479 void blk_set_queue_dying(struct request_queue *q)
481 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
484 blk_mq_wake_waiters(q);
486 struct request_list *rl;
488 blk_queue_for_each_rl(rl, q) {
490 wake_up(&rl->wait[BLK_RW_SYNC]);
491 wake_up(&rl->wait[BLK_RW_ASYNC]);
496 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
499 * blk_cleanup_queue - shutdown a request queue
500 * @q: request queue to shutdown
502 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
503 * put it. All future requests will be failed immediately with -ENODEV.
505 void blk_cleanup_queue(struct request_queue *q)
507 spinlock_t *lock = q->queue_lock;
509 /* mark @q DYING, no new request or merges will be allowed afterwards */
510 mutex_lock(&q->sysfs_lock);
511 blk_set_queue_dying(q);
515 * A dying queue is permanently in bypass mode till released. Note
516 * that, unlike blk_queue_bypass_start(), we aren't performing
517 * synchronize_rcu() after entering bypass mode to avoid the delay
518 * as some drivers create and destroy a lot of queues while
519 * probing. This is still safe because blk_release_queue() will be
520 * called only after the queue refcnt drops to zero and nothing,
521 * RCU or not, would be traversing the queue by then.
524 queue_flag_set(QUEUE_FLAG_BYPASS, q);
526 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
527 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
528 queue_flag_set(QUEUE_FLAG_DYING, q);
529 spin_unlock_irq(lock);
530 mutex_unlock(&q->sysfs_lock);
533 * Drain all requests queued before DYING marking. Set DEAD flag to
534 * prevent that q->request_fn() gets invoked after draining finished.
537 blk_mq_freeze_queue(q);
541 __blk_drain_queue(q, true);
543 queue_flag_set(QUEUE_FLAG_DEAD, q);
544 spin_unlock_irq(lock);
546 /* @q won't process any more request, flush async actions */
547 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
551 blk_mq_free_queue(q);
554 if (q->queue_lock != &q->__queue_lock)
555 q->queue_lock = &q->__queue_lock;
556 spin_unlock_irq(lock);
558 bdi_destroy(&q->backing_dev_info);
560 /* @q is and will stay empty, shutdown and put */
563 EXPORT_SYMBOL(blk_cleanup_queue);
565 /* Allocate memory local to the request queue */
566 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
568 int nid = (int)(long)data;
569 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
572 static void free_request_struct(void *element, void *unused)
574 kmem_cache_free(request_cachep, element);
577 int blk_init_rl(struct request_list *rl, struct request_queue *q,
580 if (unlikely(rl->rq_pool))
584 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
585 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
586 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
587 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
589 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
591 (void *)(long)q->node, gfp_mask,
599 void blk_exit_rl(struct request_list *rl)
602 mempool_destroy(rl->rq_pool);
605 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
607 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
609 EXPORT_SYMBOL(blk_alloc_queue);
611 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
613 struct request_queue *q;
616 q = kmem_cache_alloc_node(blk_requestq_cachep,
617 gfp_mask | __GFP_ZERO, node_id);
621 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
625 q->backing_dev_info.ra_pages =
626 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
627 q->backing_dev_info.state = 0;
628 q->backing_dev_info.capabilities = 0;
629 q->backing_dev_info.name = "block";
632 err = bdi_init(&q->backing_dev_info);
636 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
637 laptop_mode_timer_fn, (unsigned long) q);
638 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
639 INIT_LIST_HEAD(&q->queue_head);
640 INIT_LIST_HEAD(&q->timeout_list);
641 INIT_LIST_HEAD(&q->icq_list);
642 #ifdef CONFIG_BLK_CGROUP
643 INIT_LIST_HEAD(&q->blkg_list);
645 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
647 kobject_init(&q->kobj, &blk_queue_ktype);
649 mutex_init(&q->sysfs_lock);
650 spin_lock_init(&q->__queue_lock);
653 * By default initialize queue_lock to internal lock and driver can
654 * override it later if need be.
656 q->queue_lock = &q->__queue_lock;
659 * A queue starts its life with bypass turned on to avoid
660 * unnecessary bypass on/off overhead and nasty surprises during
661 * init. The initial bypass will be finished when the queue is
662 * registered by blk_register_queue().
665 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
667 init_swait_head(&q->mq_freeze_wq);
669 if (blkcg_init_queue(q))
675 bdi_destroy(&q->backing_dev_info);
677 ida_simple_remove(&blk_queue_ida, q->id);
679 kmem_cache_free(blk_requestq_cachep, q);
682 EXPORT_SYMBOL(blk_alloc_queue_node);
685 * blk_init_queue - prepare a request queue for use with a block device
686 * @rfn: The function to be called to process requests that have been
687 * placed on the queue.
688 * @lock: Request queue spin lock
691 * If a block device wishes to use the standard request handling procedures,
692 * which sorts requests and coalesces adjacent requests, then it must
693 * call blk_init_queue(). The function @rfn will be called when there
694 * are requests on the queue that need to be processed. If the device
695 * supports plugging, then @rfn may not be called immediately when requests
696 * are available on the queue, but may be called at some time later instead.
697 * Plugged queues are generally unplugged when a buffer belonging to one
698 * of the requests on the queue is needed, or due to memory pressure.
700 * @rfn is not required, or even expected, to remove all requests off the
701 * queue, but only as many as it can handle at a time. If it does leave
702 * requests on the queue, it is responsible for arranging that the requests
703 * get dealt with eventually.
705 * The queue spin lock must be held while manipulating the requests on the
706 * request queue; this lock will be taken also from interrupt context, so irq
707 * disabling is needed for it.
709 * Function returns a pointer to the initialized request queue, or %NULL if
713 * blk_init_queue() must be paired with a blk_cleanup_queue() call
714 * when the block device is deactivated (such as at module unload).
717 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
719 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
721 EXPORT_SYMBOL(blk_init_queue);
723 struct request_queue *
724 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
726 struct request_queue *uninit_q, *q;
728 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
732 q = blk_init_allocated_queue(uninit_q, rfn, lock);
734 blk_cleanup_queue(uninit_q);
738 EXPORT_SYMBOL(blk_init_queue_node);
740 static void blk_queue_bio(struct request_queue *q, struct bio *bio);
742 struct request_queue *
743 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
749 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
753 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
757 q->prep_rq_fn = NULL;
758 q->unprep_rq_fn = NULL;
759 q->queue_flags |= QUEUE_FLAG_DEFAULT;
761 /* Override internal queue lock with supplied lock pointer */
763 q->queue_lock = lock;
766 * This also sets hw/phys segments, boundary and size
768 blk_queue_make_request(q, blk_queue_bio);
770 q->sg_reserved_size = INT_MAX;
772 /* Protect q->elevator from elevator_change */
773 mutex_lock(&q->sysfs_lock);
776 if (elevator_init(q, NULL)) {
777 mutex_unlock(&q->sysfs_lock);
781 mutex_unlock(&q->sysfs_lock);
786 blk_free_flush_queue(q->fq);
789 EXPORT_SYMBOL(blk_init_allocated_queue);
791 bool blk_get_queue(struct request_queue *q)
793 if (likely(!blk_queue_dying(q))) {
800 EXPORT_SYMBOL(blk_get_queue);
802 static inline void blk_free_request(struct request_list *rl, struct request *rq)
804 if (rq->cmd_flags & REQ_ELVPRIV) {
805 elv_put_request(rl->q, rq);
807 put_io_context(rq->elv.icq->ioc);
810 mempool_free(rq, rl->rq_pool);
814 * ioc_batching returns true if the ioc is a valid batching request and
815 * should be given priority access to a request.
817 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
823 * Make sure the process is able to allocate at least 1 request
824 * even if the batch times out, otherwise we could theoretically
827 return ioc->nr_batch_requests == q->nr_batching ||
828 (ioc->nr_batch_requests > 0
829 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
833 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
834 * will cause the process to be a "batcher" on all queues in the system. This
835 * is the behaviour we want though - once it gets a wakeup it should be given
838 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
840 if (!ioc || ioc_batching(q, ioc))
843 ioc->nr_batch_requests = q->nr_batching;
844 ioc->last_waited = jiffies;
847 static void __freed_request(struct request_list *rl, int sync)
849 struct request_queue *q = rl->q;
852 * bdi isn't aware of blkcg yet. As all async IOs end up root
853 * blkcg anyway, just use root blkcg state.
855 if (rl == &q->root_rl &&
856 rl->count[sync] < queue_congestion_off_threshold(q))
857 blk_clear_queue_congested(q, sync);
859 if (rl->count[sync] + 1 <= q->nr_requests) {
860 if (waitqueue_active(&rl->wait[sync]))
861 wake_up(&rl->wait[sync]);
863 blk_clear_rl_full(rl, sync);
868 * A request has just been released. Account for it, update the full and
869 * congestion status, wake up any waiters. Called under q->queue_lock.
871 static void freed_request(struct request_list *rl, unsigned int flags)
873 struct request_queue *q = rl->q;
874 int sync = rw_is_sync(flags);
878 if (flags & REQ_ELVPRIV)
881 __freed_request(rl, sync);
883 if (unlikely(rl->starved[sync ^ 1]))
884 __freed_request(rl, sync ^ 1);
887 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
889 struct request_list *rl;
891 spin_lock_irq(q->queue_lock);
893 blk_queue_congestion_threshold(q);
895 /* congestion isn't cgroup aware and follows root blkcg for now */
898 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
899 blk_set_queue_congested(q, BLK_RW_SYNC);
900 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
901 blk_clear_queue_congested(q, BLK_RW_SYNC);
903 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
904 blk_set_queue_congested(q, BLK_RW_ASYNC);
905 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
906 blk_clear_queue_congested(q, BLK_RW_ASYNC);
908 blk_queue_for_each_rl(rl, q) {
909 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
910 blk_set_rl_full(rl, BLK_RW_SYNC);
912 blk_clear_rl_full(rl, BLK_RW_SYNC);
913 wake_up(&rl->wait[BLK_RW_SYNC]);
916 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
917 blk_set_rl_full(rl, BLK_RW_ASYNC);
919 blk_clear_rl_full(rl, BLK_RW_ASYNC);
920 wake_up(&rl->wait[BLK_RW_ASYNC]);
924 spin_unlock_irq(q->queue_lock);
929 * Determine if elevator data should be initialized when allocating the
930 * request associated with @bio.
932 static bool blk_rq_should_init_elevator(struct bio *bio)
938 * Flush requests do not use the elevator so skip initialization.
939 * This allows a request to share the flush and elevator data.
941 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
948 * rq_ioc - determine io_context for request allocation
949 * @bio: request being allocated is for this bio (can be %NULL)
951 * Determine io_context to use for request allocation for @bio. May return
952 * %NULL if %current->io_context doesn't exist.
954 static struct io_context *rq_ioc(struct bio *bio)
956 #ifdef CONFIG_BLK_CGROUP
957 if (bio && bio->bi_ioc)
960 return current->io_context;
964 * __get_request - get a free request
965 * @rl: request list to allocate from
966 * @rw_flags: RW and SYNC flags
967 * @bio: bio to allocate request for (can be %NULL)
968 * @gfp_mask: allocation mask
970 * Get a free request from @q. This function may fail under memory
971 * pressure or if @q is dead.
973 * Must be called with @q->queue_lock held and,
974 * Returns ERR_PTR on failure, with @q->queue_lock held.
975 * Returns request pointer on success, with @q->queue_lock *not held*.
977 static struct request *__get_request(struct request_list *rl, int rw_flags,
978 struct bio *bio, gfp_t gfp_mask)
980 struct request_queue *q = rl->q;
982 struct elevator_type *et = q->elevator->type;
983 struct io_context *ioc = rq_ioc(bio);
984 struct io_cq *icq = NULL;
985 const bool is_sync = rw_is_sync(rw_flags) != 0;
988 if (unlikely(blk_queue_dying(q)))
989 return ERR_PTR(-ENODEV);
991 may_queue = elv_may_queue(q, rw_flags);
992 if (may_queue == ELV_MQUEUE_NO)
995 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
996 if (rl->count[is_sync]+1 >= q->nr_requests) {
998 * The queue will fill after this allocation, so set
999 * it as full, and mark this process as "batching".
1000 * This process will be allowed to complete a batch of
1001 * requests, others will be blocked.
1003 if (!blk_rl_full(rl, is_sync)) {
1004 ioc_set_batching(q, ioc);
1005 blk_set_rl_full(rl, is_sync);
1007 if (may_queue != ELV_MQUEUE_MUST
1008 && !ioc_batching(q, ioc)) {
1010 * The queue is full and the allocating
1011 * process is not a "batcher", and not
1012 * exempted by the IO scheduler
1014 return ERR_PTR(-ENOMEM);
1019 * bdi isn't aware of blkcg yet. As all async IOs end up
1020 * root blkcg anyway, just use root blkcg state.
1022 if (rl == &q->root_rl)
1023 blk_set_queue_congested(q, is_sync);
1027 * Only allow batching queuers to allocate up to 50% over the defined
1028 * limit of requests, otherwise we could have thousands of requests
1029 * allocated with any setting of ->nr_requests
1031 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1032 return ERR_PTR(-ENOMEM);
1034 q->nr_rqs[is_sync]++;
1035 rl->count[is_sync]++;
1036 rl->starved[is_sync] = 0;
1039 * Decide whether the new request will be managed by elevator. If
1040 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1041 * prevent the current elevator from being destroyed until the new
1042 * request is freed. This guarantees icq's won't be destroyed and
1043 * makes creating new ones safe.
1045 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1046 * it will be created after releasing queue_lock.
1048 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1049 rw_flags |= REQ_ELVPRIV;
1050 q->nr_rqs_elvpriv++;
1051 if (et->icq_cache && ioc)
1052 icq = ioc_lookup_icq(ioc, q);
1055 if (blk_queue_io_stat(q))
1056 rw_flags |= REQ_IO_STAT;
1057 spin_unlock_irq(q->queue_lock);
1059 /* allocate and init request */
1060 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1065 blk_rq_set_rl(rq, rl);
1066 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1069 if (rw_flags & REQ_ELVPRIV) {
1070 if (unlikely(et->icq_cache && !icq)) {
1072 icq = ioc_create_icq(ioc, q, gfp_mask);
1078 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1081 /* @rq->elv.icq holds io_context until @rq is freed */
1083 get_io_context(icq->ioc);
1087 * ioc may be NULL here, and ioc_batching will be false. That's
1088 * OK, if the queue is under the request limit then requests need
1089 * not count toward the nr_batch_requests limit. There will always
1090 * be some limit enforced by BLK_BATCH_TIME.
1092 if (ioc_batching(q, ioc))
1093 ioc->nr_batch_requests--;
1095 trace_block_getrq(q, bio, rw_flags & 1);
1100 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1101 * and may fail indefinitely under memory pressure and thus
1102 * shouldn't stall IO. Treat this request as !elvpriv. This will
1103 * disturb iosched and blkcg but weird is bettern than dead.
1105 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1106 __func__, dev_name(q->backing_dev_info.dev));
1108 rq->cmd_flags &= ~REQ_ELVPRIV;
1111 spin_lock_irq(q->queue_lock);
1112 q->nr_rqs_elvpriv--;
1113 spin_unlock_irq(q->queue_lock);
1118 * Allocation failed presumably due to memory. Undo anything we
1119 * might have messed up.
1121 * Allocating task should really be put onto the front of the wait
1122 * queue, but this is pretty rare.
1124 spin_lock_irq(q->queue_lock);
1125 freed_request(rl, rw_flags);
1128 * in the very unlikely event that allocation failed and no
1129 * requests for this direction was pending, mark us starved so that
1130 * freeing of a request in the other direction will notice
1131 * us. another possible fix would be to split the rq mempool into
1135 if (unlikely(rl->count[is_sync] == 0))
1136 rl->starved[is_sync] = 1;
1137 return ERR_PTR(-ENOMEM);
1141 * get_request - get a free request
1142 * @q: request_queue to allocate request from
1143 * @rw_flags: RW and SYNC flags
1144 * @bio: bio to allocate request for (can be %NULL)
1145 * @gfp_mask: allocation mask
1147 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1148 * function keeps retrying under memory pressure and fails iff @q is dead.
1150 * Must be called with @q->queue_lock held and,
1151 * Returns ERR_PTR on failure, with @q->queue_lock held.
1152 * Returns request pointer on success, with @q->queue_lock *not held*.
1154 static struct request *get_request(struct request_queue *q, int rw_flags,
1155 struct bio *bio, gfp_t gfp_mask)
1157 const bool is_sync = rw_is_sync(rw_flags) != 0;
1159 struct request_list *rl;
1162 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1164 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1168 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1173 /* wait on @rl and retry */
1174 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1175 TASK_UNINTERRUPTIBLE);
1177 trace_block_sleeprq(q, bio, rw_flags & 1);
1179 spin_unlock_irq(q->queue_lock);
1183 * After sleeping, we become a "batching" process and will be able
1184 * to allocate at least one request, and up to a big batch of them
1185 * for a small period time. See ioc_batching, ioc_set_batching
1187 ioc_set_batching(q, current->io_context);
1189 spin_lock_irq(q->queue_lock);
1190 finish_wait(&rl->wait[is_sync], &wait);
1195 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1200 BUG_ON(rw != READ && rw != WRITE);
1202 /* create ioc upfront */
1203 create_io_context(gfp_mask, q->node);
1205 spin_lock_irq(q->queue_lock);
1206 rq = get_request(q, rw, NULL, gfp_mask);
1208 spin_unlock_irq(q->queue_lock);
1209 /* q->queue_lock is unlocked at this point */
1214 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1217 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1219 return blk_old_get_request(q, rw, gfp_mask);
1221 EXPORT_SYMBOL(blk_get_request);
1224 * blk_make_request - given a bio, allocate a corresponding struct request.
1225 * @q: target request queue
1226 * @bio: The bio describing the memory mappings that will be submitted for IO.
1227 * It may be a chained-bio properly constructed by block/bio layer.
1228 * @gfp_mask: gfp flags to be used for memory allocation
1230 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1231 * type commands. Where the struct request needs to be farther initialized by
1232 * the caller. It is passed a &struct bio, which describes the memory info of
1235 * The caller of blk_make_request must make sure that bi_io_vec
1236 * are set to describe the memory buffers. That bio_data_dir() will return
1237 * the needed direction of the request. (And all bio's in the passed bio-chain
1238 * are properly set accordingly)
1240 * If called under none-sleepable conditions, mapped bio buffers must not
1241 * need bouncing, by calling the appropriate masked or flagged allocator,
1242 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1245 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1246 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1247 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1248 * completion of a bio that hasn't been submitted yet, thus resulting in a
1249 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1250 * of bio_alloc(), as that avoids the mempool deadlock.
1251 * If possible a big IO should be split into smaller parts when allocation
1252 * fails. Partial allocation should not be an error, or you risk a live-lock.
1254 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1257 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1262 blk_rq_set_block_pc(rq);
1265 struct bio *bounce_bio = bio;
1268 blk_queue_bounce(q, &bounce_bio);
1269 ret = blk_rq_append_bio(q, rq, bounce_bio);
1270 if (unlikely(ret)) {
1271 blk_put_request(rq);
1272 return ERR_PTR(ret);
1278 EXPORT_SYMBOL(blk_make_request);
1281 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1282 * @rq: request to be initialized
1285 void blk_rq_set_block_pc(struct request *rq)
1287 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1289 rq->__sector = (sector_t) -1;
1290 rq->bio = rq->biotail = NULL;
1291 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1293 EXPORT_SYMBOL(blk_rq_set_block_pc);
1296 * blk_requeue_request - put a request back on queue
1297 * @q: request queue where request should be inserted
1298 * @rq: request to be inserted
1301 * Drivers often keep queueing requests until the hardware cannot accept
1302 * more, when that condition happens we need to put the request back
1303 * on the queue. Must be called with queue lock held.
1305 void blk_requeue_request(struct request_queue *q, struct request *rq)
1307 blk_delete_timer(rq);
1308 blk_clear_rq_complete(rq);
1309 trace_block_rq_requeue(q, rq);
1311 if (rq->cmd_flags & REQ_QUEUED)
1312 blk_queue_end_tag(q, rq);
1314 BUG_ON(blk_queued_rq(rq));
1316 elv_requeue_request(q, rq);
1318 EXPORT_SYMBOL(blk_requeue_request);
1320 static void add_acct_request(struct request_queue *q, struct request *rq,
1323 blk_account_io_start(rq, true);
1324 __elv_add_request(q, rq, where);
1327 static void part_round_stats_single(int cpu, struct hd_struct *part,
1332 if (now == part->stamp)
1335 inflight = part_in_flight(part);
1337 __part_stat_add(cpu, part, time_in_queue,
1338 inflight * (now - part->stamp));
1339 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1345 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1346 * @cpu: cpu number for stats access
1347 * @part: target partition
1349 * The average IO queue length and utilisation statistics are maintained
1350 * by observing the current state of the queue length and the amount of
1351 * time it has been in this state for.
1353 * Normally, that accounting is done on IO completion, but that can result
1354 * in more than a second's worth of IO being accounted for within any one
1355 * second, leading to >100% utilisation. To deal with that, we call this
1356 * function to do a round-off before returning the results when reading
1357 * /proc/diskstats. This accounts immediately for all queue usage up to
1358 * the current jiffies and restarts the counters again.
1360 void part_round_stats(int cpu, struct hd_struct *part)
1362 unsigned long now = jiffies;
1365 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1366 part_round_stats_single(cpu, part, now);
1368 EXPORT_SYMBOL_GPL(part_round_stats);
1371 static void blk_pm_put_request(struct request *rq)
1373 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1374 pm_runtime_mark_last_busy(rq->q->dev);
1377 static inline void blk_pm_put_request(struct request *rq) {}
1381 * queue lock must be held
1383 void __blk_put_request(struct request_queue *q, struct request *req)
1389 blk_mq_free_request(req);
1393 blk_pm_put_request(req);
1395 elv_completed_request(q, req);
1397 /* this is a bio leak */
1398 WARN_ON(req->bio != NULL);
1401 * Request may not have originated from ll_rw_blk. if not,
1402 * it didn't come out of our reserved rq pools
1404 if (req->cmd_flags & REQ_ALLOCED) {
1405 unsigned int flags = req->cmd_flags;
1406 struct request_list *rl = blk_rq_rl(req);
1408 BUG_ON(!list_empty(&req->queuelist));
1409 BUG_ON(ELV_ON_HASH(req));
1411 blk_free_request(rl, req);
1412 freed_request(rl, flags);
1416 EXPORT_SYMBOL_GPL(__blk_put_request);
1418 void blk_put_request(struct request *req)
1420 struct request_queue *q = req->q;
1423 blk_mq_free_request(req);
1425 unsigned long flags;
1427 spin_lock_irqsave(q->queue_lock, flags);
1428 __blk_put_request(q, req);
1429 spin_unlock_irqrestore(q->queue_lock, flags);
1432 EXPORT_SYMBOL(blk_put_request);
1435 * blk_add_request_payload - add a payload to a request
1436 * @rq: request to update
1437 * @page: page backing the payload
1438 * @len: length of the payload.
1440 * This allows to later add a payload to an already submitted request by
1441 * a block driver. The driver needs to take care of freeing the payload
1444 * Note that this is a quite horrible hack and nothing but handling of
1445 * discard requests should ever use it.
1447 void blk_add_request_payload(struct request *rq, struct page *page,
1450 struct bio *bio = rq->bio;
1452 bio->bi_io_vec->bv_page = page;
1453 bio->bi_io_vec->bv_offset = 0;
1454 bio->bi_io_vec->bv_len = len;
1456 bio->bi_iter.bi_size = len;
1458 bio->bi_phys_segments = 1;
1460 rq->__data_len = rq->resid_len = len;
1461 rq->nr_phys_segments = 1;
1463 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1465 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1468 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1470 if (!ll_back_merge_fn(q, req, bio))
1473 trace_block_bio_backmerge(q, req, bio);
1475 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1476 blk_rq_set_mixed_merge(req);
1478 req->biotail->bi_next = bio;
1480 req->__data_len += bio->bi_iter.bi_size;
1481 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1483 blk_account_io_start(req, false);
1487 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1490 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1492 if (!ll_front_merge_fn(q, req, bio))
1495 trace_block_bio_frontmerge(q, req, bio);
1497 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1498 blk_rq_set_mixed_merge(req);
1500 bio->bi_next = req->bio;
1503 req->__sector = bio->bi_iter.bi_sector;
1504 req->__data_len += bio->bi_iter.bi_size;
1505 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1507 blk_account_io_start(req, false);
1512 * blk_attempt_plug_merge - try to merge with %current's plugged list
1513 * @q: request_queue new bio is being queued at
1514 * @bio: new bio being queued
1515 * @request_count: out parameter for number of traversed plugged requests
1517 * Determine whether @bio being queued on @q can be merged with a request
1518 * on %current's plugged list. Returns %true if merge was successful,
1521 * Plugging coalesces IOs from the same issuer for the same purpose without
1522 * going through @q->queue_lock. As such it's more of an issuing mechanism
1523 * than scheduling, and the request, while may have elvpriv data, is not
1524 * added on the elevator at this point. In addition, we don't have
1525 * reliable access to the elevator outside queue lock. Only check basic
1526 * merging parameters without querying the elevator.
1528 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1530 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1531 unsigned int *request_count)
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 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1557 el_ret = blk_try_merge(rq, bio);
1558 if (el_ret == ELEVATOR_BACK_MERGE) {
1559 ret = bio_attempt_back_merge(q, rq, bio);
1562 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1563 ret = bio_attempt_front_merge(q, rq, bio);
1572 void init_request_from_bio(struct request *req, struct bio *bio)
1574 req->cmd_type = REQ_TYPE_FS;
1576 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1577 if (bio->bi_rw & REQ_RAHEAD)
1578 req->cmd_flags |= REQ_FAILFAST_MASK;
1581 req->__sector = bio->bi_iter.bi_sector;
1582 req->ioprio = bio_prio(bio);
1583 blk_rq_bio_prep(req->q, req, bio);
1586 static void blk_queue_bio(struct request_queue *q, struct bio *bio)
1588 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1589 struct blk_plug *plug;
1590 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1591 struct request *req;
1592 unsigned int request_count = 0;
1595 * low level driver can indicate that it wants pages above a
1596 * certain limit bounced to low memory (ie for highmem, or even
1597 * ISA dma in theory)
1599 blk_queue_bounce(q, &bio);
1601 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1602 bio_endio(bio, -EIO);
1606 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1607 spin_lock_irq(q->queue_lock);
1608 where = ELEVATOR_INSERT_FLUSH;
1613 * Check if we can merge with the plugged list before grabbing
1616 if (!blk_queue_nomerges(q) &&
1617 blk_attempt_plug_merge(q, bio, &request_count))
1620 spin_lock_irq(q->queue_lock);
1622 el_ret = elv_merge(q, &req, bio);
1623 if (el_ret == ELEVATOR_BACK_MERGE) {
1624 if (bio_attempt_back_merge(q, req, bio)) {
1625 elv_bio_merged(q, req, bio);
1626 if (!attempt_back_merge(q, req))
1627 elv_merged_request(q, req, el_ret);
1630 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1631 if (bio_attempt_front_merge(q, req, bio)) {
1632 elv_bio_merged(q, req, bio);
1633 if (!attempt_front_merge(q, req))
1634 elv_merged_request(q, req, el_ret);
1641 * This sync check and mask will be re-done in init_request_from_bio(),
1642 * but we need to set it earlier to expose the sync flag to the
1643 * rq allocator and io schedulers.
1645 rw_flags = bio_data_dir(bio);
1647 rw_flags |= REQ_SYNC;
1650 * Grab a free request. This is might sleep but can not fail.
1651 * Returns with the queue unlocked.
1653 req = get_request(q, rw_flags, bio, GFP_NOIO);
1655 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
1660 * After dropping the lock and possibly sleeping here, our request
1661 * may now be mergeable after it had proven unmergeable (above).
1662 * We don't worry about that case for efficiency. It won't happen
1663 * often, and the elevators are able to handle it.
1665 init_request_from_bio(req, bio);
1667 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1668 req->cpu = raw_smp_processor_id();
1670 plug = current->plug;
1673 * If this is the first request added after a plug, fire
1677 trace_block_plug(q);
1679 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1680 blk_flush_plug_list(plug, false);
1681 trace_block_plug(q);
1684 list_add_tail(&req->queuelist, &plug->list);
1685 blk_account_io_start(req, true);
1687 spin_lock_irq(q->queue_lock);
1688 add_acct_request(q, req, where);
1691 spin_unlock_irq(q->queue_lock);
1696 * If bio->bi_dev is a partition, remap the location
1698 static inline void blk_partition_remap(struct bio *bio)
1700 struct block_device *bdev = bio->bi_bdev;
1702 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1703 struct hd_struct *p = bdev->bd_part;
1705 bio->bi_iter.bi_sector += p->start_sect;
1706 bio->bi_bdev = bdev->bd_contains;
1708 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1710 bio->bi_iter.bi_sector - p->start_sect);
1714 static void handle_bad_sector(struct bio *bio)
1716 char b[BDEVNAME_SIZE];
1718 printk(KERN_INFO "attempt to access beyond end of device\n");
1719 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1720 bdevname(bio->bi_bdev, b),
1722 (unsigned long long)bio_end_sector(bio),
1723 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1725 set_bit(BIO_EOF, &bio->bi_flags);
1728 #ifdef CONFIG_FAIL_MAKE_REQUEST
1730 static DECLARE_FAULT_ATTR(fail_make_request);
1732 static int __init setup_fail_make_request(char *str)
1734 return setup_fault_attr(&fail_make_request, str);
1736 __setup("fail_make_request=", setup_fail_make_request);
1738 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1740 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1743 static int __init fail_make_request_debugfs(void)
1745 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1746 NULL, &fail_make_request);
1748 return PTR_ERR_OR_ZERO(dir);
1751 late_initcall(fail_make_request_debugfs);
1753 #else /* CONFIG_FAIL_MAKE_REQUEST */
1755 static inline bool should_fail_request(struct hd_struct *part,
1761 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1764 * Check whether this bio extends beyond the end of the device.
1766 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1773 /* Test device or partition size, when known. */
1774 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1776 sector_t sector = bio->bi_iter.bi_sector;
1778 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1780 * This may well happen - the kernel calls bread()
1781 * without checking the size of the device, e.g., when
1782 * mounting a device.
1784 handle_bad_sector(bio);
1792 static noinline_for_stack bool
1793 generic_make_request_checks(struct bio *bio)
1795 struct request_queue *q;
1796 int nr_sectors = bio_sectors(bio);
1798 char b[BDEVNAME_SIZE];
1799 struct hd_struct *part;
1803 if (bio_check_eod(bio, nr_sectors))
1806 q = bdev_get_queue(bio->bi_bdev);
1809 "generic_make_request: Trying to access "
1810 "nonexistent block-device %s (%Lu)\n",
1811 bdevname(bio->bi_bdev, b),
1812 (long long) bio->bi_iter.bi_sector);
1816 if (likely(bio_is_rw(bio) &&
1817 nr_sectors > queue_max_hw_sectors(q))) {
1818 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1819 bdevname(bio->bi_bdev, b),
1821 queue_max_hw_sectors(q));
1825 part = bio->bi_bdev->bd_part;
1826 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1827 should_fail_request(&part_to_disk(part)->part0,
1828 bio->bi_iter.bi_size))
1832 * If this device has partitions, remap block n
1833 * of partition p to block n+start(p) of the disk.
1835 blk_partition_remap(bio);
1837 if (bio_check_eod(bio, nr_sectors))
1841 * Filter flush bio's early so that make_request based
1842 * drivers without flush support don't have to worry
1845 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1846 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1853 if ((bio->bi_rw & REQ_DISCARD) &&
1854 (!blk_queue_discard(q) ||
1855 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1860 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1866 * Various block parts want %current->io_context and lazy ioc
1867 * allocation ends up trading a lot of pain for a small amount of
1868 * memory. Just allocate it upfront. This may fail and block
1869 * layer knows how to live with it.
1871 create_io_context(GFP_ATOMIC, q->node);
1873 if (blk_throtl_bio(q, bio))
1874 return false; /* throttled, will be resubmitted later */
1876 trace_block_bio_queue(q, bio);
1880 bio_endio(bio, err);
1885 * generic_make_request - hand a buffer to its device driver for I/O
1886 * @bio: The bio describing the location in memory and on the device.
1888 * generic_make_request() is used to make I/O requests of block
1889 * devices. It is passed a &struct bio, which describes the I/O that needs
1892 * generic_make_request() does not return any status. The
1893 * success/failure status of the request, along with notification of
1894 * completion, is delivered asynchronously through the bio->bi_end_io
1895 * function described (one day) else where.
1897 * The caller of generic_make_request must make sure that bi_io_vec
1898 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1899 * set to describe the device address, and the
1900 * bi_end_io and optionally bi_private are set to describe how
1901 * completion notification should be signaled.
1903 * generic_make_request and the drivers it calls may use bi_next if this
1904 * bio happens to be merged with someone else, and may resubmit the bio to
1905 * a lower device by calling into generic_make_request recursively, which
1906 * means the bio should NOT be touched after the call to ->make_request_fn.
1908 void generic_make_request(struct bio *bio)
1910 struct bio_list bio_list_on_stack;
1912 if (!generic_make_request_checks(bio))
1916 * We only want one ->make_request_fn to be active at a time, else
1917 * stack usage with stacked devices could be a problem. So use
1918 * current->bio_list to keep a list of requests submited by a
1919 * make_request_fn function. current->bio_list is also used as a
1920 * flag to say if generic_make_request is currently active in this
1921 * task or not. If it is NULL, then no make_request is active. If
1922 * it is non-NULL, then a make_request is active, and new requests
1923 * should be added at the tail
1925 if (current->bio_list) {
1926 bio_list_add(current->bio_list, bio);
1930 /* following loop may be a bit non-obvious, and so deserves some
1932 * Before entering the loop, bio->bi_next is NULL (as all callers
1933 * ensure that) so we have a list with a single bio.
1934 * We pretend that we have just taken it off a longer list, so
1935 * we assign bio_list to a pointer to the bio_list_on_stack,
1936 * thus initialising the bio_list of new bios to be
1937 * added. ->make_request() may indeed add some more bios
1938 * through a recursive call to generic_make_request. If it
1939 * did, we find a non-NULL value in bio_list and re-enter the loop
1940 * from the top. In this case we really did just take the bio
1941 * of the top of the list (no pretending) and so remove it from
1942 * bio_list, and call into ->make_request() again.
1944 BUG_ON(bio->bi_next);
1945 bio_list_init(&bio_list_on_stack);
1946 current->bio_list = &bio_list_on_stack;
1948 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1950 q->make_request_fn(q, bio);
1952 bio = bio_list_pop(current->bio_list);
1954 current->bio_list = NULL; /* deactivate */
1956 EXPORT_SYMBOL(generic_make_request);
1959 * submit_bio - submit a bio to the block device layer for I/O
1960 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1961 * @bio: The &struct bio which describes the I/O
1963 * submit_bio() is very similar in purpose to generic_make_request(), and
1964 * uses that function to do most of the work. Both are fairly rough
1965 * interfaces; @bio must be presetup and ready for I/O.
1968 void submit_bio(int rw, struct bio *bio)
1973 * If it's a regular read/write or a barrier with data attached,
1974 * go through the normal accounting stuff before submission.
1976 if (bio_has_data(bio)) {
1979 if (unlikely(rw & REQ_WRITE_SAME))
1980 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1982 count = bio_sectors(bio);
1985 count_vm_events(PGPGOUT, count);
1987 task_io_account_read(bio->bi_iter.bi_size);
1988 count_vm_events(PGPGIN, count);
1991 if (unlikely(block_dump)) {
1992 char b[BDEVNAME_SIZE];
1993 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1994 current->comm, task_pid_nr(current),
1995 (rw & WRITE) ? "WRITE" : "READ",
1996 (unsigned long long)bio->bi_iter.bi_sector,
1997 bdevname(bio->bi_bdev, b),
2002 generic_make_request(bio);
2004 EXPORT_SYMBOL(submit_bio);
2007 * blk_rq_check_limits - Helper function to check a request for the queue limit
2009 * @rq: the request being checked
2012 * @rq may have been made based on weaker limitations of upper-level queues
2013 * in request stacking drivers, and it may violate the limitation of @q.
2014 * Since the block layer and the underlying device driver trust @rq
2015 * after it is inserted to @q, it should be checked against @q before
2016 * the insertion using this generic function.
2018 * This function should also be useful for request stacking drivers
2019 * in some cases below, so export this function.
2020 * Request stacking drivers like request-based dm may change the queue
2021 * limits while requests are in the queue (e.g. dm's table swapping).
2022 * Such request stacking drivers should check those requests against
2023 * the new queue limits again when they dispatch those requests,
2024 * although such checkings are also done against the old queue limits
2025 * when submitting requests.
2027 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2029 if (!rq_mergeable(rq))
2032 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2033 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2038 * queue's settings related to segment counting like q->bounce_pfn
2039 * may differ from that of other stacking queues.
2040 * Recalculate it to check the request correctly on this queue's
2043 blk_recalc_rq_segments(rq);
2044 if (rq->nr_phys_segments > queue_max_segments(q)) {
2045 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2051 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2054 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2055 * @q: the queue to submit the request
2056 * @rq: the request being queued
2058 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2060 unsigned long flags;
2061 int where = ELEVATOR_INSERT_BACK;
2063 if (blk_rq_check_limits(q, rq))
2067 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2071 if (blk_queue_io_stat(q))
2072 blk_account_io_start(rq, true);
2073 blk_mq_insert_request(rq, false, true, true);
2077 spin_lock_irqsave(q->queue_lock, flags);
2078 if (unlikely(blk_queue_dying(q))) {
2079 spin_unlock_irqrestore(q->queue_lock, flags);
2084 * Submitting request must be dequeued before calling this function
2085 * because it will be linked to another request_queue
2087 BUG_ON(blk_queued_rq(rq));
2089 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2090 where = ELEVATOR_INSERT_FLUSH;
2092 add_acct_request(q, rq, where);
2093 if (where == ELEVATOR_INSERT_FLUSH)
2095 spin_unlock_irqrestore(q->queue_lock, flags);
2099 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2102 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2103 * @rq: request to examine
2106 * A request could be merge of IOs which require different failure
2107 * handling. This function determines the number of bytes which
2108 * can be failed from the beginning of the request without
2109 * crossing into area which need to be retried further.
2112 * The number of bytes to fail.
2115 * queue_lock must be held.
2117 unsigned int blk_rq_err_bytes(const struct request *rq)
2119 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2120 unsigned int bytes = 0;
2123 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2124 return blk_rq_bytes(rq);
2127 * Currently the only 'mixing' which can happen is between
2128 * different fastfail types. We can safely fail portions
2129 * which have all the failfast bits that the first one has -
2130 * the ones which are at least as eager to fail as the first
2133 for (bio = rq->bio; bio; bio = bio->bi_next) {
2134 if ((bio->bi_rw & ff) != ff)
2136 bytes += bio->bi_iter.bi_size;
2139 /* this could lead to infinite loop */
2140 BUG_ON(blk_rq_bytes(rq) && !bytes);
2143 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2145 void blk_account_io_completion(struct request *req, unsigned int bytes)
2147 if (blk_do_io_stat(req)) {
2148 const int rw = rq_data_dir(req);
2149 struct hd_struct *part;
2152 cpu = part_stat_lock();
2154 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2159 void blk_account_io_done(struct request *req)
2162 * Account IO completion. flush_rq isn't accounted as a
2163 * normal IO on queueing nor completion. Accounting the
2164 * containing request is enough.
2166 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2167 unsigned long duration = jiffies - req->start_time;
2168 const int rw = rq_data_dir(req);
2169 struct hd_struct *part;
2172 cpu = part_stat_lock();
2175 part_stat_inc(cpu, part, ios[rw]);
2176 part_stat_add(cpu, part, ticks[rw], duration);
2177 part_round_stats(cpu, part);
2178 part_dec_in_flight(part, rw);
2180 hd_struct_put(part);
2187 * Don't process normal requests when queue is suspended
2188 * or in the process of suspending/resuming
2190 static struct request *blk_pm_peek_request(struct request_queue *q,
2193 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2194 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2200 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2207 void blk_account_io_start(struct request *rq, bool new_io)
2209 struct hd_struct *part;
2210 int rw = rq_data_dir(rq);
2213 if (!blk_do_io_stat(rq))
2216 cpu = part_stat_lock();
2220 part_stat_inc(cpu, part, merges[rw]);
2222 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2223 if (!hd_struct_try_get(part)) {
2225 * The partition is already being removed,
2226 * the request will be accounted on the disk only
2228 * We take a reference on disk->part0 although that
2229 * partition will never be deleted, so we can treat
2230 * it as any other partition.
2232 part = &rq->rq_disk->part0;
2233 hd_struct_get(part);
2235 part_round_stats(cpu, part);
2236 part_inc_in_flight(part, rw);
2244 * blk_peek_request - peek at the top of a request queue
2245 * @q: request queue to peek at
2248 * Return the request at the top of @q. The returned request
2249 * should be started using blk_start_request() before LLD starts
2253 * Pointer to the request at the top of @q if available. Null
2257 * queue_lock must be held.
2259 struct request *blk_peek_request(struct request_queue *q)
2264 while ((rq = __elv_next_request(q)) != NULL) {
2266 rq = blk_pm_peek_request(q, rq);
2270 if (!(rq->cmd_flags & REQ_STARTED)) {
2272 * This is the first time the device driver
2273 * sees this request (possibly after
2274 * requeueing). Notify IO scheduler.
2276 if (rq->cmd_flags & REQ_SORTED)
2277 elv_activate_rq(q, rq);
2280 * just mark as started even if we don't start
2281 * it, a request that has been delayed should
2282 * not be passed by new incoming requests
2284 rq->cmd_flags |= REQ_STARTED;
2285 trace_block_rq_issue(q, rq);
2288 if (!q->boundary_rq || q->boundary_rq == rq) {
2289 q->end_sector = rq_end_sector(rq);
2290 q->boundary_rq = NULL;
2293 if (rq->cmd_flags & REQ_DONTPREP)
2296 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2298 * make sure space for the drain appears we
2299 * know we can do this because max_hw_segments
2300 * has been adjusted to be one fewer than the
2303 rq->nr_phys_segments++;
2309 ret = q->prep_rq_fn(q, rq);
2310 if (ret == BLKPREP_OK) {
2312 } else if (ret == BLKPREP_DEFER) {
2314 * the request may have been (partially) prepped.
2315 * we need to keep this request in the front to
2316 * avoid resource deadlock. REQ_STARTED will
2317 * prevent other fs requests from passing this one.
2319 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2320 !(rq->cmd_flags & REQ_DONTPREP)) {
2322 * remove the space for the drain we added
2323 * so that we don't add it again
2325 --rq->nr_phys_segments;
2330 } else if (ret == BLKPREP_KILL) {
2331 rq->cmd_flags |= REQ_QUIET;
2333 * Mark this request as started so we don't trigger
2334 * any debug logic in the end I/O path.
2336 blk_start_request(rq);
2337 __blk_end_request_all(rq, -EIO);
2339 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2346 EXPORT_SYMBOL(blk_peek_request);
2348 void blk_dequeue_request(struct request *rq)
2350 struct request_queue *q = rq->q;
2352 BUG_ON(list_empty(&rq->queuelist));
2353 BUG_ON(ELV_ON_HASH(rq));
2355 list_del_init(&rq->queuelist);
2358 * the time frame between a request being removed from the lists
2359 * and to it is freed is accounted as io that is in progress at
2362 if (blk_account_rq(rq)) {
2363 q->in_flight[rq_is_sync(rq)]++;
2364 set_io_start_time_ns(rq);
2369 * blk_start_request - start request processing on the driver
2370 * @req: request to dequeue
2373 * Dequeue @req and start timeout timer on it. This hands off the
2374 * request to the driver.
2376 * Block internal functions which don't want to start timer should
2377 * call blk_dequeue_request().
2380 * queue_lock must be held.
2382 void blk_start_request(struct request *req)
2384 blk_dequeue_request(req);
2387 * We are now handing the request to the hardware, initialize
2388 * resid_len to full count and add the timeout handler.
2390 req->resid_len = blk_rq_bytes(req);
2391 if (unlikely(blk_bidi_rq(req)))
2392 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2394 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2397 EXPORT_SYMBOL(blk_start_request);
2400 * blk_fetch_request - fetch a request from a request queue
2401 * @q: request queue to fetch a request from
2404 * Return the request at the top of @q. The request is started on
2405 * return and LLD can start processing it immediately.
2408 * Pointer to the request at the top of @q if available. Null
2412 * queue_lock must be held.
2414 struct request *blk_fetch_request(struct request_queue *q)
2418 rq = blk_peek_request(q);
2420 blk_start_request(rq);
2423 EXPORT_SYMBOL(blk_fetch_request);
2426 * blk_update_request - Special helper function for request stacking drivers
2427 * @req: the request being processed
2428 * @error: %0 for success, < %0 for error
2429 * @nr_bytes: number of bytes to complete @req
2432 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2433 * the request structure even if @req doesn't have leftover.
2434 * If @req has leftover, sets it up for the next range of segments.
2436 * This special helper function is only for request stacking drivers
2437 * (e.g. request-based dm) so that they can handle partial completion.
2438 * Actual device drivers should use blk_end_request instead.
2440 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2441 * %false return from this function.
2444 * %false - this request doesn't have any more data
2445 * %true - this request has more data
2447 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2451 trace_block_rq_complete(req->q, req, nr_bytes);
2457 * For fs requests, rq is just carrier of independent bio's
2458 * and each partial completion should be handled separately.
2459 * Reset per-request error on each partial completion.
2461 * TODO: tj: This is too subtle. It would be better to let
2462 * low level drivers do what they see fit.
2464 if (req->cmd_type == REQ_TYPE_FS)
2467 if (error && req->cmd_type == REQ_TYPE_FS &&
2468 !(req->cmd_flags & REQ_QUIET)) {
2473 error_type = "recoverable transport";
2476 error_type = "critical target";
2479 error_type = "critical nexus";
2482 error_type = "timeout";
2485 error_type = "critical space allocation";
2488 error_type = "critical medium";
2495 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2496 __func__, error_type, req->rq_disk ?
2497 req->rq_disk->disk_name : "?",
2498 (unsigned long long)blk_rq_pos(req));
2502 blk_account_io_completion(req, nr_bytes);
2506 struct bio *bio = req->bio;
2507 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2509 if (bio_bytes == bio->bi_iter.bi_size)
2510 req->bio = bio->bi_next;
2512 req_bio_endio(req, bio, bio_bytes, error);
2514 total_bytes += bio_bytes;
2515 nr_bytes -= bio_bytes;
2526 * Reset counters so that the request stacking driver
2527 * can find how many bytes remain in the request
2530 req->__data_len = 0;
2534 req->__data_len -= total_bytes;
2536 /* update sector only for requests with clear definition of sector */
2537 if (req->cmd_type == REQ_TYPE_FS)
2538 req->__sector += total_bytes >> 9;
2540 /* mixed attributes always follow the first bio */
2541 if (req->cmd_flags & REQ_MIXED_MERGE) {
2542 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2543 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2547 * If total number of sectors is less than the first segment
2548 * size, something has gone terribly wrong.
2550 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2551 blk_dump_rq_flags(req, "request botched");
2552 req->__data_len = blk_rq_cur_bytes(req);
2555 /* recalculate the number of segments */
2556 blk_recalc_rq_segments(req);
2560 EXPORT_SYMBOL_GPL(blk_update_request);
2562 static bool blk_update_bidi_request(struct request *rq, int error,
2563 unsigned int nr_bytes,
2564 unsigned int bidi_bytes)
2566 if (blk_update_request(rq, error, nr_bytes))
2569 /* Bidi request must be completed as a whole */
2570 if (unlikely(blk_bidi_rq(rq)) &&
2571 blk_update_request(rq->next_rq, error, bidi_bytes))
2574 if (blk_queue_add_random(rq->q))
2575 add_disk_randomness(rq->rq_disk);
2581 * blk_unprep_request - unprepare a request
2584 * This function makes a request ready for complete resubmission (or
2585 * completion). It happens only after all error handling is complete,
2586 * so represents the appropriate moment to deallocate any resources
2587 * that were allocated to the request in the prep_rq_fn. The queue
2588 * lock is held when calling this.
2590 void blk_unprep_request(struct request *req)
2592 struct request_queue *q = req->q;
2594 req->cmd_flags &= ~REQ_DONTPREP;
2595 if (q->unprep_rq_fn)
2596 q->unprep_rq_fn(q, req);
2598 EXPORT_SYMBOL_GPL(blk_unprep_request);
2601 * queue lock must be held
2603 void blk_finish_request(struct request *req, int error)
2605 if (req->cmd_flags & REQ_QUEUED)
2606 blk_queue_end_tag(req->q, req);
2608 BUG_ON(blk_queued_rq(req));
2610 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2611 laptop_io_completion(&req->q->backing_dev_info);
2613 blk_delete_timer(req);
2615 if (req->cmd_flags & REQ_DONTPREP)
2616 blk_unprep_request(req);
2618 blk_account_io_done(req);
2621 req->end_io(req, error);
2623 if (blk_bidi_rq(req))
2624 __blk_put_request(req->next_rq->q, req->next_rq);
2626 __blk_put_request(req->q, req);
2629 EXPORT_SYMBOL(blk_finish_request);
2632 * blk_end_bidi_request - Complete a bidi request
2633 * @rq: the request to complete
2634 * @error: %0 for success, < %0 for error
2635 * @nr_bytes: number of bytes to complete @rq
2636 * @bidi_bytes: number of bytes to complete @rq->next_rq
2639 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2640 * Drivers that supports bidi can safely call this member for any
2641 * type of request, bidi or uni. In the later case @bidi_bytes is
2645 * %false - we are done with this request
2646 * %true - still buffers pending for this request
2648 static bool blk_end_bidi_request(struct request *rq, int error,
2649 unsigned int nr_bytes, unsigned int bidi_bytes)
2651 struct request_queue *q = rq->q;
2652 unsigned long flags;
2654 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2657 spin_lock_irqsave(q->queue_lock, flags);
2658 blk_finish_request(rq, error);
2659 spin_unlock_irqrestore(q->queue_lock, flags);
2665 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2666 * @rq: the request to complete
2667 * @error: %0 for success, < %0 for error
2668 * @nr_bytes: number of bytes to complete @rq
2669 * @bidi_bytes: number of bytes to complete @rq->next_rq
2672 * Identical to blk_end_bidi_request() except that queue lock is
2673 * assumed to be locked on entry and remains so on return.
2676 * %false - we are done with this request
2677 * %true - still buffers pending for this request
2679 bool __blk_end_bidi_request(struct request *rq, int error,
2680 unsigned int nr_bytes, unsigned int bidi_bytes)
2682 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2685 blk_finish_request(rq, error);
2691 * blk_end_request - Helper function for drivers to complete the request.
2692 * @rq: the request being processed
2693 * @error: %0 for success, < %0 for error
2694 * @nr_bytes: number of bytes to complete
2697 * Ends I/O on a number of bytes attached to @rq.
2698 * If @rq has leftover, sets it up for the next range of segments.
2701 * %false - we are done with this request
2702 * %true - still buffers pending for this request
2704 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2706 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2708 EXPORT_SYMBOL(blk_end_request);
2711 * blk_end_request_all - Helper function for drives to finish the request.
2712 * @rq: the request to finish
2713 * @error: %0 for success, < %0 for error
2716 * Completely finish @rq.
2718 void blk_end_request_all(struct request *rq, int error)
2721 unsigned int bidi_bytes = 0;
2723 if (unlikely(blk_bidi_rq(rq)))
2724 bidi_bytes = blk_rq_bytes(rq->next_rq);
2726 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2729 EXPORT_SYMBOL(blk_end_request_all);
2732 * blk_end_request_cur - Helper function to finish the current request chunk.
2733 * @rq: the request to finish the current chunk for
2734 * @error: %0 for success, < %0 for error
2737 * Complete the current consecutively mapped chunk from @rq.
2740 * %false - we are done with this request
2741 * %true - still buffers pending for this request
2743 bool blk_end_request_cur(struct request *rq, int error)
2745 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2747 EXPORT_SYMBOL(blk_end_request_cur);
2750 * blk_end_request_err - Finish a request till the next failure boundary.
2751 * @rq: the request to finish till the next failure boundary for
2752 * @error: must be negative errno
2755 * Complete @rq till the next failure boundary.
2758 * %false - we are done with this request
2759 * %true - still buffers pending for this request
2761 bool blk_end_request_err(struct request *rq, int error)
2763 WARN_ON(error >= 0);
2764 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2766 EXPORT_SYMBOL_GPL(blk_end_request_err);
2769 * __blk_end_request - Helper function for drivers to complete the request.
2770 * @rq: the request being processed
2771 * @error: %0 for success, < %0 for error
2772 * @nr_bytes: number of bytes to complete
2775 * Must be called with queue lock held unlike blk_end_request().
2778 * %false - we are done with this request
2779 * %true - still buffers pending for this request
2781 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2783 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2785 EXPORT_SYMBOL(__blk_end_request);
2788 * __blk_end_request_all - Helper function for drives to finish the request.
2789 * @rq: the request to finish
2790 * @error: %0 for success, < %0 for error
2793 * Completely finish @rq. Must be called with queue lock held.
2795 void __blk_end_request_all(struct request *rq, int error)
2798 unsigned int bidi_bytes = 0;
2800 if (unlikely(blk_bidi_rq(rq)))
2801 bidi_bytes = blk_rq_bytes(rq->next_rq);
2803 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2806 EXPORT_SYMBOL(__blk_end_request_all);
2809 * __blk_end_request_cur - Helper function to finish the current request chunk.
2810 * @rq: the request to finish the current chunk for
2811 * @error: %0 for success, < %0 for error
2814 * Complete the current consecutively mapped chunk from @rq. Must
2815 * be called with queue lock held.
2818 * %false - we are done with this request
2819 * %true - still buffers pending for this request
2821 bool __blk_end_request_cur(struct request *rq, int error)
2823 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2825 EXPORT_SYMBOL(__blk_end_request_cur);
2828 * __blk_end_request_err - Finish a request till the next failure boundary.
2829 * @rq: the request to finish till the next failure boundary for
2830 * @error: must be negative errno
2833 * Complete @rq till the next failure boundary. Must be called
2834 * with queue lock held.
2837 * %false - we are done with this request
2838 * %true - still buffers pending for this request
2840 bool __blk_end_request_err(struct request *rq, int error)
2842 WARN_ON(error >= 0);
2843 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2845 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2847 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2850 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2851 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2853 if (bio_has_data(bio))
2854 rq->nr_phys_segments = bio_phys_segments(q, bio);
2856 rq->__data_len = bio->bi_iter.bi_size;
2857 rq->bio = rq->biotail = bio;
2860 rq->rq_disk = bio->bi_bdev->bd_disk;
2863 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2865 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2866 * @rq: the request to be flushed
2869 * Flush all pages in @rq.
2871 void rq_flush_dcache_pages(struct request *rq)
2873 struct req_iterator iter;
2874 struct bio_vec bvec;
2876 rq_for_each_segment(bvec, rq, iter)
2877 flush_dcache_page(bvec.bv_page);
2879 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2883 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2884 * @q : the queue of the device being checked
2887 * Check if underlying low-level drivers of a device are busy.
2888 * If the drivers want to export their busy state, they must set own
2889 * exporting function using blk_queue_lld_busy() first.
2891 * Basically, this function is used only by request stacking drivers
2892 * to stop dispatching requests to underlying devices when underlying
2893 * devices are busy. This behavior helps more I/O merging on the queue
2894 * of the request stacking driver and prevents I/O throughput regression
2895 * on burst I/O load.
2898 * 0 - Not busy (The request stacking driver should dispatch request)
2899 * 1 - Busy (The request stacking driver should stop dispatching request)
2901 int blk_lld_busy(struct request_queue *q)
2904 return q->lld_busy_fn(q);
2908 EXPORT_SYMBOL_GPL(blk_lld_busy);
2911 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2912 * @rq: the clone request to be cleaned up
2915 * Free all bios in @rq for a cloned request.
2917 void blk_rq_unprep_clone(struct request *rq)
2921 while ((bio = rq->bio) != NULL) {
2922 rq->bio = bio->bi_next;
2927 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2930 * Copy attributes of the original request to the clone request.
2931 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2933 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2935 dst->cpu = src->cpu;
2936 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2937 dst->cmd_type = src->cmd_type;
2938 dst->__sector = blk_rq_pos(src);
2939 dst->__data_len = blk_rq_bytes(src);
2940 dst->nr_phys_segments = src->nr_phys_segments;
2941 dst->ioprio = src->ioprio;
2942 dst->extra_len = src->extra_len;
2946 * blk_rq_prep_clone - Helper function to setup clone request
2947 * @rq: the request to be setup
2948 * @rq_src: original request to be cloned
2949 * @bs: bio_set that bios for clone are allocated from
2950 * @gfp_mask: memory allocation mask for bio
2951 * @bio_ctr: setup function to be called for each clone bio.
2952 * Returns %0 for success, non %0 for failure.
2953 * @data: private data to be passed to @bio_ctr
2956 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2957 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2958 * are not copied, and copying such parts is the caller's responsibility.
2959 * Also, pages which the original bios are pointing to are not copied
2960 * and the cloned bios just point same pages.
2961 * So cloned bios must be completed before original bios, which means
2962 * the caller must complete @rq before @rq_src.
2964 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2965 struct bio_set *bs, gfp_t gfp_mask,
2966 int (*bio_ctr)(struct bio *, struct bio *, void *),
2969 struct bio *bio, *bio_src;
2974 __rq_for_each_bio(bio_src, rq_src) {
2975 bio = bio_clone_fast(bio_src, gfp_mask, bs);
2979 if (bio_ctr && bio_ctr(bio, bio_src, data))
2983 rq->biotail->bi_next = bio;
2986 rq->bio = rq->biotail = bio;
2989 __blk_rq_prep_clone(rq, rq_src);
2996 blk_rq_unprep_clone(rq);
3000 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3002 int kblockd_schedule_work(struct work_struct *work)
3004 return queue_work(kblockd_workqueue, work);
3006 EXPORT_SYMBOL(kblockd_schedule_work);
3008 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3009 unsigned long delay)
3011 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3013 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3015 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3016 unsigned long delay)
3018 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3020 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3023 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3024 * @plug: The &struct blk_plug that needs to be initialized
3027 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3028 * pending I/O should the task end up blocking between blk_start_plug() and
3029 * blk_finish_plug(). This is important from a performance perspective, but
3030 * also ensures that we don't deadlock. For instance, if the task is blocking
3031 * for a memory allocation, memory reclaim could end up wanting to free a
3032 * page belonging to that request that is currently residing in our private
3033 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3034 * this kind of deadlock.
3036 void blk_start_plug(struct blk_plug *plug)
3038 struct task_struct *tsk = current;
3040 INIT_LIST_HEAD(&plug->list);
3041 INIT_LIST_HEAD(&plug->mq_list);
3042 INIT_LIST_HEAD(&plug->cb_list);
3045 * If this is a nested plug, don't actually assign it. It will be
3046 * flushed on its own.
3050 * Store ordering should not be needed here, since a potential
3051 * preempt will imply a full memory barrier
3056 EXPORT_SYMBOL(blk_start_plug);
3058 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3060 struct request *rqa = container_of(a, struct request, queuelist);
3061 struct request *rqb = container_of(b, struct request, queuelist);
3063 return !(rqa->q < rqb->q ||
3064 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3068 * If 'from_schedule' is true, then postpone the dispatch of requests
3069 * until a safe kblockd context. We due this to avoid accidental big
3070 * additional stack usage in driver dispatch, in places where the originally
3071 * plugger did not intend it.
3073 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3075 __releases(q->queue_lock)
3077 trace_block_unplug(q, depth, !from_schedule);
3080 blk_run_queue_async(q);
3083 spin_unlock_irq(q->queue_lock);
3086 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3088 LIST_HEAD(callbacks);
3090 while (!list_empty(&plug->cb_list)) {
3091 list_splice_init(&plug->cb_list, &callbacks);
3093 while (!list_empty(&callbacks)) {
3094 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3097 list_del(&cb->list);
3098 cb->callback(cb, from_schedule);
3103 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3106 struct blk_plug *plug = current->plug;
3107 struct blk_plug_cb *cb;
3112 list_for_each_entry(cb, &plug->cb_list, list)
3113 if (cb->callback == unplug && cb->data == data)
3116 /* Not currently on the callback list */
3117 BUG_ON(size < sizeof(*cb));
3118 cb = kzalloc(size, GFP_ATOMIC);
3121 cb->callback = unplug;
3122 list_add(&cb->list, &plug->cb_list);
3126 EXPORT_SYMBOL(blk_check_plugged);
3128 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3130 struct request_queue *q;
3135 flush_plug_callbacks(plug, from_schedule);
3137 if (!list_empty(&plug->mq_list))
3138 blk_mq_flush_plug_list(plug, from_schedule);
3140 if (list_empty(&plug->list))
3143 list_splice_init(&plug->list, &list);
3145 list_sort(NULL, &list, plug_rq_cmp);
3150 while (!list_empty(&list)) {
3151 rq = list_entry_rq(list.next);
3152 list_del_init(&rq->queuelist);
3156 * This drops the queue lock
3159 queue_unplugged(q, depth, from_schedule);
3162 spin_lock_irq(q->queue_lock);
3166 * Short-circuit if @q is dead
3168 if (unlikely(blk_queue_dying(q))) {
3169 __blk_end_request_all(rq, -ENODEV);
3174 * rq is already accounted, so use raw insert
3176 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3177 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3179 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3185 * This drops the queue lock
3188 queue_unplugged(q, depth, from_schedule);
3191 void blk_finish_plug(struct blk_plug *plug)
3193 blk_flush_plug_list(plug, false);
3195 if (plug == current->plug)
3196 current->plug = NULL;
3198 EXPORT_SYMBOL(blk_finish_plug);
3202 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3203 * @q: the queue of the device
3204 * @dev: the device the queue belongs to
3207 * Initialize runtime-PM-related fields for @q and start auto suspend for
3208 * @dev. Drivers that want to take advantage of request-based runtime PM
3209 * should call this function after @dev has been initialized, and its
3210 * request queue @q has been allocated, and runtime PM for it can not happen
3211 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3212 * cases, driver should call this function before any I/O has taken place.
3214 * This function takes care of setting up using auto suspend for the device,
3215 * the autosuspend delay is set to -1 to make runtime suspend impossible
3216 * until an updated value is either set by user or by driver. Drivers do
3217 * not need to touch other autosuspend settings.
3219 * The block layer runtime PM is request based, so only works for drivers
3220 * that use request as their IO unit instead of those directly use bio's.
3222 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3225 q->rpm_status = RPM_ACTIVE;
3226 pm_runtime_set_autosuspend_delay(q->dev, -1);
3227 pm_runtime_use_autosuspend(q->dev);
3229 EXPORT_SYMBOL(blk_pm_runtime_init);
3232 * blk_pre_runtime_suspend - Pre runtime suspend check
3233 * @q: the queue of the device
3236 * This function will check if runtime suspend is allowed for the device
3237 * by examining if there are any requests pending in the queue. If there
3238 * are requests pending, the device can not be runtime suspended; otherwise,
3239 * the queue's status will be updated to SUSPENDING and the driver can
3240 * proceed to suspend the device.
3242 * For the not allowed case, we mark last busy for the device so that
3243 * runtime PM core will try to autosuspend it some time later.
3245 * This function should be called near the start of the device's
3246 * runtime_suspend callback.
3249 * 0 - OK to runtime suspend the device
3250 * -EBUSY - Device should not be runtime suspended
3252 int blk_pre_runtime_suspend(struct request_queue *q)
3256 spin_lock_irq(q->queue_lock);
3257 if (q->nr_pending) {
3259 pm_runtime_mark_last_busy(q->dev);
3261 q->rpm_status = RPM_SUSPENDING;
3263 spin_unlock_irq(q->queue_lock);
3266 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3269 * blk_post_runtime_suspend - Post runtime suspend processing
3270 * @q: the queue of the device
3271 * @err: return value of the device's runtime_suspend function
3274 * Update the queue's runtime status according to the return value of the
3275 * device's runtime suspend function and mark last busy for the device so
3276 * that PM core will try to auto suspend the device at a later time.
3278 * This function should be called near the end of the device's
3279 * runtime_suspend callback.
3281 void blk_post_runtime_suspend(struct request_queue *q, int err)
3283 spin_lock_irq(q->queue_lock);
3285 q->rpm_status = RPM_SUSPENDED;
3287 q->rpm_status = RPM_ACTIVE;
3288 pm_runtime_mark_last_busy(q->dev);
3290 spin_unlock_irq(q->queue_lock);
3292 EXPORT_SYMBOL(blk_post_runtime_suspend);
3295 * blk_pre_runtime_resume - Pre runtime resume processing
3296 * @q: the queue of the device
3299 * Update the queue's runtime status to RESUMING in preparation for the
3300 * runtime resume of the device.
3302 * This function should be called near the start of the device's
3303 * runtime_resume callback.
3305 void blk_pre_runtime_resume(struct request_queue *q)
3307 spin_lock_irq(q->queue_lock);
3308 q->rpm_status = RPM_RESUMING;
3309 spin_unlock_irq(q->queue_lock);
3311 EXPORT_SYMBOL(blk_pre_runtime_resume);
3314 * blk_post_runtime_resume - Post runtime resume processing
3315 * @q: the queue of the device
3316 * @err: return value of the device's runtime_resume function
3319 * Update the queue's runtime status according to the return value of the
3320 * device's runtime_resume function. If it is successfully resumed, process
3321 * the requests that are queued into the device's queue when it is resuming
3322 * and then mark last busy and initiate autosuspend for it.
3324 * This function should be called near the end of the device's
3325 * runtime_resume callback.
3327 void blk_post_runtime_resume(struct request_queue *q, int err)
3329 spin_lock_irq(q->queue_lock);
3331 q->rpm_status = RPM_ACTIVE;
3333 pm_runtime_mark_last_busy(q->dev);
3334 pm_request_autosuspend(q->dev);
3336 q->rpm_status = RPM_SUSPENDED;
3338 spin_unlock_irq(q->queue_lock);
3340 EXPORT_SYMBOL(blk_post_runtime_resume);
3343 int __init blk_dev_init(void)
3345 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3346 sizeof(((struct request *)0)->cmd_flags));
3348 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3349 kblockd_workqueue = alloc_workqueue("kblockd",
3350 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3351 if (!kblockd_workqueue)
3352 panic("Failed to create kblockd\n");
3354 request_cachep = kmem_cache_create("blkdev_requests",
3355 sizeof(struct request), 0, SLAB_PANIC, NULL);
3357 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3358 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);