2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
36 #define CREATE_TRACE_POINTS
37 #include <trace/events/block.h>
40 #include "blk-cgroup.h"
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
49 DEFINE_IDA(blk_queue_ida);
52 * For the allocated request tables
54 struct kmem_cache *request_cachep = NULL;
57 * For queue allocation
59 struct kmem_cache *blk_requestq_cachep;
62 * Controlling structure to kblockd
64 static struct workqueue_struct *kblockd_workqueue;
66 void blk_queue_congestion_threshold(struct request_queue *q)
70 nr = q->nr_requests - (q->nr_requests / 8) + 1;
71 if (nr > q->nr_requests)
73 q->nr_congestion_on = nr;
75 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
78 q->nr_congestion_off = nr;
82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
85 * Locates the passed device's request queue and returns the address of its
86 * backing_dev_info. This function can only be called if @bdev is opened
87 * and the return value is never NULL.
89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
91 struct request_queue *q = bdev_get_queue(bdev);
93 return &q->backing_dev_info;
95 EXPORT_SYMBOL(blk_get_backing_dev_info);
97 void blk_rq_init(struct request_queue *q, struct request *rq)
99 memset(rq, 0, sizeof(*rq));
101 INIT_LIST_HEAD(&rq->queuelist);
102 INIT_LIST_HEAD(&rq->timeout_list);
105 rq->__sector = (sector_t) -1;
106 INIT_HLIST_NODE(&rq->hash);
107 RB_CLEAR_NODE(&rq->rb_node);
109 rq->cmd_len = BLK_MAX_CDB;
111 rq->start_time = jiffies;
112 set_start_time_ns(rq);
115 EXPORT_SYMBOL(blk_rq_init);
117 static void req_bio_endio(struct request *rq, struct bio *bio,
118 unsigned int nbytes, int error)
121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
122 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
125 if (unlikely(rq->cmd_flags & REQ_QUIET))
126 set_bit(BIO_QUIET, &bio->bi_flags);
128 bio_advance(bio, nbytes);
130 /* don't actually finish bio if it's part of flush sequence */
131 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
132 bio_endio(bio, error);
135 void blk_dump_rq_flags(struct request *rq, char *msg)
139 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
140 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
141 (unsigned long long) rq->cmd_flags);
143 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
144 (unsigned long long)blk_rq_pos(rq),
145 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
146 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
147 rq->bio, rq->biotail, blk_rq_bytes(rq));
149 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
150 printk(KERN_INFO " cdb: ");
151 for (bit = 0; bit < BLK_MAX_CDB; bit++)
152 printk("%02x ", rq->cmd[bit]);
156 EXPORT_SYMBOL(blk_dump_rq_flags);
158 static void blk_delay_work(struct work_struct *work)
160 struct request_queue *q;
162 q = container_of(work, struct request_queue, delay_work.work);
163 spin_lock_irq(q->queue_lock);
165 spin_unlock_irq(q->queue_lock);
169 * blk_delay_queue - restart queueing after defined interval
170 * @q: The &struct request_queue in question
171 * @msecs: Delay in msecs
174 * Sometimes queueing needs to be postponed for a little while, to allow
175 * resources to come back. This function will make sure that queueing is
176 * restarted around the specified time. Queue lock must be held.
178 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
180 if (likely(!blk_queue_dead(q)))
181 queue_delayed_work(kblockd_workqueue, &q->delay_work,
182 msecs_to_jiffies(msecs));
184 EXPORT_SYMBOL(blk_delay_queue);
187 * blk_start_queue - restart a previously stopped queue
188 * @q: The &struct request_queue in question
191 * blk_start_queue() will clear the stop flag on the queue, and call
192 * the request_fn for the queue if it was in a stopped state when
193 * entered. Also see blk_stop_queue(). Queue lock must be held.
195 void blk_start_queue(struct request_queue *q)
197 WARN_ON(!irqs_disabled());
199 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
202 EXPORT_SYMBOL(blk_start_queue);
205 * blk_stop_queue - stop a queue
206 * @q: The &struct request_queue in question
209 * The Linux block layer assumes that a block driver will consume all
210 * entries on the request queue when the request_fn strategy is called.
211 * Often this will not happen, because of hardware limitations (queue
212 * depth settings). If a device driver gets a 'queue full' response,
213 * or if it simply chooses not to queue more I/O at one point, it can
214 * call this function to prevent the request_fn from being called until
215 * the driver has signalled it's ready to go again. This happens by calling
216 * blk_start_queue() to restart queue operations. Queue lock must be held.
218 void blk_stop_queue(struct request_queue *q)
220 cancel_delayed_work(&q->delay_work);
221 queue_flag_set(QUEUE_FLAG_STOPPED, q);
223 EXPORT_SYMBOL(blk_stop_queue);
226 * blk_sync_queue - cancel any pending callbacks on a queue
230 * The block layer may perform asynchronous callback activity
231 * on a queue, such as calling the unplug function after a timeout.
232 * A block device may call blk_sync_queue to ensure that any
233 * such activity is cancelled, thus allowing it to release resources
234 * that the callbacks might use. The caller must already have made sure
235 * that its ->make_request_fn will not re-add plugging prior to calling
238 * This function does not cancel any asynchronous activity arising
239 * out of elevator or throttling code. That would require elevator_exit()
240 * and blkcg_exit_queue() to be called with queue lock initialized.
243 void blk_sync_queue(struct request_queue *q)
245 del_timer_sync(&q->timeout);
248 struct blk_mq_hw_ctx *hctx;
251 queue_for_each_hw_ctx(q, hctx, i) {
252 cancel_delayed_work_sync(&hctx->run_work);
253 cancel_delayed_work_sync(&hctx->delay_work);
256 cancel_delayed_work_sync(&q->delay_work);
259 EXPORT_SYMBOL(blk_sync_queue);
262 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
263 * @q: The queue to run
266 * Invoke request handling on a queue if there are any pending requests.
267 * May be used to restart request handling after a request has completed.
268 * This variant runs the queue whether or not the queue has been
269 * stopped. Must be called with the queue lock held and interrupts
270 * disabled. See also @blk_run_queue.
272 inline void __blk_run_queue_uncond(struct request_queue *q)
274 if (unlikely(blk_queue_dead(q)))
278 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
279 * the queue lock internally. As a result multiple threads may be
280 * running such a request function concurrently. Keep track of the
281 * number of active request_fn invocations such that blk_drain_queue()
282 * can wait until all these request_fn calls have finished.
284 q->request_fn_active++;
286 q->request_fn_active--;
288 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
291 * __blk_run_queue - run a single device queue
292 * @q: The queue to run
295 * See @blk_run_queue. This variant must be called with the queue lock
296 * held and interrupts disabled.
298 void __blk_run_queue(struct request_queue *q)
300 if (unlikely(blk_queue_stopped(q)))
303 __blk_run_queue_uncond(q);
305 EXPORT_SYMBOL(__blk_run_queue);
308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
313 * of us. The caller must hold the queue lock.
315 void blk_run_queue_async(struct request_queue *q)
317 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
318 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
320 EXPORT_SYMBOL(blk_run_queue_async);
323 * blk_run_queue - run a single device queue
324 * @q: The queue to run
327 * Invoke request handling on this queue, if it has pending work to do.
328 * May be used to restart queueing when a request has completed.
330 void blk_run_queue(struct request_queue *q)
334 spin_lock_irqsave(q->queue_lock, flags);
336 spin_unlock_irqrestore(q->queue_lock, flags);
338 EXPORT_SYMBOL(blk_run_queue);
340 void blk_put_queue(struct request_queue *q)
342 kobject_put(&q->kobj);
344 EXPORT_SYMBOL(blk_put_queue);
347 * __blk_drain_queue - drain requests from request_queue
349 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
351 * Drain requests from @q. If @drain_all is set, all requests are drained.
352 * If not, only ELVPRIV requests are drained. The caller is responsible
353 * for ensuring that no new requests which need to be drained are queued.
355 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
356 __releases(q->queue_lock)
357 __acquires(q->queue_lock)
361 lockdep_assert_held(q->queue_lock);
367 * The caller might be trying to drain @q before its
368 * elevator is initialized.
371 elv_drain_elevator(q);
373 blkcg_drain_queue(q);
376 * This function might be called on a queue which failed
377 * driver init after queue creation or is not yet fully
378 * active yet. Some drivers (e.g. fd and loop) get unhappy
379 * in such cases. Kick queue iff dispatch queue has
380 * something on it and @q has request_fn set.
382 if (!list_empty(&q->queue_head) && q->request_fn)
385 drain |= q->nr_rqs_elvpriv;
386 drain |= q->request_fn_active;
389 * Unfortunately, requests are queued at and tracked from
390 * multiple places and there's no single counter which can
391 * be drained. Check all the queues and counters.
394 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
395 drain |= !list_empty(&q->queue_head);
396 for (i = 0; i < 2; i++) {
397 drain |= q->nr_rqs[i];
398 drain |= q->in_flight[i];
400 drain |= !list_empty(&fq->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)
440 spin_lock_irq(q->queue_lock);
442 queue_flag_set(QUEUE_FLAG_BYPASS, q);
443 spin_unlock_irq(q->queue_lock);
446 * Queues start drained. Skip actual draining till init is
447 * complete. This avoids lenghty delays during queue init which
448 * can happen many times during boot.
450 if (blk_queue_init_done(q)) {
451 spin_lock_irq(q->queue_lock);
452 __blk_drain_queue(q, false);
453 spin_unlock_irq(q->queue_lock);
455 /* ensure blk_queue_bypass() is %true inside RCU read lock */
459 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
462 * blk_queue_bypass_end - leave queue bypass mode
463 * @q: queue of interest
465 * Leave bypass mode and restore the normal queueing behavior.
467 void blk_queue_bypass_end(struct request_queue *q)
469 spin_lock_irq(q->queue_lock);
470 if (!--q->bypass_depth)
471 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
472 WARN_ON_ONCE(q->bypass_depth < 0);
473 spin_unlock_irq(q->queue_lock);
475 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
477 void blk_set_queue_dying(struct request_queue *q)
479 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
482 blk_mq_wake_waiters(q);
484 struct request_list *rl;
486 blk_queue_for_each_rl(rl, q) {
488 wake_up(&rl->wait[BLK_RW_SYNC]);
489 wake_up(&rl->wait[BLK_RW_ASYNC]);
494 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
497 * blk_cleanup_queue - shutdown a request queue
498 * @q: request queue to shutdown
500 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
501 * put it. All future requests will be failed immediately with -ENODEV.
503 void blk_cleanup_queue(struct request_queue *q)
505 spinlock_t *lock = q->queue_lock;
507 /* mark @q DYING, no new request or merges will be allowed afterwards */
508 mutex_lock(&q->sysfs_lock);
509 blk_set_queue_dying(q);
513 * A dying queue is permanently in bypass mode till released. Note
514 * that, unlike blk_queue_bypass_start(), we aren't performing
515 * synchronize_rcu() after entering bypass mode to avoid the delay
516 * as some drivers create and destroy a lot of queues while
517 * probing. This is still safe because blk_release_queue() will be
518 * called only after the queue refcnt drops to zero and nothing,
519 * RCU or not, would be traversing the queue by then.
522 queue_flag_set(QUEUE_FLAG_BYPASS, q);
524 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
525 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
526 queue_flag_set(QUEUE_FLAG_DYING, q);
527 spin_unlock_irq(lock);
528 mutex_unlock(&q->sysfs_lock);
531 * Drain all requests queued before DYING marking. Set DEAD flag to
532 * prevent that q->request_fn() gets invoked after draining finished.
535 blk_mq_freeze_queue(q);
539 __blk_drain_queue(q, true);
541 queue_flag_set(QUEUE_FLAG_DEAD, q);
542 spin_unlock_irq(lock);
544 /* @q won't process any more request, flush async actions */
545 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
549 blk_mq_free_queue(q);
552 if (q->queue_lock != &q->__queue_lock)
553 q->queue_lock = &q->__queue_lock;
554 spin_unlock_irq(lock);
556 bdi_destroy(&q->backing_dev_info);
558 /* @q is and will stay empty, shutdown and put */
561 EXPORT_SYMBOL(blk_cleanup_queue);
563 /* Allocate memory local to the request queue */
564 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
566 int nid = (int)(long)data;
567 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
570 static void free_request_struct(void *element, void *unused)
572 kmem_cache_free(request_cachep, element);
575 int blk_init_rl(struct request_list *rl, struct request_queue *q,
578 if (unlikely(rl->rq_pool))
582 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
583 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
584 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
585 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
587 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
589 (void *)(long)q->node, gfp_mask,
597 void blk_exit_rl(struct request_list *rl)
600 mempool_destroy(rl->rq_pool);
603 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
605 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
607 EXPORT_SYMBOL(blk_alloc_queue);
609 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
611 struct request_queue *q;
614 q = kmem_cache_alloc_node(blk_requestq_cachep,
615 gfp_mask | __GFP_ZERO, node_id);
619 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
623 q->backing_dev_info.ra_pages =
624 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
625 q->backing_dev_info.state = 0;
626 q->backing_dev_info.capabilities = 0;
627 q->backing_dev_info.name = "block";
630 err = bdi_init(&q->backing_dev_info);
634 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
635 laptop_mode_timer_fn, (unsigned long) q);
636 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
637 INIT_LIST_HEAD(&q->queue_head);
638 INIT_LIST_HEAD(&q->timeout_list);
639 INIT_LIST_HEAD(&q->icq_list);
640 #ifdef CONFIG_BLK_CGROUP
641 INIT_LIST_HEAD(&q->blkg_list);
643 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
645 kobject_init(&q->kobj, &blk_queue_ktype);
647 mutex_init(&q->sysfs_lock);
648 spin_lock_init(&q->__queue_lock);
651 * By default initialize queue_lock to internal lock and driver can
652 * override it later if need be.
654 q->queue_lock = &q->__queue_lock;
657 * A queue starts its life with bypass turned on to avoid
658 * unnecessary bypass on/off overhead and nasty surprises during
659 * init. The initial bypass will be finished when the queue is
660 * registered by blk_register_queue().
663 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
665 init_waitqueue_head(&q->mq_freeze_wq);
667 if (blkcg_init_queue(q))
673 bdi_destroy(&q->backing_dev_info);
675 ida_simple_remove(&blk_queue_ida, q->id);
677 kmem_cache_free(blk_requestq_cachep, q);
680 EXPORT_SYMBOL(blk_alloc_queue_node);
683 * blk_init_queue - prepare a request queue for use with a block device
684 * @rfn: The function to be called to process requests that have been
685 * placed on the queue.
686 * @lock: Request queue spin lock
689 * If a block device wishes to use the standard request handling procedures,
690 * which sorts requests and coalesces adjacent requests, then it must
691 * call blk_init_queue(). The function @rfn will be called when there
692 * are requests on the queue that need to be processed. If the device
693 * supports plugging, then @rfn may not be called immediately when requests
694 * are available on the queue, but may be called at some time later instead.
695 * Plugged queues are generally unplugged when a buffer belonging to one
696 * of the requests on the queue is needed, or due to memory pressure.
698 * @rfn is not required, or even expected, to remove all requests off the
699 * queue, but only as many as it can handle at a time. If it does leave
700 * requests on the queue, it is responsible for arranging that the requests
701 * get dealt with eventually.
703 * The queue spin lock must be held while manipulating the requests on the
704 * request queue; this lock will be taken also from interrupt context, so irq
705 * disabling is needed for it.
707 * Function returns a pointer to the initialized request queue, or %NULL if
711 * blk_init_queue() must be paired with a blk_cleanup_queue() call
712 * when the block device is deactivated (such as at module unload).
715 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
717 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
719 EXPORT_SYMBOL(blk_init_queue);
721 struct request_queue *
722 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
724 struct request_queue *uninit_q, *q;
726 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
730 q = blk_init_allocated_queue(uninit_q, rfn, lock);
732 blk_cleanup_queue(uninit_q);
736 EXPORT_SYMBOL(blk_init_queue_node);
738 static void blk_queue_bio(struct request_queue *q, struct bio *bio);
740 struct request_queue *
741 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
747 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
751 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
755 q->prep_rq_fn = NULL;
756 q->unprep_rq_fn = NULL;
757 q->queue_flags |= QUEUE_FLAG_DEFAULT;
759 /* Override internal queue lock with supplied lock pointer */
761 q->queue_lock = lock;
764 * This also sets hw/phys segments, boundary and size
766 blk_queue_make_request(q, blk_queue_bio);
768 q->sg_reserved_size = INT_MAX;
770 /* Protect q->elevator from elevator_change */
771 mutex_lock(&q->sysfs_lock);
774 if (elevator_init(q, NULL)) {
775 mutex_unlock(&q->sysfs_lock);
779 mutex_unlock(&q->sysfs_lock);
784 blk_free_flush_queue(q->fq);
787 EXPORT_SYMBOL(blk_init_allocated_queue);
789 bool blk_get_queue(struct request_queue *q)
791 if (likely(!blk_queue_dying(q))) {
798 EXPORT_SYMBOL(blk_get_queue);
800 static inline void blk_free_request(struct request_list *rl, struct request *rq)
802 if (rq->cmd_flags & REQ_ELVPRIV) {
803 elv_put_request(rl->q, rq);
805 put_io_context(rq->elv.icq->ioc);
808 mempool_free(rq, rl->rq_pool);
812 * ioc_batching returns true if the ioc is a valid batching request and
813 * should be given priority access to a request.
815 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
821 * Make sure the process is able to allocate at least 1 request
822 * even if the batch times out, otherwise we could theoretically
825 return ioc->nr_batch_requests == q->nr_batching ||
826 (ioc->nr_batch_requests > 0
827 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
831 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
832 * will cause the process to be a "batcher" on all queues in the system. This
833 * is the behaviour we want though - once it gets a wakeup it should be given
836 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
838 if (!ioc || ioc_batching(q, ioc))
841 ioc->nr_batch_requests = q->nr_batching;
842 ioc->last_waited = jiffies;
845 static void __freed_request(struct request_list *rl, int sync)
847 struct request_queue *q = rl->q;
850 * bdi isn't aware of blkcg yet. As all async IOs end up root
851 * blkcg anyway, just use root blkcg state.
853 if (rl == &q->root_rl &&
854 rl->count[sync] < queue_congestion_off_threshold(q))
855 blk_clear_queue_congested(q, sync);
857 if (rl->count[sync] + 1 <= q->nr_requests) {
858 if (waitqueue_active(&rl->wait[sync]))
859 wake_up(&rl->wait[sync]);
861 blk_clear_rl_full(rl, sync);
866 * A request has just been released. Account for it, update the full and
867 * congestion status, wake up any waiters. Called under q->queue_lock.
869 static void freed_request(struct request_list *rl, unsigned int flags)
871 struct request_queue *q = rl->q;
872 int sync = rw_is_sync(flags);
876 if (flags & REQ_ELVPRIV)
879 __freed_request(rl, sync);
881 if (unlikely(rl->starved[sync ^ 1]))
882 __freed_request(rl, sync ^ 1);
885 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
887 struct request_list *rl;
889 spin_lock_irq(q->queue_lock);
891 blk_queue_congestion_threshold(q);
893 /* congestion isn't cgroup aware and follows root blkcg for now */
896 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
897 blk_set_queue_congested(q, BLK_RW_SYNC);
898 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
899 blk_clear_queue_congested(q, BLK_RW_SYNC);
901 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
902 blk_set_queue_congested(q, BLK_RW_ASYNC);
903 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
904 blk_clear_queue_congested(q, BLK_RW_ASYNC);
906 blk_queue_for_each_rl(rl, q) {
907 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
908 blk_set_rl_full(rl, BLK_RW_SYNC);
910 blk_clear_rl_full(rl, BLK_RW_SYNC);
911 wake_up(&rl->wait[BLK_RW_SYNC]);
914 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
915 blk_set_rl_full(rl, BLK_RW_ASYNC);
917 blk_clear_rl_full(rl, BLK_RW_ASYNC);
918 wake_up(&rl->wait[BLK_RW_ASYNC]);
922 spin_unlock_irq(q->queue_lock);
927 * Determine if elevator data should be initialized when allocating the
928 * request associated with @bio.
930 static bool blk_rq_should_init_elevator(struct bio *bio)
936 * Flush requests do not use the elevator so skip initialization.
937 * This allows a request to share the flush and elevator data.
939 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
946 * rq_ioc - determine io_context for request allocation
947 * @bio: request being allocated is for this bio (can be %NULL)
949 * Determine io_context to use for request allocation for @bio. May return
950 * %NULL if %current->io_context doesn't exist.
952 static struct io_context *rq_ioc(struct bio *bio)
954 #ifdef CONFIG_BLK_CGROUP
955 if (bio && bio->bi_ioc)
958 return current->io_context;
962 * __get_request - get a free request
963 * @rl: request list to allocate from
964 * @rw_flags: RW and SYNC flags
965 * @bio: bio to allocate request for (can be %NULL)
966 * @gfp_mask: allocation mask
968 * Get a free request from @q. This function may fail under memory
969 * pressure or if @q is dead.
971 * Must be called with @q->queue_lock held and,
972 * Returns ERR_PTR on failure, with @q->queue_lock held.
973 * Returns request pointer on success, with @q->queue_lock *not held*.
975 static struct request *__get_request(struct request_list *rl, int rw_flags,
976 struct bio *bio, gfp_t gfp_mask)
978 struct request_queue *q = rl->q;
980 struct elevator_type *et = q->elevator->type;
981 struct io_context *ioc = rq_ioc(bio);
982 struct io_cq *icq = NULL;
983 const bool is_sync = rw_is_sync(rw_flags) != 0;
986 if (unlikely(blk_queue_dying(q)))
987 return ERR_PTR(-ENODEV);
989 may_queue = elv_may_queue(q, rw_flags);
990 if (may_queue == ELV_MQUEUE_NO)
993 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
994 if (rl->count[is_sync]+1 >= q->nr_requests) {
996 * The queue will fill after this allocation, so set
997 * it as full, and mark this process as "batching".
998 * This process will be allowed to complete a batch of
999 * requests, others will be blocked.
1001 if (!blk_rl_full(rl, is_sync)) {
1002 ioc_set_batching(q, ioc);
1003 blk_set_rl_full(rl, is_sync);
1005 if (may_queue != ELV_MQUEUE_MUST
1006 && !ioc_batching(q, ioc)) {
1008 * The queue is full and the allocating
1009 * process is not a "batcher", and not
1010 * exempted by the IO scheduler
1012 return ERR_PTR(-ENOMEM);
1017 * bdi isn't aware of blkcg yet. As all async IOs end up
1018 * root blkcg anyway, just use root blkcg state.
1020 if (rl == &q->root_rl)
1021 blk_set_queue_congested(q, is_sync);
1025 * Only allow batching queuers to allocate up to 50% over the defined
1026 * limit of requests, otherwise we could have thousands of requests
1027 * allocated with any setting of ->nr_requests
1029 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1030 return ERR_PTR(-ENOMEM);
1032 q->nr_rqs[is_sync]++;
1033 rl->count[is_sync]++;
1034 rl->starved[is_sync] = 0;
1037 * Decide whether the new request will be managed by elevator. If
1038 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1039 * prevent the current elevator from being destroyed until the new
1040 * request is freed. This guarantees icq's won't be destroyed and
1041 * makes creating new ones safe.
1043 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1044 * it will be created after releasing queue_lock.
1046 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1047 rw_flags |= REQ_ELVPRIV;
1048 q->nr_rqs_elvpriv++;
1049 if (et->icq_cache && ioc)
1050 icq = ioc_lookup_icq(ioc, q);
1053 if (blk_queue_io_stat(q))
1054 rw_flags |= REQ_IO_STAT;
1055 spin_unlock_irq(q->queue_lock);
1057 /* allocate and init request */
1058 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1063 blk_rq_set_rl(rq, rl);
1064 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1067 if (rw_flags & REQ_ELVPRIV) {
1068 if (unlikely(et->icq_cache && !icq)) {
1070 icq = ioc_create_icq(ioc, q, gfp_mask);
1076 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1079 /* @rq->elv.icq holds io_context until @rq is freed */
1081 get_io_context(icq->ioc);
1085 * ioc may be NULL here, and ioc_batching will be false. That's
1086 * OK, if the queue is under the request limit then requests need
1087 * not count toward the nr_batch_requests limit. There will always
1088 * be some limit enforced by BLK_BATCH_TIME.
1090 if (ioc_batching(q, ioc))
1091 ioc->nr_batch_requests--;
1093 trace_block_getrq(q, bio, rw_flags & 1);
1098 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1099 * and may fail indefinitely under memory pressure and thus
1100 * shouldn't stall IO. Treat this request as !elvpriv. This will
1101 * disturb iosched and blkcg but weird is bettern than dead.
1103 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1104 __func__, dev_name(q->backing_dev_info.dev));
1106 rq->cmd_flags &= ~REQ_ELVPRIV;
1109 spin_lock_irq(q->queue_lock);
1110 q->nr_rqs_elvpriv--;
1111 spin_unlock_irq(q->queue_lock);
1116 * Allocation failed presumably due to memory. Undo anything we
1117 * might have messed up.
1119 * Allocating task should really be put onto the front of the wait
1120 * queue, but this is pretty rare.
1122 spin_lock_irq(q->queue_lock);
1123 freed_request(rl, rw_flags);
1126 * in the very unlikely event that allocation failed and no
1127 * requests for this direction was pending, mark us starved so that
1128 * freeing of a request in the other direction will notice
1129 * us. another possible fix would be to split the rq mempool into
1133 if (unlikely(rl->count[is_sync] == 0))
1134 rl->starved[is_sync] = 1;
1135 return ERR_PTR(-ENOMEM);
1139 * get_request - get a free request
1140 * @q: request_queue to allocate request from
1141 * @rw_flags: RW and SYNC flags
1142 * @bio: bio to allocate request for (can be %NULL)
1143 * @gfp_mask: allocation mask
1145 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1146 * function keeps retrying under memory pressure and fails iff @q is dead.
1148 * Must be called with @q->queue_lock held and,
1149 * Returns ERR_PTR on failure, with @q->queue_lock held.
1150 * Returns request pointer on success, with @q->queue_lock *not held*.
1152 static struct request *get_request(struct request_queue *q, int rw_flags,
1153 struct bio *bio, gfp_t gfp_mask)
1155 const bool is_sync = rw_is_sync(rw_flags) != 0;
1157 struct request_list *rl;
1160 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1162 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1166 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1171 /* wait on @rl and retry */
1172 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1173 TASK_UNINTERRUPTIBLE);
1175 trace_block_sleeprq(q, bio, rw_flags & 1);
1177 spin_unlock_irq(q->queue_lock);
1181 * After sleeping, we become a "batching" process and will be able
1182 * to allocate at least one request, and up to a big batch of them
1183 * for a small period time. See ioc_batching, ioc_set_batching
1185 ioc_set_batching(q, current->io_context);
1187 spin_lock_irq(q->queue_lock);
1188 finish_wait(&rl->wait[is_sync], &wait);
1193 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1198 BUG_ON(rw != READ && rw != WRITE);
1200 /* create ioc upfront */
1201 create_io_context(gfp_mask, q->node);
1203 spin_lock_irq(q->queue_lock);
1204 rq = get_request(q, rw, NULL, gfp_mask);
1206 spin_unlock_irq(q->queue_lock);
1207 /* q->queue_lock is unlocked at this point */
1212 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1215 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1217 return blk_old_get_request(q, rw, gfp_mask);
1219 EXPORT_SYMBOL(blk_get_request);
1222 * blk_make_request - given a bio, allocate a corresponding struct request.
1223 * @q: target request queue
1224 * @bio: The bio describing the memory mappings that will be submitted for IO.
1225 * It may be a chained-bio properly constructed by block/bio layer.
1226 * @gfp_mask: gfp flags to be used for memory allocation
1228 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1229 * type commands. Where the struct request needs to be farther initialized by
1230 * the caller. It is passed a &struct bio, which describes the memory info of
1233 * The caller of blk_make_request must make sure that bi_io_vec
1234 * are set to describe the memory buffers. That bio_data_dir() will return
1235 * the needed direction of the request. (And all bio's in the passed bio-chain
1236 * are properly set accordingly)
1238 * If called under none-sleepable conditions, mapped bio buffers must not
1239 * need bouncing, by calling the appropriate masked or flagged allocator,
1240 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1243 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1244 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1245 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1246 * completion of a bio that hasn't been submitted yet, thus resulting in a
1247 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1248 * of bio_alloc(), as that avoids the mempool deadlock.
1249 * If possible a big IO should be split into smaller parts when allocation
1250 * fails. Partial allocation should not be an error, or you risk a live-lock.
1252 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1255 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1260 blk_rq_set_block_pc(rq);
1263 struct bio *bounce_bio = bio;
1266 blk_queue_bounce(q, &bounce_bio);
1267 ret = blk_rq_append_bio(q, rq, bounce_bio);
1268 if (unlikely(ret)) {
1269 blk_put_request(rq);
1270 return ERR_PTR(ret);
1276 EXPORT_SYMBOL(blk_make_request);
1279 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1280 * @rq: request to be initialized
1283 void blk_rq_set_block_pc(struct request *rq)
1285 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1287 rq->__sector = (sector_t) -1;
1288 rq->bio = rq->biotail = NULL;
1289 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1291 EXPORT_SYMBOL(blk_rq_set_block_pc);
1294 * blk_requeue_request - put a request back on queue
1295 * @q: request queue where request should be inserted
1296 * @rq: request to be inserted
1299 * Drivers often keep queueing requests until the hardware cannot accept
1300 * more, when that condition happens we need to put the request back
1301 * on the queue. Must be called with queue lock held.
1303 void blk_requeue_request(struct request_queue *q, struct request *rq)
1305 blk_delete_timer(rq);
1306 blk_clear_rq_complete(rq);
1307 trace_block_rq_requeue(q, rq);
1309 if (rq->cmd_flags & REQ_QUEUED)
1310 blk_queue_end_tag(q, rq);
1312 BUG_ON(blk_queued_rq(rq));
1314 elv_requeue_request(q, rq);
1316 EXPORT_SYMBOL(blk_requeue_request);
1318 static void add_acct_request(struct request_queue *q, struct request *rq,
1321 blk_account_io_start(rq, true);
1322 __elv_add_request(q, rq, where);
1325 static void part_round_stats_single(int cpu, struct hd_struct *part,
1330 if (now == part->stamp)
1333 inflight = part_in_flight(part);
1335 __part_stat_add(cpu, part, time_in_queue,
1336 inflight * (now - part->stamp));
1337 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1343 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1344 * @cpu: cpu number for stats access
1345 * @part: target partition
1347 * The average IO queue length and utilisation statistics are maintained
1348 * by observing the current state of the queue length and the amount of
1349 * time it has been in this state for.
1351 * Normally, that accounting is done on IO completion, but that can result
1352 * in more than a second's worth of IO being accounted for within any one
1353 * second, leading to >100% utilisation. To deal with that, we call this
1354 * function to do a round-off before returning the results when reading
1355 * /proc/diskstats. This accounts immediately for all queue usage up to
1356 * the current jiffies and restarts the counters again.
1358 void part_round_stats(int cpu, struct hd_struct *part)
1360 unsigned long now = jiffies;
1363 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1364 part_round_stats_single(cpu, part, now);
1366 EXPORT_SYMBOL_GPL(part_round_stats);
1369 static void blk_pm_put_request(struct request *rq)
1371 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1372 pm_runtime_mark_last_busy(rq->q->dev);
1375 static inline void blk_pm_put_request(struct request *rq) {}
1379 * queue lock must be held
1381 void __blk_put_request(struct request_queue *q, struct request *req)
1387 blk_mq_free_request(req);
1391 blk_pm_put_request(req);
1393 elv_completed_request(q, req);
1395 /* this is a bio leak */
1396 WARN_ON(req->bio != NULL);
1399 * Request may not have originated from ll_rw_blk. if not,
1400 * it didn't come out of our reserved rq pools
1402 if (req->cmd_flags & REQ_ALLOCED) {
1403 unsigned int flags = req->cmd_flags;
1404 struct request_list *rl = blk_rq_rl(req);
1406 BUG_ON(!list_empty(&req->queuelist));
1407 BUG_ON(ELV_ON_HASH(req));
1409 blk_free_request(rl, req);
1410 freed_request(rl, flags);
1414 EXPORT_SYMBOL_GPL(__blk_put_request);
1416 void blk_put_request(struct request *req)
1418 struct request_queue *q = req->q;
1421 blk_mq_free_request(req);
1423 unsigned long flags;
1425 spin_lock_irqsave(q->queue_lock, flags);
1426 __blk_put_request(q, req);
1427 spin_unlock_irqrestore(q->queue_lock, flags);
1430 EXPORT_SYMBOL(blk_put_request);
1433 * blk_add_request_payload - add a payload to a request
1434 * @rq: request to update
1435 * @page: page backing the payload
1436 * @len: length of the payload.
1438 * This allows to later add a payload to an already submitted request by
1439 * a block driver. The driver needs to take care of freeing the payload
1442 * Note that this is a quite horrible hack and nothing but handling of
1443 * discard requests should ever use it.
1445 void blk_add_request_payload(struct request *rq, struct page *page,
1448 struct bio *bio = rq->bio;
1450 bio->bi_io_vec->bv_page = page;
1451 bio->bi_io_vec->bv_offset = 0;
1452 bio->bi_io_vec->bv_len = len;
1454 bio->bi_iter.bi_size = len;
1456 bio->bi_phys_segments = 1;
1458 rq->__data_len = rq->resid_len = len;
1459 rq->nr_phys_segments = 1;
1461 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1463 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1466 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1468 if (!ll_back_merge_fn(q, req, bio))
1471 trace_block_bio_backmerge(q, req, bio);
1473 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1474 blk_rq_set_mixed_merge(req);
1476 req->biotail->bi_next = bio;
1478 req->__data_len += bio->bi_iter.bi_size;
1479 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1481 blk_account_io_start(req, false);
1485 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1488 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1490 if (!ll_front_merge_fn(q, req, bio))
1493 trace_block_bio_frontmerge(q, req, bio);
1495 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1496 blk_rq_set_mixed_merge(req);
1498 bio->bi_next = req->bio;
1501 req->__sector = bio->bi_iter.bi_sector;
1502 req->__data_len += bio->bi_iter.bi_size;
1503 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1505 blk_account_io_start(req, false);
1510 * blk_attempt_plug_merge - try to merge with %current's plugged list
1511 * @q: request_queue new bio is being queued at
1512 * @bio: new bio being queued
1513 * @request_count: out parameter for number of traversed plugged requests
1515 * Determine whether @bio being queued on @q can be merged with a request
1516 * on %current's plugged list. Returns %true if merge was successful,
1519 * Plugging coalesces IOs from the same issuer for the same purpose without
1520 * going through @q->queue_lock. As such it's more of an issuing mechanism
1521 * than scheduling, and the request, while may have elvpriv data, is not
1522 * added on the elevator at this point. In addition, we don't have
1523 * reliable access to the elevator outside queue lock. Only check basic
1524 * merging parameters without querying the elevator.
1526 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1528 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1529 unsigned int *request_count,
1530 struct request **same_queue_rq)
1532 struct blk_plug *plug;
1535 struct list_head *plug_list;
1537 plug = current->plug;
1543 plug_list = &plug->mq_list;
1545 plug_list = &plug->list;
1547 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1553 * Only blk-mq multiple hardware queues case checks the
1554 * rq in the same queue, there should be only one such
1558 *same_queue_rq = rq;
1561 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1564 el_ret = blk_try_merge(rq, bio);
1565 if (el_ret == ELEVATOR_BACK_MERGE) {
1566 ret = bio_attempt_back_merge(q, rq, bio);
1569 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1570 ret = bio_attempt_front_merge(q, rq, bio);
1579 void init_request_from_bio(struct request *req, struct bio *bio)
1581 req->cmd_type = REQ_TYPE_FS;
1583 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1584 if (bio->bi_rw & REQ_RAHEAD)
1585 req->cmd_flags |= REQ_FAILFAST_MASK;
1588 req->__sector = bio->bi_iter.bi_sector;
1589 req->ioprio = bio_prio(bio);
1590 blk_rq_bio_prep(req->q, req, bio);
1593 static void blk_queue_bio(struct request_queue *q, struct bio *bio)
1595 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1596 struct blk_plug *plug;
1597 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1598 struct request *req;
1599 unsigned int request_count = 0;
1602 * low level driver can indicate that it wants pages above a
1603 * certain limit bounced to low memory (ie for highmem, or even
1604 * ISA dma in theory)
1606 blk_queue_bounce(q, &bio);
1608 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1609 bio_endio(bio, -EIO);
1613 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1614 spin_lock_irq(q->queue_lock);
1615 where = ELEVATOR_INSERT_FLUSH;
1620 * Check if we can merge with the plugged list before grabbing
1623 if (!blk_queue_nomerges(q) &&
1624 blk_attempt_plug_merge(q, bio, &request_count, NULL))
1627 spin_lock_irq(q->queue_lock);
1629 el_ret = elv_merge(q, &req, bio);
1630 if (el_ret == ELEVATOR_BACK_MERGE) {
1631 if (bio_attempt_back_merge(q, req, bio)) {
1632 elv_bio_merged(q, req, bio);
1633 if (!attempt_back_merge(q, req))
1634 elv_merged_request(q, req, el_ret);
1637 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1638 if (bio_attempt_front_merge(q, req, bio)) {
1639 elv_bio_merged(q, req, bio);
1640 if (!attempt_front_merge(q, req))
1641 elv_merged_request(q, req, el_ret);
1648 * This sync check and mask will be re-done in init_request_from_bio(),
1649 * but we need to set it earlier to expose the sync flag to the
1650 * rq allocator and io schedulers.
1652 rw_flags = bio_data_dir(bio);
1654 rw_flags |= REQ_SYNC;
1657 * Grab a free request. This is might sleep but can not fail.
1658 * Returns with the queue unlocked.
1660 req = get_request(q, rw_flags, bio, GFP_NOIO);
1662 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
1667 * After dropping the lock and possibly sleeping here, our request
1668 * may now be mergeable after it had proven unmergeable (above).
1669 * We don't worry about that case for efficiency. It won't happen
1670 * often, and the elevators are able to handle it.
1672 init_request_from_bio(req, bio);
1674 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1675 req->cpu = raw_smp_processor_id();
1677 plug = current->plug;
1680 * If this is the first request added after a plug, fire
1684 trace_block_plug(q);
1686 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1687 blk_flush_plug_list(plug, false);
1688 trace_block_plug(q);
1691 list_add_tail(&req->queuelist, &plug->list);
1692 blk_account_io_start(req, true);
1694 spin_lock_irq(q->queue_lock);
1695 add_acct_request(q, req, where);
1698 spin_unlock_irq(q->queue_lock);
1703 * If bio->bi_dev is a partition, remap the location
1705 static inline void blk_partition_remap(struct bio *bio)
1707 struct block_device *bdev = bio->bi_bdev;
1709 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1710 struct hd_struct *p = bdev->bd_part;
1712 bio->bi_iter.bi_sector += p->start_sect;
1713 bio->bi_bdev = bdev->bd_contains;
1715 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1717 bio->bi_iter.bi_sector - p->start_sect);
1721 static void handle_bad_sector(struct bio *bio)
1723 char b[BDEVNAME_SIZE];
1725 printk(KERN_INFO "attempt to access beyond end of device\n");
1726 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1727 bdevname(bio->bi_bdev, b),
1729 (unsigned long long)bio_end_sector(bio),
1730 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1733 #ifdef CONFIG_FAIL_MAKE_REQUEST
1735 static DECLARE_FAULT_ATTR(fail_make_request);
1737 static int __init setup_fail_make_request(char *str)
1739 return setup_fault_attr(&fail_make_request, str);
1741 __setup("fail_make_request=", setup_fail_make_request);
1743 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1745 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1748 static int __init fail_make_request_debugfs(void)
1750 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1751 NULL, &fail_make_request);
1753 return PTR_ERR_OR_ZERO(dir);
1756 late_initcall(fail_make_request_debugfs);
1758 #else /* CONFIG_FAIL_MAKE_REQUEST */
1760 static inline bool should_fail_request(struct hd_struct *part,
1766 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1769 * Check whether this bio extends beyond the end of the device.
1771 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1778 /* Test device or partition size, when known. */
1779 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1781 sector_t sector = bio->bi_iter.bi_sector;
1783 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1785 * This may well happen - the kernel calls bread()
1786 * without checking the size of the device, e.g., when
1787 * mounting a device.
1789 handle_bad_sector(bio);
1797 static noinline_for_stack bool
1798 generic_make_request_checks(struct bio *bio)
1800 struct request_queue *q;
1801 int nr_sectors = bio_sectors(bio);
1803 char b[BDEVNAME_SIZE];
1804 struct hd_struct *part;
1808 if (bio_check_eod(bio, nr_sectors))
1811 q = bdev_get_queue(bio->bi_bdev);
1814 "generic_make_request: Trying to access "
1815 "nonexistent block-device %s (%Lu)\n",
1816 bdevname(bio->bi_bdev, b),
1817 (long long) bio->bi_iter.bi_sector);
1821 if (likely(bio_is_rw(bio) &&
1822 nr_sectors > queue_max_hw_sectors(q))) {
1823 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1824 bdevname(bio->bi_bdev, b),
1826 queue_max_hw_sectors(q));
1830 part = bio->bi_bdev->bd_part;
1831 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1832 should_fail_request(&part_to_disk(part)->part0,
1833 bio->bi_iter.bi_size))
1837 * If this device has partitions, remap block n
1838 * of partition p to block n+start(p) of the disk.
1840 blk_partition_remap(bio);
1842 if (bio_check_eod(bio, nr_sectors))
1846 * Filter flush bio's early so that make_request based
1847 * drivers without flush support don't have to worry
1850 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1851 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1858 if ((bio->bi_rw & REQ_DISCARD) &&
1859 (!blk_queue_discard(q) ||
1860 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1865 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1871 * Various block parts want %current->io_context and lazy ioc
1872 * allocation ends up trading a lot of pain for a small amount of
1873 * memory. Just allocate it upfront. This may fail and block
1874 * layer knows how to live with it.
1876 create_io_context(GFP_ATOMIC, q->node);
1878 if (blk_throtl_bio(q, bio))
1879 return false; /* throttled, will be resubmitted later */
1881 trace_block_bio_queue(q, bio);
1885 bio_endio(bio, err);
1890 * generic_make_request - hand a buffer to its device driver for I/O
1891 * @bio: The bio describing the location in memory and on the device.
1893 * generic_make_request() is used to make I/O requests of block
1894 * devices. It is passed a &struct bio, which describes the I/O that needs
1897 * generic_make_request() does not return any status. The
1898 * success/failure status of the request, along with notification of
1899 * completion, is delivered asynchronously through the bio->bi_end_io
1900 * function described (one day) else where.
1902 * The caller of generic_make_request must make sure that bi_io_vec
1903 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1904 * set to describe the device address, and the
1905 * bi_end_io and optionally bi_private are set to describe how
1906 * completion notification should be signaled.
1908 * generic_make_request and the drivers it calls may use bi_next if this
1909 * bio happens to be merged with someone else, and may resubmit the bio to
1910 * a lower device by calling into generic_make_request recursively, which
1911 * means the bio should NOT be touched after the call to ->make_request_fn.
1913 void generic_make_request(struct bio *bio)
1915 struct bio_list bio_list_on_stack;
1917 if (!generic_make_request_checks(bio))
1921 * We only want one ->make_request_fn to be active at a time, else
1922 * stack usage with stacked devices could be a problem. So use
1923 * current->bio_list to keep a list of requests submited by a
1924 * make_request_fn function. current->bio_list is also used as a
1925 * flag to say if generic_make_request is currently active in this
1926 * task or not. If it is NULL, then no make_request is active. If
1927 * it is non-NULL, then a make_request is active, and new requests
1928 * should be added at the tail
1930 if (current->bio_list) {
1931 bio_list_add(current->bio_list, bio);
1935 /* following loop may be a bit non-obvious, and so deserves some
1937 * Before entering the loop, bio->bi_next is NULL (as all callers
1938 * ensure that) so we have a list with a single bio.
1939 * We pretend that we have just taken it off a longer list, so
1940 * we assign bio_list to a pointer to the bio_list_on_stack,
1941 * thus initialising the bio_list of new bios to be
1942 * added. ->make_request() may indeed add some more bios
1943 * through a recursive call to generic_make_request. If it
1944 * did, we find a non-NULL value in bio_list and re-enter the loop
1945 * from the top. In this case we really did just take the bio
1946 * of the top of the list (no pretending) and so remove it from
1947 * bio_list, and call into ->make_request() again.
1949 BUG_ON(bio->bi_next);
1950 bio_list_init(&bio_list_on_stack);
1951 current->bio_list = &bio_list_on_stack;
1953 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1955 q->make_request_fn(q, bio);
1957 bio = bio_list_pop(current->bio_list);
1959 current->bio_list = NULL; /* deactivate */
1961 EXPORT_SYMBOL(generic_make_request);
1964 * submit_bio - submit a bio to the block device layer for I/O
1965 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1966 * @bio: The &struct bio which describes the I/O
1968 * submit_bio() is very similar in purpose to generic_make_request(), and
1969 * uses that function to do most of the work. Both are fairly rough
1970 * interfaces; @bio must be presetup and ready for I/O.
1973 void submit_bio(int rw, struct bio *bio)
1978 * If it's a regular read/write or a barrier with data attached,
1979 * go through the normal accounting stuff before submission.
1981 if (bio_has_data(bio)) {
1984 if (unlikely(rw & REQ_WRITE_SAME))
1985 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1987 count = bio_sectors(bio);
1990 count_vm_events(PGPGOUT, count);
1992 task_io_account_read(bio->bi_iter.bi_size);
1993 count_vm_events(PGPGIN, count);
1996 if (unlikely(block_dump)) {
1997 char b[BDEVNAME_SIZE];
1998 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1999 current->comm, task_pid_nr(current),
2000 (rw & WRITE) ? "WRITE" : "READ",
2001 (unsigned long long)bio->bi_iter.bi_sector,
2002 bdevname(bio->bi_bdev, b),
2007 generic_make_request(bio);
2009 EXPORT_SYMBOL(submit_bio);
2012 * blk_rq_check_limits - Helper function to check a request for the queue limit
2014 * @rq: the request being checked
2017 * @rq may have been made based on weaker limitations of upper-level queues
2018 * in request stacking drivers, and it may violate the limitation of @q.
2019 * Since the block layer and the underlying device driver trust @rq
2020 * after it is inserted to @q, it should be checked against @q before
2021 * the insertion using this generic function.
2023 * This function should also be useful for request stacking drivers
2024 * in some cases below, so export this function.
2025 * Request stacking drivers like request-based dm may change the queue
2026 * limits while requests are in the queue (e.g. dm's table swapping).
2027 * Such request stacking drivers should check those requests against
2028 * the new queue limits again when they dispatch those requests,
2029 * although such checkings are also done against the old queue limits
2030 * when submitting requests.
2032 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2034 if (!rq_mergeable(rq))
2037 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2038 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2043 * queue's settings related to segment counting like q->bounce_pfn
2044 * may differ from that of other stacking queues.
2045 * Recalculate it to check the request correctly on this queue's
2048 blk_recalc_rq_segments(rq);
2049 if (rq->nr_phys_segments > queue_max_segments(q)) {
2050 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2056 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2059 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2060 * @q: the queue to submit the request
2061 * @rq: the request being queued
2063 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2065 unsigned long flags;
2066 int where = ELEVATOR_INSERT_BACK;
2068 if (blk_rq_check_limits(q, rq))
2072 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2076 if (blk_queue_io_stat(q))
2077 blk_account_io_start(rq, true);
2078 blk_mq_insert_request(rq, false, true, true);
2082 spin_lock_irqsave(q->queue_lock, flags);
2083 if (unlikely(blk_queue_dying(q))) {
2084 spin_unlock_irqrestore(q->queue_lock, flags);
2089 * Submitting request must be dequeued before calling this function
2090 * because it will be linked to another request_queue
2092 BUG_ON(blk_queued_rq(rq));
2094 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2095 where = ELEVATOR_INSERT_FLUSH;
2097 add_acct_request(q, rq, where);
2098 if (where == ELEVATOR_INSERT_FLUSH)
2100 spin_unlock_irqrestore(q->queue_lock, flags);
2104 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2107 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2108 * @rq: request to examine
2111 * A request could be merge of IOs which require different failure
2112 * handling. This function determines the number of bytes which
2113 * can be failed from the beginning of the request without
2114 * crossing into area which need to be retried further.
2117 * The number of bytes to fail.
2120 * queue_lock must be held.
2122 unsigned int blk_rq_err_bytes(const struct request *rq)
2124 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2125 unsigned int bytes = 0;
2128 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2129 return blk_rq_bytes(rq);
2132 * Currently the only 'mixing' which can happen is between
2133 * different fastfail types. We can safely fail portions
2134 * which have all the failfast bits that the first one has -
2135 * the ones which are at least as eager to fail as the first
2138 for (bio = rq->bio; bio; bio = bio->bi_next) {
2139 if ((bio->bi_rw & ff) != ff)
2141 bytes += bio->bi_iter.bi_size;
2144 /* this could lead to infinite loop */
2145 BUG_ON(blk_rq_bytes(rq) && !bytes);
2148 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2150 void blk_account_io_completion(struct request *req, unsigned int bytes)
2152 if (blk_do_io_stat(req)) {
2153 const int rw = rq_data_dir(req);
2154 struct hd_struct *part;
2157 cpu = part_stat_lock();
2159 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2164 void blk_account_io_done(struct request *req)
2167 * Account IO completion. flush_rq isn't accounted as a
2168 * normal IO on queueing nor completion. Accounting the
2169 * containing request is enough.
2171 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2172 unsigned long duration = jiffies - req->start_time;
2173 const int rw = rq_data_dir(req);
2174 struct hd_struct *part;
2177 cpu = part_stat_lock();
2180 part_stat_inc(cpu, part, ios[rw]);
2181 part_stat_add(cpu, part, ticks[rw], duration);
2182 part_round_stats(cpu, part);
2183 part_dec_in_flight(part, rw);
2185 hd_struct_put(part);
2192 * Don't process normal requests when queue is suspended
2193 * or in the process of suspending/resuming
2195 static struct request *blk_pm_peek_request(struct request_queue *q,
2198 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2199 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2205 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2212 void blk_account_io_start(struct request *rq, bool new_io)
2214 struct hd_struct *part;
2215 int rw = rq_data_dir(rq);
2218 if (!blk_do_io_stat(rq))
2221 cpu = part_stat_lock();
2225 part_stat_inc(cpu, part, merges[rw]);
2227 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2228 if (!hd_struct_try_get(part)) {
2230 * The partition is already being removed,
2231 * the request will be accounted on the disk only
2233 * We take a reference on disk->part0 although that
2234 * partition will never be deleted, so we can treat
2235 * it as any other partition.
2237 part = &rq->rq_disk->part0;
2238 hd_struct_get(part);
2240 part_round_stats(cpu, part);
2241 part_inc_in_flight(part, rw);
2249 * blk_peek_request - peek at the top of a request queue
2250 * @q: request queue to peek at
2253 * Return the request at the top of @q. The returned request
2254 * should be started using blk_start_request() before LLD starts
2258 * Pointer to the request at the top of @q if available. Null
2262 * queue_lock must be held.
2264 struct request *blk_peek_request(struct request_queue *q)
2269 while ((rq = __elv_next_request(q)) != NULL) {
2271 rq = blk_pm_peek_request(q, rq);
2275 if (!(rq->cmd_flags & REQ_STARTED)) {
2277 * This is the first time the device driver
2278 * sees this request (possibly after
2279 * requeueing). Notify IO scheduler.
2281 if (rq->cmd_flags & REQ_SORTED)
2282 elv_activate_rq(q, rq);
2285 * just mark as started even if we don't start
2286 * it, a request that has been delayed should
2287 * not be passed by new incoming requests
2289 rq->cmd_flags |= REQ_STARTED;
2290 trace_block_rq_issue(q, rq);
2293 if (!q->boundary_rq || q->boundary_rq == rq) {
2294 q->end_sector = rq_end_sector(rq);
2295 q->boundary_rq = NULL;
2298 if (rq->cmd_flags & REQ_DONTPREP)
2301 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2303 * make sure space for the drain appears we
2304 * know we can do this because max_hw_segments
2305 * has been adjusted to be one fewer than the
2308 rq->nr_phys_segments++;
2314 ret = q->prep_rq_fn(q, rq);
2315 if (ret == BLKPREP_OK) {
2317 } else if (ret == BLKPREP_DEFER) {
2319 * the request may have been (partially) prepped.
2320 * we need to keep this request in the front to
2321 * avoid resource deadlock. REQ_STARTED will
2322 * prevent other fs requests from passing this one.
2324 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2325 !(rq->cmd_flags & REQ_DONTPREP)) {
2327 * remove the space for the drain we added
2328 * so that we don't add it again
2330 --rq->nr_phys_segments;
2335 } else if (ret == BLKPREP_KILL) {
2336 rq->cmd_flags |= REQ_QUIET;
2338 * Mark this request as started so we don't trigger
2339 * any debug logic in the end I/O path.
2341 blk_start_request(rq);
2342 __blk_end_request_all(rq, -EIO);
2344 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2351 EXPORT_SYMBOL(blk_peek_request);
2353 void blk_dequeue_request(struct request *rq)
2355 struct request_queue *q = rq->q;
2357 BUG_ON(list_empty(&rq->queuelist));
2358 BUG_ON(ELV_ON_HASH(rq));
2360 list_del_init(&rq->queuelist);
2363 * the time frame between a request being removed from the lists
2364 * and to it is freed is accounted as io that is in progress at
2367 if (blk_account_rq(rq)) {
2368 q->in_flight[rq_is_sync(rq)]++;
2369 set_io_start_time_ns(rq);
2374 * blk_start_request - start request processing on the driver
2375 * @req: request to dequeue
2378 * Dequeue @req and start timeout timer on it. This hands off the
2379 * request to the driver.
2381 * Block internal functions which don't want to start timer should
2382 * call blk_dequeue_request().
2385 * queue_lock must be held.
2387 void blk_start_request(struct request *req)
2389 blk_dequeue_request(req);
2392 * We are now handing the request to the hardware, initialize
2393 * resid_len to full count and add the timeout handler.
2395 req->resid_len = blk_rq_bytes(req);
2396 if (unlikely(blk_bidi_rq(req)))
2397 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2399 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2402 EXPORT_SYMBOL(blk_start_request);
2405 * blk_fetch_request - fetch a request from a request queue
2406 * @q: request queue to fetch a request from
2409 * Return the request at the top of @q. The request is started on
2410 * return and LLD can start processing it immediately.
2413 * Pointer to the request at the top of @q if available. Null
2417 * queue_lock must be held.
2419 struct request *blk_fetch_request(struct request_queue *q)
2423 rq = blk_peek_request(q);
2425 blk_start_request(rq);
2428 EXPORT_SYMBOL(blk_fetch_request);
2431 * blk_update_request - Special helper function for request stacking drivers
2432 * @req: the request being processed
2433 * @error: %0 for success, < %0 for error
2434 * @nr_bytes: number of bytes to complete @req
2437 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2438 * the request structure even if @req doesn't have leftover.
2439 * If @req has leftover, sets it up for the next range of segments.
2441 * This special helper function is only for request stacking drivers
2442 * (e.g. request-based dm) so that they can handle partial completion.
2443 * Actual device drivers should use blk_end_request instead.
2445 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2446 * %false return from this function.
2449 * %false - this request doesn't have any more data
2450 * %true - this request has more data
2452 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2456 trace_block_rq_complete(req->q, req, nr_bytes);
2462 * For fs requests, rq is just carrier of independent bio's
2463 * and each partial completion should be handled separately.
2464 * Reset per-request error on each partial completion.
2466 * TODO: tj: This is too subtle. It would be better to let
2467 * low level drivers do what they see fit.
2469 if (req->cmd_type == REQ_TYPE_FS)
2472 if (error && req->cmd_type == REQ_TYPE_FS &&
2473 !(req->cmd_flags & REQ_QUIET)) {
2478 error_type = "recoverable transport";
2481 error_type = "critical target";
2484 error_type = "critical nexus";
2487 error_type = "timeout";
2490 error_type = "critical space allocation";
2493 error_type = "critical medium";
2500 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2501 __func__, error_type, req->rq_disk ?
2502 req->rq_disk->disk_name : "?",
2503 (unsigned long long)blk_rq_pos(req));
2507 blk_account_io_completion(req, nr_bytes);
2511 struct bio *bio = req->bio;
2512 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2514 if (bio_bytes == bio->bi_iter.bi_size)
2515 req->bio = bio->bi_next;
2517 req_bio_endio(req, bio, bio_bytes, error);
2519 total_bytes += bio_bytes;
2520 nr_bytes -= bio_bytes;
2531 * Reset counters so that the request stacking driver
2532 * can find how many bytes remain in the request
2535 req->__data_len = 0;
2539 req->__data_len -= total_bytes;
2541 /* update sector only for requests with clear definition of sector */
2542 if (req->cmd_type == REQ_TYPE_FS)
2543 req->__sector += total_bytes >> 9;
2545 /* mixed attributes always follow the first bio */
2546 if (req->cmd_flags & REQ_MIXED_MERGE) {
2547 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2548 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2552 * If total number of sectors is less than the first segment
2553 * size, something has gone terribly wrong.
2555 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2556 blk_dump_rq_flags(req, "request botched");
2557 req->__data_len = blk_rq_cur_bytes(req);
2560 /* recalculate the number of segments */
2561 blk_recalc_rq_segments(req);
2565 EXPORT_SYMBOL_GPL(blk_update_request);
2567 static bool blk_update_bidi_request(struct request *rq, int error,
2568 unsigned int nr_bytes,
2569 unsigned int bidi_bytes)
2571 if (blk_update_request(rq, error, nr_bytes))
2574 /* Bidi request must be completed as a whole */
2575 if (unlikely(blk_bidi_rq(rq)) &&
2576 blk_update_request(rq->next_rq, error, bidi_bytes))
2579 if (blk_queue_add_random(rq->q))
2580 add_disk_randomness(rq->rq_disk);
2586 * blk_unprep_request - unprepare a request
2589 * This function makes a request ready for complete resubmission (or
2590 * completion). It happens only after all error handling is complete,
2591 * so represents the appropriate moment to deallocate any resources
2592 * that were allocated to the request in the prep_rq_fn. The queue
2593 * lock is held when calling this.
2595 void blk_unprep_request(struct request *req)
2597 struct request_queue *q = req->q;
2599 req->cmd_flags &= ~REQ_DONTPREP;
2600 if (q->unprep_rq_fn)
2601 q->unprep_rq_fn(q, req);
2603 EXPORT_SYMBOL_GPL(blk_unprep_request);
2606 * queue lock must be held
2608 void blk_finish_request(struct request *req, int error)
2610 if (req->cmd_flags & REQ_QUEUED)
2611 blk_queue_end_tag(req->q, req);
2613 BUG_ON(blk_queued_rq(req));
2615 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2616 laptop_io_completion(&req->q->backing_dev_info);
2618 blk_delete_timer(req);
2620 if (req->cmd_flags & REQ_DONTPREP)
2621 blk_unprep_request(req);
2623 blk_account_io_done(req);
2626 req->end_io(req, error);
2628 if (blk_bidi_rq(req))
2629 __blk_put_request(req->next_rq->q, req->next_rq);
2631 __blk_put_request(req->q, req);
2634 EXPORT_SYMBOL(blk_finish_request);
2637 * blk_end_bidi_request - Complete a bidi request
2638 * @rq: the request to complete
2639 * @error: %0 for success, < %0 for error
2640 * @nr_bytes: number of bytes to complete @rq
2641 * @bidi_bytes: number of bytes to complete @rq->next_rq
2644 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2645 * Drivers that supports bidi can safely call this member for any
2646 * type of request, bidi or uni. In the later case @bidi_bytes is
2650 * %false - we are done with this request
2651 * %true - still buffers pending for this request
2653 static bool blk_end_bidi_request(struct request *rq, int error,
2654 unsigned int nr_bytes, unsigned int bidi_bytes)
2656 struct request_queue *q = rq->q;
2657 unsigned long flags;
2659 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2662 spin_lock_irqsave(q->queue_lock, flags);
2663 blk_finish_request(rq, error);
2664 spin_unlock_irqrestore(q->queue_lock, flags);
2670 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2671 * @rq: the request to complete
2672 * @error: %0 for success, < %0 for error
2673 * @nr_bytes: number of bytes to complete @rq
2674 * @bidi_bytes: number of bytes to complete @rq->next_rq
2677 * Identical to blk_end_bidi_request() except that queue lock is
2678 * assumed to be locked on entry and remains so on return.
2681 * %false - we are done with this request
2682 * %true - still buffers pending for this request
2684 bool __blk_end_bidi_request(struct request *rq, int error,
2685 unsigned int nr_bytes, unsigned int bidi_bytes)
2687 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2690 blk_finish_request(rq, error);
2696 * blk_end_request - Helper function for drivers to complete the request.
2697 * @rq: the request being processed
2698 * @error: %0 for success, < %0 for error
2699 * @nr_bytes: number of bytes to complete
2702 * Ends I/O on a number of bytes attached to @rq.
2703 * If @rq has leftover, sets it up for the next range of segments.
2706 * %false - we are done with this request
2707 * %true - still buffers pending for this request
2709 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2711 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2713 EXPORT_SYMBOL(blk_end_request);
2716 * blk_end_request_all - Helper function for drives to finish the request.
2717 * @rq: the request to finish
2718 * @error: %0 for success, < %0 for error
2721 * Completely finish @rq.
2723 void blk_end_request_all(struct request *rq, int error)
2726 unsigned int bidi_bytes = 0;
2728 if (unlikely(blk_bidi_rq(rq)))
2729 bidi_bytes = blk_rq_bytes(rq->next_rq);
2731 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2734 EXPORT_SYMBOL(blk_end_request_all);
2737 * blk_end_request_cur - Helper function to finish the current request chunk.
2738 * @rq: the request to finish the current chunk for
2739 * @error: %0 for success, < %0 for error
2742 * Complete the current consecutively mapped chunk from @rq.
2745 * %false - we are done with this request
2746 * %true - still buffers pending for this request
2748 bool blk_end_request_cur(struct request *rq, int error)
2750 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2752 EXPORT_SYMBOL(blk_end_request_cur);
2755 * blk_end_request_err - Finish a request till the next failure boundary.
2756 * @rq: the request to finish till the next failure boundary for
2757 * @error: must be negative errno
2760 * Complete @rq till the next failure boundary.
2763 * %false - we are done with this request
2764 * %true - still buffers pending for this request
2766 bool blk_end_request_err(struct request *rq, int error)
2768 WARN_ON(error >= 0);
2769 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2771 EXPORT_SYMBOL_GPL(blk_end_request_err);
2774 * __blk_end_request - Helper function for drivers to complete the request.
2775 * @rq: the request being processed
2776 * @error: %0 for success, < %0 for error
2777 * @nr_bytes: number of bytes to complete
2780 * Must be called with queue lock held unlike blk_end_request().
2783 * %false - we are done with this request
2784 * %true - still buffers pending for this request
2786 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2788 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2790 EXPORT_SYMBOL(__blk_end_request);
2793 * __blk_end_request_all - Helper function for drives to finish the request.
2794 * @rq: the request to finish
2795 * @error: %0 for success, < %0 for error
2798 * Completely finish @rq. Must be called with queue lock held.
2800 void __blk_end_request_all(struct request *rq, int error)
2803 unsigned int bidi_bytes = 0;
2805 if (unlikely(blk_bidi_rq(rq)))
2806 bidi_bytes = blk_rq_bytes(rq->next_rq);
2808 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2811 EXPORT_SYMBOL(__blk_end_request_all);
2814 * __blk_end_request_cur - Helper function to finish the current request chunk.
2815 * @rq: the request to finish the current chunk for
2816 * @error: %0 for success, < %0 for error
2819 * Complete the current consecutively mapped chunk from @rq. Must
2820 * be called with queue lock held.
2823 * %false - we are done with this request
2824 * %true - still buffers pending for this request
2826 bool __blk_end_request_cur(struct request *rq, int error)
2828 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2830 EXPORT_SYMBOL(__blk_end_request_cur);
2833 * __blk_end_request_err - Finish a request till the next failure boundary.
2834 * @rq: the request to finish till the next failure boundary for
2835 * @error: must be negative errno
2838 * Complete @rq till the next failure boundary. Must be called
2839 * with queue lock held.
2842 * %false - we are done with this request
2843 * %true - still buffers pending for this request
2845 bool __blk_end_request_err(struct request *rq, int error)
2847 WARN_ON(error >= 0);
2848 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2850 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2852 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2855 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2856 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2858 if (bio_has_data(bio))
2859 rq->nr_phys_segments = bio_phys_segments(q, bio);
2861 rq->__data_len = bio->bi_iter.bi_size;
2862 rq->bio = rq->biotail = bio;
2865 rq->rq_disk = bio->bi_bdev->bd_disk;
2868 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2870 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2871 * @rq: the request to be flushed
2874 * Flush all pages in @rq.
2876 void rq_flush_dcache_pages(struct request *rq)
2878 struct req_iterator iter;
2879 struct bio_vec bvec;
2881 rq_for_each_segment(bvec, rq, iter)
2882 flush_dcache_page(bvec.bv_page);
2884 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2888 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2889 * @q : the queue of the device being checked
2892 * Check if underlying low-level drivers of a device are busy.
2893 * If the drivers want to export their busy state, they must set own
2894 * exporting function using blk_queue_lld_busy() first.
2896 * Basically, this function is used only by request stacking drivers
2897 * to stop dispatching requests to underlying devices when underlying
2898 * devices are busy. This behavior helps more I/O merging on the queue
2899 * of the request stacking driver and prevents I/O throughput regression
2900 * on burst I/O load.
2903 * 0 - Not busy (The request stacking driver should dispatch request)
2904 * 1 - Busy (The request stacking driver should stop dispatching request)
2906 int blk_lld_busy(struct request_queue *q)
2909 return q->lld_busy_fn(q);
2913 EXPORT_SYMBOL_GPL(blk_lld_busy);
2916 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2917 * @rq: the clone request to be cleaned up
2920 * Free all bios in @rq for a cloned request.
2922 void blk_rq_unprep_clone(struct request *rq)
2926 while ((bio = rq->bio) != NULL) {
2927 rq->bio = bio->bi_next;
2932 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2935 * Copy attributes of the original request to the clone request.
2936 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2938 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2940 dst->cpu = src->cpu;
2941 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2942 dst->cmd_type = src->cmd_type;
2943 dst->__sector = blk_rq_pos(src);
2944 dst->__data_len = blk_rq_bytes(src);
2945 dst->nr_phys_segments = src->nr_phys_segments;
2946 dst->ioprio = src->ioprio;
2947 dst->extra_len = src->extra_len;
2951 * blk_rq_prep_clone - Helper function to setup clone request
2952 * @rq: the request to be setup
2953 * @rq_src: original request to be cloned
2954 * @bs: bio_set that bios for clone are allocated from
2955 * @gfp_mask: memory allocation mask for bio
2956 * @bio_ctr: setup function to be called for each clone bio.
2957 * Returns %0 for success, non %0 for failure.
2958 * @data: private data to be passed to @bio_ctr
2961 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2962 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2963 * are not copied, and copying such parts is the caller's responsibility.
2964 * Also, pages which the original bios are pointing to are not copied
2965 * and the cloned bios just point same pages.
2966 * So cloned bios must be completed before original bios, which means
2967 * the caller must complete @rq before @rq_src.
2969 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2970 struct bio_set *bs, gfp_t gfp_mask,
2971 int (*bio_ctr)(struct bio *, struct bio *, void *),
2974 struct bio *bio, *bio_src;
2979 __rq_for_each_bio(bio_src, rq_src) {
2980 bio = bio_clone_fast(bio_src, gfp_mask, bs);
2984 if (bio_ctr && bio_ctr(bio, bio_src, data))
2988 rq->biotail->bi_next = bio;
2991 rq->bio = rq->biotail = bio;
2994 __blk_rq_prep_clone(rq, rq_src);
3001 blk_rq_unprep_clone(rq);
3005 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3007 int kblockd_schedule_work(struct work_struct *work)
3009 return queue_work(kblockd_workqueue, work);
3011 EXPORT_SYMBOL(kblockd_schedule_work);
3013 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3014 unsigned long delay)
3016 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3018 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3020 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3021 unsigned long delay)
3023 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3025 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3028 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3029 * @plug: The &struct blk_plug that needs to be initialized
3032 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3033 * pending I/O should the task end up blocking between blk_start_plug() and
3034 * blk_finish_plug(). This is important from a performance perspective, but
3035 * also ensures that we don't deadlock. For instance, if the task is blocking
3036 * for a memory allocation, memory reclaim could end up wanting to free a
3037 * page belonging to that request that is currently residing in our private
3038 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3039 * this kind of deadlock.
3041 void blk_start_plug(struct blk_plug *plug)
3043 struct task_struct *tsk = current;
3046 * If this is a nested plug, don't actually assign it.
3051 INIT_LIST_HEAD(&plug->list);
3052 INIT_LIST_HEAD(&plug->mq_list);
3053 INIT_LIST_HEAD(&plug->cb_list);
3055 * Store ordering should not be needed here, since a potential
3056 * preempt will imply a full memory barrier
3060 EXPORT_SYMBOL(blk_start_plug);
3062 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3064 struct request *rqa = container_of(a, struct request, queuelist);
3065 struct request *rqb = container_of(b, struct request, queuelist);
3067 return !(rqa->q < rqb->q ||
3068 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3072 * If 'from_schedule' is true, then postpone the dispatch of requests
3073 * until a safe kblockd context. We due this to avoid accidental big
3074 * additional stack usage in driver dispatch, in places where the originally
3075 * plugger did not intend it.
3077 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3079 __releases(q->queue_lock)
3081 trace_block_unplug(q, depth, !from_schedule);
3084 blk_run_queue_async(q);
3087 spin_unlock(q->queue_lock);
3090 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3092 LIST_HEAD(callbacks);
3094 while (!list_empty(&plug->cb_list)) {
3095 list_splice_init(&plug->cb_list, &callbacks);
3097 while (!list_empty(&callbacks)) {
3098 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3101 list_del(&cb->list);
3102 cb->callback(cb, from_schedule);
3107 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3110 struct blk_plug *plug = current->plug;
3111 struct blk_plug_cb *cb;
3116 list_for_each_entry(cb, &plug->cb_list, list)
3117 if (cb->callback == unplug && cb->data == data)
3120 /* Not currently on the callback list */
3121 BUG_ON(size < sizeof(*cb));
3122 cb = kzalloc(size, GFP_ATOMIC);
3125 cb->callback = unplug;
3126 list_add(&cb->list, &plug->cb_list);
3130 EXPORT_SYMBOL(blk_check_plugged);
3132 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3134 struct request_queue *q;
3135 unsigned long flags;
3140 flush_plug_callbacks(plug, from_schedule);
3142 if (!list_empty(&plug->mq_list))
3143 blk_mq_flush_plug_list(plug, from_schedule);
3145 if (list_empty(&plug->list))
3148 list_splice_init(&plug->list, &list);
3150 list_sort(NULL, &list, plug_rq_cmp);
3156 * Save and disable interrupts here, to avoid doing it for every
3157 * queue lock we have to take.
3159 local_irq_save(flags);
3160 while (!list_empty(&list)) {
3161 rq = list_entry_rq(list.next);
3162 list_del_init(&rq->queuelist);
3166 * This drops the queue lock
3169 queue_unplugged(q, depth, from_schedule);
3172 spin_lock(q->queue_lock);
3176 * Short-circuit if @q is dead
3178 if (unlikely(blk_queue_dying(q))) {
3179 __blk_end_request_all(rq, -ENODEV);
3184 * rq is already accounted, so use raw insert
3186 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3187 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3189 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3195 * This drops the queue lock
3198 queue_unplugged(q, depth, from_schedule);
3200 local_irq_restore(flags);
3203 void blk_finish_plug(struct blk_plug *plug)
3205 if (plug != current->plug)
3207 blk_flush_plug_list(plug, false);
3209 current->plug = NULL;
3211 EXPORT_SYMBOL(blk_finish_plug);
3215 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3216 * @q: the queue of the device
3217 * @dev: the device the queue belongs to
3220 * Initialize runtime-PM-related fields for @q and start auto suspend for
3221 * @dev. Drivers that want to take advantage of request-based runtime PM
3222 * should call this function after @dev has been initialized, and its
3223 * request queue @q has been allocated, and runtime PM for it can not happen
3224 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3225 * cases, driver should call this function before any I/O has taken place.
3227 * This function takes care of setting up using auto suspend for the device,
3228 * the autosuspend delay is set to -1 to make runtime suspend impossible
3229 * until an updated value is either set by user or by driver. Drivers do
3230 * not need to touch other autosuspend settings.
3232 * The block layer runtime PM is request based, so only works for drivers
3233 * that use request as their IO unit instead of those directly use bio's.
3235 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3238 q->rpm_status = RPM_ACTIVE;
3239 pm_runtime_set_autosuspend_delay(q->dev, -1);
3240 pm_runtime_use_autosuspend(q->dev);
3242 EXPORT_SYMBOL(blk_pm_runtime_init);
3245 * blk_pre_runtime_suspend - Pre runtime suspend check
3246 * @q: the queue of the device
3249 * This function will check if runtime suspend is allowed for the device
3250 * by examining if there are any requests pending in the queue. If there
3251 * are requests pending, the device can not be runtime suspended; otherwise,
3252 * the queue's status will be updated to SUSPENDING and the driver can
3253 * proceed to suspend the device.
3255 * For the not allowed case, we mark last busy for the device so that
3256 * runtime PM core will try to autosuspend it some time later.
3258 * This function should be called near the start of the device's
3259 * runtime_suspend callback.
3262 * 0 - OK to runtime suspend the device
3263 * -EBUSY - Device should not be runtime suspended
3265 int blk_pre_runtime_suspend(struct request_queue *q)
3269 spin_lock_irq(q->queue_lock);
3270 if (q->nr_pending) {
3272 pm_runtime_mark_last_busy(q->dev);
3274 q->rpm_status = RPM_SUSPENDING;
3276 spin_unlock_irq(q->queue_lock);
3279 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3282 * blk_post_runtime_suspend - Post runtime suspend processing
3283 * @q: the queue of the device
3284 * @err: return value of the device's runtime_suspend function
3287 * Update the queue's runtime status according to the return value of the
3288 * device's runtime suspend function and mark last busy for the device so
3289 * that PM core will try to auto suspend the device at a later time.
3291 * This function should be called near the end of the device's
3292 * runtime_suspend callback.
3294 void blk_post_runtime_suspend(struct request_queue *q, int err)
3296 spin_lock_irq(q->queue_lock);
3298 q->rpm_status = RPM_SUSPENDED;
3300 q->rpm_status = RPM_ACTIVE;
3301 pm_runtime_mark_last_busy(q->dev);
3303 spin_unlock_irq(q->queue_lock);
3305 EXPORT_SYMBOL(blk_post_runtime_suspend);
3308 * blk_pre_runtime_resume - Pre runtime resume processing
3309 * @q: the queue of the device
3312 * Update the queue's runtime status to RESUMING in preparation for the
3313 * runtime resume of the device.
3315 * This function should be called near the start of the device's
3316 * runtime_resume callback.
3318 void blk_pre_runtime_resume(struct request_queue *q)
3320 spin_lock_irq(q->queue_lock);
3321 q->rpm_status = RPM_RESUMING;
3322 spin_unlock_irq(q->queue_lock);
3324 EXPORT_SYMBOL(blk_pre_runtime_resume);
3327 * blk_post_runtime_resume - Post runtime resume processing
3328 * @q: the queue of the device
3329 * @err: return value of the device's runtime_resume function
3332 * Update the queue's runtime status according to the return value of the
3333 * device's runtime_resume function. If it is successfully resumed, process
3334 * the requests that are queued into the device's queue when it is resuming
3335 * and then mark last busy and initiate autosuspend for it.
3337 * This function should be called near the end of the device's
3338 * runtime_resume callback.
3340 void blk_post_runtime_resume(struct request_queue *q, int err)
3342 spin_lock_irq(q->queue_lock);
3344 q->rpm_status = RPM_ACTIVE;
3346 pm_runtime_mark_last_busy(q->dev);
3347 pm_request_autosuspend(q->dev);
3349 q->rpm_status = RPM_SUSPENDED;
3351 spin_unlock_irq(q->queue_lock);
3353 EXPORT_SYMBOL(blk_post_runtime_resume);
3356 int __init blk_dev_init(void)
3358 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3359 sizeof(((struct request *)0)->cmd_flags));
3361 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3362 kblockd_workqueue = alloc_workqueue("kblockd",
3363 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3364 if (!kblockd_workqueue)
3365 panic("Failed to create kblockd\n");
3367 request_cachep = kmem_cache_create("blkdev_requests",
3368 sizeof(struct request), 0, SLAB_PANIC, NULL);
3370 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3371 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);