2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/block.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 static void blk_clear_congested(struct request_list *rl, int sync)
68 #ifdef CONFIG_CGROUP_WRITEBACK
69 clear_wb_congested(rl->blkg->wb_congested, sync);
72 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
73 * flip its congestion state for events on other blkcgs.
75 if (rl == &rl->q->root_rl)
76 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
80 static void blk_set_congested(struct request_list *rl, int sync)
82 #ifdef CONFIG_CGROUP_WRITEBACK
83 set_wb_congested(rl->blkg->wb_congested, sync);
85 /* see blk_clear_congested() */
86 if (rl == &rl->q->root_rl)
87 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
91 void blk_queue_congestion_threshold(struct request_queue *q)
95 nr = q->nr_requests - (q->nr_requests / 8) + 1;
96 if (nr > q->nr_requests)
98 q->nr_congestion_on = nr;
100 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103 q->nr_congestion_off = nr;
107 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * Locates the passed device's request queue and returns the address of its
111 * backing_dev_info. This function can only be called if @bdev is opened
112 * and the return value is never NULL.
114 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
116 struct request_queue *q = bdev_get_queue(bdev);
118 return &q->backing_dev_info;
120 EXPORT_SYMBOL(blk_get_backing_dev_info);
122 void blk_rq_init(struct request_queue *q, struct request *rq)
124 memset(rq, 0, sizeof(*rq));
126 INIT_LIST_HEAD(&rq->queuelist);
127 INIT_LIST_HEAD(&rq->timeout_list);
130 rq->__sector = (sector_t) -1;
131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
134 rq->cmd_len = BLK_MAX_CDB;
136 rq->start_time = jiffies;
137 set_start_time_ns(rq);
140 EXPORT_SYMBOL(blk_rq_init);
142 static void req_bio_endio(struct request *rq, struct bio *bio,
143 unsigned int nbytes, int error)
146 bio->bi_error = error;
148 if (unlikely(rq->cmd_flags & REQ_QUIET))
149 bio_set_flag(bio, BIO_QUIET);
151 bio_advance(bio, nbytes);
153 /* don't actually finish bio if it's part of flush sequence */
154 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
158 void blk_dump_rq_flags(struct request *rq, char *msg)
162 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
163 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
164 (unsigned long long) rq->cmd_flags);
166 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
167 (unsigned long long)blk_rq_pos(rq),
168 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
169 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
170 rq->bio, rq->biotail, blk_rq_bytes(rq));
172 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
173 printk(KERN_INFO " cdb: ");
174 for (bit = 0; bit < BLK_MAX_CDB; bit++)
175 printk("%02x ", rq->cmd[bit]);
179 EXPORT_SYMBOL(blk_dump_rq_flags);
181 static void blk_delay_work(struct work_struct *work)
183 struct request_queue *q;
185 q = container_of(work, struct request_queue, delay_work.work);
186 spin_lock_irq(q->queue_lock);
188 spin_unlock_irq(q->queue_lock);
192 * blk_delay_queue - restart queueing after defined interval
193 * @q: The &struct request_queue in question
194 * @msecs: Delay in msecs
197 * Sometimes queueing needs to be postponed for a little while, to allow
198 * resources to come back. This function will make sure that queueing is
199 * restarted around the specified time. Queue lock must be held.
201 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
203 if (likely(!blk_queue_dead(q)))
204 queue_delayed_work(kblockd_workqueue, &q->delay_work,
205 msecs_to_jiffies(msecs));
207 EXPORT_SYMBOL(blk_delay_queue);
210 * blk_start_queue - restart a previously stopped queue
211 * @q: The &struct request_queue in question
214 * blk_start_queue() will clear the stop flag on the queue, and call
215 * the request_fn for the queue if it was in a stopped state when
216 * entered. Also see blk_stop_queue(). Queue lock must be held.
218 void blk_start_queue(struct request_queue *q)
220 WARN_ON(!irqs_disabled());
222 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
225 EXPORT_SYMBOL(blk_start_queue);
228 * blk_stop_queue - stop a queue
229 * @q: The &struct request_queue in question
232 * The Linux block layer assumes that a block driver will consume all
233 * entries on the request queue when the request_fn strategy is called.
234 * Often this will not happen, because of hardware limitations (queue
235 * depth settings). If a device driver gets a 'queue full' response,
236 * or if it simply chooses not to queue more I/O at one point, it can
237 * call this function to prevent the request_fn from being called until
238 * the driver has signalled it's ready to go again. This happens by calling
239 * blk_start_queue() to restart queue operations. Queue lock must be held.
241 void blk_stop_queue(struct request_queue *q)
243 cancel_delayed_work(&q->delay_work);
244 queue_flag_set(QUEUE_FLAG_STOPPED, q);
246 EXPORT_SYMBOL(blk_stop_queue);
249 * blk_sync_queue - cancel any pending callbacks on a queue
253 * The block layer may perform asynchronous callback activity
254 * on a queue, such as calling the unplug function after a timeout.
255 * A block device may call blk_sync_queue to ensure that any
256 * such activity is cancelled, thus allowing it to release resources
257 * that the callbacks might use. The caller must already have made sure
258 * that its ->make_request_fn will not re-add plugging prior to calling
261 * This function does not cancel any asynchronous activity arising
262 * out of elevator or throttling code. That would require elevator_exit()
263 * and blkcg_exit_queue() to be called with queue lock initialized.
266 void blk_sync_queue(struct request_queue *q)
268 del_timer_sync(&q->timeout);
271 struct blk_mq_hw_ctx *hctx;
274 queue_for_each_hw_ctx(q, hctx, i) {
275 cancel_delayed_work_sync(&hctx->run_work);
276 cancel_delayed_work_sync(&hctx->delay_work);
279 cancel_delayed_work_sync(&q->delay_work);
282 EXPORT_SYMBOL(blk_sync_queue);
285 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
286 * @q: The queue to run
289 * Invoke request handling on a queue if there are any pending requests.
290 * May be used to restart request handling after a request has completed.
291 * This variant runs the queue whether or not the queue has been
292 * stopped. Must be called with the queue lock held and interrupts
293 * disabled. See also @blk_run_queue.
295 inline void __blk_run_queue_uncond(struct request_queue *q)
297 if (unlikely(blk_queue_dead(q)))
301 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
302 * the queue lock internally. As a result multiple threads may be
303 * running such a request function concurrently. Keep track of the
304 * number of active request_fn invocations such that blk_drain_queue()
305 * can wait until all these request_fn calls have finished.
307 q->request_fn_active++;
309 q->request_fn_active--;
311 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
314 * __blk_run_queue - run a single device queue
315 * @q: The queue to run
318 * See @blk_run_queue. This variant must be called with the queue lock
319 * held and interrupts disabled.
321 void __blk_run_queue(struct request_queue *q)
323 if (unlikely(blk_queue_stopped(q)))
326 __blk_run_queue_uncond(q);
328 EXPORT_SYMBOL(__blk_run_queue);
331 * blk_run_queue_async - run a single device queue in workqueue context
332 * @q: The queue to run
335 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
336 * of us. The caller must hold the queue lock.
338 void blk_run_queue_async(struct request_queue *q)
340 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
341 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
343 EXPORT_SYMBOL(blk_run_queue_async);
346 * blk_run_queue - run a single device queue
347 * @q: The queue to run
350 * Invoke request handling on this queue, if it has pending work to do.
351 * May be used to restart queueing when a request has completed.
353 void blk_run_queue(struct request_queue *q)
357 spin_lock_irqsave(q->queue_lock, flags);
359 spin_unlock_irqrestore(q->queue_lock, flags);
361 EXPORT_SYMBOL(blk_run_queue);
363 void blk_put_queue(struct request_queue *q)
365 kobject_put(&q->kobj);
367 EXPORT_SYMBOL(blk_put_queue);
370 * __blk_drain_queue - drain requests from request_queue
372 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
374 * Drain requests from @q. If @drain_all is set, all requests are drained.
375 * If not, only ELVPRIV requests are drained. The caller is responsible
376 * for ensuring that no new requests which need to be drained are queued.
378 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
379 __releases(q->queue_lock)
380 __acquires(q->queue_lock)
384 lockdep_assert_held(q->queue_lock);
390 * The caller might be trying to drain @q before its
391 * elevator is initialized.
394 elv_drain_elevator(q);
396 blkcg_drain_queue(q);
399 * This function might be called on a queue which failed
400 * driver init after queue creation or is not yet fully
401 * active yet. Some drivers (e.g. fd and loop) get unhappy
402 * in such cases. Kick queue iff dispatch queue has
403 * something on it and @q has request_fn set.
405 if (!list_empty(&q->queue_head) && q->request_fn)
408 drain |= q->nr_rqs_elvpriv;
409 drain |= q->request_fn_active;
412 * Unfortunately, requests are queued at and tracked from
413 * multiple places and there's no single counter which can
414 * be drained. Check all the queues and counters.
417 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
418 drain |= !list_empty(&q->queue_head);
419 for (i = 0; i < 2; i++) {
420 drain |= q->nr_rqs[i];
421 drain |= q->in_flight[i];
423 drain |= !list_empty(&fq->flush_queue[i]);
430 spin_unlock_irq(q->queue_lock);
434 spin_lock_irq(q->queue_lock);
438 * With queue marked dead, any woken up waiter will fail the
439 * allocation path, so the wakeup chaining is lost and we're
440 * left with hung waiters. We need to wake up those waiters.
443 struct request_list *rl;
445 blk_queue_for_each_rl(rl, q)
446 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
447 wake_up_all(&rl->wait[i]);
452 * blk_queue_bypass_start - enter queue bypass mode
453 * @q: queue of interest
455 * In bypass mode, only the dispatch FIFO queue of @q is used. This
456 * function makes @q enter bypass mode and drains all requests which were
457 * throttled or issued before. On return, it's guaranteed that no request
458 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
459 * inside queue or RCU read lock.
461 void blk_queue_bypass_start(struct request_queue *q)
463 spin_lock_irq(q->queue_lock);
465 queue_flag_set(QUEUE_FLAG_BYPASS, q);
466 spin_unlock_irq(q->queue_lock);
469 * Queues start drained. Skip actual draining till init is
470 * complete. This avoids lenghty delays during queue init which
471 * can happen many times during boot.
473 if (blk_queue_init_done(q)) {
474 spin_lock_irq(q->queue_lock);
475 __blk_drain_queue(q, false);
476 spin_unlock_irq(q->queue_lock);
478 /* ensure blk_queue_bypass() is %true inside RCU read lock */
482 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
485 * blk_queue_bypass_end - leave queue bypass mode
486 * @q: queue of interest
488 * Leave bypass mode and restore the normal queueing behavior.
490 void blk_queue_bypass_end(struct request_queue *q)
492 spin_lock_irq(q->queue_lock);
493 if (!--q->bypass_depth)
494 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
495 WARN_ON_ONCE(q->bypass_depth < 0);
496 spin_unlock_irq(q->queue_lock);
498 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
500 void blk_set_queue_dying(struct request_queue *q)
502 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
505 blk_mq_wake_waiters(q);
507 struct request_list *rl;
509 blk_queue_for_each_rl(rl, q) {
511 wake_up(&rl->wait[BLK_RW_SYNC]);
512 wake_up(&rl->wait[BLK_RW_ASYNC]);
517 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
520 * blk_cleanup_queue - shutdown a request queue
521 * @q: request queue to shutdown
523 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
524 * put it. All future requests will be failed immediately with -ENODEV.
526 void blk_cleanup_queue(struct request_queue *q)
528 spinlock_t *lock = q->queue_lock;
530 /* mark @q DYING, no new request or merges will be allowed afterwards */
531 mutex_lock(&q->sysfs_lock);
532 blk_set_queue_dying(q);
536 * A dying queue is permanently in bypass mode till released. Note
537 * that, unlike blk_queue_bypass_start(), we aren't performing
538 * synchronize_rcu() after entering bypass mode to avoid the delay
539 * as some drivers create and destroy a lot of queues while
540 * probing. This is still safe because blk_release_queue() will be
541 * called only after the queue refcnt drops to zero and nothing,
542 * RCU or not, would be traversing the queue by then.
545 queue_flag_set(QUEUE_FLAG_BYPASS, q);
547 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
548 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
549 queue_flag_set(QUEUE_FLAG_DYING, q);
550 spin_unlock_irq(lock);
551 mutex_unlock(&q->sysfs_lock);
554 * Drain all requests queued before DYING marking. Set DEAD flag to
555 * prevent that q->request_fn() gets invoked after draining finished.
558 blk_mq_freeze_queue(q);
562 __blk_drain_queue(q, true);
564 queue_flag_set(QUEUE_FLAG_DEAD, q);
565 spin_unlock_irq(lock);
567 /* @q won't process any more request, flush async actions */
568 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
572 blk_mq_free_queue(q);
575 if (q->queue_lock != &q->__queue_lock)
576 q->queue_lock = &q->__queue_lock;
577 spin_unlock_irq(lock);
579 bdi_destroy(&q->backing_dev_info);
581 /* @q is and will stay empty, shutdown and put */
584 EXPORT_SYMBOL(blk_cleanup_queue);
586 /* Allocate memory local to the request queue */
587 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
589 int nid = (int)(long)data;
590 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
593 static void free_request_struct(void *element, void *unused)
595 kmem_cache_free(request_cachep, element);
598 int blk_init_rl(struct request_list *rl, struct request_queue *q,
601 if (unlikely(rl->rq_pool))
605 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
606 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
607 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
608 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
610 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
612 (void *)(long)q->node, gfp_mask,
620 void blk_exit_rl(struct request_list *rl)
623 mempool_destroy(rl->rq_pool);
626 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
628 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
630 EXPORT_SYMBOL(blk_alloc_queue);
632 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
634 struct request_queue *q;
637 q = kmem_cache_alloc_node(blk_requestq_cachep,
638 gfp_mask | __GFP_ZERO, node_id);
642 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
646 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
650 q->backing_dev_info.ra_pages =
651 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
652 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
653 q->backing_dev_info.name = "block";
656 err = bdi_init(&q->backing_dev_info);
660 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
661 laptop_mode_timer_fn, (unsigned long) q);
662 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
663 INIT_LIST_HEAD(&q->queue_head);
664 INIT_LIST_HEAD(&q->timeout_list);
665 INIT_LIST_HEAD(&q->icq_list);
666 #ifdef CONFIG_BLK_CGROUP
667 INIT_LIST_HEAD(&q->blkg_list);
669 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
671 kobject_init(&q->kobj, &blk_queue_ktype);
673 mutex_init(&q->sysfs_lock);
674 spin_lock_init(&q->__queue_lock);
677 * By default initialize queue_lock to internal lock and driver can
678 * override it later if need be.
680 q->queue_lock = &q->__queue_lock;
683 * A queue starts its life with bypass turned on to avoid
684 * unnecessary bypass on/off overhead and nasty surprises during
685 * init. The initial bypass will be finished when the queue is
686 * registered by blk_register_queue().
689 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
691 init_waitqueue_head(&q->mq_freeze_wq);
693 if (blkcg_init_queue(q))
699 bdi_destroy(&q->backing_dev_info);
701 bioset_free(q->bio_split);
703 ida_simple_remove(&blk_queue_ida, q->id);
705 kmem_cache_free(blk_requestq_cachep, q);
708 EXPORT_SYMBOL(blk_alloc_queue_node);
711 * blk_init_queue - prepare a request queue for use with a block device
712 * @rfn: The function to be called to process requests that have been
713 * placed on the queue.
714 * @lock: Request queue spin lock
717 * If a block device wishes to use the standard request handling procedures,
718 * which sorts requests and coalesces adjacent requests, then it must
719 * call blk_init_queue(). The function @rfn will be called when there
720 * are requests on the queue that need to be processed. If the device
721 * supports plugging, then @rfn may not be called immediately when requests
722 * are available on the queue, but may be called at some time later instead.
723 * Plugged queues are generally unplugged when a buffer belonging to one
724 * of the requests on the queue is needed, or due to memory pressure.
726 * @rfn is not required, or even expected, to remove all requests off the
727 * queue, but only as many as it can handle at a time. If it does leave
728 * requests on the queue, it is responsible for arranging that the requests
729 * get dealt with eventually.
731 * The queue spin lock must be held while manipulating the requests on the
732 * request queue; this lock will be taken also from interrupt context, so irq
733 * disabling is needed for it.
735 * Function returns a pointer to the initialized request queue, or %NULL if
739 * blk_init_queue() must be paired with a blk_cleanup_queue() call
740 * when the block device is deactivated (such as at module unload).
743 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
745 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
747 EXPORT_SYMBOL(blk_init_queue);
749 struct request_queue *
750 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
752 struct request_queue *uninit_q, *q;
754 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
758 q = blk_init_allocated_queue(uninit_q, rfn, lock);
760 blk_cleanup_queue(uninit_q);
764 EXPORT_SYMBOL(blk_init_queue_node);
766 static void blk_queue_bio(struct request_queue *q, struct bio *bio);
768 struct request_queue *
769 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
775 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
779 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
783 q->prep_rq_fn = NULL;
784 q->unprep_rq_fn = NULL;
785 q->queue_flags |= QUEUE_FLAG_DEFAULT;
787 /* Override internal queue lock with supplied lock pointer */
789 q->queue_lock = lock;
792 * This also sets hw/phys segments, boundary and size
794 blk_queue_make_request(q, blk_queue_bio);
796 q->sg_reserved_size = INT_MAX;
798 /* Protect q->elevator from elevator_change */
799 mutex_lock(&q->sysfs_lock);
802 if (elevator_init(q, NULL)) {
803 mutex_unlock(&q->sysfs_lock);
807 mutex_unlock(&q->sysfs_lock);
812 blk_free_flush_queue(q->fq);
815 EXPORT_SYMBOL(blk_init_allocated_queue);
817 bool blk_get_queue(struct request_queue *q)
819 if (likely(!blk_queue_dying(q))) {
826 EXPORT_SYMBOL(blk_get_queue);
828 static inline void blk_free_request(struct request_list *rl, struct request *rq)
830 if (rq->cmd_flags & REQ_ELVPRIV) {
831 elv_put_request(rl->q, rq);
833 put_io_context(rq->elv.icq->ioc);
836 mempool_free(rq, rl->rq_pool);
840 * ioc_batching returns true if the ioc is a valid batching request and
841 * should be given priority access to a request.
843 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
849 * Make sure the process is able to allocate at least 1 request
850 * even if the batch times out, otherwise we could theoretically
853 return ioc->nr_batch_requests == q->nr_batching ||
854 (ioc->nr_batch_requests > 0
855 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
859 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
860 * will cause the process to be a "batcher" on all queues in the system. This
861 * is the behaviour we want though - once it gets a wakeup it should be given
864 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
866 if (!ioc || ioc_batching(q, ioc))
869 ioc->nr_batch_requests = q->nr_batching;
870 ioc->last_waited = jiffies;
873 static void __freed_request(struct request_list *rl, int sync)
875 struct request_queue *q = rl->q;
877 if (rl->count[sync] < queue_congestion_off_threshold(q))
878 blk_clear_congested(rl, sync);
880 if (rl->count[sync] + 1 <= q->nr_requests) {
881 if (waitqueue_active(&rl->wait[sync]))
882 wake_up(&rl->wait[sync]);
884 blk_clear_rl_full(rl, sync);
889 * A request has just been released. Account for it, update the full and
890 * congestion status, wake up any waiters. Called under q->queue_lock.
892 static void freed_request(struct request_list *rl, unsigned int flags)
894 struct request_queue *q = rl->q;
895 int sync = rw_is_sync(flags);
899 if (flags & REQ_ELVPRIV)
902 __freed_request(rl, sync);
904 if (unlikely(rl->starved[sync ^ 1]))
905 __freed_request(rl, sync ^ 1);
908 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
910 struct request_list *rl;
911 int on_thresh, off_thresh;
913 spin_lock_irq(q->queue_lock);
915 blk_queue_congestion_threshold(q);
916 on_thresh = queue_congestion_on_threshold(q);
917 off_thresh = queue_congestion_off_threshold(q);
919 blk_queue_for_each_rl(rl, q) {
920 if (rl->count[BLK_RW_SYNC] >= on_thresh)
921 blk_set_congested(rl, BLK_RW_SYNC);
922 else if (rl->count[BLK_RW_SYNC] < off_thresh)
923 blk_clear_congested(rl, BLK_RW_SYNC);
925 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
926 blk_set_congested(rl, BLK_RW_ASYNC);
927 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
928 blk_clear_congested(rl, BLK_RW_ASYNC);
930 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
931 blk_set_rl_full(rl, BLK_RW_SYNC);
933 blk_clear_rl_full(rl, BLK_RW_SYNC);
934 wake_up(&rl->wait[BLK_RW_SYNC]);
937 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
938 blk_set_rl_full(rl, BLK_RW_ASYNC);
940 blk_clear_rl_full(rl, BLK_RW_ASYNC);
941 wake_up(&rl->wait[BLK_RW_ASYNC]);
945 spin_unlock_irq(q->queue_lock);
950 * Determine if elevator data should be initialized when allocating the
951 * request associated with @bio.
953 static bool blk_rq_should_init_elevator(struct bio *bio)
959 * Flush requests do not use the elevator so skip initialization.
960 * This allows a request to share the flush and elevator data.
962 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
969 * rq_ioc - determine io_context for request allocation
970 * @bio: request being allocated is for this bio (can be %NULL)
972 * Determine io_context to use for request allocation for @bio. May return
973 * %NULL if %current->io_context doesn't exist.
975 static struct io_context *rq_ioc(struct bio *bio)
977 #ifdef CONFIG_BLK_CGROUP
978 if (bio && bio->bi_ioc)
981 return current->io_context;
985 * __get_request - get a free request
986 * @rl: request list to allocate from
987 * @rw_flags: RW and SYNC flags
988 * @bio: bio to allocate request for (can be %NULL)
989 * @gfp_mask: allocation mask
991 * Get a free request from @q. This function may fail under memory
992 * pressure or if @q is dead.
994 * Must be called with @q->queue_lock held and,
995 * Returns ERR_PTR on failure, with @q->queue_lock held.
996 * Returns request pointer on success, with @q->queue_lock *not held*.
998 static struct request *__get_request(struct request_list *rl, int rw_flags,
999 struct bio *bio, gfp_t gfp_mask)
1001 struct request_queue *q = rl->q;
1003 struct elevator_type *et = q->elevator->type;
1004 struct io_context *ioc = rq_ioc(bio);
1005 struct io_cq *icq = NULL;
1006 const bool is_sync = rw_is_sync(rw_flags) != 0;
1009 if (unlikely(blk_queue_dying(q)))
1010 return ERR_PTR(-ENODEV);
1012 may_queue = elv_may_queue(q, rw_flags);
1013 if (may_queue == ELV_MQUEUE_NO)
1016 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1017 if (rl->count[is_sync]+1 >= q->nr_requests) {
1019 * The queue will fill after this allocation, so set
1020 * it as full, and mark this process as "batching".
1021 * This process will be allowed to complete a batch of
1022 * requests, others will be blocked.
1024 if (!blk_rl_full(rl, is_sync)) {
1025 ioc_set_batching(q, ioc);
1026 blk_set_rl_full(rl, is_sync);
1028 if (may_queue != ELV_MQUEUE_MUST
1029 && !ioc_batching(q, ioc)) {
1031 * The queue is full and the allocating
1032 * process is not a "batcher", and not
1033 * exempted by the IO scheduler
1035 return ERR_PTR(-ENOMEM);
1039 blk_set_congested(rl, is_sync);
1043 * Only allow batching queuers to allocate up to 50% over the defined
1044 * limit of requests, otherwise we could have thousands of requests
1045 * allocated with any setting of ->nr_requests
1047 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1048 return ERR_PTR(-ENOMEM);
1050 q->nr_rqs[is_sync]++;
1051 rl->count[is_sync]++;
1052 rl->starved[is_sync] = 0;
1055 * Decide whether the new request will be managed by elevator. If
1056 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1057 * prevent the current elevator from being destroyed until the new
1058 * request is freed. This guarantees icq's won't be destroyed and
1059 * makes creating new ones safe.
1061 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1062 * it will be created after releasing queue_lock.
1064 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1065 rw_flags |= REQ_ELVPRIV;
1066 q->nr_rqs_elvpriv++;
1067 if (et->icq_cache && ioc)
1068 icq = ioc_lookup_icq(ioc, q);
1071 if (blk_queue_io_stat(q))
1072 rw_flags |= REQ_IO_STAT;
1073 spin_unlock_irq(q->queue_lock);
1075 /* allocate and init request */
1076 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1081 blk_rq_set_rl(rq, rl);
1082 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1085 if (rw_flags & REQ_ELVPRIV) {
1086 if (unlikely(et->icq_cache && !icq)) {
1088 icq = ioc_create_icq(ioc, q, gfp_mask);
1094 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1097 /* @rq->elv.icq holds io_context until @rq is freed */
1099 get_io_context(icq->ioc);
1103 * ioc may be NULL here, and ioc_batching will be false. That's
1104 * OK, if the queue is under the request limit then requests need
1105 * not count toward the nr_batch_requests limit. There will always
1106 * be some limit enforced by BLK_BATCH_TIME.
1108 if (ioc_batching(q, ioc))
1109 ioc->nr_batch_requests--;
1111 trace_block_getrq(q, bio, rw_flags & 1);
1116 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1117 * and may fail indefinitely under memory pressure and thus
1118 * shouldn't stall IO. Treat this request as !elvpriv. This will
1119 * disturb iosched and blkcg but weird is bettern than dead.
1121 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1122 __func__, dev_name(q->backing_dev_info.dev));
1124 rq->cmd_flags &= ~REQ_ELVPRIV;
1127 spin_lock_irq(q->queue_lock);
1128 q->nr_rqs_elvpriv--;
1129 spin_unlock_irq(q->queue_lock);
1134 * Allocation failed presumably due to memory. Undo anything we
1135 * might have messed up.
1137 * Allocating task should really be put onto the front of the wait
1138 * queue, but this is pretty rare.
1140 spin_lock_irq(q->queue_lock);
1141 freed_request(rl, rw_flags);
1144 * in the very unlikely event that allocation failed and no
1145 * requests for this direction was pending, mark us starved so that
1146 * freeing of a request in the other direction will notice
1147 * us. another possible fix would be to split the rq mempool into
1151 if (unlikely(rl->count[is_sync] == 0))
1152 rl->starved[is_sync] = 1;
1153 return ERR_PTR(-ENOMEM);
1157 * get_request - get a free request
1158 * @q: request_queue to allocate request from
1159 * @rw_flags: RW and SYNC flags
1160 * @bio: bio to allocate request for (can be %NULL)
1161 * @gfp_mask: allocation mask
1163 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1164 * function keeps retrying under memory pressure and fails iff @q is dead.
1166 * Must be called with @q->queue_lock held and,
1167 * Returns ERR_PTR on failure, with @q->queue_lock held.
1168 * Returns request pointer on success, with @q->queue_lock *not held*.
1170 static struct request *get_request(struct request_queue *q, int rw_flags,
1171 struct bio *bio, gfp_t gfp_mask)
1173 const bool is_sync = rw_is_sync(rw_flags) != 0;
1175 struct request_list *rl;
1178 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1180 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1184 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1189 /* wait on @rl and retry */
1190 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1191 TASK_UNINTERRUPTIBLE);
1193 trace_block_sleeprq(q, bio, rw_flags & 1);
1195 spin_unlock_irq(q->queue_lock);
1199 * After sleeping, we become a "batching" process and will be able
1200 * to allocate at least one request, and up to a big batch of them
1201 * for a small period time. See ioc_batching, ioc_set_batching
1203 ioc_set_batching(q, current->io_context);
1205 spin_lock_irq(q->queue_lock);
1206 finish_wait(&rl->wait[is_sync], &wait);
1211 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1216 BUG_ON(rw != READ && rw != WRITE);
1218 /* create ioc upfront */
1219 create_io_context(gfp_mask, q->node);
1221 spin_lock_irq(q->queue_lock);
1222 rq = get_request(q, rw, NULL, gfp_mask);
1224 spin_unlock_irq(q->queue_lock);
1225 /* q->queue_lock is unlocked at this point */
1230 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1233 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1235 return blk_old_get_request(q, rw, gfp_mask);
1237 EXPORT_SYMBOL(blk_get_request);
1240 * blk_make_request - given a bio, allocate a corresponding struct request.
1241 * @q: target request queue
1242 * @bio: The bio describing the memory mappings that will be submitted for IO.
1243 * It may be a chained-bio properly constructed by block/bio layer.
1244 * @gfp_mask: gfp flags to be used for memory allocation
1246 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1247 * type commands. Where the struct request needs to be farther initialized by
1248 * the caller. It is passed a &struct bio, which describes the memory info of
1251 * The caller of blk_make_request must make sure that bi_io_vec
1252 * are set to describe the memory buffers. That bio_data_dir() will return
1253 * the needed direction of the request. (And all bio's in the passed bio-chain
1254 * are properly set accordingly)
1256 * If called under none-sleepable conditions, mapped bio buffers must not
1257 * need bouncing, by calling the appropriate masked or flagged allocator,
1258 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1261 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1262 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1263 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1264 * completion of a bio that hasn't been submitted yet, thus resulting in a
1265 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1266 * of bio_alloc(), as that avoids the mempool deadlock.
1267 * If possible a big IO should be split into smaller parts when allocation
1268 * fails. Partial allocation should not be an error, or you risk a live-lock.
1270 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1273 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1278 blk_rq_set_block_pc(rq);
1281 struct bio *bounce_bio = bio;
1284 blk_queue_bounce(q, &bounce_bio);
1285 ret = blk_rq_append_bio(q, rq, bounce_bio);
1286 if (unlikely(ret)) {
1287 blk_put_request(rq);
1288 return ERR_PTR(ret);
1294 EXPORT_SYMBOL(blk_make_request);
1297 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1298 * @rq: request to be initialized
1301 void blk_rq_set_block_pc(struct request *rq)
1303 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1305 rq->__sector = (sector_t) -1;
1306 rq->bio = rq->biotail = NULL;
1307 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1309 EXPORT_SYMBOL(blk_rq_set_block_pc);
1312 * blk_requeue_request - put a request back on queue
1313 * @q: request queue where request should be inserted
1314 * @rq: request to be inserted
1317 * Drivers often keep queueing requests until the hardware cannot accept
1318 * more, when that condition happens we need to put the request back
1319 * on the queue. Must be called with queue lock held.
1321 void blk_requeue_request(struct request_queue *q, struct request *rq)
1323 blk_delete_timer(rq);
1324 blk_clear_rq_complete(rq);
1325 trace_block_rq_requeue(q, rq);
1327 if (rq->cmd_flags & REQ_QUEUED)
1328 blk_queue_end_tag(q, rq);
1330 BUG_ON(blk_queued_rq(rq));
1332 elv_requeue_request(q, rq);
1334 EXPORT_SYMBOL(blk_requeue_request);
1336 static void add_acct_request(struct request_queue *q, struct request *rq,
1339 blk_account_io_start(rq, true);
1340 __elv_add_request(q, rq, where);
1343 static void part_round_stats_single(int cpu, struct hd_struct *part,
1348 if (now == part->stamp)
1351 inflight = part_in_flight(part);
1353 __part_stat_add(cpu, part, time_in_queue,
1354 inflight * (now - part->stamp));
1355 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1361 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1362 * @cpu: cpu number for stats access
1363 * @part: target partition
1365 * The average IO queue length and utilisation statistics are maintained
1366 * by observing the current state of the queue length and the amount of
1367 * time it has been in this state for.
1369 * Normally, that accounting is done on IO completion, but that can result
1370 * in more than a second's worth of IO being accounted for within any one
1371 * second, leading to >100% utilisation. To deal with that, we call this
1372 * function to do a round-off before returning the results when reading
1373 * /proc/diskstats. This accounts immediately for all queue usage up to
1374 * the current jiffies and restarts the counters again.
1376 void part_round_stats(int cpu, struct hd_struct *part)
1378 unsigned long now = jiffies;
1381 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1382 part_round_stats_single(cpu, part, now);
1384 EXPORT_SYMBOL_GPL(part_round_stats);
1387 static void blk_pm_put_request(struct request *rq)
1389 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1390 pm_runtime_mark_last_busy(rq->q->dev);
1393 static inline void blk_pm_put_request(struct request *rq) {}
1397 * queue lock must be held
1399 void __blk_put_request(struct request_queue *q, struct request *req)
1405 blk_mq_free_request(req);
1409 blk_pm_put_request(req);
1411 elv_completed_request(q, req);
1413 /* this is a bio leak */
1414 WARN_ON(req->bio != NULL);
1417 * Request may not have originated from ll_rw_blk. if not,
1418 * it didn't come out of our reserved rq pools
1420 if (req->cmd_flags & REQ_ALLOCED) {
1421 unsigned int flags = req->cmd_flags;
1422 struct request_list *rl = blk_rq_rl(req);
1424 BUG_ON(!list_empty(&req->queuelist));
1425 BUG_ON(ELV_ON_HASH(req));
1427 blk_free_request(rl, req);
1428 freed_request(rl, flags);
1432 EXPORT_SYMBOL_GPL(__blk_put_request);
1434 void blk_put_request(struct request *req)
1436 struct request_queue *q = req->q;
1439 blk_mq_free_request(req);
1441 unsigned long flags;
1443 spin_lock_irqsave(q->queue_lock, flags);
1444 __blk_put_request(q, req);
1445 spin_unlock_irqrestore(q->queue_lock, flags);
1448 EXPORT_SYMBOL(blk_put_request);
1451 * blk_add_request_payload - add a payload to a request
1452 * @rq: request to update
1453 * @page: page backing the payload
1454 * @len: length of the payload.
1456 * This allows to later add a payload to an already submitted request by
1457 * a block driver. The driver needs to take care of freeing the payload
1460 * Note that this is a quite horrible hack and nothing but handling of
1461 * discard requests should ever use it.
1463 void blk_add_request_payload(struct request *rq, struct page *page,
1466 struct bio *bio = rq->bio;
1468 bio->bi_io_vec->bv_page = page;
1469 bio->bi_io_vec->bv_offset = 0;
1470 bio->bi_io_vec->bv_len = len;
1472 bio->bi_iter.bi_size = len;
1474 bio->bi_phys_segments = 1;
1476 rq->__data_len = rq->resid_len = len;
1477 rq->nr_phys_segments = 1;
1479 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1481 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1484 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1486 if (!ll_back_merge_fn(q, req, bio))
1489 trace_block_bio_backmerge(q, req, bio);
1491 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1492 blk_rq_set_mixed_merge(req);
1494 req->biotail->bi_next = bio;
1496 req->__data_len += bio->bi_iter.bi_size;
1497 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1499 blk_account_io_start(req, false);
1503 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1506 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1508 if (!ll_front_merge_fn(q, req, bio))
1511 trace_block_bio_frontmerge(q, req, bio);
1513 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1514 blk_rq_set_mixed_merge(req);
1516 bio->bi_next = req->bio;
1519 req->__sector = bio->bi_iter.bi_sector;
1520 req->__data_len += bio->bi_iter.bi_size;
1521 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1523 blk_account_io_start(req, false);
1528 * blk_attempt_plug_merge - try to merge with %current's plugged list
1529 * @q: request_queue new bio is being queued at
1530 * @bio: new bio being queued
1531 * @request_count: out parameter for number of traversed plugged requests
1533 * Determine whether @bio being queued on @q can be merged with a request
1534 * on %current's plugged list. Returns %true if merge was successful,
1537 * Plugging coalesces IOs from the same issuer for the same purpose without
1538 * going through @q->queue_lock. As such it's more of an issuing mechanism
1539 * than scheduling, and the request, while may have elvpriv data, is not
1540 * added on the elevator at this point. In addition, we don't have
1541 * reliable access to the elevator outside queue lock. Only check basic
1542 * merging parameters without querying the elevator.
1544 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1546 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1547 unsigned int *request_count,
1548 struct request **same_queue_rq)
1550 struct blk_plug *plug;
1553 struct list_head *plug_list;
1555 plug = current->plug;
1561 plug_list = &plug->mq_list;
1563 plug_list = &plug->list;
1565 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1571 * Only blk-mq multiple hardware queues case checks the
1572 * rq in the same queue, there should be only one such
1576 *same_queue_rq = rq;
1579 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1582 el_ret = blk_try_merge(rq, bio);
1583 if (el_ret == ELEVATOR_BACK_MERGE) {
1584 ret = bio_attempt_back_merge(q, rq, bio);
1587 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1588 ret = bio_attempt_front_merge(q, rq, bio);
1597 void init_request_from_bio(struct request *req, struct bio *bio)
1599 req->cmd_type = REQ_TYPE_FS;
1601 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1602 if (bio->bi_rw & REQ_RAHEAD)
1603 req->cmd_flags |= REQ_FAILFAST_MASK;
1606 req->__sector = bio->bi_iter.bi_sector;
1607 req->ioprio = bio_prio(bio);
1608 blk_rq_bio_prep(req->q, req, bio);
1611 static void blk_queue_bio(struct request_queue *q, struct bio *bio)
1613 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1614 struct blk_plug *plug;
1615 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1616 struct request *req;
1617 unsigned int request_count = 0;
1619 blk_queue_split(q, &bio, q->bio_split);
1622 * low level driver can indicate that it wants pages above a
1623 * certain limit bounced to low memory (ie for highmem, or even
1624 * ISA dma in theory)
1626 blk_queue_bounce(q, &bio);
1628 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1629 bio->bi_error = -EIO;
1634 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1635 spin_lock_irq(q->queue_lock);
1636 where = ELEVATOR_INSERT_FLUSH;
1641 * Check if we can merge with the plugged list before grabbing
1644 if (!blk_queue_nomerges(q) &&
1645 blk_attempt_plug_merge(q, bio, &request_count, NULL))
1648 spin_lock_irq(q->queue_lock);
1650 el_ret = elv_merge(q, &req, bio);
1651 if (el_ret == ELEVATOR_BACK_MERGE) {
1652 if (bio_attempt_back_merge(q, req, bio)) {
1653 elv_bio_merged(q, req, bio);
1654 if (!attempt_back_merge(q, req))
1655 elv_merged_request(q, req, el_ret);
1658 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1659 if (bio_attempt_front_merge(q, req, bio)) {
1660 elv_bio_merged(q, req, bio);
1661 if (!attempt_front_merge(q, req))
1662 elv_merged_request(q, req, el_ret);
1669 * This sync check and mask will be re-done in init_request_from_bio(),
1670 * but we need to set it earlier to expose the sync flag to the
1671 * rq allocator and io schedulers.
1673 rw_flags = bio_data_dir(bio);
1675 rw_flags |= REQ_SYNC;
1678 * Grab a free request. This is might sleep but can not fail.
1679 * Returns with the queue unlocked.
1681 req = get_request(q, rw_flags, bio, GFP_NOIO);
1683 bio->bi_error = PTR_ERR(req);
1689 * After dropping the lock and possibly sleeping here, our request
1690 * may now be mergeable after it had proven unmergeable (above).
1691 * We don't worry about that case for efficiency. It won't happen
1692 * often, and the elevators are able to handle it.
1694 init_request_from_bio(req, bio);
1696 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1697 req->cpu = raw_smp_processor_id();
1699 plug = current->plug;
1702 * If this is the first request added after a plug, fire
1706 trace_block_plug(q);
1708 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1709 blk_flush_plug_list(plug, false);
1710 trace_block_plug(q);
1713 list_add_tail(&req->queuelist, &plug->list);
1714 blk_account_io_start(req, true);
1716 spin_lock_irq(q->queue_lock);
1717 add_acct_request(q, req, where);
1720 spin_unlock_irq(q->queue_lock);
1725 * If bio->bi_dev is a partition, remap the location
1727 static inline void blk_partition_remap(struct bio *bio)
1729 struct block_device *bdev = bio->bi_bdev;
1731 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1732 struct hd_struct *p = bdev->bd_part;
1734 bio->bi_iter.bi_sector += p->start_sect;
1735 bio->bi_bdev = bdev->bd_contains;
1737 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1739 bio->bi_iter.bi_sector - p->start_sect);
1743 static void handle_bad_sector(struct bio *bio)
1745 char b[BDEVNAME_SIZE];
1747 printk(KERN_INFO "attempt to access beyond end of device\n");
1748 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1749 bdevname(bio->bi_bdev, b),
1751 (unsigned long long)bio_end_sector(bio),
1752 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1755 #ifdef CONFIG_FAIL_MAKE_REQUEST
1757 static DECLARE_FAULT_ATTR(fail_make_request);
1759 static int __init setup_fail_make_request(char *str)
1761 return setup_fault_attr(&fail_make_request, str);
1763 __setup("fail_make_request=", setup_fail_make_request);
1765 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1767 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1770 static int __init fail_make_request_debugfs(void)
1772 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1773 NULL, &fail_make_request);
1775 return PTR_ERR_OR_ZERO(dir);
1778 late_initcall(fail_make_request_debugfs);
1780 #else /* CONFIG_FAIL_MAKE_REQUEST */
1782 static inline bool should_fail_request(struct hd_struct *part,
1788 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1791 * Check whether this bio extends beyond the end of the device.
1793 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1800 /* Test device or partition size, when known. */
1801 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1803 sector_t sector = bio->bi_iter.bi_sector;
1805 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1807 * This may well happen - the kernel calls bread()
1808 * without checking the size of the device, e.g., when
1809 * mounting a device.
1811 handle_bad_sector(bio);
1819 static noinline_for_stack bool
1820 generic_make_request_checks(struct bio *bio)
1822 struct request_queue *q;
1823 int nr_sectors = bio_sectors(bio);
1825 char b[BDEVNAME_SIZE];
1826 struct hd_struct *part;
1830 if (bio_check_eod(bio, nr_sectors))
1833 q = bdev_get_queue(bio->bi_bdev);
1836 "generic_make_request: Trying to access "
1837 "nonexistent block-device %s (%Lu)\n",
1838 bdevname(bio->bi_bdev, b),
1839 (long long) bio->bi_iter.bi_sector);
1843 part = bio->bi_bdev->bd_part;
1844 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1845 should_fail_request(&part_to_disk(part)->part0,
1846 bio->bi_iter.bi_size))
1850 * If this device has partitions, remap block n
1851 * of partition p to block n+start(p) of the disk.
1853 blk_partition_remap(bio);
1855 if (bio_check_eod(bio, nr_sectors))
1859 * Filter flush bio's early so that make_request based
1860 * drivers without flush support don't have to worry
1863 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1864 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1871 if ((bio->bi_rw & REQ_DISCARD) &&
1872 (!blk_queue_discard(q) ||
1873 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1878 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1884 * Various block parts want %current->io_context and lazy ioc
1885 * allocation ends up trading a lot of pain for a small amount of
1886 * memory. Just allocate it upfront. This may fail and block
1887 * layer knows how to live with it.
1889 create_io_context(GFP_ATOMIC, q->node);
1891 if (!blkcg_bio_issue_check(q, bio))
1894 trace_block_bio_queue(q, bio);
1898 bio->bi_error = err;
1904 * generic_make_request - hand a buffer to its device driver for I/O
1905 * @bio: The bio describing the location in memory and on the device.
1907 * generic_make_request() is used to make I/O requests of block
1908 * devices. It is passed a &struct bio, which describes the I/O that needs
1911 * generic_make_request() does not return any status. The
1912 * success/failure status of the request, along with notification of
1913 * completion, is delivered asynchronously through the bio->bi_end_io
1914 * function described (one day) else where.
1916 * The caller of generic_make_request must make sure that bi_io_vec
1917 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1918 * set to describe the device address, and the
1919 * bi_end_io and optionally bi_private are set to describe how
1920 * completion notification should be signaled.
1922 * generic_make_request and the drivers it calls may use bi_next if this
1923 * bio happens to be merged with someone else, and may resubmit the bio to
1924 * a lower device by calling into generic_make_request recursively, which
1925 * means the bio should NOT be touched after the call to ->make_request_fn.
1927 void generic_make_request(struct bio *bio)
1929 struct bio_list bio_list_on_stack;
1931 if (!generic_make_request_checks(bio))
1935 * We only want one ->make_request_fn to be active at a time, else
1936 * stack usage with stacked devices could be a problem. So use
1937 * current->bio_list to keep a list of requests submited by a
1938 * make_request_fn function. current->bio_list is also used as a
1939 * flag to say if generic_make_request is currently active in this
1940 * task or not. If it is NULL, then no make_request is active. If
1941 * it is non-NULL, then a make_request is active, and new requests
1942 * should be added at the tail
1944 if (current->bio_list) {
1945 bio_list_add(current->bio_list, bio);
1949 /* following loop may be a bit non-obvious, and so deserves some
1951 * Before entering the loop, bio->bi_next is NULL (as all callers
1952 * ensure that) so we have a list with a single bio.
1953 * We pretend that we have just taken it off a longer list, so
1954 * we assign bio_list to a pointer to the bio_list_on_stack,
1955 * thus initialising the bio_list of new bios to be
1956 * added. ->make_request() may indeed add some more bios
1957 * through a recursive call to generic_make_request. If it
1958 * did, we find a non-NULL value in bio_list and re-enter the loop
1959 * from the top. In this case we really did just take the bio
1960 * of the top of the list (no pretending) and so remove it from
1961 * bio_list, and call into ->make_request() again.
1963 BUG_ON(bio->bi_next);
1964 bio_list_init(&bio_list_on_stack);
1965 current->bio_list = &bio_list_on_stack;
1967 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1969 q->make_request_fn(q, bio);
1971 bio = bio_list_pop(current->bio_list);
1973 current->bio_list = NULL; /* deactivate */
1975 EXPORT_SYMBOL(generic_make_request);
1978 * submit_bio - submit a bio to the block device layer for I/O
1979 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1980 * @bio: The &struct bio which describes the I/O
1982 * submit_bio() is very similar in purpose to generic_make_request(), and
1983 * uses that function to do most of the work. Both are fairly rough
1984 * interfaces; @bio must be presetup and ready for I/O.
1987 void submit_bio(int rw, struct bio *bio)
1992 * If it's a regular read/write or a barrier with data attached,
1993 * go through the normal accounting stuff before submission.
1995 if (bio_has_data(bio)) {
1998 if (unlikely(rw & REQ_WRITE_SAME))
1999 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2001 count = bio_sectors(bio);
2004 count_vm_events(PGPGOUT, count);
2006 task_io_account_read(bio->bi_iter.bi_size);
2007 count_vm_events(PGPGIN, count);
2010 if (unlikely(block_dump)) {
2011 char b[BDEVNAME_SIZE];
2012 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2013 current->comm, task_pid_nr(current),
2014 (rw & WRITE) ? "WRITE" : "READ",
2015 (unsigned long long)bio->bi_iter.bi_sector,
2016 bdevname(bio->bi_bdev, b),
2021 generic_make_request(bio);
2023 EXPORT_SYMBOL(submit_bio);
2026 * blk_rq_check_limits - Helper function to check a request for the queue limit
2028 * @rq: the request being checked
2031 * @rq may have been made based on weaker limitations of upper-level queues
2032 * in request stacking drivers, and it may violate the limitation of @q.
2033 * Since the block layer and the underlying device driver trust @rq
2034 * after it is inserted to @q, it should be checked against @q before
2035 * the insertion using this generic function.
2037 * This function should also be useful for request stacking drivers
2038 * in some cases below, so export this function.
2039 * Request stacking drivers like request-based dm may change the queue
2040 * limits while requests are in the queue (e.g. dm's table swapping).
2041 * Such request stacking drivers should check those requests against
2042 * the new queue limits again when they dispatch those requests,
2043 * although such checkings are also done against the old queue limits
2044 * when submitting requests.
2046 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2048 if (!rq_mergeable(rq))
2051 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2052 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2057 * queue's settings related to segment counting like q->bounce_pfn
2058 * may differ from that of other stacking queues.
2059 * Recalculate it to check the request correctly on this queue's
2062 blk_recalc_rq_segments(rq);
2063 if (rq->nr_phys_segments > queue_max_segments(q)) {
2064 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2070 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2073 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2074 * @q: the queue to submit the request
2075 * @rq: the request being queued
2077 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2079 unsigned long flags;
2080 int where = ELEVATOR_INSERT_BACK;
2082 if (blk_rq_check_limits(q, rq))
2086 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2090 if (blk_queue_io_stat(q))
2091 blk_account_io_start(rq, true);
2092 blk_mq_insert_request(rq, false, true, true);
2096 spin_lock_irqsave(q->queue_lock, flags);
2097 if (unlikely(blk_queue_dying(q))) {
2098 spin_unlock_irqrestore(q->queue_lock, flags);
2103 * Submitting request must be dequeued before calling this function
2104 * because it will be linked to another request_queue
2106 BUG_ON(blk_queued_rq(rq));
2108 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2109 where = ELEVATOR_INSERT_FLUSH;
2111 add_acct_request(q, rq, where);
2112 if (where == ELEVATOR_INSERT_FLUSH)
2114 spin_unlock_irqrestore(q->queue_lock, flags);
2118 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2121 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2122 * @rq: request to examine
2125 * A request could be merge of IOs which require different failure
2126 * handling. This function determines the number of bytes which
2127 * can be failed from the beginning of the request without
2128 * crossing into area which need to be retried further.
2131 * The number of bytes to fail.
2134 * queue_lock must be held.
2136 unsigned int blk_rq_err_bytes(const struct request *rq)
2138 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2139 unsigned int bytes = 0;
2142 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2143 return blk_rq_bytes(rq);
2146 * Currently the only 'mixing' which can happen is between
2147 * different fastfail types. We can safely fail portions
2148 * which have all the failfast bits that the first one has -
2149 * the ones which are at least as eager to fail as the first
2152 for (bio = rq->bio; bio; bio = bio->bi_next) {
2153 if ((bio->bi_rw & ff) != ff)
2155 bytes += bio->bi_iter.bi_size;
2158 /* this could lead to infinite loop */
2159 BUG_ON(blk_rq_bytes(rq) && !bytes);
2162 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2164 void blk_account_io_completion(struct request *req, unsigned int bytes)
2166 if (blk_do_io_stat(req)) {
2167 const int rw = rq_data_dir(req);
2168 struct hd_struct *part;
2171 cpu = part_stat_lock();
2173 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2178 void blk_account_io_done(struct request *req)
2181 * Account IO completion. flush_rq isn't accounted as a
2182 * normal IO on queueing nor completion. Accounting the
2183 * containing request is enough.
2185 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2186 unsigned long duration = jiffies - req->start_time;
2187 const int rw = rq_data_dir(req);
2188 struct hd_struct *part;
2191 cpu = part_stat_lock();
2194 part_stat_inc(cpu, part, ios[rw]);
2195 part_stat_add(cpu, part, ticks[rw], duration);
2196 part_round_stats(cpu, part);
2197 part_dec_in_flight(part, rw);
2199 hd_struct_put(part);
2206 * Don't process normal requests when queue is suspended
2207 * or in the process of suspending/resuming
2209 static struct request *blk_pm_peek_request(struct request_queue *q,
2212 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2213 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2219 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2226 void blk_account_io_start(struct request *rq, bool new_io)
2228 struct hd_struct *part;
2229 int rw = rq_data_dir(rq);
2232 if (!blk_do_io_stat(rq))
2235 cpu = part_stat_lock();
2239 part_stat_inc(cpu, part, merges[rw]);
2241 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2242 if (!hd_struct_try_get(part)) {
2244 * The partition is already being removed,
2245 * the request will be accounted on the disk only
2247 * We take a reference on disk->part0 although that
2248 * partition will never be deleted, so we can treat
2249 * it as any other partition.
2251 part = &rq->rq_disk->part0;
2252 hd_struct_get(part);
2254 part_round_stats(cpu, part);
2255 part_inc_in_flight(part, rw);
2263 * blk_peek_request - peek at the top of a request queue
2264 * @q: request queue to peek at
2267 * Return the request at the top of @q. The returned request
2268 * should be started using blk_start_request() before LLD starts
2272 * Pointer to the request at the top of @q if available. Null
2276 * queue_lock must be held.
2278 struct request *blk_peek_request(struct request_queue *q)
2283 while ((rq = __elv_next_request(q)) != NULL) {
2285 rq = blk_pm_peek_request(q, rq);
2289 if (!(rq->cmd_flags & REQ_STARTED)) {
2291 * This is the first time the device driver
2292 * sees this request (possibly after
2293 * requeueing). Notify IO scheduler.
2295 if (rq->cmd_flags & REQ_SORTED)
2296 elv_activate_rq(q, rq);
2299 * just mark as started even if we don't start
2300 * it, a request that has been delayed should
2301 * not be passed by new incoming requests
2303 rq->cmd_flags |= REQ_STARTED;
2304 trace_block_rq_issue(q, rq);
2307 if (!q->boundary_rq || q->boundary_rq == rq) {
2308 q->end_sector = rq_end_sector(rq);
2309 q->boundary_rq = NULL;
2312 if (rq->cmd_flags & REQ_DONTPREP)
2315 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2317 * make sure space for the drain appears we
2318 * know we can do this because max_hw_segments
2319 * has been adjusted to be one fewer than the
2322 rq->nr_phys_segments++;
2328 ret = q->prep_rq_fn(q, rq);
2329 if (ret == BLKPREP_OK) {
2331 } else if (ret == BLKPREP_DEFER) {
2333 * the request may have been (partially) prepped.
2334 * we need to keep this request in the front to
2335 * avoid resource deadlock. REQ_STARTED will
2336 * prevent other fs requests from passing this one.
2338 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2339 !(rq->cmd_flags & REQ_DONTPREP)) {
2341 * remove the space for the drain we added
2342 * so that we don't add it again
2344 --rq->nr_phys_segments;
2349 } else if (ret == BLKPREP_KILL) {
2350 rq->cmd_flags |= REQ_QUIET;
2352 * Mark this request as started so we don't trigger
2353 * any debug logic in the end I/O path.
2355 blk_start_request(rq);
2356 __blk_end_request_all(rq, -EIO);
2358 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2365 EXPORT_SYMBOL(blk_peek_request);
2367 void blk_dequeue_request(struct request *rq)
2369 struct request_queue *q = rq->q;
2371 BUG_ON(list_empty(&rq->queuelist));
2372 BUG_ON(ELV_ON_HASH(rq));
2374 list_del_init(&rq->queuelist);
2377 * the time frame between a request being removed from the lists
2378 * and to it is freed is accounted as io that is in progress at
2381 if (blk_account_rq(rq)) {
2382 q->in_flight[rq_is_sync(rq)]++;
2383 set_io_start_time_ns(rq);
2388 * blk_start_request - start request processing on the driver
2389 * @req: request to dequeue
2392 * Dequeue @req and start timeout timer on it. This hands off the
2393 * request to the driver.
2395 * Block internal functions which don't want to start timer should
2396 * call blk_dequeue_request().
2399 * queue_lock must be held.
2401 void blk_start_request(struct request *req)
2403 blk_dequeue_request(req);
2406 * We are now handing the request to the hardware, initialize
2407 * resid_len to full count and add the timeout handler.
2409 req->resid_len = blk_rq_bytes(req);
2410 if (unlikely(blk_bidi_rq(req)))
2411 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2413 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2416 EXPORT_SYMBOL(blk_start_request);
2419 * blk_fetch_request - fetch a request from a request queue
2420 * @q: request queue to fetch a request from
2423 * Return the request at the top of @q. The request is started on
2424 * return and LLD can start processing it immediately.
2427 * Pointer to the request at the top of @q if available. Null
2431 * queue_lock must be held.
2433 struct request *blk_fetch_request(struct request_queue *q)
2437 rq = blk_peek_request(q);
2439 blk_start_request(rq);
2442 EXPORT_SYMBOL(blk_fetch_request);
2445 * blk_update_request - Special helper function for request stacking drivers
2446 * @req: the request being processed
2447 * @error: %0 for success, < %0 for error
2448 * @nr_bytes: number of bytes to complete @req
2451 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2452 * the request structure even if @req doesn't have leftover.
2453 * If @req has leftover, sets it up for the next range of segments.
2455 * This special helper function is only for request stacking drivers
2456 * (e.g. request-based dm) so that they can handle partial completion.
2457 * Actual device drivers should use blk_end_request instead.
2459 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2460 * %false return from this function.
2463 * %false - this request doesn't have any more data
2464 * %true - this request has more data
2466 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2470 trace_block_rq_complete(req->q, req, nr_bytes);
2476 * For fs requests, rq is just carrier of independent bio's
2477 * and each partial completion should be handled separately.
2478 * Reset per-request error on each partial completion.
2480 * TODO: tj: This is too subtle. It would be better to let
2481 * low level drivers do what they see fit.
2483 if (req->cmd_type == REQ_TYPE_FS)
2486 if (error && req->cmd_type == REQ_TYPE_FS &&
2487 !(req->cmd_flags & REQ_QUIET)) {
2492 error_type = "recoverable transport";
2495 error_type = "critical target";
2498 error_type = "critical nexus";
2501 error_type = "timeout";
2504 error_type = "critical space allocation";
2507 error_type = "critical medium";
2514 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2515 __func__, error_type, req->rq_disk ?
2516 req->rq_disk->disk_name : "?",
2517 (unsigned long long)blk_rq_pos(req));
2521 blk_account_io_completion(req, nr_bytes);
2525 struct bio *bio = req->bio;
2526 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2528 if (bio_bytes == bio->bi_iter.bi_size)
2529 req->bio = bio->bi_next;
2531 req_bio_endio(req, bio, bio_bytes, error);
2533 total_bytes += bio_bytes;
2534 nr_bytes -= bio_bytes;
2545 * Reset counters so that the request stacking driver
2546 * can find how many bytes remain in the request
2549 req->__data_len = 0;
2553 req->__data_len -= total_bytes;
2555 /* update sector only for requests with clear definition of sector */
2556 if (req->cmd_type == REQ_TYPE_FS)
2557 req->__sector += total_bytes >> 9;
2559 /* mixed attributes always follow the first bio */
2560 if (req->cmd_flags & REQ_MIXED_MERGE) {
2561 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2562 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2566 * If total number of sectors is less than the first segment
2567 * size, something has gone terribly wrong.
2569 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2570 blk_dump_rq_flags(req, "request botched");
2571 req->__data_len = blk_rq_cur_bytes(req);
2574 /* recalculate the number of segments */
2575 blk_recalc_rq_segments(req);
2579 EXPORT_SYMBOL_GPL(blk_update_request);
2581 static bool blk_update_bidi_request(struct request *rq, int error,
2582 unsigned int nr_bytes,
2583 unsigned int bidi_bytes)
2585 if (blk_update_request(rq, error, nr_bytes))
2588 /* Bidi request must be completed as a whole */
2589 if (unlikely(blk_bidi_rq(rq)) &&
2590 blk_update_request(rq->next_rq, error, bidi_bytes))
2593 if (blk_queue_add_random(rq->q))
2594 add_disk_randomness(rq->rq_disk);
2600 * blk_unprep_request - unprepare a request
2603 * This function makes a request ready for complete resubmission (or
2604 * completion). It happens only after all error handling is complete,
2605 * so represents the appropriate moment to deallocate any resources
2606 * that were allocated to the request in the prep_rq_fn. The queue
2607 * lock is held when calling this.
2609 void blk_unprep_request(struct request *req)
2611 struct request_queue *q = req->q;
2613 req->cmd_flags &= ~REQ_DONTPREP;
2614 if (q->unprep_rq_fn)
2615 q->unprep_rq_fn(q, req);
2617 EXPORT_SYMBOL_GPL(blk_unprep_request);
2620 * queue lock must be held
2622 void blk_finish_request(struct request *req, int error)
2624 if (req->cmd_flags & REQ_QUEUED)
2625 blk_queue_end_tag(req->q, req);
2627 BUG_ON(blk_queued_rq(req));
2629 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2630 laptop_io_completion(&req->q->backing_dev_info);
2632 blk_delete_timer(req);
2634 if (req->cmd_flags & REQ_DONTPREP)
2635 blk_unprep_request(req);
2637 blk_account_io_done(req);
2640 req->end_io(req, error);
2642 if (blk_bidi_rq(req))
2643 __blk_put_request(req->next_rq->q, req->next_rq);
2645 __blk_put_request(req->q, req);
2648 EXPORT_SYMBOL(blk_finish_request);
2651 * blk_end_bidi_request - Complete a bidi request
2652 * @rq: the request to complete
2653 * @error: %0 for success, < %0 for error
2654 * @nr_bytes: number of bytes to complete @rq
2655 * @bidi_bytes: number of bytes to complete @rq->next_rq
2658 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2659 * Drivers that supports bidi can safely call this member for any
2660 * type of request, bidi or uni. In the later case @bidi_bytes is
2664 * %false - we are done with this request
2665 * %true - still buffers pending for this request
2667 static bool blk_end_bidi_request(struct request *rq, int error,
2668 unsigned int nr_bytes, unsigned int bidi_bytes)
2670 struct request_queue *q = rq->q;
2671 unsigned long flags;
2673 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2676 spin_lock_irqsave(q->queue_lock, flags);
2677 blk_finish_request(rq, error);
2678 spin_unlock_irqrestore(q->queue_lock, flags);
2684 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2685 * @rq: the request to complete
2686 * @error: %0 for success, < %0 for error
2687 * @nr_bytes: number of bytes to complete @rq
2688 * @bidi_bytes: number of bytes to complete @rq->next_rq
2691 * Identical to blk_end_bidi_request() except that queue lock is
2692 * assumed to be locked on entry and remains so on return.
2695 * %false - we are done with this request
2696 * %true - still buffers pending for this request
2698 bool __blk_end_bidi_request(struct request *rq, int error,
2699 unsigned int nr_bytes, unsigned int bidi_bytes)
2701 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2704 blk_finish_request(rq, error);
2710 * blk_end_request - Helper function for drivers to complete the request.
2711 * @rq: the request being processed
2712 * @error: %0 for success, < %0 for error
2713 * @nr_bytes: number of bytes to complete
2716 * Ends I/O on a number of bytes attached to @rq.
2717 * If @rq has leftover, sets it up for the next range of segments.
2720 * %false - we are done with this request
2721 * %true - still buffers pending for this request
2723 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2725 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2727 EXPORT_SYMBOL(blk_end_request);
2730 * blk_end_request_all - Helper function for drives to finish the request.
2731 * @rq: the request to finish
2732 * @error: %0 for success, < %0 for error
2735 * Completely finish @rq.
2737 void blk_end_request_all(struct request *rq, int error)
2740 unsigned int bidi_bytes = 0;
2742 if (unlikely(blk_bidi_rq(rq)))
2743 bidi_bytes = blk_rq_bytes(rq->next_rq);
2745 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2748 EXPORT_SYMBOL(blk_end_request_all);
2751 * blk_end_request_cur - Helper function to finish the current request chunk.
2752 * @rq: the request to finish the current chunk for
2753 * @error: %0 for success, < %0 for error
2756 * Complete the current consecutively mapped chunk from @rq.
2759 * %false - we are done with this request
2760 * %true - still buffers pending for this request
2762 bool blk_end_request_cur(struct request *rq, int error)
2764 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2766 EXPORT_SYMBOL(blk_end_request_cur);
2769 * blk_end_request_err - Finish a request till the next failure boundary.
2770 * @rq: the request to finish till the next failure boundary for
2771 * @error: must be negative errno
2774 * Complete @rq till the next failure boundary.
2777 * %false - we are done with this request
2778 * %true - still buffers pending for this request
2780 bool blk_end_request_err(struct request *rq, int error)
2782 WARN_ON(error >= 0);
2783 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2785 EXPORT_SYMBOL_GPL(blk_end_request_err);
2788 * __blk_end_request - Helper function for drivers to complete the request.
2789 * @rq: the request being processed
2790 * @error: %0 for success, < %0 for error
2791 * @nr_bytes: number of bytes to complete
2794 * Must be called with queue lock held unlike blk_end_request().
2797 * %false - we are done with this request
2798 * %true - still buffers pending for this request
2800 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2802 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2804 EXPORT_SYMBOL(__blk_end_request);
2807 * __blk_end_request_all - Helper function for drives to finish the request.
2808 * @rq: the request to finish
2809 * @error: %0 for success, < %0 for error
2812 * Completely finish @rq. Must be called with queue lock held.
2814 void __blk_end_request_all(struct request *rq, int error)
2817 unsigned int bidi_bytes = 0;
2819 if (unlikely(blk_bidi_rq(rq)))
2820 bidi_bytes = blk_rq_bytes(rq->next_rq);
2822 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2825 EXPORT_SYMBOL(__blk_end_request_all);
2828 * __blk_end_request_cur - Helper function to finish the current request chunk.
2829 * @rq: the request to finish the current chunk for
2830 * @error: %0 for success, < %0 for error
2833 * Complete the current consecutively mapped chunk from @rq. Must
2834 * be called 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_cur(struct request *rq, int error)
2842 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2844 EXPORT_SYMBOL(__blk_end_request_cur);
2847 * __blk_end_request_err - Finish a request till the next failure boundary.
2848 * @rq: the request to finish till the next failure boundary for
2849 * @error: must be negative errno
2852 * Complete @rq till the next failure boundary. Must be called
2853 * with queue lock held.
2856 * %false - we are done with this request
2857 * %true - still buffers pending for this request
2859 bool __blk_end_request_err(struct request *rq, int error)
2861 WARN_ON(error >= 0);
2862 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2864 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2866 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2869 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2870 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2872 if (bio_has_data(bio))
2873 rq->nr_phys_segments = bio_phys_segments(q, bio);
2875 rq->__data_len = bio->bi_iter.bi_size;
2876 rq->bio = rq->biotail = bio;
2879 rq->rq_disk = bio->bi_bdev->bd_disk;
2882 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2884 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2885 * @rq: the request to be flushed
2888 * Flush all pages in @rq.
2890 void rq_flush_dcache_pages(struct request *rq)
2892 struct req_iterator iter;
2893 struct bio_vec bvec;
2895 rq_for_each_segment(bvec, rq, iter)
2896 flush_dcache_page(bvec.bv_page);
2898 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2902 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2903 * @q : the queue of the device being checked
2906 * Check if underlying low-level drivers of a device are busy.
2907 * If the drivers want to export their busy state, they must set own
2908 * exporting function using blk_queue_lld_busy() first.
2910 * Basically, this function is used only by request stacking drivers
2911 * to stop dispatching requests to underlying devices when underlying
2912 * devices are busy. This behavior helps more I/O merging on the queue
2913 * of the request stacking driver and prevents I/O throughput regression
2914 * on burst I/O load.
2917 * 0 - Not busy (The request stacking driver should dispatch request)
2918 * 1 - Busy (The request stacking driver should stop dispatching request)
2920 int blk_lld_busy(struct request_queue *q)
2923 return q->lld_busy_fn(q);
2927 EXPORT_SYMBOL_GPL(blk_lld_busy);
2930 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2931 * @rq: the clone request to be cleaned up
2934 * Free all bios in @rq for a cloned request.
2936 void blk_rq_unprep_clone(struct request *rq)
2940 while ((bio = rq->bio) != NULL) {
2941 rq->bio = bio->bi_next;
2946 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2949 * Copy attributes of the original request to the clone request.
2950 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2952 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2954 dst->cpu = src->cpu;
2955 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2956 dst->cmd_type = src->cmd_type;
2957 dst->__sector = blk_rq_pos(src);
2958 dst->__data_len = blk_rq_bytes(src);
2959 dst->nr_phys_segments = src->nr_phys_segments;
2960 dst->ioprio = src->ioprio;
2961 dst->extra_len = src->extra_len;
2965 * blk_rq_prep_clone - Helper function to setup clone request
2966 * @rq: the request to be setup
2967 * @rq_src: original request to be cloned
2968 * @bs: bio_set that bios for clone are allocated from
2969 * @gfp_mask: memory allocation mask for bio
2970 * @bio_ctr: setup function to be called for each clone bio.
2971 * Returns %0 for success, non %0 for failure.
2972 * @data: private data to be passed to @bio_ctr
2975 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2976 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2977 * are not copied, and copying such parts is the caller's responsibility.
2978 * Also, pages which the original bios are pointing to are not copied
2979 * and the cloned bios just point same pages.
2980 * So cloned bios must be completed before original bios, which means
2981 * the caller must complete @rq before @rq_src.
2983 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2984 struct bio_set *bs, gfp_t gfp_mask,
2985 int (*bio_ctr)(struct bio *, struct bio *, void *),
2988 struct bio *bio, *bio_src;
2993 __rq_for_each_bio(bio_src, rq_src) {
2994 bio = bio_clone_fast(bio_src, gfp_mask, bs);
2998 if (bio_ctr && bio_ctr(bio, bio_src, data))
3002 rq->biotail->bi_next = bio;
3005 rq->bio = rq->biotail = bio;
3008 __blk_rq_prep_clone(rq, rq_src);
3015 blk_rq_unprep_clone(rq);
3019 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3021 int kblockd_schedule_work(struct work_struct *work)
3023 return queue_work(kblockd_workqueue, work);
3025 EXPORT_SYMBOL(kblockd_schedule_work);
3027 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3028 unsigned long delay)
3030 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3032 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3034 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3035 unsigned long delay)
3037 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3039 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3042 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3043 * @plug: The &struct blk_plug that needs to be initialized
3046 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3047 * pending I/O should the task end up blocking between blk_start_plug() and
3048 * blk_finish_plug(). This is important from a performance perspective, but
3049 * also ensures that we don't deadlock. For instance, if the task is blocking
3050 * for a memory allocation, memory reclaim could end up wanting to free a
3051 * page belonging to that request that is currently residing in our private
3052 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3053 * this kind of deadlock.
3055 void blk_start_plug(struct blk_plug *plug)
3057 struct task_struct *tsk = current;
3060 * If this is a nested plug, don't actually assign it.
3065 INIT_LIST_HEAD(&plug->list);
3066 INIT_LIST_HEAD(&plug->mq_list);
3067 INIT_LIST_HEAD(&plug->cb_list);
3069 * Store ordering should not be needed here, since a potential
3070 * preempt will imply a full memory barrier
3074 EXPORT_SYMBOL(blk_start_plug);
3076 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3078 struct request *rqa = container_of(a, struct request, queuelist);
3079 struct request *rqb = container_of(b, struct request, queuelist);
3081 return !(rqa->q < rqb->q ||
3082 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3086 * If 'from_schedule' is true, then postpone the dispatch of requests
3087 * until a safe kblockd context. We due this to avoid accidental big
3088 * additional stack usage in driver dispatch, in places where the originally
3089 * plugger did not intend it.
3091 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3093 __releases(q->queue_lock)
3095 trace_block_unplug(q, depth, !from_schedule);
3098 blk_run_queue_async(q);
3101 spin_unlock(q->queue_lock);
3104 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3106 LIST_HEAD(callbacks);
3108 while (!list_empty(&plug->cb_list)) {
3109 list_splice_init(&plug->cb_list, &callbacks);
3111 while (!list_empty(&callbacks)) {
3112 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3115 list_del(&cb->list);
3116 cb->callback(cb, from_schedule);
3121 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3124 struct blk_plug *plug = current->plug;
3125 struct blk_plug_cb *cb;
3130 list_for_each_entry(cb, &plug->cb_list, list)
3131 if (cb->callback == unplug && cb->data == data)
3134 /* Not currently on the callback list */
3135 BUG_ON(size < sizeof(*cb));
3136 cb = kzalloc(size, GFP_ATOMIC);
3139 cb->callback = unplug;
3140 list_add(&cb->list, &plug->cb_list);
3144 EXPORT_SYMBOL(blk_check_plugged);
3146 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3148 struct request_queue *q;
3149 unsigned long flags;
3154 flush_plug_callbacks(plug, from_schedule);
3156 if (!list_empty(&plug->mq_list))
3157 blk_mq_flush_plug_list(plug, from_schedule);
3159 if (list_empty(&plug->list))
3162 list_splice_init(&plug->list, &list);
3164 list_sort(NULL, &list, plug_rq_cmp);
3170 * Save and disable interrupts here, to avoid doing it for every
3171 * queue lock we have to take.
3173 local_irq_save(flags);
3174 while (!list_empty(&list)) {
3175 rq = list_entry_rq(list.next);
3176 list_del_init(&rq->queuelist);
3180 * This drops the queue lock
3183 queue_unplugged(q, depth, from_schedule);
3186 spin_lock(q->queue_lock);
3190 * Short-circuit if @q is dead
3192 if (unlikely(blk_queue_dying(q))) {
3193 __blk_end_request_all(rq, -ENODEV);
3198 * rq is already accounted, so use raw insert
3200 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3201 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3203 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3209 * This drops the queue lock
3212 queue_unplugged(q, depth, from_schedule);
3214 local_irq_restore(flags);
3217 void blk_finish_plug(struct blk_plug *plug)
3219 if (plug != current->plug)
3221 blk_flush_plug_list(plug, false);
3223 current->plug = NULL;
3225 EXPORT_SYMBOL(blk_finish_plug);
3229 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3230 * @q: the queue of the device
3231 * @dev: the device the queue belongs to
3234 * Initialize runtime-PM-related fields for @q and start auto suspend for
3235 * @dev. Drivers that want to take advantage of request-based runtime PM
3236 * should call this function after @dev has been initialized, and its
3237 * request queue @q has been allocated, and runtime PM for it can not happen
3238 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3239 * cases, driver should call this function before any I/O has taken place.
3241 * This function takes care of setting up using auto suspend for the device,
3242 * the autosuspend delay is set to -1 to make runtime suspend impossible
3243 * until an updated value is either set by user or by driver. Drivers do
3244 * not need to touch other autosuspend settings.
3246 * The block layer runtime PM is request based, so only works for drivers
3247 * that use request as their IO unit instead of those directly use bio's.
3249 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3252 q->rpm_status = RPM_ACTIVE;
3253 pm_runtime_set_autosuspend_delay(q->dev, -1);
3254 pm_runtime_use_autosuspend(q->dev);
3256 EXPORT_SYMBOL(blk_pm_runtime_init);
3259 * blk_pre_runtime_suspend - Pre runtime suspend check
3260 * @q: the queue of the device
3263 * This function will check if runtime suspend is allowed for the device
3264 * by examining if there are any requests pending in the queue. If there
3265 * are requests pending, the device can not be runtime suspended; otherwise,
3266 * the queue's status will be updated to SUSPENDING and the driver can
3267 * proceed to suspend the device.
3269 * For the not allowed case, we mark last busy for the device so that
3270 * runtime PM core will try to autosuspend it some time later.
3272 * This function should be called near the start of the device's
3273 * runtime_suspend callback.
3276 * 0 - OK to runtime suspend the device
3277 * -EBUSY - Device should not be runtime suspended
3279 int blk_pre_runtime_suspend(struct request_queue *q)
3283 spin_lock_irq(q->queue_lock);
3284 if (q->nr_pending) {
3286 pm_runtime_mark_last_busy(q->dev);
3288 q->rpm_status = RPM_SUSPENDING;
3290 spin_unlock_irq(q->queue_lock);
3293 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3296 * blk_post_runtime_suspend - Post runtime suspend processing
3297 * @q: the queue of the device
3298 * @err: return value of the device's runtime_suspend function
3301 * Update the queue's runtime status according to the return value of the
3302 * device's runtime suspend function and mark last busy for the device so
3303 * that PM core will try to auto suspend the device at a later time.
3305 * This function should be called near the end of the device's
3306 * runtime_suspend callback.
3308 void blk_post_runtime_suspend(struct request_queue *q, int err)
3310 spin_lock_irq(q->queue_lock);
3312 q->rpm_status = RPM_SUSPENDED;
3314 q->rpm_status = RPM_ACTIVE;
3315 pm_runtime_mark_last_busy(q->dev);
3317 spin_unlock_irq(q->queue_lock);
3319 EXPORT_SYMBOL(blk_post_runtime_suspend);
3322 * blk_pre_runtime_resume - Pre runtime resume processing
3323 * @q: the queue of the device
3326 * Update the queue's runtime status to RESUMING in preparation for the
3327 * runtime resume of the device.
3329 * This function should be called near the start of the device's
3330 * runtime_resume callback.
3332 void blk_pre_runtime_resume(struct request_queue *q)
3334 spin_lock_irq(q->queue_lock);
3335 q->rpm_status = RPM_RESUMING;
3336 spin_unlock_irq(q->queue_lock);
3338 EXPORT_SYMBOL(blk_pre_runtime_resume);
3341 * blk_post_runtime_resume - Post runtime resume processing
3342 * @q: the queue of the device
3343 * @err: return value of the device's runtime_resume function
3346 * Update the queue's runtime status according to the return value of the
3347 * device's runtime_resume function. If it is successfully resumed, process
3348 * the requests that are queued into the device's queue when it is resuming
3349 * and then mark last busy and initiate autosuspend for it.
3351 * This function should be called near the end of the device's
3352 * runtime_resume callback.
3354 void blk_post_runtime_resume(struct request_queue *q, int err)
3356 spin_lock_irq(q->queue_lock);
3358 q->rpm_status = RPM_ACTIVE;
3360 pm_runtime_mark_last_busy(q->dev);
3361 pm_request_autosuspend(q->dev);
3363 q->rpm_status = RPM_SUSPENDED;
3365 spin_unlock_irq(q->queue_lock);
3367 EXPORT_SYMBOL(blk_post_runtime_resume);
3370 int __init blk_dev_init(void)
3372 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3373 FIELD_SIZEOF(struct request, cmd_flags));
3375 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3376 kblockd_workqueue = alloc_workqueue("kblockd",
3377 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3378 if (!kblockd_workqueue)
3379 panic("Failed to create kblockd\n");
3381 request_cachep = kmem_cache_create("blkdev_requests",
3382 sizeof(struct request), 0, SLAB_PANIC, NULL);
3384 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3385 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);