2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
36 #define CREATE_TRACE_POINTS
37 #include <trace/events/block.h>
40 #include "blk-cgroup.h"
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
48 DEFINE_IDA(blk_queue_ida);
51 * For the allocated request tables
53 struct kmem_cache *request_cachep = NULL;
56 * For queue allocation
58 struct kmem_cache *blk_requestq_cachep;
61 * Controlling structure to kblockd
63 static struct workqueue_struct *kblockd_workqueue;
65 void blk_queue_congestion_threshold(struct request_queue *q)
69 nr = q->nr_requests - (q->nr_requests / 8) + 1;
70 if (nr > q->nr_requests)
72 q->nr_congestion_on = nr;
74 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
77 q->nr_congestion_off = nr;
81 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
84 * Locates the passed device's request queue and returns the address of its
87 * Will return NULL if the request queue cannot be located.
89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
91 struct backing_dev_info *ret = NULL;
92 struct request_queue *q = bdev_get_queue(bdev);
95 ret = &q->backing_dev_info;
98 EXPORT_SYMBOL(blk_get_backing_dev_info);
100 void blk_rq_init(struct request_queue *q, struct request *rq)
102 memset(rq, 0, sizeof(*rq));
104 INIT_LIST_HEAD(&rq->queuelist);
105 INIT_LIST_HEAD(&rq->timeout_list);
108 rq->__sector = (sector_t) -1;
109 INIT_HLIST_NODE(&rq->hash);
110 RB_CLEAR_NODE(&rq->rb_node);
112 rq->cmd_len = BLK_MAX_CDB;
114 rq->start_time = jiffies;
115 set_start_time_ns(rq);
118 EXPORT_SYMBOL(blk_rq_init);
120 static void req_bio_endio(struct request *rq, struct bio *bio,
121 unsigned int nbytes, int error)
124 clear_bit(BIO_UPTODATE, &bio->bi_flags);
125 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
128 if (unlikely(rq->cmd_flags & REQ_QUIET))
129 set_bit(BIO_QUIET, &bio->bi_flags);
131 bio_advance(bio, nbytes);
133 /* don't actually finish bio if it's part of flush sequence */
134 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
135 bio_endio(bio, error);
138 void blk_dump_rq_flags(struct request *rq, char *msg)
142 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
143 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
144 (unsigned long long) rq->cmd_flags);
146 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
147 (unsigned long long)blk_rq_pos(rq),
148 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
149 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
150 rq->bio, rq->biotail, blk_rq_bytes(rq));
152 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
153 printk(KERN_INFO " cdb: ");
154 for (bit = 0; bit < BLK_MAX_CDB; bit++)
155 printk("%02x ", rq->cmd[bit]);
159 EXPORT_SYMBOL(blk_dump_rq_flags);
161 static void blk_delay_work(struct work_struct *work)
163 struct request_queue *q;
165 q = container_of(work, struct request_queue, delay_work.work);
166 spin_lock_irq(q->queue_lock);
168 spin_unlock_irq(q->queue_lock);
172 * blk_delay_queue - restart queueing after defined interval
173 * @q: The &struct request_queue in question
174 * @msecs: Delay in msecs
177 * Sometimes queueing needs to be postponed for a little while, to allow
178 * resources to come back. This function will make sure that queueing is
179 * restarted around the specified time. Queue lock must be held.
181 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
183 if (likely(!blk_queue_dead(q)))
184 queue_delayed_work(kblockd_workqueue, &q->delay_work,
185 msecs_to_jiffies(msecs));
187 EXPORT_SYMBOL(blk_delay_queue);
190 * blk_start_queue - restart a previously stopped queue
191 * @q: The &struct request_queue in question
194 * blk_start_queue() will clear the stop flag on the queue, and call
195 * the request_fn for the queue if it was in a stopped state when
196 * entered. Also see blk_stop_queue(). Queue lock must be held.
198 void blk_start_queue(struct request_queue *q)
200 WARN_ON(!irqs_disabled());
202 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
205 EXPORT_SYMBOL(blk_start_queue);
208 * blk_stop_queue - stop a queue
209 * @q: The &struct request_queue in question
212 * The Linux block layer assumes that a block driver will consume all
213 * entries on the request queue when the request_fn strategy is called.
214 * Often this will not happen, because of hardware limitations (queue
215 * depth settings). If a device driver gets a 'queue full' response,
216 * or if it simply chooses not to queue more I/O at one point, it can
217 * call this function to prevent the request_fn from being called until
218 * the driver has signalled it's ready to go again. This happens by calling
219 * blk_start_queue() to restart queue operations. Queue lock must be held.
221 void blk_stop_queue(struct request_queue *q)
223 cancel_delayed_work(&q->delay_work);
224 queue_flag_set(QUEUE_FLAG_STOPPED, q);
226 EXPORT_SYMBOL(blk_stop_queue);
229 * blk_sync_queue - cancel any pending callbacks on a queue
233 * The block layer may perform asynchronous callback activity
234 * on a queue, such as calling the unplug function after a timeout.
235 * A block device may call blk_sync_queue to ensure that any
236 * such activity is cancelled, thus allowing it to release resources
237 * that the callbacks might use. The caller must already have made sure
238 * that its ->make_request_fn will not re-add plugging prior to calling
241 * This function does not cancel any asynchronous activity arising
242 * out of elevator or throttling code. That would require elevaotor_exit()
243 * and blkcg_exit_queue() to be called with queue lock initialized.
246 void blk_sync_queue(struct request_queue *q)
248 del_timer_sync(&q->timeout);
251 struct blk_mq_hw_ctx *hctx;
254 queue_for_each_hw_ctx(q, hctx, i) {
255 cancel_delayed_work_sync(&hctx->run_work);
256 cancel_delayed_work_sync(&hctx->delay_work);
259 cancel_delayed_work_sync(&q->delay_work);
262 EXPORT_SYMBOL(blk_sync_queue);
265 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
266 * @q: The queue to run
269 * Invoke request handling on a queue if there are any pending requests.
270 * May be used to restart request handling after a request has completed.
271 * This variant runs the queue whether or not the queue has been
272 * stopped. Must be called with the queue lock held and interrupts
273 * disabled. See also @blk_run_queue.
275 inline void __blk_run_queue_uncond(struct request_queue *q)
277 if (unlikely(blk_queue_dead(q)))
281 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
282 * the queue lock internally. As a result multiple threads may be
283 * running such a request function concurrently. Keep track of the
284 * number of active request_fn invocations such that blk_drain_queue()
285 * can wait until all these request_fn calls have finished.
287 q->request_fn_active++;
289 q->request_fn_active--;
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue *q)
302 if (unlikely(blk_queue_stopped(q)))
305 __blk_run_queue_uncond(q);
307 EXPORT_SYMBOL(__blk_run_queue);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 * of us. The caller must hold the queue lock.
317 void blk_run_queue_async(struct request_queue *q)
319 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
320 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
322 EXPORT_SYMBOL(blk_run_queue_async);
325 * blk_run_queue - run a single device queue
326 * @q: The queue to run
329 * Invoke request handling on this queue, if it has pending work to do.
330 * May be used to restart queueing when a request has completed.
332 void blk_run_queue(struct request_queue *q)
336 spin_lock_irqsave(q->queue_lock, flags);
338 spin_unlock_irqrestore(q->queue_lock, flags);
340 EXPORT_SYMBOL(blk_run_queue);
342 void blk_put_queue(struct request_queue *q)
344 kobject_put(&q->kobj);
346 EXPORT_SYMBOL(blk_put_queue);
349 * __blk_drain_queue - drain requests from request_queue
351 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
353 * Drain requests from @q. If @drain_all is set, all requests are drained.
354 * If not, only ELVPRIV requests are drained. The caller is responsible
355 * for ensuring that no new requests which need to be drained are queued.
357 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
358 __releases(q->queue_lock)
359 __acquires(q->queue_lock)
363 lockdep_assert_held(q->queue_lock);
369 * The caller might be trying to drain @q before its
370 * elevator is initialized.
373 elv_drain_elevator(q);
375 blkcg_drain_queue(q);
378 * This function might be called on a queue which failed
379 * driver init after queue creation or is not yet fully
380 * active yet. Some drivers (e.g. fd and loop) get unhappy
381 * in such cases. Kick queue iff dispatch queue has
382 * something on it and @q has request_fn set.
384 if (!list_empty(&q->queue_head) && q->request_fn)
387 drain |= q->nr_rqs_elvpriv;
388 drain |= q->request_fn_active;
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
396 drain |= !list_empty(&q->queue_head);
397 for (i = 0; i < 2; i++) {
398 drain |= q->nr_rqs[i];
399 drain |= q->in_flight[i];
400 drain |= !list_empty(&q->flush_queue[i]);
407 spin_unlock_irq(q->queue_lock);
411 spin_lock_irq(q->queue_lock);
415 * With queue marked dead, any woken up waiter will fail the
416 * allocation path, so the wakeup chaining is lost and we're
417 * left with hung waiters. We need to wake up those waiters.
420 struct request_list *rl;
422 blk_queue_for_each_rl(rl, q)
423 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
424 wake_up_all(&rl->wait[i]);
429 * blk_queue_bypass_start - enter queue bypass mode
430 * @q: queue of interest
432 * In bypass mode, only the dispatch FIFO queue of @q is used. This
433 * function makes @q enter bypass mode and drains all requests which were
434 * throttled or issued before. On return, it's guaranteed that no request
435 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
436 * inside queue or RCU read lock.
438 void blk_queue_bypass_start(struct request_queue *q)
442 spin_lock_irq(q->queue_lock);
443 drain = !q->bypass_depth++;
444 queue_flag_set(QUEUE_FLAG_BYPASS, q);
445 spin_unlock_irq(q->queue_lock);
448 spin_lock_irq(q->queue_lock);
449 __blk_drain_queue(q, false);
450 spin_unlock_irq(q->queue_lock);
452 /* ensure blk_queue_bypass() is %true inside RCU read lock */
456 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
459 * blk_queue_bypass_end - leave queue bypass mode
460 * @q: queue of interest
462 * Leave bypass mode and restore the normal queueing behavior.
464 void blk_queue_bypass_end(struct request_queue *q)
466 spin_lock_irq(q->queue_lock);
467 if (!--q->bypass_depth)
468 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
469 WARN_ON_ONCE(q->bypass_depth < 0);
470 spin_unlock_irq(q->queue_lock);
472 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
475 * blk_cleanup_queue - shutdown a request queue
476 * @q: request queue to shutdown
478 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
479 * put it. All future requests will be failed immediately with -ENODEV.
481 void blk_cleanup_queue(struct request_queue *q)
483 spinlock_t *lock = q->queue_lock;
485 /* mark @q DYING, no new request or merges will be allowed afterwards */
486 mutex_lock(&q->sysfs_lock);
487 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
491 * A dying queue is permanently in bypass mode till released. Note
492 * that, unlike blk_queue_bypass_start(), we aren't performing
493 * synchronize_rcu() after entering bypass mode to avoid the delay
494 * as some drivers create and destroy a lot of queues while
495 * probing. This is still safe because blk_release_queue() will be
496 * called only after the queue refcnt drops to zero and nothing,
497 * RCU or not, would be traversing the queue by then.
500 queue_flag_set(QUEUE_FLAG_BYPASS, q);
502 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
503 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
504 queue_flag_set(QUEUE_FLAG_DYING, q);
505 spin_unlock_irq(lock);
506 mutex_unlock(&q->sysfs_lock);
509 * Drain all requests queued before DYING marking. Set DEAD flag to
510 * prevent that q->request_fn() gets invoked after draining finished.
513 blk_mq_drain_queue(q);
517 __blk_drain_queue(q, true);
519 queue_flag_set(QUEUE_FLAG_DEAD, q);
520 spin_unlock_irq(lock);
522 /* @q won't process any more request, flush async actions */
523 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
527 if (q->queue_lock != &q->__queue_lock)
528 q->queue_lock = &q->__queue_lock;
529 spin_unlock_irq(lock);
531 /* @q is and will stay empty, shutdown and put */
534 EXPORT_SYMBOL(blk_cleanup_queue);
536 int blk_init_rl(struct request_list *rl, struct request_queue *q,
539 if (unlikely(rl->rq_pool))
543 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
544 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
545 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
546 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
548 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
549 mempool_free_slab, request_cachep,
557 void blk_exit_rl(struct request_list *rl)
560 mempool_destroy(rl->rq_pool);
563 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
565 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
567 EXPORT_SYMBOL(blk_alloc_queue);
569 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
571 struct request_queue *q;
574 q = kmem_cache_alloc_node(blk_requestq_cachep,
575 gfp_mask | __GFP_ZERO, node_id);
579 if (percpu_counter_init(&q->mq_usage_counter, 0))
582 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
586 q->backing_dev_info.ra_pages =
587 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
588 q->backing_dev_info.state = 0;
589 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
590 q->backing_dev_info.name = "block";
593 err = bdi_init(&q->backing_dev_info);
597 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
598 laptop_mode_timer_fn, (unsigned long) q);
599 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
600 INIT_LIST_HEAD(&q->queue_head);
601 INIT_LIST_HEAD(&q->timeout_list);
602 INIT_LIST_HEAD(&q->icq_list);
603 #ifdef CONFIG_BLK_CGROUP
604 INIT_LIST_HEAD(&q->blkg_list);
606 INIT_LIST_HEAD(&q->flush_queue[0]);
607 INIT_LIST_HEAD(&q->flush_queue[1]);
608 INIT_LIST_HEAD(&q->flush_data_in_flight);
609 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
611 kobject_init(&q->kobj, &blk_queue_ktype);
613 mutex_init(&q->sysfs_lock);
614 spin_lock_init(&q->__queue_lock);
617 * By default initialize queue_lock to internal lock and driver can
618 * override it later if need be.
620 q->queue_lock = &q->__queue_lock;
623 * A queue starts its life with bypass turned on to avoid
624 * unnecessary bypass on/off overhead and nasty surprises during
625 * init. The initial bypass will be finished when the queue is
626 * registered by blk_register_queue().
629 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
631 init_waitqueue_head(&q->mq_freeze_wq);
633 if (blkcg_init_queue(q))
639 bdi_destroy(&q->backing_dev_info);
641 ida_simple_remove(&blk_queue_ida, q->id);
643 percpu_counter_destroy(&q->mq_usage_counter);
645 kmem_cache_free(blk_requestq_cachep, q);
648 EXPORT_SYMBOL(blk_alloc_queue_node);
651 * blk_init_queue - prepare a request queue for use with a block device
652 * @rfn: The function to be called to process requests that have been
653 * placed on the queue.
654 * @lock: Request queue spin lock
657 * If a block device wishes to use the standard request handling procedures,
658 * which sorts requests and coalesces adjacent requests, then it must
659 * call blk_init_queue(). The function @rfn will be called when there
660 * are requests on the queue that need to be processed. If the device
661 * supports plugging, then @rfn may not be called immediately when requests
662 * are available on the queue, but may be called at some time later instead.
663 * Plugged queues are generally unplugged when a buffer belonging to one
664 * of the requests on the queue is needed, or due to memory pressure.
666 * @rfn is not required, or even expected, to remove all requests off the
667 * queue, but only as many as it can handle at a time. If it does leave
668 * requests on the queue, it is responsible for arranging that the requests
669 * get dealt with eventually.
671 * The queue spin lock must be held while manipulating the requests on the
672 * request queue; this lock will be taken also from interrupt context, so irq
673 * disabling is needed for it.
675 * Function returns a pointer to the initialized request queue, or %NULL if
679 * blk_init_queue() must be paired with a blk_cleanup_queue() call
680 * when the block device is deactivated (such as at module unload).
683 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
685 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
687 EXPORT_SYMBOL(blk_init_queue);
689 struct request_queue *
690 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
692 struct request_queue *uninit_q, *q;
694 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
698 q = blk_init_allocated_queue(uninit_q, rfn, lock);
700 blk_cleanup_queue(uninit_q);
704 EXPORT_SYMBOL(blk_init_queue_node);
706 struct request_queue *
707 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
713 q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
717 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
721 q->prep_rq_fn = NULL;
722 q->unprep_rq_fn = NULL;
723 q->queue_flags |= QUEUE_FLAG_DEFAULT;
725 /* Override internal queue lock with supplied lock pointer */
727 q->queue_lock = lock;
730 * This also sets hw/phys segments, boundary and size
732 blk_queue_make_request(q, blk_queue_bio);
734 q->sg_reserved_size = INT_MAX;
736 /* Protect q->elevator from elevator_change */
737 mutex_lock(&q->sysfs_lock);
740 if (elevator_init(q, NULL)) {
741 mutex_unlock(&q->sysfs_lock);
745 mutex_unlock(&q->sysfs_lock);
753 EXPORT_SYMBOL(blk_init_allocated_queue);
755 bool blk_get_queue(struct request_queue *q)
757 if (likely(!blk_queue_dying(q))) {
764 EXPORT_SYMBOL(blk_get_queue);
766 static inline void blk_free_request(struct request_list *rl, struct request *rq)
768 if (rq->cmd_flags & REQ_ELVPRIV) {
769 elv_put_request(rl->q, rq);
771 put_io_context(rq->elv.icq->ioc);
774 mempool_free(rq, rl->rq_pool);
778 * ioc_batching returns true if the ioc is a valid batching request and
779 * should be given priority access to a request.
781 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
787 * Make sure the process is able to allocate at least 1 request
788 * even if the batch times out, otherwise we could theoretically
791 return ioc->nr_batch_requests == q->nr_batching ||
792 (ioc->nr_batch_requests > 0
793 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
797 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
798 * will cause the process to be a "batcher" on all queues in the system. This
799 * is the behaviour we want though - once it gets a wakeup it should be given
802 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
804 if (!ioc || ioc_batching(q, ioc))
807 ioc->nr_batch_requests = q->nr_batching;
808 ioc->last_waited = jiffies;
811 static void __freed_request(struct request_list *rl, int sync)
813 struct request_queue *q = rl->q;
816 * bdi isn't aware of blkcg yet. As all async IOs end up root
817 * blkcg anyway, just use root blkcg state.
819 if (rl == &q->root_rl &&
820 rl->count[sync] < queue_congestion_off_threshold(q))
821 blk_clear_queue_congested(q, sync);
823 if (rl->count[sync] + 1 <= q->nr_requests) {
824 if (waitqueue_active(&rl->wait[sync]))
825 wake_up(&rl->wait[sync]);
827 blk_clear_rl_full(rl, sync);
832 * A request has just been released. Account for it, update the full and
833 * congestion status, wake up any waiters. Called under q->queue_lock.
835 static void freed_request(struct request_list *rl, unsigned int flags)
837 struct request_queue *q = rl->q;
838 int sync = rw_is_sync(flags);
842 if (flags & REQ_ELVPRIV)
845 __freed_request(rl, sync);
847 if (unlikely(rl->starved[sync ^ 1]))
848 __freed_request(rl, sync ^ 1);
852 * Determine if elevator data should be initialized when allocating the
853 * request associated with @bio.
855 static bool blk_rq_should_init_elevator(struct bio *bio)
861 * Flush requests do not use the elevator so skip initialization.
862 * This allows a request to share the flush and elevator data.
864 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
871 * rq_ioc - determine io_context for request allocation
872 * @bio: request being allocated is for this bio (can be %NULL)
874 * Determine io_context to use for request allocation for @bio. May return
875 * %NULL if %current->io_context doesn't exist.
877 static struct io_context *rq_ioc(struct bio *bio)
879 #ifdef CONFIG_BLK_CGROUP
880 if (bio && bio->bi_ioc)
883 return current->io_context;
887 * __get_request - get a free request
888 * @rl: request list to allocate from
889 * @rw_flags: RW and SYNC flags
890 * @bio: bio to allocate request for (can be %NULL)
891 * @gfp_mask: allocation mask
893 * Get a free request from @q. This function may fail under memory
894 * pressure or if @q is dead.
896 * Must be callled with @q->queue_lock held and,
897 * Returns %NULL on failure, with @q->queue_lock held.
898 * Returns !%NULL on success, with @q->queue_lock *not held*.
900 static struct request *__get_request(struct request_list *rl, int rw_flags,
901 struct bio *bio, gfp_t gfp_mask)
903 struct request_queue *q = rl->q;
905 struct elevator_type *et = q->elevator->type;
906 struct io_context *ioc = rq_ioc(bio);
907 struct io_cq *icq = NULL;
908 const bool is_sync = rw_is_sync(rw_flags) != 0;
911 if (unlikely(blk_queue_dying(q)))
914 may_queue = elv_may_queue(q, rw_flags);
915 if (may_queue == ELV_MQUEUE_NO)
918 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
919 if (rl->count[is_sync]+1 >= q->nr_requests) {
921 * The queue will fill after this allocation, so set
922 * it as full, and mark this process as "batching".
923 * This process will be allowed to complete a batch of
924 * requests, others will be blocked.
926 if (!blk_rl_full(rl, is_sync)) {
927 ioc_set_batching(q, ioc);
928 blk_set_rl_full(rl, is_sync);
930 if (may_queue != ELV_MQUEUE_MUST
931 && !ioc_batching(q, ioc)) {
933 * The queue is full and the allocating
934 * process is not a "batcher", and not
935 * exempted by the IO scheduler
942 * bdi isn't aware of blkcg yet. As all async IOs end up
943 * root blkcg anyway, just use root blkcg state.
945 if (rl == &q->root_rl)
946 blk_set_queue_congested(q, is_sync);
950 * Only allow batching queuers to allocate up to 50% over the defined
951 * limit of requests, otherwise we could have thousands of requests
952 * allocated with any setting of ->nr_requests
954 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
957 q->nr_rqs[is_sync]++;
958 rl->count[is_sync]++;
959 rl->starved[is_sync] = 0;
962 * Decide whether the new request will be managed by elevator. If
963 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
964 * prevent the current elevator from being destroyed until the new
965 * request is freed. This guarantees icq's won't be destroyed and
966 * makes creating new ones safe.
968 * Also, lookup icq while holding queue_lock. If it doesn't exist,
969 * it will be created after releasing queue_lock.
971 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
972 rw_flags |= REQ_ELVPRIV;
974 if (et->icq_cache && ioc)
975 icq = ioc_lookup_icq(ioc, q);
978 if (blk_queue_io_stat(q))
979 rw_flags |= REQ_IO_STAT;
980 spin_unlock_irq(q->queue_lock);
982 /* allocate and init request */
983 rq = mempool_alloc(rl->rq_pool, gfp_mask);
988 blk_rq_set_rl(rq, rl);
989 rq->cmd_flags = rw_flags | REQ_ALLOCED;
992 if (rw_flags & REQ_ELVPRIV) {
993 if (unlikely(et->icq_cache && !icq)) {
995 icq = ioc_create_icq(ioc, q, gfp_mask);
1001 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1004 /* @rq->elv.icq holds io_context until @rq is freed */
1006 get_io_context(icq->ioc);
1010 * ioc may be NULL here, and ioc_batching will be false. That's
1011 * OK, if the queue is under the request limit then requests need
1012 * not count toward the nr_batch_requests limit. There will always
1013 * be some limit enforced by BLK_BATCH_TIME.
1015 if (ioc_batching(q, ioc))
1016 ioc->nr_batch_requests--;
1018 trace_block_getrq(q, bio, rw_flags & 1);
1023 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1024 * and may fail indefinitely under memory pressure and thus
1025 * shouldn't stall IO. Treat this request as !elvpriv. This will
1026 * disturb iosched and blkcg but weird is bettern than dead.
1028 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1029 dev_name(q->backing_dev_info.dev));
1031 rq->cmd_flags &= ~REQ_ELVPRIV;
1034 spin_lock_irq(q->queue_lock);
1035 q->nr_rqs_elvpriv--;
1036 spin_unlock_irq(q->queue_lock);
1041 * Allocation failed presumably due to memory. Undo anything we
1042 * might have messed up.
1044 * Allocating task should really be put onto the front of the wait
1045 * queue, but this is pretty rare.
1047 spin_lock_irq(q->queue_lock);
1048 freed_request(rl, rw_flags);
1051 * in the very unlikely event that allocation failed and no
1052 * requests for this direction was pending, mark us starved so that
1053 * freeing of a request in the other direction will notice
1054 * us. another possible fix would be to split the rq mempool into
1058 if (unlikely(rl->count[is_sync] == 0))
1059 rl->starved[is_sync] = 1;
1064 * get_request - get a free request
1065 * @q: request_queue to allocate request from
1066 * @rw_flags: RW and SYNC flags
1067 * @bio: bio to allocate request for (can be %NULL)
1068 * @gfp_mask: allocation mask
1070 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1071 * function keeps retrying under memory pressure and fails iff @q is dead.
1073 * Must be callled with @q->queue_lock held and,
1074 * Returns %NULL on failure, with @q->queue_lock held.
1075 * Returns !%NULL on success, with @q->queue_lock *not held*.
1077 static struct request *get_request(struct request_queue *q, int rw_flags,
1078 struct bio *bio, gfp_t gfp_mask)
1080 const bool is_sync = rw_is_sync(rw_flags) != 0;
1082 struct request_list *rl;
1085 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1087 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1091 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1096 /* wait on @rl and retry */
1097 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1098 TASK_UNINTERRUPTIBLE);
1100 trace_block_sleeprq(q, bio, rw_flags & 1);
1102 spin_unlock_irq(q->queue_lock);
1106 * After sleeping, we become a "batching" process and will be able
1107 * to allocate at least one request, and up to a big batch of them
1108 * for a small period time. See ioc_batching, ioc_set_batching
1110 ioc_set_batching(q, current->io_context);
1112 spin_lock_irq(q->queue_lock);
1113 finish_wait(&rl->wait[is_sync], &wait);
1118 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1123 BUG_ON(rw != READ && rw != WRITE);
1125 /* create ioc upfront */
1126 create_io_context(gfp_mask, q->node);
1128 spin_lock_irq(q->queue_lock);
1129 rq = get_request(q, rw, NULL, gfp_mask);
1131 spin_unlock_irq(q->queue_lock);
1132 /* q->queue_lock is unlocked at this point */
1137 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1140 return blk_mq_alloc_request(q, rw, gfp_mask);
1142 return blk_old_get_request(q, rw, gfp_mask);
1144 EXPORT_SYMBOL(blk_get_request);
1147 * blk_make_request - given a bio, allocate a corresponding struct request.
1148 * @q: target request queue
1149 * @bio: The bio describing the memory mappings that will be submitted for IO.
1150 * It may be a chained-bio properly constructed by block/bio layer.
1151 * @gfp_mask: gfp flags to be used for memory allocation
1153 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1154 * type commands. Where the struct request needs to be farther initialized by
1155 * the caller. It is passed a &struct bio, which describes the memory info of
1158 * The caller of blk_make_request must make sure that bi_io_vec
1159 * are set to describe the memory buffers. That bio_data_dir() will return
1160 * the needed direction of the request. (And all bio's in the passed bio-chain
1161 * are properly set accordingly)
1163 * If called under none-sleepable conditions, mapped bio buffers must not
1164 * need bouncing, by calling the appropriate masked or flagged allocator,
1165 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1168 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1169 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1170 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1171 * completion of a bio that hasn't been submitted yet, thus resulting in a
1172 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1173 * of bio_alloc(), as that avoids the mempool deadlock.
1174 * If possible a big IO should be split into smaller parts when allocation
1175 * fails. Partial allocation should not be an error, or you risk a live-lock.
1177 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1180 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1183 return ERR_PTR(-ENOMEM);
1186 struct bio *bounce_bio = bio;
1189 blk_queue_bounce(q, &bounce_bio);
1190 ret = blk_rq_append_bio(q, rq, bounce_bio);
1191 if (unlikely(ret)) {
1192 blk_put_request(rq);
1193 return ERR_PTR(ret);
1199 EXPORT_SYMBOL(blk_make_request);
1202 * blk_requeue_request - put a request back on queue
1203 * @q: request queue where request should be inserted
1204 * @rq: request to be inserted
1207 * Drivers often keep queueing requests until the hardware cannot accept
1208 * more, when that condition happens we need to put the request back
1209 * on the queue. Must be called with queue lock held.
1211 void blk_requeue_request(struct request_queue *q, struct request *rq)
1213 blk_delete_timer(rq);
1214 blk_clear_rq_complete(rq);
1215 trace_block_rq_requeue(q, rq);
1217 if (blk_rq_tagged(rq))
1218 blk_queue_end_tag(q, rq);
1220 BUG_ON(blk_queued_rq(rq));
1222 elv_requeue_request(q, rq);
1224 EXPORT_SYMBOL(blk_requeue_request);
1226 static void add_acct_request(struct request_queue *q, struct request *rq,
1229 blk_account_io_start(rq, true);
1230 __elv_add_request(q, rq, where);
1233 static void part_round_stats_single(int cpu, struct hd_struct *part,
1236 if (now == part->stamp)
1239 if (part_in_flight(part)) {
1240 __part_stat_add(cpu, part, time_in_queue,
1241 part_in_flight(part) * (now - part->stamp));
1242 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1248 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1249 * @cpu: cpu number for stats access
1250 * @part: target partition
1252 * The average IO queue length and utilisation statistics are maintained
1253 * by observing the current state of the queue length and the amount of
1254 * time it has been in this state for.
1256 * Normally, that accounting is done on IO completion, but that can result
1257 * in more than a second's worth of IO being accounted for within any one
1258 * second, leading to >100% utilisation. To deal with that, we call this
1259 * function to do a round-off before returning the results when reading
1260 * /proc/diskstats. This accounts immediately for all queue usage up to
1261 * the current jiffies and restarts the counters again.
1263 void part_round_stats(int cpu, struct hd_struct *part)
1265 unsigned long now = jiffies;
1268 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1269 part_round_stats_single(cpu, part, now);
1271 EXPORT_SYMBOL_GPL(part_round_stats);
1273 #ifdef CONFIG_PM_RUNTIME
1274 static void blk_pm_put_request(struct request *rq)
1276 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1277 pm_runtime_mark_last_busy(rq->q->dev);
1280 static inline void blk_pm_put_request(struct request *rq) {}
1284 * queue lock must be held
1286 void __blk_put_request(struct request_queue *q, struct request *req)
1292 blk_mq_free_request(req);
1296 blk_pm_put_request(req);
1298 elv_completed_request(q, req);
1300 /* this is a bio leak */
1301 WARN_ON(req->bio != NULL);
1304 * Request may not have originated from ll_rw_blk. if not,
1305 * it didn't come out of our reserved rq pools
1307 if (req->cmd_flags & REQ_ALLOCED) {
1308 unsigned int flags = req->cmd_flags;
1309 struct request_list *rl = blk_rq_rl(req);
1311 BUG_ON(!list_empty(&req->queuelist));
1312 BUG_ON(ELV_ON_HASH(req));
1314 blk_free_request(rl, req);
1315 freed_request(rl, flags);
1319 EXPORT_SYMBOL_GPL(__blk_put_request);
1321 void blk_put_request(struct request *req)
1323 struct request_queue *q = req->q;
1326 blk_mq_free_request(req);
1328 unsigned long flags;
1330 spin_lock_irqsave(q->queue_lock, flags);
1331 __blk_put_request(q, req);
1332 spin_unlock_irqrestore(q->queue_lock, flags);
1335 EXPORT_SYMBOL(blk_put_request);
1338 * blk_add_request_payload - add a payload to a request
1339 * @rq: request to update
1340 * @page: page backing the payload
1341 * @len: length of the payload.
1343 * This allows to later add a payload to an already submitted request by
1344 * a block driver. The driver needs to take care of freeing the payload
1347 * Note that this is a quite horrible hack and nothing but handling of
1348 * discard requests should ever use it.
1350 void blk_add_request_payload(struct request *rq, struct page *page,
1353 struct bio *bio = rq->bio;
1355 bio->bi_io_vec->bv_page = page;
1356 bio->bi_io_vec->bv_offset = 0;
1357 bio->bi_io_vec->bv_len = len;
1359 bio->bi_iter.bi_size = len;
1361 bio->bi_phys_segments = 1;
1363 rq->__data_len = rq->resid_len = len;
1364 rq->nr_phys_segments = 1;
1366 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1368 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1371 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1373 if (!ll_back_merge_fn(q, req, bio))
1376 trace_block_bio_backmerge(q, req, bio);
1378 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1379 blk_rq_set_mixed_merge(req);
1381 req->biotail->bi_next = bio;
1383 req->__data_len += bio->bi_iter.bi_size;
1384 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1386 blk_account_io_start(req, false);
1390 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1393 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1395 if (!ll_front_merge_fn(q, req, bio))
1398 trace_block_bio_frontmerge(q, req, bio);
1400 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1401 blk_rq_set_mixed_merge(req);
1403 bio->bi_next = req->bio;
1406 req->__sector = bio->bi_iter.bi_sector;
1407 req->__data_len += bio->bi_iter.bi_size;
1408 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1410 blk_account_io_start(req, false);
1415 * blk_attempt_plug_merge - try to merge with %current's plugged list
1416 * @q: request_queue new bio is being queued at
1417 * @bio: new bio being queued
1418 * @request_count: out parameter for number of traversed plugged requests
1420 * Determine whether @bio being queued on @q can be merged with a request
1421 * on %current's plugged list. Returns %true if merge was successful,
1424 * Plugging coalesces IOs from the same issuer for the same purpose without
1425 * going through @q->queue_lock. As such it's more of an issuing mechanism
1426 * than scheduling, and the request, while may have elvpriv data, is not
1427 * added on the elevator at this point. In addition, we don't have
1428 * reliable access to the elevator outside queue lock. Only check basic
1429 * merging parameters without querying the elevator.
1431 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1432 unsigned int *request_count)
1434 struct blk_plug *plug;
1437 struct list_head *plug_list;
1439 if (blk_queue_nomerges(q))
1442 plug = current->plug;
1448 plug_list = &plug->mq_list;
1450 plug_list = &plug->list;
1452 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1458 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1461 el_ret = blk_try_merge(rq, bio);
1462 if (el_ret == ELEVATOR_BACK_MERGE) {
1463 ret = bio_attempt_back_merge(q, rq, bio);
1466 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1467 ret = bio_attempt_front_merge(q, rq, bio);
1476 void init_request_from_bio(struct request *req, struct bio *bio)
1478 req->cmd_type = REQ_TYPE_FS;
1480 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1481 if (bio->bi_rw & REQ_RAHEAD)
1482 req->cmd_flags |= REQ_FAILFAST_MASK;
1485 req->__sector = bio->bi_iter.bi_sector;
1486 req->ioprio = bio_prio(bio);
1487 blk_rq_bio_prep(req->q, req, bio);
1490 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1492 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1493 struct blk_plug *plug;
1494 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1495 struct request *req;
1496 unsigned int request_count = 0;
1499 * low level driver can indicate that it wants pages above a
1500 * certain limit bounced to low memory (ie for highmem, or even
1501 * ISA dma in theory)
1503 blk_queue_bounce(q, &bio);
1505 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1506 bio_endio(bio, -EIO);
1510 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1511 spin_lock_irq(q->queue_lock);
1512 where = ELEVATOR_INSERT_FLUSH;
1517 * Check if we can merge with the plugged list before grabbing
1520 if (blk_attempt_plug_merge(q, bio, &request_count))
1523 spin_lock_irq(q->queue_lock);
1525 el_ret = elv_merge(q, &req, bio);
1526 if (el_ret == ELEVATOR_BACK_MERGE) {
1527 if (bio_attempt_back_merge(q, req, bio)) {
1528 elv_bio_merged(q, req, bio);
1529 if (!attempt_back_merge(q, req))
1530 elv_merged_request(q, req, el_ret);
1533 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1534 if (bio_attempt_front_merge(q, req, bio)) {
1535 elv_bio_merged(q, req, bio);
1536 if (!attempt_front_merge(q, req))
1537 elv_merged_request(q, req, el_ret);
1544 * This sync check and mask will be re-done in init_request_from_bio(),
1545 * but we need to set it earlier to expose the sync flag to the
1546 * rq allocator and io schedulers.
1548 rw_flags = bio_data_dir(bio);
1550 rw_flags |= REQ_SYNC;
1553 * Grab a free request. This is might sleep but can not fail.
1554 * Returns with the queue unlocked.
1556 req = get_request(q, rw_flags, bio, GFP_NOIO);
1557 if (unlikely(!req)) {
1558 bio_endio(bio, -ENODEV); /* @q is dead */
1563 * After dropping the lock and possibly sleeping here, our request
1564 * may now be mergeable after it had proven unmergeable (above).
1565 * We don't worry about that case for efficiency. It won't happen
1566 * often, and the elevators are able to handle it.
1568 init_request_from_bio(req, bio);
1570 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1571 req->cpu = raw_smp_processor_id();
1573 plug = current->plug;
1576 * If this is the first request added after a plug, fire
1580 trace_block_plug(q);
1582 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1583 blk_flush_plug_list(plug, false);
1584 trace_block_plug(q);
1587 list_add_tail(&req->queuelist, &plug->list);
1588 blk_account_io_start(req, true);
1590 spin_lock_irq(q->queue_lock);
1591 add_acct_request(q, req, where);
1594 spin_unlock_irq(q->queue_lock);
1597 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1600 * If bio->bi_dev is a partition, remap the location
1602 static inline void blk_partition_remap(struct bio *bio)
1604 struct block_device *bdev = bio->bi_bdev;
1606 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1607 struct hd_struct *p = bdev->bd_part;
1609 bio->bi_iter.bi_sector += p->start_sect;
1610 bio->bi_bdev = bdev->bd_contains;
1612 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1614 bio->bi_iter.bi_sector - p->start_sect);
1618 static void handle_bad_sector(struct bio *bio)
1620 char b[BDEVNAME_SIZE];
1622 printk(KERN_INFO "attempt to access beyond end of device\n");
1623 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1624 bdevname(bio->bi_bdev, b),
1626 (unsigned long long)bio_end_sector(bio),
1627 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1629 set_bit(BIO_EOF, &bio->bi_flags);
1632 #ifdef CONFIG_FAIL_MAKE_REQUEST
1634 static DECLARE_FAULT_ATTR(fail_make_request);
1636 static int __init setup_fail_make_request(char *str)
1638 return setup_fault_attr(&fail_make_request, str);
1640 __setup("fail_make_request=", setup_fail_make_request);
1642 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1644 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1647 static int __init fail_make_request_debugfs(void)
1649 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1650 NULL, &fail_make_request);
1652 return PTR_ERR_OR_ZERO(dir);
1655 late_initcall(fail_make_request_debugfs);
1657 #else /* CONFIG_FAIL_MAKE_REQUEST */
1659 static inline bool should_fail_request(struct hd_struct *part,
1665 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1668 * Check whether this bio extends beyond the end of the device.
1670 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1677 /* Test device or partition size, when known. */
1678 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1680 sector_t sector = bio->bi_iter.bi_sector;
1682 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1684 * This may well happen - the kernel calls bread()
1685 * without checking the size of the device, e.g., when
1686 * mounting a device.
1688 handle_bad_sector(bio);
1696 static noinline_for_stack bool
1697 generic_make_request_checks(struct bio *bio)
1699 struct request_queue *q;
1700 int nr_sectors = bio_sectors(bio);
1702 char b[BDEVNAME_SIZE];
1703 struct hd_struct *part;
1707 if (bio_check_eod(bio, nr_sectors))
1710 q = bdev_get_queue(bio->bi_bdev);
1713 "generic_make_request: Trying to access "
1714 "nonexistent block-device %s (%Lu)\n",
1715 bdevname(bio->bi_bdev, b),
1716 (long long) bio->bi_iter.bi_sector);
1720 if (likely(bio_is_rw(bio) &&
1721 nr_sectors > queue_max_hw_sectors(q))) {
1722 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1723 bdevname(bio->bi_bdev, b),
1725 queue_max_hw_sectors(q));
1729 part = bio->bi_bdev->bd_part;
1730 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1731 should_fail_request(&part_to_disk(part)->part0,
1732 bio->bi_iter.bi_size))
1736 * If this device has partitions, remap block n
1737 * of partition p to block n+start(p) of the disk.
1739 blk_partition_remap(bio);
1741 if (bio_check_eod(bio, nr_sectors))
1745 * Filter flush bio's early so that make_request based
1746 * drivers without flush support don't have to worry
1749 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1750 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1757 if ((bio->bi_rw & REQ_DISCARD) &&
1758 (!blk_queue_discard(q) ||
1759 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1764 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1770 * Various block parts want %current->io_context and lazy ioc
1771 * allocation ends up trading a lot of pain for a small amount of
1772 * memory. Just allocate it upfront. This may fail and block
1773 * layer knows how to live with it.
1775 create_io_context(GFP_ATOMIC, q->node);
1777 if (blk_throtl_bio(q, bio))
1778 return false; /* throttled, will be resubmitted later */
1780 trace_block_bio_queue(q, bio);
1784 bio_endio(bio, err);
1789 * generic_make_request - hand a buffer to its device driver for I/O
1790 * @bio: The bio describing the location in memory and on the device.
1792 * generic_make_request() is used to make I/O requests of block
1793 * devices. It is passed a &struct bio, which describes the I/O that needs
1796 * generic_make_request() does not return any status. The
1797 * success/failure status of the request, along with notification of
1798 * completion, is delivered asynchronously through the bio->bi_end_io
1799 * function described (one day) else where.
1801 * The caller of generic_make_request must make sure that bi_io_vec
1802 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1803 * set to describe the device address, and the
1804 * bi_end_io and optionally bi_private are set to describe how
1805 * completion notification should be signaled.
1807 * generic_make_request and the drivers it calls may use bi_next if this
1808 * bio happens to be merged with someone else, and may resubmit the bio to
1809 * a lower device by calling into generic_make_request recursively, which
1810 * means the bio should NOT be touched after the call to ->make_request_fn.
1812 void generic_make_request(struct bio *bio)
1814 struct bio_list bio_list_on_stack;
1816 if (!generic_make_request_checks(bio))
1820 * We only want one ->make_request_fn to be active at a time, else
1821 * stack usage with stacked devices could be a problem. So use
1822 * current->bio_list to keep a list of requests submited by a
1823 * make_request_fn function. current->bio_list is also used as a
1824 * flag to say if generic_make_request is currently active in this
1825 * task or not. If it is NULL, then no make_request is active. If
1826 * it is non-NULL, then a make_request is active, and new requests
1827 * should be added at the tail
1829 if (current->bio_list) {
1830 bio_list_add(current->bio_list, bio);
1834 /* following loop may be a bit non-obvious, and so deserves some
1836 * Before entering the loop, bio->bi_next is NULL (as all callers
1837 * ensure that) so we have a list with a single bio.
1838 * We pretend that we have just taken it off a longer list, so
1839 * we assign bio_list to a pointer to the bio_list_on_stack,
1840 * thus initialising the bio_list of new bios to be
1841 * added. ->make_request() may indeed add some more bios
1842 * through a recursive call to generic_make_request. If it
1843 * did, we find a non-NULL value in bio_list and re-enter the loop
1844 * from the top. In this case we really did just take the bio
1845 * of the top of the list (no pretending) and so remove it from
1846 * bio_list, and call into ->make_request() again.
1848 BUG_ON(bio->bi_next);
1849 bio_list_init(&bio_list_on_stack);
1850 current->bio_list = &bio_list_on_stack;
1852 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1854 q->make_request_fn(q, bio);
1856 bio = bio_list_pop(current->bio_list);
1858 current->bio_list = NULL; /* deactivate */
1860 EXPORT_SYMBOL(generic_make_request);
1863 * submit_bio - submit a bio to the block device layer for I/O
1864 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1865 * @bio: The &struct bio which describes the I/O
1867 * submit_bio() is very similar in purpose to generic_make_request(), and
1868 * uses that function to do most of the work. Both are fairly rough
1869 * interfaces; @bio must be presetup and ready for I/O.
1872 void submit_bio(int rw, struct bio *bio)
1877 * If it's a regular read/write or a barrier with data attached,
1878 * go through the normal accounting stuff before submission.
1880 if (bio_has_data(bio)) {
1883 if (unlikely(rw & REQ_WRITE_SAME))
1884 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1886 count = bio_sectors(bio);
1889 count_vm_events(PGPGOUT, count);
1891 task_io_account_read(bio->bi_iter.bi_size);
1892 count_vm_events(PGPGIN, count);
1895 if (unlikely(block_dump)) {
1896 char b[BDEVNAME_SIZE];
1897 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1898 current->comm, task_pid_nr(current),
1899 (rw & WRITE) ? "WRITE" : "READ",
1900 (unsigned long long)bio->bi_iter.bi_sector,
1901 bdevname(bio->bi_bdev, b),
1906 generic_make_request(bio);
1908 EXPORT_SYMBOL(submit_bio);
1911 * blk_rq_check_limits - Helper function to check a request for the queue limit
1913 * @rq: the request being checked
1916 * @rq may have been made based on weaker limitations of upper-level queues
1917 * in request stacking drivers, and it may violate the limitation of @q.
1918 * Since the block layer and the underlying device driver trust @rq
1919 * after it is inserted to @q, it should be checked against @q before
1920 * the insertion using this generic function.
1922 * This function should also be useful for request stacking drivers
1923 * in some cases below, so export this function.
1924 * Request stacking drivers like request-based dm may change the queue
1925 * limits while requests are in the queue (e.g. dm's table swapping).
1926 * Such request stacking drivers should check those requests against
1927 * the new queue limits again when they dispatch those requests,
1928 * although such checkings are also done against the old queue limits
1929 * when submitting requests.
1931 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1933 if (!rq_mergeable(rq))
1936 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1937 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1942 * queue's settings related to segment counting like q->bounce_pfn
1943 * may differ from that of other stacking queues.
1944 * Recalculate it to check the request correctly on this queue's
1947 blk_recalc_rq_segments(rq);
1948 if (rq->nr_phys_segments > queue_max_segments(q)) {
1949 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1955 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1958 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1959 * @q: the queue to submit the request
1960 * @rq: the request being queued
1962 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1964 unsigned long flags;
1965 int where = ELEVATOR_INSERT_BACK;
1967 if (blk_rq_check_limits(q, rq))
1971 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1974 spin_lock_irqsave(q->queue_lock, flags);
1975 if (unlikely(blk_queue_dying(q))) {
1976 spin_unlock_irqrestore(q->queue_lock, flags);
1981 * Submitting request must be dequeued before calling this function
1982 * because it will be linked to another request_queue
1984 BUG_ON(blk_queued_rq(rq));
1986 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1987 where = ELEVATOR_INSERT_FLUSH;
1989 add_acct_request(q, rq, where);
1990 if (where == ELEVATOR_INSERT_FLUSH)
1992 spin_unlock_irqrestore(q->queue_lock, flags);
1996 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1999 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2000 * @rq: request to examine
2003 * A request could be merge of IOs which require different failure
2004 * handling. This function determines the number of bytes which
2005 * can be failed from the beginning of the request without
2006 * crossing into area which need to be retried further.
2009 * The number of bytes to fail.
2012 * queue_lock must be held.
2014 unsigned int blk_rq_err_bytes(const struct request *rq)
2016 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2017 unsigned int bytes = 0;
2020 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2021 return blk_rq_bytes(rq);
2024 * Currently the only 'mixing' which can happen is between
2025 * different fastfail types. We can safely fail portions
2026 * which have all the failfast bits that the first one has -
2027 * the ones which are at least as eager to fail as the first
2030 for (bio = rq->bio; bio; bio = bio->bi_next) {
2031 if ((bio->bi_rw & ff) != ff)
2033 bytes += bio->bi_iter.bi_size;
2036 /* this could lead to infinite loop */
2037 BUG_ON(blk_rq_bytes(rq) && !bytes);
2040 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2042 void blk_account_io_completion(struct request *req, unsigned int bytes)
2044 if (blk_do_io_stat(req)) {
2045 const int rw = rq_data_dir(req);
2046 struct hd_struct *part;
2049 cpu = part_stat_lock();
2051 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2056 void blk_account_io_done(struct request *req)
2059 * Account IO completion. flush_rq isn't accounted as a
2060 * normal IO on queueing nor completion. Accounting the
2061 * containing request is enough.
2063 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2064 unsigned long duration = jiffies - req->start_time;
2065 const int rw = rq_data_dir(req);
2066 struct hd_struct *part;
2069 cpu = part_stat_lock();
2072 part_stat_inc(cpu, part, ios[rw]);
2073 part_stat_add(cpu, part, ticks[rw], duration);
2074 part_round_stats(cpu, part);
2075 part_dec_in_flight(part, rw);
2077 hd_struct_put(part);
2082 #ifdef CONFIG_PM_RUNTIME
2084 * Don't process normal requests when queue is suspended
2085 * or in the process of suspending/resuming
2087 static struct request *blk_pm_peek_request(struct request_queue *q,
2090 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2091 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2097 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2104 void blk_account_io_start(struct request *rq, bool new_io)
2106 struct hd_struct *part;
2107 int rw = rq_data_dir(rq);
2110 if (!blk_do_io_stat(rq))
2113 cpu = part_stat_lock();
2117 part_stat_inc(cpu, part, merges[rw]);
2119 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2120 if (!hd_struct_try_get(part)) {
2122 * The partition is already being removed,
2123 * the request will be accounted on the disk only
2125 * We take a reference on disk->part0 although that
2126 * partition will never be deleted, so we can treat
2127 * it as any other partition.
2129 part = &rq->rq_disk->part0;
2130 hd_struct_get(part);
2132 part_round_stats(cpu, part);
2133 part_inc_in_flight(part, rw);
2141 * blk_peek_request - peek at the top of a request queue
2142 * @q: request queue to peek at
2145 * Return the request at the top of @q. The returned request
2146 * should be started using blk_start_request() before LLD starts
2150 * Pointer to the request at the top of @q if available. Null
2154 * queue_lock must be held.
2156 struct request *blk_peek_request(struct request_queue *q)
2161 while ((rq = __elv_next_request(q)) != NULL) {
2163 rq = blk_pm_peek_request(q, rq);
2167 if (!(rq->cmd_flags & REQ_STARTED)) {
2169 * This is the first time the device driver
2170 * sees this request (possibly after
2171 * requeueing). Notify IO scheduler.
2173 if (rq->cmd_flags & REQ_SORTED)
2174 elv_activate_rq(q, rq);
2177 * just mark as started even if we don't start
2178 * it, a request that has been delayed should
2179 * not be passed by new incoming requests
2181 rq->cmd_flags |= REQ_STARTED;
2182 trace_block_rq_issue(q, rq);
2185 if (!q->boundary_rq || q->boundary_rq == rq) {
2186 q->end_sector = rq_end_sector(rq);
2187 q->boundary_rq = NULL;
2190 if (rq->cmd_flags & REQ_DONTPREP)
2193 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2195 * make sure space for the drain appears we
2196 * know we can do this because max_hw_segments
2197 * has been adjusted to be one fewer than the
2200 rq->nr_phys_segments++;
2206 ret = q->prep_rq_fn(q, rq);
2207 if (ret == BLKPREP_OK) {
2209 } else if (ret == BLKPREP_DEFER) {
2211 * the request may have been (partially) prepped.
2212 * we need to keep this request in the front to
2213 * avoid resource deadlock. REQ_STARTED will
2214 * prevent other fs requests from passing this one.
2216 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2217 !(rq->cmd_flags & REQ_DONTPREP)) {
2219 * remove the space for the drain we added
2220 * so that we don't add it again
2222 --rq->nr_phys_segments;
2227 } else if (ret == BLKPREP_KILL) {
2228 rq->cmd_flags |= REQ_QUIET;
2230 * Mark this request as started so we don't trigger
2231 * any debug logic in the end I/O path.
2233 blk_start_request(rq);
2234 __blk_end_request_all(rq, -EIO);
2236 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2243 EXPORT_SYMBOL(blk_peek_request);
2245 void blk_dequeue_request(struct request *rq)
2247 struct request_queue *q = rq->q;
2249 BUG_ON(list_empty(&rq->queuelist));
2250 BUG_ON(ELV_ON_HASH(rq));
2252 list_del_init(&rq->queuelist);
2255 * the time frame between a request being removed from the lists
2256 * and to it is freed is accounted as io that is in progress at
2259 if (blk_account_rq(rq)) {
2260 q->in_flight[rq_is_sync(rq)]++;
2261 set_io_start_time_ns(rq);
2266 * blk_start_request - start request processing on the driver
2267 * @req: request to dequeue
2270 * Dequeue @req and start timeout timer on it. This hands off the
2271 * request to the driver.
2273 * Block internal functions which don't want to start timer should
2274 * call blk_dequeue_request().
2277 * queue_lock must be held.
2279 void blk_start_request(struct request *req)
2281 blk_dequeue_request(req);
2284 * We are now handing the request to the hardware, initialize
2285 * resid_len to full count and add the timeout handler.
2287 req->resid_len = blk_rq_bytes(req);
2288 if (unlikely(blk_bidi_rq(req)))
2289 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2291 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2294 EXPORT_SYMBOL(blk_start_request);
2297 * blk_fetch_request - fetch a request from a request queue
2298 * @q: request queue to fetch a request from
2301 * Return the request at the top of @q. The request is started on
2302 * return and LLD can start processing it immediately.
2305 * Pointer to the request at the top of @q if available. Null
2309 * queue_lock must be held.
2311 struct request *blk_fetch_request(struct request_queue *q)
2315 rq = blk_peek_request(q);
2317 blk_start_request(rq);
2320 EXPORT_SYMBOL(blk_fetch_request);
2323 * blk_update_request - Special helper function for request stacking drivers
2324 * @req: the request being processed
2325 * @error: %0 for success, < %0 for error
2326 * @nr_bytes: number of bytes to complete @req
2329 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2330 * the request structure even if @req doesn't have leftover.
2331 * If @req has leftover, sets it up for the next range of segments.
2333 * This special helper function is only for request stacking drivers
2334 * (e.g. request-based dm) so that they can handle partial completion.
2335 * Actual device drivers should use blk_end_request instead.
2337 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2338 * %false return from this function.
2341 * %false - this request doesn't have any more data
2342 * %true - this request has more data
2344 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2351 trace_block_rq_complete(req->q, req, nr_bytes);
2354 * For fs requests, rq is just carrier of independent bio's
2355 * and each partial completion should be handled separately.
2356 * Reset per-request error on each partial completion.
2358 * TODO: tj: This is too subtle. It would be better to let
2359 * low level drivers do what they see fit.
2361 if (req->cmd_type == REQ_TYPE_FS)
2364 if (error && req->cmd_type == REQ_TYPE_FS &&
2365 !(req->cmd_flags & REQ_QUIET)) {
2370 error_type = "recoverable transport";
2373 error_type = "critical target";
2376 error_type = "critical nexus";
2379 error_type = "timeout";
2382 error_type = "critical space allocation";
2385 error_type = "critical medium";
2392 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2393 error_type, req->rq_disk ?
2394 req->rq_disk->disk_name : "?",
2395 (unsigned long long)blk_rq_pos(req));
2399 blk_account_io_completion(req, nr_bytes);
2403 struct bio *bio = req->bio;
2404 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2406 if (bio_bytes == bio->bi_iter.bi_size)
2407 req->bio = bio->bi_next;
2409 req_bio_endio(req, bio, bio_bytes, error);
2411 total_bytes += bio_bytes;
2412 nr_bytes -= bio_bytes;
2423 * Reset counters so that the request stacking driver
2424 * can find how many bytes remain in the request
2427 req->__data_len = 0;
2431 req->__data_len -= total_bytes;
2433 /* update sector only for requests with clear definition of sector */
2434 if (req->cmd_type == REQ_TYPE_FS)
2435 req->__sector += total_bytes >> 9;
2437 /* mixed attributes always follow the first bio */
2438 if (req->cmd_flags & REQ_MIXED_MERGE) {
2439 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2440 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2444 * If total number of sectors is less than the first segment
2445 * size, something has gone terribly wrong.
2447 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2448 blk_dump_rq_flags(req, "request botched");
2449 req->__data_len = blk_rq_cur_bytes(req);
2452 /* recalculate the number of segments */
2453 blk_recalc_rq_segments(req);
2457 EXPORT_SYMBOL_GPL(blk_update_request);
2459 static bool blk_update_bidi_request(struct request *rq, int error,
2460 unsigned int nr_bytes,
2461 unsigned int bidi_bytes)
2463 if (blk_update_request(rq, error, nr_bytes))
2466 /* Bidi request must be completed as a whole */
2467 if (unlikely(blk_bidi_rq(rq)) &&
2468 blk_update_request(rq->next_rq, error, bidi_bytes))
2471 if (blk_queue_add_random(rq->q))
2472 add_disk_randomness(rq->rq_disk);
2478 * blk_unprep_request - unprepare a request
2481 * This function makes a request ready for complete resubmission (or
2482 * completion). It happens only after all error handling is complete,
2483 * so represents the appropriate moment to deallocate any resources
2484 * that were allocated to the request in the prep_rq_fn. The queue
2485 * lock is held when calling this.
2487 void blk_unprep_request(struct request *req)
2489 struct request_queue *q = req->q;
2491 req->cmd_flags &= ~REQ_DONTPREP;
2492 if (q->unprep_rq_fn)
2493 q->unprep_rq_fn(q, req);
2495 EXPORT_SYMBOL_GPL(blk_unprep_request);
2498 * queue lock must be held
2500 static void blk_finish_request(struct request *req, int error)
2502 if (blk_rq_tagged(req))
2503 blk_queue_end_tag(req->q, req);
2505 BUG_ON(blk_queued_rq(req));
2507 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2508 laptop_io_completion(&req->q->backing_dev_info);
2510 blk_delete_timer(req);
2512 if (req->cmd_flags & REQ_DONTPREP)
2513 blk_unprep_request(req);
2515 blk_account_io_done(req);
2518 req->end_io(req, error);
2520 if (blk_bidi_rq(req))
2521 __blk_put_request(req->next_rq->q, req->next_rq);
2523 __blk_put_request(req->q, req);
2528 * blk_end_bidi_request - Complete a bidi request
2529 * @rq: the request to complete
2530 * @error: %0 for success, < %0 for error
2531 * @nr_bytes: number of bytes to complete @rq
2532 * @bidi_bytes: number of bytes to complete @rq->next_rq
2535 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2536 * Drivers that supports bidi can safely call this member for any
2537 * type of request, bidi or uni. In the later case @bidi_bytes is
2541 * %false - we are done with this request
2542 * %true - still buffers pending for this request
2544 static bool blk_end_bidi_request(struct request *rq, int error,
2545 unsigned int nr_bytes, unsigned int bidi_bytes)
2547 struct request_queue *q = rq->q;
2548 unsigned long flags;
2550 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2553 spin_lock_irqsave(q->queue_lock, flags);
2554 blk_finish_request(rq, error);
2555 spin_unlock_irqrestore(q->queue_lock, flags);
2561 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2562 * @rq: the request to complete
2563 * @error: %0 for success, < %0 for error
2564 * @nr_bytes: number of bytes to complete @rq
2565 * @bidi_bytes: number of bytes to complete @rq->next_rq
2568 * Identical to blk_end_bidi_request() except that queue lock is
2569 * assumed to be locked on entry and remains so on return.
2572 * %false - we are done with this request
2573 * %true - still buffers pending for this request
2575 bool __blk_end_bidi_request(struct request *rq, int error,
2576 unsigned int nr_bytes, unsigned int bidi_bytes)
2578 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2581 blk_finish_request(rq, error);
2587 * blk_end_request - Helper function for drivers to complete the request.
2588 * @rq: the request being processed
2589 * @error: %0 for success, < %0 for error
2590 * @nr_bytes: number of bytes to complete
2593 * Ends I/O on a number of bytes attached to @rq.
2594 * If @rq has leftover, sets it up for the next range of segments.
2597 * %false - we are done with this request
2598 * %true - still buffers pending for this request
2600 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2602 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2604 EXPORT_SYMBOL(blk_end_request);
2607 * blk_end_request_all - Helper function for drives to finish the request.
2608 * @rq: the request to finish
2609 * @error: %0 for success, < %0 for error
2612 * Completely finish @rq.
2614 void blk_end_request_all(struct request *rq, int error)
2617 unsigned int bidi_bytes = 0;
2619 if (unlikely(blk_bidi_rq(rq)))
2620 bidi_bytes = blk_rq_bytes(rq->next_rq);
2622 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2625 EXPORT_SYMBOL(blk_end_request_all);
2628 * blk_end_request_cur - Helper function to finish the current request chunk.
2629 * @rq: the request to finish the current chunk for
2630 * @error: %0 for success, < %0 for error
2633 * Complete the current consecutively mapped chunk from @rq.
2636 * %false - we are done with this request
2637 * %true - still buffers pending for this request
2639 bool blk_end_request_cur(struct request *rq, int error)
2641 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2643 EXPORT_SYMBOL(blk_end_request_cur);
2646 * blk_end_request_err - Finish a request till the next failure boundary.
2647 * @rq: the request to finish till the next failure boundary for
2648 * @error: must be negative errno
2651 * Complete @rq till the next failure boundary.
2654 * %false - we are done with this request
2655 * %true - still buffers pending for this request
2657 bool blk_end_request_err(struct request *rq, int error)
2659 WARN_ON(error >= 0);
2660 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2662 EXPORT_SYMBOL_GPL(blk_end_request_err);
2665 * __blk_end_request - Helper function for drivers to complete the request.
2666 * @rq: the request being processed
2667 * @error: %0 for success, < %0 for error
2668 * @nr_bytes: number of bytes to complete
2671 * Must be called with queue lock held unlike blk_end_request().
2674 * %false - we are done with this request
2675 * %true - still buffers pending for this request
2677 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2679 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2681 EXPORT_SYMBOL(__blk_end_request);
2684 * __blk_end_request_all - Helper function for drives to finish the request.
2685 * @rq: the request to finish
2686 * @error: %0 for success, < %0 for error
2689 * Completely finish @rq. Must be called with queue lock held.
2691 void __blk_end_request_all(struct request *rq, int error)
2694 unsigned int bidi_bytes = 0;
2696 if (unlikely(blk_bidi_rq(rq)))
2697 bidi_bytes = blk_rq_bytes(rq->next_rq);
2699 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2702 EXPORT_SYMBOL(__blk_end_request_all);
2705 * __blk_end_request_cur - Helper function to finish the current request chunk.
2706 * @rq: the request to finish the current chunk for
2707 * @error: %0 for success, < %0 for error
2710 * Complete the current consecutively mapped chunk from @rq. Must
2711 * be called with queue lock held.
2714 * %false - we are done with this request
2715 * %true - still buffers pending for this request
2717 bool __blk_end_request_cur(struct request *rq, int error)
2719 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2721 EXPORT_SYMBOL(__blk_end_request_cur);
2724 * __blk_end_request_err - Finish a request till the next failure boundary.
2725 * @rq: the request to finish till the next failure boundary for
2726 * @error: must be negative errno
2729 * Complete @rq till the next failure boundary. Must be called
2730 * with queue lock held.
2733 * %false - we are done with this request
2734 * %true - still buffers pending for this request
2736 bool __blk_end_request_err(struct request *rq, int error)
2738 WARN_ON(error >= 0);
2739 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2741 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2743 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2746 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2747 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2749 if (bio_has_data(bio))
2750 rq->nr_phys_segments = bio_phys_segments(q, bio);
2752 rq->__data_len = bio->bi_iter.bi_size;
2753 rq->bio = rq->biotail = bio;
2756 rq->rq_disk = bio->bi_bdev->bd_disk;
2759 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2761 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2762 * @rq: the request to be flushed
2765 * Flush all pages in @rq.
2767 void rq_flush_dcache_pages(struct request *rq)
2769 struct req_iterator iter;
2770 struct bio_vec bvec;
2772 rq_for_each_segment(bvec, rq, iter)
2773 flush_dcache_page(bvec.bv_page);
2775 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2779 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2780 * @q : the queue of the device being checked
2783 * Check if underlying low-level drivers of a device are busy.
2784 * If the drivers want to export their busy state, they must set own
2785 * exporting function using blk_queue_lld_busy() first.
2787 * Basically, this function is used only by request stacking drivers
2788 * to stop dispatching requests to underlying devices when underlying
2789 * devices are busy. This behavior helps more I/O merging on the queue
2790 * of the request stacking driver and prevents I/O throughput regression
2791 * on burst I/O load.
2794 * 0 - Not busy (The request stacking driver should dispatch request)
2795 * 1 - Busy (The request stacking driver should stop dispatching request)
2797 int blk_lld_busy(struct request_queue *q)
2800 return q->lld_busy_fn(q);
2804 EXPORT_SYMBOL_GPL(blk_lld_busy);
2807 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2808 * @rq: the clone request to be cleaned up
2811 * Free all bios in @rq for a cloned request.
2813 void blk_rq_unprep_clone(struct request *rq)
2817 while ((bio = rq->bio) != NULL) {
2818 rq->bio = bio->bi_next;
2823 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2826 * Copy attributes of the original request to the clone request.
2827 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2829 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2831 dst->cpu = src->cpu;
2832 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2833 dst->cmd_type = src->cmd_type;
2834 dst->__sector = blk_rq_pos(src);
2835 dst->__data_len = blk_rq_bytes(src);
2836 dst->nr_phys_segments = src->nr_phys_segments;
2837 dst->ioprio = src->ioprio;
2838 dst->extra_len = src->extra_len;
2842 * blk_rq_prep_clone - Helper function to setup clone request
2843 * @rq: the request to be setup
2844 * @rq_src: original request to be cloned
2845 * @bs: bio_set that bios for clone are allocated from
2846 * @gfp_mask: memory allocation mask for bio
2847 * @bio_ctr: setup function to be called for each clone bio.
2848 * Returns %0 for success, non %0 for failure.
2849 * @data: private data to be passed to @bio_ctr
2852 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2853 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2854 * are not copied, and copying such parts is the caller's responsibility.
2855 * Also, pages which the original bios are pointing to are not copied
2856 * and the cloned bios just point same pages.
2857 * So cloned bios must be completed before original bios, which means
2858 * the caller must complete @rq before @rq_src.
2860 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2861 struct bio_set *bs, gfp_t gfp_mask,
2862 int (*bio_ctr)(struct bio *, struct bio *, void *),
2865 struct bio *bio, *bio_src;
2870 blk_rq_init(NULL, rq);
2872 __rq_for_each_bio(bio_src, rq_src) {
2873 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2877 if (bio_ctr && bio_ctr(bio, bio_src, data))
2881 rq->biotail->bi_next = bio;
2884 rq->bio = rq->biotail = bio;
2887 __blk_rq_prep_clone(rq, rq_src);
2894 blk_rq_unprep_clone(rq);
2898 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2900 int kblockd_schedule_work(struct work_struct *work)
2902 return queue_work(kblockd_workqueue, work);
2904 EXPORT_SYMBOL(kblockd_schedule_work);
2906 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
2907 unsigned long delay)
2909 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2911 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2913 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
2914 unsigned long delay)
2916 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
2918 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
2920 #define PLUG_MAGIC 0x91827364
2923 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2924 * @plug: The &struct blk_plug that needs to be initialized
2927 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2928 * pending I/O should the task end up blocking between blk_start_plug() and
2929 * blk_finish_plug(). This is important from a performance perspective, but
2930 * also ensures that we don't deadlock. For instance, if the task is blocking
2931 * for a memory allocation, memory reclaim could end up wanting to free a
2932 * page belonging to that request that is currently residing in our private
2933 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2934 * this kind of deadlock.
2936 void blk_start_plug(struct blk_plug *plug)
2938 struct task_struct *tsk = current;
2940 plug->magic = PLUG_MAGIC;
2941 INIT_LIST_HEAD(&plug->list);
2942 INIT_LIST_HEAD(&plug->mq_list);
2943 INIT_LIST_HEAD(&plug->cb_list);
2946 * If this is a nested plug, don't actually assign it. It will be
2947 * flushed on its own.
2951 * Store ordering should not be needed here, since a potential
2952 * preempt will imply a full memory barrier
2957 EXPORT_SYMBOL(blk_start_plug);
2959 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2961 struct request *rqa = container_of(a, struct request, queuelist);
2962 struct request *rqb = container_of(b, struct request, queuelist);
2964 return !(rqa->q < rqb->q ||
2965 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2969 * If 'from_schedule' is true, then postpone the dispatch of requests
2970 * until a safe kblockd context. We due this to avoid accidental big
2971 * additional stack usage in driver dispatch, in places where the originally
2972 * plugger did not intend it.
2974 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2976 __releases(q->queue_lock)
2978 trace_block_unplug(q, depth, !from_schedule);
2981 blk_run_queue_async(q);
2984 spin_unlock(q->queue_lock);
2987 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2989 LIST_HEAD(callbacks);
2991 while (!list_empty(&plug->cb_list)) {
2992 list_splice_init(&plug->cb_list, &callbacks);
2994 while (!list_empty(&callbacks)) {
2995 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2998 list_del(&cb->list);
2999 cb->callback(cb, from_schedule);
3004 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3007 struct blk_plug *plug = current->plug;
3008 struct blk_plug_cb *cb;
3013 list_for_each_entry(cb, &plug->cb_list, list)
3014 if (cb->callback == unplug && cb->data == data)
3017 /* Not currently on the callback list */
3018 BUG_ON(size < sizeof(*cb));
3019 cb = kzalloc(size, GFP_ATOMIC);
3022 cb->callback = unplug;
3023 list_add(&cb->list, &plug->cb_list);
3027 EXPORT_SYMBOL(blk_check_plugged);
3029 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3031 struct request_queue *q;
3032 unsigned long flags;
3037 BUG_ON(plug->magic != PLUG_MAGIC);
3039 flush_plug_callbacks(plug, from_schedule);
3041 if (!list_empty(&plug->mq_list))
3042 blk_mq_flush_plug_list(plug, from_schedule);
3044 if (list_empty(&plug->list))
3047 list_splice_init(&plug->list, &list);
3049 list_sort(NULL, &list, plug_rq_cmp);
3055 * Save and disable interrupts here, to avoid doing it for every
3056 * queue lock we have to take.
3058 local_irq_save(flags);
3059 while (!list_empty(&list)) {
3060 rq = list_entry_rq(list.next);
3061 list_del_init(&rq->queuelist);
3065 * This drops the queue lock
3068 queue_unplugged(q, depth, from_schedule);
3071 spin_lock(q->queue_lock);
3075 * Short-circuit if @q is dead
3077 if (unlikely(blk_queue_dying(q))) {
3078 __blk_end_request_all(rq, -ENODEV);
3083 * rq is already accounted, so use raw insert
3085 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3086 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3088 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3094 * This drops the queue lock
3097 queue_unplugged(q, depth, from_schedule);
3099 local_irq_restore(flags);
3102 void blk_finish_plug(struct blk_plug *plug)
3104 blk_flush_plug_list(plug, false);
3106 if (plug == current->plug)
3107 current->plug = NULL;
3109 EXPORT_SYMBOL(blk_finish_plug);
3111 #ifdef CONFIG_PM_RUNTIME
3113 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3114 * @q: the queue of the device
3115 * @dev: the device the queue belongs to
3118 * Initialize runtime-PM-related fields for @q and start auto suspend for
3119 * @dev. Drivers that want to take advantage of request-based runtime PM
3120 * should call this function after @dev has been initialized, and its
3121 * request queue @q has been allocated, and runtime PM for it can not happen
3122 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3123 * cases, driver should call this function before any I/O has taken place.
3125 * This function takes care of setting up using auto suspend for the device,
3126 * the autosuspend delay is set to -1 to make runtime suspend impossible
3127 * until an updated value is either set by user or by driver. Drivers do
3128 * not need to touch other autosuspend settings.
3130 * The block layer runtime PM is request based, so only works for drivers
3131 * that use request as their IO unit instead of those directly use bio's.
3133 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3136 q->rpm_status = RPM_ACTIVE;
3137 pm_runtime_set_autosuspend_delay(q->dev, -1);
3138 pm_runtime_use_autosuspend(q->dev);
3140 EXPORT_SYMBOL(blk_pm_runtime_init);
3143 * blk_pre_runtime_suspend - Pre runtime suspend check
3144 * @q: the queue of the device
3147 * This function will check if runtime suspend is allowed for the device
3148 * by examining if there are any requests pending in the queue. If there
3149 * are requests pending, the device can not be runtime suspended; otherwise,
3150 * the queue's status will be updated to SUSPENDING and the driver can
3151 * proceed to suspend the device.
3153 * For the not allowed case, we mark last busy for the device so that
3154 * runtime PM core will try to autosuspend it some time later.
3156 * This function should be called near the start of the device's
3157 * runtime_suspend callback.
3160 * 0 - OK to runtime suspend the device
3161 * -EBUSY - Device should not be runtime suspended
3163 int blk_pre_runtime_suspend(struct request_queue *q)
3167 spin_lock_irq(q->queue_lock);
3168 if (q->nr_pending) {
3170 pm_runtime_mark_last_busy(q->dev);
3172 q->rpm_status = RPM_SUSPENDING;
3174 spin_unlock_irq(q->queue_lock);
3177 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3180 * blk_post_runtime_suspend - Post runtime suspend processing
3181 * @q: the queue of the device
3182 * @err: return value of the device's runtime_suspend function
3185 * Update the queue's runtime status according to the return value of the
3186 * device's runtime suspend function and mark last busy for the device so
3187 * that PM core will try to auto suspend the device at a later time.
3189 * This function should be called near the end of the device's
3190 * runtime_suspend callback.
3192 void blk_post_runtime_suspend(struct request_queue *q, int err)
3194 spin_lock_irq(q->queue_lock);
3196 q->rpm_status = RPM_SUSPENDED;
3198 q->rpm_status = RPM_ACTIVE;
3199 pm_runtime_mark_last_busy(q->dev);
3201 spin_unlock_irq(q->queue_lock);
3203 EXPORT_SYMBOL(blk_post_runtime_suspend);
3206 * blk_pre_runtime_resume - Pre runtime resume processing
3207 * @q: the queue of the device
3210 * Update the queue's runtime status to RESUMING in preparation for the
3211 * runtime resume of the device.
3213 * This function should be called near the start of the device's
3214 * runtime_resume callback.
3216 void blk_pre_runtime_resume(struct request_queue *q)
3218 spin_lock_irq(q->queue_lock);
3219 q->rpm_status = RPM_RESUMING;
3220 spin_unlock_irq(q->queue_lock);
3222 EXPORT_SYMBOL(blk_pre_runtime_resume);
3225 * blk_post_runtime_resume - Post runtime resume processing
3226 * @q: the queue of the device
3227 * @err: return value of the device's runtime_resume function
3230 * Update the queue's runtime status according to the return value of the
3231 * device's runtime_resume function. If it is successfully resumed, process
3232 * the requests that are queued into the device's queue when it is resuming
3233 * and then mark last busy and initiate autosuspend for it.
3235 * This function should be called near the end of the device's
3236 * runtime_resume callback.
3238 void blk_post_runtime_resume(struct request_queue *q, int err)
3240 spin_lock_irq(q->queue_lock);
3242 q->rpm_status = RPM_ACTIVE;
3244 pm_runtime_mark_last_busy(q->dev);
3245 pm_request_autosuspend(q->dev);
3247 q->rpm_status = RPM_SUSPENDED;
3249 spin_unlock_irq(q->queue_lock);
3251 EXPORT_SYMBOL(blk_post_runtime_resume);
3254 int __init blk_dev_init(void)
3256 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3257 sizeof(((struct request *)0)->cmd_flags));
3259 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3260 kblockd_workqueue = alloc_workqueue("kblockd",
3261 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3262 WQ_POWER_EFFICIENT, 0);
3263 if (!kblockd_workqueue)
3264 panic("Failed to create kblockd\n");
3266 request_cachep = kmem_cache_create("blkdev_requests",
3267 sizeof(struct request), 0, SLAB_PANIC, NULL);
3269 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3270 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);