2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
36 #define CREATE_TRACE_POINTS
37 #include <trace/events/block.h>
40 #include "blk-cgroup.h"
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
49 DEFINE_IDA(blk_queue_ida);
52 * For the allocated request tables
54 struct kmem_cache *request_cachep = NULL;
57 * For queue allocation
59 struct kmem_cache *blk_requestq_cachep;
62 * Controlling structure to kblockd
64 static struct workqueue_struct *kblockd_workqueue;
66 void blk_queue_congestion_threshold(struct request_queue *q)
70 nr = q->nr_requests - (q->nr_requests / 8) + 1;
71 if (nr > q->nr_requests)
73 q->nr_congestion_on = nr;
75 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
78 q->nr_congestion_off = nr;
82 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
85 * Locates the passed device's request queue and returns the address of its
86 * backing_dev_info. This function can only be called if @bdev is opened
87 * and the return value is never NULL.
89 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
91 struct request_queue *q = bdev_get_queue(bdev);
93 return &q->backing_dev_info;
95 EXPORT_SYMBOL(blk_get_backing_dev_info);
97 void blk_rq_init(struct request_queue *q, struct request *rq)
99 memset(rq, 0, sizeof(*rq));
101 INIT_LIST_HEAD(&rq->queuelist);
102 INIT_LIST_HEAD(&rq->timeout_list);
105 rq->__sector = (sector_t) -1;
106 INIT_HLIST_NODE(&rq->hash);
107 RB_CLEAR_NODE(&rq->rb_node);
109 rq->cmd_len = BLK_MAX_CDB;
111 rq->start_time = jiffies;
112 set_start_time_ns(rq);
115 EXPORT_SYMBOL(blk_rq_init);
117 static void req_bio_endio(struct request *rq, struct bio *bio,
118 unsigned int nbytes, int error)
121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
122 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
125 if (unlikely(rq->cmd_flags & REQ_QUIET))
126 set_bit(BIO_QUIET, &bio->bi_flags);
128 bio_advance(bio, nbytes);
130 /* don't actually finish bio if it's part of flush sequence */
131 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
132 bio_endio(bio, error);
135 void blk_dump_rq_flags(struct request *rq, char *msg)
139 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
140 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
141 (unsigned long long) rq->cmd_flags);
143 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
144 (unsigned long long)blk_rq_pos(rq),
145 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
146 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
147 rq->bio, rq->biotail, blk_rq_bytes(rq));
149 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
150 printk(KERN_INFO " cdb: ");
151 for (bit = 0; bit < BLK_MAX_CDB; bit++)
152 printk("%02x ", rq->cmd[bit]);
156 EXPORT_SYMBOL(blk_dump_rq_flags);
158 static void blk_delay_work(struct work_struct *work)
160 struct request_queue *q;
162 q = container_of(work, struct request_queue, delay_work.work);
163 spin_lock_irq(q->queue_lock);
165 spin_unlock_irq(q->queue_lock);
169 * blk_delay_queue - restart queueing after defined interval
170 * @q: The &struct request_queue in question
171 * @msecs: Delay in msecs
174 * Sometimes queueing needs to be postponed for a little while, to allow
175 * resources to come back. This function will make sure that queueing is
176 * restarted around the specified time. Queue lock must be held.
178 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
180 if (likely(!blk_queue_dead(q)))
181 queue_delayed_work(kblockd_workqueue, &q->delay_work,
182 msecs_to_jiffies(msecs));
184 EXPORT_SYMBOL(blk_delay_queue);
187 * blk_start_queue - restart a previously stopped queue
188 * @q: The &struct request_queue in question
191 * blk_start_queue() will clear the stop flag on the queue, and call
192 * the request_fn for the queue if it was in a stopped state when
193 * entered. Also see blk_stop_queue(). Queue lock must be held.
195 void blk_start_queue(struct request_queue *q)
197 WARN_ON(!irqs_disabled());
199 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
202 EXPORT_SYMBOL(blk_start_queue);
205 * blk_stop_queue - stop a queue
206 * @q: The &struct request_queue in question
209 * The Linux block layer assumes that a block driver will consume all
210 * entries on the request queue when the request_fn strategy is called.
211 * Often this will not happen, because of hardware limitations (queue
212 * depth settings). If a device driver gets a 'queue full' response,
213 * or if it simply chooses not to queue more I/O at one point, it can
214 * call this function to prevent the request_fn from being called until
215 * the driver has signalled it's ready to go again. This happens by calling
216 * blk_start_queue() to restart queue operations. Queue lock must be held.
218 void blk_stop_queue(struct request_queue *q)
220 cancel_delayed_work(&q->delay_work);
221 queue_flag_set(QUEUE_FLAG_STOPPED, q);
223 EXPORT_SYMBOL(blk_stop_queue);
226 * blk_sync_queue - cancel any pending callbacks on a queue
230 * The block layer may perform asynchronous callback activity
231 * on a queue, such as calling the unplug function after a timeout.
232 * A block device may call blk_sync_queue to ensure that any
233 * such activity is cancelled, thus allowing it to release resources
234 * that the callbacks might use. The caller must already have made sure
235 * that its ->make_request_fn will not re-add plugging prior to calling
238 * This function does not cancel any asynchronous activity arising
239 * out of elevator or throttling code. That would require elevator_exit()
240 * and blkcg_exit_queue() to be called with queue lock initialized.
243 void blk_sync_queue(struct request_queue *q)
245 del_timer_sync(&q->timeout);
248 struct blk_mq_hw_ctx *hctx;
251 queue_for_each_hw_ctx(q, hctx, i) {
252 cancel_delayed_work_sync(&hctx->run_work);
253 cancel_delayed_work_sync(&hctx->delay_work);
256 cancel_delayed_work_sync(&q->delay_work);
259 EXPORT_SYMBOL(blk_sync_queue);
262 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
263 * @q: The queue to run
266 * Invoke request handling on a queue if there are any pending requests.
267 * May be used to restart request handling after a request has completed.
268 * This variant runs the queue whether or not the queue has been
269 * stopped. Must be called with the queue lock held and interrupts
270 * disabled. See also @blk_run_queue.
272 inline void __blk_run_queue_uncond(struct request_queue *q)
274 if (unlikely(blk_queue_dead(q)))
278 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
279 * the queue lock internally. As a result multiple threads may be
280 * running such a request function concurrently. Keep track of the
281 * number of active request_fn invocations such that blk_drain_queue()
282 * can wait until all these request_fn calls have finished.
284 q->request_fn_active++;
286 q->request_fn_active--;
290 * __blk_run_queue - run a single device queue
291 * @q: The queue to run
294 * See @blk_run_queue. This variant must be called with the queue lock
295 * held and interrupts disabled.
297 void __blk_run_queue(struct request_queue *q)
299 if (unlikely(blk_queue_stopped(q)))
302 __blk_run_queue_uncond(q);
304 EXPORT_SYMBOL(__blk_run_queue);
307 * blk_run_queue_async - run a single device queue in workqueue context
308 * @q: The queue to run
311 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
312 * of us. The caller must hold the queue lock.
314 void blk_run_queue_async(struct request_queue *q)
316 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
317 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
319 EXPORT_SYMBOL(blk_run_queue_async);
322 * blk_run_queue - run a single device queue
323 * @q: The queue to run
326 * Invoke request handling on this queue, if it has pending work to do.
327 * May be used to restart queueing when a request has completed.
329 void blk_run_queue(struct request_queue *q)
333 spin_lock_irqsave(q->queue_lock, flags);
335 spin_unlock_irqrestore(q->queue_lock, flags);
337 EXPORT_SYMBOL(blk_run_queue);
339 void blk_put_queue(struct request_queue *q)
341 kobject_put(&q->kobj);
343 EXPORT_SYMBOL(blk_put_queue);
346 * __blk_drain_queue - drain requests from request_queue
348 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
350 * Drain requests from @q. If @drain_all is set, all requests are drained.
351 * If not, only ELVPRIV requests are drained. The caller is responsible
352 * for ensuring that no new requests which need to be drained are queued.
354 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
355 __releases(q->queue_lock)
356 __acquires(q->queue_lock)
360 lockdep_assert_held(q->queue_lock);
366 * The caller might be trying to drain @q before its
367 * elevator is initialized.
370 elv_drain_elevator(q);
372 blkcg_drain_queue(q);
375 * This function might be called on a queue which failed
376 * driver init after queue creation or is not yet fully
377 * active yet. Some drivers (e.g. fd and loop) get unhappy
378 * in such cases. Kick queue iff dispatch queue has
379 * something on it and @q has request_fn set.
381 if (!list_empty(&q->queue_head) && q->request_fn)
384 drain |= q->nr_rqs_elvpriv;
385 drain |= q->request_fn_active;
388 * Unfortunately, requests are queued at and tracked from
389 * multiple places and there's no single counter which can
390 * be drained. Check all the queues and counters.
393 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
394 drain |= !list_empty(&q->queue_head);
395 for (i = 0; i < 2; i++) {
396 drain |= q->nr_rqs[i];
397 drain |= q->in_flight[i];
399 drain |= !list_empty(&fq->flush_queue[i]);
406 spin_unlock_irq(q->queue_lock);
410 spin_lock_irq(q->queue_lock);
414 * With queue marked dead, any woken up waiter will fail the
415 * allocation path, so the wakeup chaining is lost and we're
416 * left with hung waiters. We need to wake up those waiters.
419 struct request_list *rl;
421 blk_queue_for_each_rl(rl, q)
422 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
423 wake_up_all(&rl->wait[i]);
428 * blk_queue_bypass_start - enter queue bypass mode
429 * @q: queue of interest
431 * In bypass mode, only the dispatch FIFO queue of @q is used. This
432 * function makes @q enter bypass mode and drains all requests which were
433 * throttled or issued before. On return, it's guaranteed that no request
434 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
435 * inside queue or RCU read lock.
437 void blk_queue_bypass_start(struct request_queue *q)
439 spin_lock_irq(q->queue_lock);
441 queue_flag_set(QUEUE_FLAG_BYPASS, q);
442 spin_unlock_irq(q->queue_lock);
445 * Queues start drained. Skip actual draining till init is
446 * complete. This avoids lenghty delays during queue init which
447 * can happen many times during boot.
449 if (blk_queue_init_done(q)) {
450 spin_lock_irq(q->queue_lock);
451 __blk_drain_queue(q, false);
452 spin_unlock_irq(q->queue_lock);
454 /* ensure blk_queue_bypass() is %true inside RCU read lock */
458 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
461 * blk_queue_bypass_end - leave queue bypass mode
462 * @q: queue of interest
464 * Leave bypass mode and restore the normal queueing behavior.
466 void blk_queue_bypass_end(struct request_queue *q)
468 spin_lock_irq(q->queue_lock);
469 if (!--q->bypass_depth)
470 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
471 WARN_ON_ONCE(q->bypass_depth < 0);
472 spin_unlock_irq(q->queue_lock);
474 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
476 void blk_set_queue_dying(struct request_queue *q)
478 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
481 blk_mq_wake_waiters(q);
483 struct request_list *rl;
485 blk_queue_for_each_rl(rl, q) {
487 wake_up(&rl->wait[BLK_RW_SYNC]);
488 wake_up(&rl->wait[BLK_RW_ASYNC]);
493 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
496 * blk_cleanup_queue - shutdown a request queue
497 * @q: request queue to shutdown
499 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
500 * put it. All future requests will be failed immediately with -ENODEV.
502 void blk_cleanup_queue(struct request_queue *q)
504 spinlock_t *lock = q->queue_lock;
506 /* mark @q DYING, no new request or merges will be allowed afterwards */
507 mutex_lock(&q->sysfs_lock);
508 blk_set_queue_dying(q);
512 * A dying queue is permanently in bypass mode till released. Note
513 * that, unlike blk_queue_bypass_start(), we aren't performing
514 * synchronize_rcu() after entering bypass mode to avoid the delay
515 * as some drivers create and destroy a lot of queues while
516 * probing. This is still safe because blk_release_queue() will be
517 * called only after the queue refcnt drops to zero and nothing,
518 * RCU or not, would be traversing the queue by then.
521 queue_flag_set(QUEUE_FLAG_BYPASS, q);
523 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
524 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
525 queue_flag_set(QUEUE_FLAG_DYING, q);
526 spin_unlock_irq(lock);
527 mutex_unlock(&q->sysfs_lock);
530 * Drain all requests queued before DYING marking. Set DEAD flag to
531 * prevent that q->request_fn() gets invoked after draining finished.
534 blk_mq_freeze_queue(q);
538 __blk_drain_queue(q, true);
540 queue_flag_set(QUEUE_FLAG_DEAD, q);
541 spin_unlock_irq(lock);
543 /* @q won't process any more request, flush async actions */
544 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
548 blk_mq_free_queue(q);
551 if (q->queue_lock != &q->__queue_lock)
552 q->queue_lock = &q->__queue_lock;
553 spin_unlock_irq(lock);
555 /* @q is and will stay empty, shutdown and put */
558 EXPORT_SYMBOL(blk_cleanup_queue);
560 /* Allocate memory local to the request queue */
561 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
563 int nid = (int)(long)data;
564 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
567 static void free_request_struct(void *element, void *unused)
569 kmem_cache_free(request_cachep, element);
572 int blk_init_rl(struct request_list *rl, struct request_queue *q,
575 if (unlikely(rl->rq_pool))
579 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
580 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
581 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
582 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
584 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
586 (void *)(long)q->node, gfp_mask,
594 void blk_exit_rl(struct request_list *rl)
597 mempool_destroy(rl->rq_pool);
600 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
602 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
604 EXPORT_SYMBOL(blk_alloc_queue);
606 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
608 struct request_queue *q;
611 q = kmem_cache_alloc_node(blk_requestq_cachep,
612 gfp_mask | __GFP_ZERO, node_id);
616 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
620 q->backing_dev_info.ra_pages =
621 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
622 q->backing_dev_info.state = 0;
623 q->backing_dev_info.capabilities = 0;
624 q->backing_dev_info.name = "block";
627 err = bdi_init(&q->backing_dev_info);
631 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
632 laptop_mode_timer_fn, (unsigned long) q);
633 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
634 INIT_LIST_HEAD(&q->queue_head);
635 INIT_LIST_HEAD(&q->timeout_list);
636 INIT_LIST_HEAD(&q->icq_list);
637 #ifdef CONFIG_BLK_CGROUP
638 INIT_LIST_HEAD(&q->blkg_list);
640 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
642 kobject_init(&q->kobj, &blk_queue_ktype);
644 mutex_init(&q->sysfs_lock);
645 spin_lock_init(&q->__queue_lock);
648 * By default initialize queue_lock to internal lock and driver can
649 * override it later if need be.
651 q->queue_lock = &q->__queue_lock;
654 * A queue starts its life with bypass turned on to avoid
655 * unnecessary bypass on/off overhead and nasty surprises during
656 * init. The initial bypass will be finished when the queue is
657 * registered by blk_register_queue().
660 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
662 init_waitqueue_head(&q->mq_freeze_wq);
664 if (blkcg_init_queue(q))
670 bdi_destroy(&q->backing_dev_info);
672 ida_simple_remove(&blk_queue_ida, q->id);
674 kmem_cache_free(blk_requestq_cachep, q);
677 EXPORT_SYMBOL(blk_alloc_queue_node);
680 * blk_init_queue - prepare a request queue for use with a block device
681 * @rfn: The function to be called to process requests that have been
682 * placed on the queue.
683 * @lock: Request queue spin lock
686 * If a block device wishes to use the standard request handling procedures,
687 * which sorts requests and coalesces adjacent requests, then it must
688 * call blk_init_queue(). The function @rfn will be called when there
689 * are requests on the queue that need to be processed. If the device
690 * supports plugging, then @rfn may not be called immediately when requests
691 * are available on the queue, but may be called at some time later instead.
692 * Plugged queues are generally unplugged when a buffer belonging to one
693 * of the requests on the queue is needed, or due to memory pressure.
695 * @rfn is not required, or even expected, to remove all requests off the
696 * queue, but only as many as it can handle at a time. If it does leave
697 * requests on the queue, it is responsible for arranging that the requests
698 * get dealt with eventually.
700 * The queue spin lock must be held while manipulating the requests on the
701 * request queue; this lock will be taken also from interrupt context, so irq
702 * disabling is needed for it.
704 * Function returns a pointer to the initialized request queue, or %NULL if
708 * blk_init_queue() must be paired with a blk_cleanup_queue() call
709 * when the block device is deactivated (such as at module unload).
712 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
714 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
716 EXPORT_SYMBOL(blk_init_queue);
718 struct request_queue *
719 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
721 struct request_queue *uninit_q, *q;
723 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
727 q = blk_init_allocated_queue(uninit_q, rfn, lock);
729 blk_cleanup_queue(uninit_q);
733 EXPORT_SYMBOL(blk_init_queue_node);
735 static void blk_queue_bio(struct request_queue *q, struct bio *bio);
737 struct request_queue *
738 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
744 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
748 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
752 q->prep_rq_fn = NULL;
753 q->unprep_rq_fn = NULL;
754 q->queue_flags |= QUEUE_FLAG_DEFAULT;
756 /* Override internal queue lock with supplied lock pointer */
758 q->queue_lock = lock;
761 * This also sets hw/phys segments, boundary and size
763 blk_queue_make_request(q, blk_queue_bio);
765 q->sg_reserved_size = INT_MAX;
767 /* Protect q->elevator from elevator_change */
768 mutex_lock(&q->sysfs_lock);
771 if (elevator_init(q, NULL)) {
772 mutex_unlock(&q->sysfs_lock);
776 mutex_unlock(&q->sysfs_lock);
781 blk_free_flush_queue(q->fq);
784 EXPORT_SYMBOL(blk_init_allocated_queue);
786 bool blk_get_queue(struct request_queue *q)
788 if (likely(!blk_queue_dying(q))) {
795 EXPORT_SYMBOL(blk_get_queue);
797 static inline void blk_free_request(struct request_list *rl, struct request *rq)
799 if (rq->cmd_flags & REQ_ELVPRIV) {
800 elv_put_request(rl->q, rq);
802 put_io_context(rq->elv.icq->ioc);
805 mempool_free(rq, rl->rq_pool);
809 * ioc_batching returns true if the ioc is a valid batching request and
810 * should be given priority access to a request.
812 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
818 * Make sure the process is able to allocate at least 1 request
819 * even if the batch times out, otherwise we could theoretically
822 return ioc->nr_batch_requests == q->nr_batching ||
823 (ioc->nr_batch_requests > 0
824 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
828 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
829 * will cause the process to be a "batcher" on all queues in the system. This
830 * is the behaviour we want though - once it gets a wakeup it should be given
833 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
835 if (!ioc || ioc_batching(q, ioc))
838 ioc->nr_batch_requests = q->nr_batching;
839 ioc->last_waited = jiffies;
842 static void __freed_request(struct request_list *rl, int sync)
844 struct request_queue *q = rl->q;
847 * bdi isn't aware of blkcg yet. As all async IOs end up root
848 * blkcg anyway, just use root blkcg state.
850 if (rl == &q->root_rl &&
851 rl->count[sync] < queue_congestion_off_threshold(q))
852 blk_clear_queue_congested(q, sync);
854 if (rl->count[sync] + 1 <= q->nr_requests) {
855 if (waitqueue_active(&rl->wait[sync]))
856 wake_up(&rl->wait[sync]);
858 blk_clear_rl_full(rl, sync);
863 * A request has just been released. Account for it, update the full and
864 * congestion status, wake up any waiters. Called under q->queue_lock.
866 static void freed_request(struct request_list *rl, unsigned int flags)
868 struct request_queue *q = rl->q;
869 int sync = rw_is_sync(flags);
873 if (flags & REQ_ELVPRIV)
876 __freed_request(rl, sync);
878 if (unlikely(rl->starved[sync ^ 1]))
879 __freed_request(rl, sync ^ 1);
882 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
884 struct request_list *rl;
886 spin_lock_irq(q->queue_lock);
888 blk_queue_congestion_threshold(q);
890 /* congestion isn't cgroup aware and follows root blkcg for now */
893 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
894 blk_set_queue_congested(q, BLK_RW_SYNC);
895 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
896 blk_clear_queue_congested(q, BLK_RW_SYNC);
898 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
899 blk_set_queue_congested(q, BLK_RW_ASYNC);
900 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
901 blk_clear_queue_congested(q, BLK_RW_ASYNC);
903 blk_queue_for_each_rl(rl, q) {
904 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
905 blk_set_rl_full(rl, BLK_RW_SYNC);
907 blk_clear_rl_full(rl, BLK_RW_SYNC);
908 wake_up(&rl->wait[BLK_RW_SYNC]);
911 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
912 blk_set_rl_full(rl, BLK_RW_ASYNC);
914 blk_clear_rl_full(rl, BLK_RW_ASYNC);
915 wake_up(&rl->wait[BLK_RW_ASYNC]);
919 spin_unlock_irq(q->queue_lock);
924 * Determine if elevator data should be initialized when allocating the
925 * request associated with @bio.
927 static bool blk_rq_should_init_elevator(struct bio *bio)
933 * Flush requests do not use the elevator so skip initialization.
934 * This allows a request to share the flush and elevator data.
936 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
943 * rq_ioc - determine io_context for request allocation
944 * @bio: request being allocated is for this bio (can be %NULL)
946 * Determine io_context to use for request allocation for @bio. May return
947 * %NULL if %current->io_context doesn't exist.
949 static struct io_context *rq_ioc(struct bio *bio)
951 #ifdef CONFIG_BLK_CGROUP
952 if (bio && bio->bi_ioc)
955 return current->io_context;
959 * __get_request - get a free request
960 * @rl: request list to allocate from
961 * @rw_flags: RW and SYNC flags
962 * @bio: bio to allocate request for (can be %NULL)
963 * @gfp_mask: allocation mask
965 * Get a free request from @q. This function may fail under memory
966 * pressure or if @q is dead.
968 * Must be called with @q->queue_lock held and,
969 * Returns ERR_PTR on failure, with @q->queue_lock held.
970 * Returns request pointer on success, with @q->queue_lock *not held*.
972 static struct request *__get_request(struct request_list *rl, int rw_flags,
973 struct bio *bio, gfp_t gfp_mask)
975 struct request_queue *q = rl->q;
977 struct elevator_type *et = q->elevator->type;
978 struct io_context *ioc = rq_ioc(bio);
979 struct io_cq *icq = NULL;
980 const bool is_sync = rw_is_sync(rw_flags) != 0;
983 if (unlikely(blk_queue_dying(q)))
984 return ERR_PTR(-ENODEV);
986 may_queue = elv_may_queue(q, rw_flags);
987 if (may_queue == ELV_MQUEUE_NO)
990 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
991 if (rl->count[is_sync]+1 >= q->nr_requests) {
993 * The queue will fill after this allocation, so set
994 * it as full, and mark this process as "batching".
995 * This process will be allowed to complete a batch of
996 * requests, others will be blocked.
998 if (!blk_rl_full(rl, is_sync)) {
999 ioc_set_batching(q, ioc);
1000 blk_set_rl_full(rl, is_sync);
1002 if (may_queue != ELV_MQUEUE_MUST
1003 && !ioc_batching(q, ioc)) {
1005 * The queue is full and the allocating
1006 * process is not a "batcher", and not
1007 * exempted by the IO scheduler
1009 return ERR_PTR(-ENOMEM);
1014 * bdi isn't aware of blkcg yet. As all async IOs end up
1015 * root blkcg anyway, just use root blkcg state.
1017 if (rl == &q->root_rl)
1018 blk_set_queue_congested(q, is_sync);
1022 * Only allow batching queuers to allocate up to 50% over the defined
1023 * limit of requests, otherwise we could have thousands of requests
1024 * allocated with any setting of ->nr_requests
1026 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1027 return ERR_PTR(-ENOMEM);
1029 q->nr_rqs[is_sync]++;
1030 rl->count[is_sync]++;
1031 rl->starved[is_sync] = 0;
1034 * Decide whether the new request will be managed by elevator. If
1035 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1036 * prevent the current elevator from being destroyed until the new
1037 * request is freed. This guarantees icq's won't be destroyed and
1038 * makes creating new ones safe.
1040 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1041 * it will be created after releasing queue_lock.
1043 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1044 rw_flags |= REQ_ELVPRIV;
1045 q->nr_rqs_elvpriv++;
1046 if (et->icq_cache && ioc)
1047 icq = ioc_lookup_icq(ioc, q);
1050 if (blk_queue_io_stat(q))
1051 rw_flags |= REQ_IO_STAT;
1052 spin_unlock_irq(q->queue_lock);
1054 /* allocate and init request */
1055 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1060 blk_rq_set_rl(rq, rl);
1061 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1064 if (rw_flags & REQ_ELVPRIV) {
1065 if (unlikely(et->icq_cache && !icq)) {
1067 icq = ioc_create_icq(ioc, q, gfp_mask);
1073 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1076 /* @rq->elv.icq holds io_context until @rq is freed */
1078 get_io_context(icq->ioc);
1082 * ioc may be NULL here, and ioc_batching will be false. That's
1083 * OK, if the queue is under the request limit then requests need
1084 * not count toward the nr_batch_requests limit. There will always
1085 * be some limit enforced by BLK_BATCH_TIME.
1087 if (ioc_batching(q, ioc))
1088 ioc->nr_batch_requests--;
1090 trace_block_getrq(q, bio, rw_flags & 1);
1095 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1096 * and may fail indefinitely under memory pressure and thus
1097 * shouldn't stall IO. Treat this request as !elvpriv. This will
1098 * disturb iosched and blkcg but weird is bettern than dead.
1100 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1101 __func__, dev_name(q->backing_dev_info.dev));
1103 rq->cmd_flags &= ~REQ_ELVPRIV;
1106 spin_lock_irq(q->queue_lock);
1107 q->nr_rqs_elvpriv--;
1108 spin_unlock_irq(q->queue_lock);
1113 * Allocation failed presumably due to memory. Undo anything we
1114 * might have messed up.
1116 * Allocating task should really be put onto the front of the wait
1117 * queue, but this is pretty rare.
1119 spin_lock_irq(q->queue_lock);
1120 freed_request(rl, rw_flags);
1123 * in the very unlikely event that allocation failed and no
1124 * requests for this direction was pending, mark us starved so that
1125 * freeing of a request in the other direction will notice
1126 * us. another possible fix would be to split the rq mempool into
1130 if (unlikely(rl->count[is_sync] == 0))
1131 rl->starved[is_sync] = 1;
1132 return ERR_PTR(-ENOMEM);
1136 * get_request - get a free request
1137 * @q: request_queue to allocate request from
1138 * @rw_flags: RW and SYNC flags
1139 * @bio: bio to allocate request for (can be %NULL)
1140 * @gfp_mask: allocation mask
1142 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1143 * function keeps retrying under memory pressure and fails iff @q is dead.
1145 * Must be called with @q->queue_lock held and,
1146 * Returns ERR_PTR on failure, with @q->queue_lock held.
1147 * Returns request pointer on success, with @q->queue_lock *not held*.
1149 static struct request *get_request(struct request_queue *q, int rw_flags,
1150 struct bio *bio, gfp_t gfp_mask)
1152 const bool is_sync = rw_is_sync(rw_flags) != 0;
1154 struct request_list *rl;
1157 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1159 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1163 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1168 /* wait on @rl and retry */
1169 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1170 TASK_UNINTERRUPTIBLE);
1172 trace_block_sleeprq(q, bio, rw_flags & 1);
1174 spin_unlock_irq(q->queue_lock);
1178 * After sleeping, we become a "batching" process and will be able
1179 * to allocate at least one request, and up to a big batch of them
1180 * for a small period time. See ioc_batching, ioc_set_batching
1182 ioc_set_batching(q, current->io_context);
1184 spin_lock_irq(q->queue_lock);
1185 finish_wait(&rl->wait[is_sync], &wait);
1190 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1195 BUG_ON(rw != READ && rw != WRITE);
1197 /* create ioc upfront */
1198 create_io_context(gfp_mask, q->node);
1200 spin_lock_irq(q->queue_lock);
1201 rq = get_request(q, rw, NULL, gfp_mask);
1203 spin_unlock_irq(q->queue_lock);
1204 /* q->queue_lock is unlocked at this point */
1209 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1212 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1214 return blk_old_get_request(q, rw, gfp_mask);
1216 EXPORT_SYMBOL(blk_get_request);
1219 * blk_make_request - given a bio, allocate a corresponding struct request.
1220 * @q: target request queue
1221 * @bio: The bio describing the memory mappings that will be submitted for IO.
1222 * It may be a chained-bio properly constructed by block/bio layer.
1223 * @gfp_mask: gfp flags to be used for memory allocation
1225 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1226 * type commands. Where the struct request needs to be farther initialized by
1227 * the caller. It is passed a &struct bio, which describes the memory info of
1230 * The caller of blk_make_request must make sure that bi_io_vec
1231 * are set to describe the memory buffers. That bio_data_dir() will return
1232 * the needed direction of the request. (And all bio's in the passed bio-chain
1233 * are properly set accordingly)
1235 * If called under none-sleepable conditions, mapped bio buffers must not
1236 * need bouncing, by calling the appropriate masked or flagged allocator,
1237 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1240 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1241 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1242 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1243 * completion of a bio that hasn't been submitted yet, thus resulting in a
1244 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1245 * of bio_alloc(), as that avoids the mempool deadlock.
1246 * If possible a big IO should be split into smaller parts when allocation
1247 * fails. Partial allocation should not be an error, or you risk a live-lock.
1249 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1252 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1257 blk_rq_set_block_pc(rq);
1260 struct bio *bounce_bio = bio;
1263 blk_queue_bounce(q, &bounce_bio);
1264 ret = blk_rq_append_bio(q, rq, bounce_bio);
1265 if (unlikely(ret)) {
1266 blk_put_request(rq);
1267 return ERR_PTR(ret);
1273 EXPORT_SYMBOL(blk_make_request);
1276 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1277 * @rq: request to be initialized
1280 void blk_rq_set_block_pc(struct request *rq)
1282 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1284 rq->__sector = (sector_t) -1;
1285 rq->bio = rq->biotail = NULL;
1286 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1288 EXPORT_SYMBOL(blk_rq_set_block_pc);
1291 * blk_requeue_request - put a request back on queue
1292 * @q: request queue where request should be inserted
1293 * @rq: request to be inserted
1296 * Drivers often keep queueing requests until the hardware cannot accept
1297 * more, when that condition happens we need to put the request back
1298 * on the queue. Must be called with queue lock held.
1300 void blk_requeue_request(struct request_queue *q, struct request *rq)
1302 blk_delete_timer(rq);
1303 blk_clear_rq_complete(rq);
1304 trace_block_rq_requeue(q, rq);
1306 if (rq->cmd_flags & REQ_QUEUED)
1307 blk_queue_end_tag(q, rq);
1309 BUG_ON(blk_queued_rq(rq));
1311 elv_requeue_request(q, rq);
1313 EXPORT_SYMBOL(blk_requeue_request);
1315 static void add_acct_request(struct request_queue *q, struct request *rq,
1318 blk_account_io_start(rq, true);
1319 __elv_add_request(q, rq, where);
1322 static void part_round_stats_single(int cpu, struct hd_struct *part,
1327 if (now == part->stamp)
1330 inflight = part_in_flight(part);
1332 __part_stat_add(cpu, part, time_in_queue,
1333 inflight * (now - part->stamp));
1334 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1340 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1341 * @cpu: cpu number for stats access
1342 * @part: target partition
1344 * The average IO queue length and utilisation statistics are maintained
1345 * by observing the current state of the queue length and the amount of
1346 * time it has been in this state for.
1348 * Normally, that accounting is done on IO completion, but that can result
1349 * in more than a second's worth of IO being accounted for within any one
1350 * second, leading to >100% utilisation. To deal with that, we call this
1351 * function to do a round-off before returning the results when reading
1352 * /proc/diskstats. This accounts immediately for all queue usage up to
1353 * the current jiffies and restarts the counters again.
1355 void part_round_stats(int cpu, struct hd_struct *part)
1357 unsigned long now = jiffies;
1360 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1361 part_round_stats_single(cpu, part, now);
1363 EXPORT_SYMBOL_GPL(part_round_stats);
1366 static void blk_pm_put_request(struct request *rq)
1368 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1369 pm_runtime_mark_last_busy(rq->q->dev);
1372 static inline void blk_pm_put_request(struct request *rq) {}
1376 * queue lock must be held
1378 void __blk_put_request(struct request_queue *q, struct request *req)
1384 blk_mq_free_request(req);
1388 blk_pm_put_request(req);
1390 elv_completed_request(q, req);
1392 /* this is a bio leak */
1393 WARN_ON(req->bio != NULL);
1396 * Request may not have originated from ll_rw_blk. if not,
1397 * it didn't come out of our reserved rq pools
1399 if (req->cmd_flags & REQ_ALLOCED) {
1400 unsigned int flags = req->cmd_flags;
1401 struct request_list *rl = blk_rq_rl(req);
1403 BUG_ON(!list_empty(&req->queuelist));
1404 BUG_ON(ELV_ON_HASH(req));
1406 blk_free_request(rl, req);
1407 freed_request(rl, flags);
1411 EXPORT_SYMBOL_GPL(__blk_put_request);
1413 void blk_put_request(struct request *req)
1415 struct request_queue *q = req->q;
1418 blk_mq_free_request(req);
1420 unsigned long flags;
1422 spin_lock_irqsave(q->queue_lock, flags);
1423 __blk_put_request(q, req);
1424 spin_unlock_irqrestore(q->queue_lock, flags);
1427 EXPORT_SYMBOL(blk_put_request);
1430 * blk_add_request_payload - add a payload to a request
1431 * @rq: request to update
1432 * @page: page backing the payload
1433 * @len: length of the payload.
1435 * This allows to later add a payload to an already submitted request by
1436 * a block driver. The driver needs to take care of freeing the payload
1439 * Note that this is a quite horrible hack and nothing but handling of
1440 * discard requests should ever use it.
1442 void blk_add_request_payload(struct request *rq, struct page *page,
1445 struct bio *bio = rq->bio;
1447 bio->bi_io_vec->bv_page = page;
1448 bio->bi_io_vec->bv_offset = 0;
1449 bio->bi_io_vec->bv_len = len;
1451 bio->bi_iter.bi_size = len;
1453 bio->bi_phys_segments = 1;
1455 rq->__data_len = rq->resid_len = len;
1456 rq->nr_phys_segments = 1;
1458 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1460 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1463 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1465 if (!ll_back_merge_fn(q, req, bio))
1468 trace_block_bio_backmerge(q, req, bio);
1470 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1471 blk_rq_set_mixed_merge(req);
1473 req->biotail->bi_next = bio;
1475 req->__data_len += bio->bi_iter.bi_size;
1476 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1478 blk_account_io_start(req, false);
1482 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1485 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1487 if (!ll_front_merge_fn(q, req, bio))
1490 trace_block_bio_frontmerge(q, req, bio);
1492 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1493 blk_rq_set_mixed_merge(req);
1495 bio->bi_next = req->bio;
1498 req->__sector = bio->bi_iter.bi_sector;
1499 req->__data_len += bio->bi_iter.bi_size;
1500 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1502 blk_account_io_start(req, false);
1507 * blk_attempt_plug_merge - try to merge with %current's plugged list
1508 * @q: request_queue new bio is being queued at
1509 * @bio: new bio being queued
1510 * @request_count: out parameter for number of traversed plugged requests
1512 * Determine whether @bio being queued on @q can be merged with a request
1513 * on %current's plugged list. Returns %true if merge was successful,
1516 * Plugging coalesces IOs from the same issuer for the same purpose without
1517 * going through @q->queue_lock. As such it's more of an issuing mechanism
1518 * than scheduling, and the request, while may have elvpriv data, is not
1519 * added on the elevator at this point. In addition, we don't have
1520 * reliable access to the elevator outside queue lock. Only check basic
1521 * merging parameters without querying the elevator.
1523 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1525 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1526 unsigned int *request_count)
1528 struct blk_plug *plug;
1531 struct list_head *plug_list;
1533 plug = current->plug;
1539 plug_list = &plug->mq_list;
1541 plug_list = &plug->list;
1543 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1549 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1552 el_ret = blk_try_merge(rq, bio);
1553 if (el_ret == ELEVATOR_BACK_MERGE) {
1554 ret = bio_attempt_back_merge(q, rq, bio);
1557 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1558 ret = bio_attempt_front_merge(q, rq, bio);
1567 void init_request_from_bio(struct request *req, struct bio *bio)
1569 req->cmd_type = REQ_TYPE_FS;
1571 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1572 if (bio->bi_rw & REQ_RAHEAD)
1573 req->cmd_flags |= REQ_FAILFAST_MASK;
1576 req->__sector = bio->bi_iter.bi_sector;
1577 req->ioprio = bio_prio(bio);
1578 blk_rq_bio_prep(req->q, req, bio);
1581 static void blk_queue_bio(struct request_queue *q, struct bio *bio)
1583 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1584 struct blk_plug *plug;
1585 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1586 struct request *req;
1587 unsigned int request_count = 0;
1590 * low level driver can indicate that it wants pages above a
1591 * certain limit bounced to low memory (ie for highmem, or even
1592 * ISA dma in theory)
1594 blk_queue_bounce(q, &bio);
1596 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1597 bio_endio(bio, -EIO);
1601 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1602 spin_lock_irq(q->queue_lock);
1603 where = ELEVATOR_INSERT_FLUSH;
1608 * Check if we can merge with the plugged list before grabbing
1611 if (!blk_queue_nomerges(q) &&
1612 blk_attempt_plug_merge(q, bio, &request_count))
1615 spin_lock_irq(q->queue_lock);
1617 el_ret = elv_merge(q, &req, bio);
1618 if (el_ret == ELEVATOR_BACK_MERGE) {
1619 if (bio_attempt_back_merge(q, req, bio)) {
1620 elv_bio_merged(q, req, bio);
1621 if (!attempt_back_merge(q, req))
1622 elv_merged_request(q, req, el_ret);
1625 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1626 if (bio_attempt_front_merge(q, req, bio)) {
1627 elv_bio_merged(q, req, bio);
1628 if (!attempt_front_merge(q, req))
1629 elv_merged_request(q, req, el_ret);
1636 * This sync check and mask will be re-done in init_request_from_bio(),
1637 * but we need to set it earlier to expose the sync flag to the
1638 * rq allocator and io schedulers.
1640 rw_flags = bio_data_dir(bio);
1642 rw_flags |= REQ_SYNC;
1645 * Grab a free request. This is might sleep but can not fail.
1646 * Returns with the queue unlocked.
1648 req = get_request(q, rw_flags, bio, GFP_NOIO);
1650 bio_endio(bio, PTR_ERR(req)); /* @q is dead */
1655 * After dropping the lock and possibly sleeping here, our request
1656 * may now be mergeable after it had proven unmergeable (above).
1657 * We don't worry about that case for efficiency. It won't happen
1658 * often, and the elevators are able to handle it.
1660 init_request_from_bio(req, bio);
1662 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1663 req->cpu = raw_smp_processor_id();
1665 plug = current->plug;
1668 * If this is the first request added after a plug, fire
1672 trace_block_plug(q);
1674 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1675 blk_flush_plug_list(plug, false);
1676 trace_block_plug(q);
1679 list_add_tail(&req->queuelist, &plug->list);
1680 blk_account_io_start(req, true);
1682 spin_lock_irq(q->queue_lock);
1683 add_acct_request(q, req, where);
1686 spin_unlock_irq(q->queue_lock);
1691 * If bio->bi_dev is a partition, remap the location
1693 static inline void blk_partition_remap(struct bio *bio)
1695 struct block_device *bdev = bio->bi_bdev;
1697 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1698 struct hd_struct *p = bdev->bd_part;
1700 bio->bi_iter.bi_sector += p->start_sect;
1701 bio->bi_bdev = bdev->bd_contains;
1703 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1705 bio->bi_iter.bi_sector - p->start_sect);
1709 static void handle_bad_sector(struct bio *bio)
1711 char b[BDEVNAME_SIZE];
1713 printk(KERN_INFO "attempt to access beyond end of device\n");
1714 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1715 bdevname(bio->bi_bdev, b),
1717 (unsigned long long)bio_end_sector(bio),
1718 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1720 set_bit(BIO_EOF, &bio->bi_flags);
1723 #ifdef CONFIG_FAIL_MAKE_REQUEST
1725 static DECLARE_FAULT_ATTR(fail_make_request);
1727 static int __init setup_fail_make_request(char *str)
1729 return setup_fault_attr(&fail_make_request, str);
1731 __setup("fail_make_request=", setup_fail_make_request);
1733 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1735 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1738 static int __init fail_make_request_debugfs(void)
1740 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1741 NULL, &fail_make_request);
1743 return PTR_ERR_OR_ZERO(dir);
1746 late_initcall(fail_make_request_debugfs);
1748 #else /* CONFIG_FAIL_MAKE_REQUEST */
1750 static inline bool should_fail_request(struct hd_struct *part,
1756 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1759 * Check whether this bio extends beyond the end of the device.
1761 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1768 /* Test device or partition size, when known. */
1769 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1771 sector_t sector = bio->bi_iter.bi_sector;
1773 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1775 * This may well happen - the kernel calls bread()
1776 * without checking the size of the device, e.g., when
1777 * mounting a device.
1779 handle_bad_sector(bio);
1787 static noinline_for_stack bool
1788 generic_make_request_checks(struct bio *bio)
1790 struct request_queue *q;
1791 int nr_sectors = bio_sectors(bio);
1793 char b[BDEVNAME_SIZE];
1794 struct hd_struct *part;
1798 if (bio_check_eod(bio, nr_sectors))
1801 q = bdev_get_queue(bio->bi_bdev);
1804 "generic_make_request: Trying to access "
1805 "nonexistent block-device %s (%Lu)\n",
1806 bdevname(bio->bi_bdev, b),
1807 (long long) bio->bi_iter.bi_sector);
1811 if (likely(bio_is_rw(bio) &&
1812 nr_sectors > queue_max_hw_sectors(q))) {
1813 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1814 bdevname(bio->bi_bdev, b),
1816 queue_max_hw_sectors(q));
1820 part = bio->bi_bdev->bd_part;
1821 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1822 should_fail_request(&part_to_disk(part)->part0,
1823 bio->bi_iter.bi_size))
1827 * If this device has partitions, remap block n
1828 * of partition p to block n+start(p) of the disk.
1830 blk_partition_remap(bio);
1832 if (bio_check_eod(bio, nr_sectors))
1836 * Filter flush bio's early so that make_request based
1837 * drivers without flush support don't have to worry
1840 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1841 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1848 if ((bio->bi_rw & REQ_DISCARD) &&
1849 (!blk_queue_discard(q) ||
1850 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1855 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1861 * Various block parts want %current->io_context and lazy ioc
1862 * allocation ends up trading a lot of pain for a small amount of
1863 * memory. Just allocate it upfront. This may fail and block
1864 * layer knows how to live with it.
1866 create_io_context(GFP_ATOMIC, q->node);
1868 if (blk_throtl_bio(q, bio))
1869 return false; /* throttled, will be resubmitted later */
1871 trace_block_bio_queue(q, bio);
1875 bio_endio(bio, err);
1880 * generic_make_request - hand a buffer to its device driver for I/O
1881 * @bio: The bio describing the location in memory and on the device.
1883 * generic_make_request() is used to make I/O requests of block
1884 * devices. It is passed a &struct bio, which describes the I/O that needs
1887 * generic_make_request() does not return any status. The
1888 * success/failure status of the request, along with notification of
1889 * completion, is delivered asynchronously through the bio->bi_end_io
1890 * function described (one day) else where.
1892 * The caller of generic_make_request must make sure that bi_io_vec
1893 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1894 * set to describe the device address, and the
1895 * bi_end_io and optionally bi_private are set to describe how
1896 * completion notification should be signaled.
1898 * generic_make_request and the drivers it calls may use bi_next if this
1899 * bio happens to be merged with someone else, and may resubmit the bio to
1900 * a lower device by calling into generic_make_request recursively, which
1901 * means the bio should NOT be touched after the call to ->make_request_fn.
1903 void generic_make_request(struct bio *bio)
1905 struct bio_list bio_list_on_stack;
1907 if (!generic_make_request_checks(bio))
1911 * We only want one ->make_request_fn to be active at a time, else
1912 * stack usage with stacked devices could be a problem. So use
1913 * current->bio_list to keep a list of requests submited by a
1914 * make_request_fn function. current->bio_list is also used as a
1915 * flag to say if generic_make_request is currently active in this
1916 * task or not. If it is NULL, then no make_request is active. If
1917 * it is non-NULL, then a make_request is active, and new requests
1918 * should be added at the tail
1920 if (current->bio_list) {
1921 bio_list_add(current->bio_list, bio);
1925 /* following loop may be a bit non-obvious, and so deserves some
1927 * Before entering the loop, bio->bi_next is NULL (as all callers
1928 * ensure that) so we have a list with a single bio.
1929 * We pretend that we have just taken it off a longer list, so
1930 * we assign bio_list to a pointer to the bio_list_on_stack,
1931 * thus initialising the bio_list of new bios to be
1932 * added. ->make_request() may indeed add some more bios
1933 * through a recursive call to generic_make_request. If it
1934 * did, we find a non-NULL value in bio_list and re-enter the loop
1935 * from the top. In this case we really did just take the bio
1936 * of the top of the list (no pretending) and so remove it from
1937 * bio_list, and call into ->make_request() again.
1939 BUG_ON(bio->bi_next);
1940 bio_list_init(&bio_list_on_stack);
1941 current->bio_list = &bio_list_on_stack;
1943 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1945 q->make_request_fn(q, bio);
1947 bio = bio_list_pop(current->bio_list);
1949 current->bio_list = NULL; /* deactivate */
1951 EXPORT_SYMBOL(generic_make_request);
1954 * submit_bio - submit a bio to the block device layer for I/O
1955 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1956 * @bio: The &struct bio which describes the I/O
1958 * submit_bio() is very similar in purpose to generic_make_request(), and
1959 * uses that function to do most of the work. Both are fairly rough
1960 * interfaces; @bio must be presetup and ready for I/O.
1963 void submit_bio(int rw, struct bio *bio)
1968 * If it's a regular read/write or a barrier with data attached,
1969 * go through the normal accounting stuff before submission.
1971 if (bio_has_data(bio)) {
1974 if (unlikely(rw & REQ_WRITE_SAME))
1975 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1977 count = bio_sectors(bio);
1980 count_vm_events(PGPGOUT, count);
1982 task_io_account_read(bio->bi_iter.bi_size);
1983 count_vm_events(PGPGIN, count);
1986 if (unlikely(block_dump)) {
1987 char b[BDEVNAME_SIZE];
1988 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1989 current->comm, task_pid_nr(current),
1990 (rw & WRITE) ? "WRITE" : "READ",
1991 (unsigned long long)bio->bi_iter.bi_sector,
1992 bdevname(bio->bi_bdev, b),
1997 generic_make_request(bio);
1999 EXPORT_SYMBOL(submit_bio);
2002 * blk_rq_check_limits - Helper function to check a request for the queue limit
2004 * @rq: the request being checked
2007 * @rq may have been made based on weaker limitations of upper-level queues
2008 * in request stacking drivers, and it may violate the limitation of @q.
2009 * Since the block layer and the underlying device driver trust @rq
2010 * after it is inserted to @q, it should be checked against @q before
2011 * the insertion using this generic function.
2013 * This function should also be useful for request stacking drivers
2014 * in some cases below, so export this function.
2015 * Request stacking drivers like request-based dm may change the queue
2016 * limits while requests are in the queue (e.g. dm's table swapping).
2017 * Such request stacking drivers should check those requests against
2018 * the new queue limits again when they dispatch those requests,
2019 * although such checkings are also done against the old queue limits
2020 * when submitting requests.
2022 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
2024 if (!rq_mergeable(rq))
2027 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2028 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2033 * queue's settings related to segment counting like q->bounce_pfn
2034 * may differ from that of other stacking queues.
2035 * Recalculate it to check the request correctly on this queue's
2038 blk_recalc_rq_segments(rq);
2039 if (rq->nr_phys_segments > queue_max_segments(q)) {
2040 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2046 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
2049 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2050 * @q: the queue to submit the request
2051 * @rq: the request being queued
2053 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2055 unsigned long flags;
2056 int where = ELEVATOR_INSERT_BACK;
2058 if (blk_rq_check_limits(q, rq))
2062 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2066 if (blk_queue_io_stat(q))
2067 blk_account_io_start(rq, true);
2068 blk_mq_insert_request(rq, false, true, true);
2072 spin_lock_irqsave(q->queue_lock, flags);
2073 if (unlikely(blk_queue_dying(q))) {
2074 spin_unlock_irqrestore(q->queue_lock, flags);
2079 * Submitting request must be dequeued before calling this function
2080 * because it will be linked to another request_queue
2082 BUG_ON(blk_queued_rq(rq));
2084 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2085 where = ELEVATOR_INSERT_FLUSH;
2087 add_acct_request(q, rq, where);
2088 if (where == ELEVATOR_INSERT_FLUSH)
2090 spin_unlock_irqrestore(q->queue_lock, flags);
2094 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2097 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2098 * @rq: request to examine
2101 * A request could be merge of IOs which require different failure
2102 * handling. This function determines the number of bytes which
2103 * can be failed from the beginning of the request without
2104 * crossing into area which need to be retried further.
2107 * The number of bytes to fail.
2110 * queue_lock must be held.
2112 unsigned int blk_rq_err_bytes(const struct request *rq)
2114 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2115 unsigned int bytes = 0;
2118 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2119 return blk_rq_bytes(rq);
2122 * Currently the only 'mixing' which can happen is between
2123 * different fastfail types. We can safely fail portions
2124 * which have all the failfast bits that the first one has -
2125 * the ones which are at least as eager to fail as the first
2128 for (bio = rq->bio; bio; bio = bio->bi_next) {
2129 if ((bio->bi_rw & ff) != ff)
2131 bytes += bio->bi_iter.bi_size;
2134 /* this could lead to infinite loop */
2135 BUG_ON(blk_rq_bytes(rq) && !bytes);
2138 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2140 void blk_account_io_completion(struct request *req, unsigned int bytes)
2142 if (blk_do_io_stat(req)) {
2143 const int rw = rq_data_dir(req);
2144 struct hd_struct *part;
2147 cpu = part_stat_lock();
2149 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2154 void blk_account_io_done(struct request *req)
2157 * Account IO completion. flush_rq isn't accounted as a
2158 * normal IO on queueing nor completion. Accounting the
2159 * containing request is enough.
2161 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2162 unsigned long duration = jiffies - req->start_time;
2163 const int rw = rq_data_dir(req);
2164 struct hd_struct *part;
2167 cpu = part_stat_lock();
2170 part_stat_inc(cpu, part, ios[rw]);
2171 part_stat_add(cpu, part, ticks[rw], duration);
2172 part_round_stats(cpu, part);
2173 part_dec_in_flight(part, rw);
2175 hd_struct_put(part);
2182 * Don't process normal requests when queue is suspended
2183 * or in the process of suspending/resuming
2185 static struct request *blk_pm_peek_request(struct request_queue *q,
2188 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2189 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2195 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2202 void blk_account_io_start(struct request *rq, bool new_io)
2204 struct hd_struct *part;
2205 int rw = rq_data_dir(rq);
2208 if (!blk_do_io_stat(rq))
2211 cpu = part_stat_lock();
2215 part_stat_inc(cpu, part, merges[rw]);
2217 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2218 if (!hd_struct_try_get(part)) {
2220 * The partition is already being removed,
2221 * the request will be accounted on the disk only
2223 * We take a reference on disk->part0 although that
2224 * partition will never be deleted, so we can treat
2225 * it as any other partition.
2227 part = &rq->rq_disk->part0;
2228 hd_struct_get(part);
2230 part_round_stats(cpu, part);
2231 part_inc_in_flight(part, rw);
2239 * blk_peek_request - peek at the top of a request queue
2240 * @q: request queue to peek at
2243 * Return the request at the top of @q. The returned request
2244 * should be started using blk_start_request() before LLD starts
2248 * Pointer to the request at the top of @q if available. Null
2252 * queue_lock must be held.
2254 struct request *blk_peek_request(struct request_queue *q)
2259 while ((rq = __elv_next_request(q)) != NULL) {
2261 rq = blk_pm_peek_request(q, rq);
2265 if (!(rq->cmd_flags & REQ_STARTED)) {
2267 * This is the first time the device driver
2268 * sees this request (possibly after
2269 * requeueing). Notify IO scheduler.
2271 if (rq->cmd_flags & REQ_SORTED)
2272 elv_activate_rq(q, rq);
2275 * just mark as started even if we don't start
2276 * it, a request that has been delayed should
2277 * not be passed by new incoming requests
2279 rq->cmd_flags |= REQ_STARTED;
2280 trace_block_rq_issue(q, rq);
2283 if (!q->boundary_rq || q->boundary_rq == rq) {
2284 q->end_sector = rq_end_sector(rq);
2285 q->boundary_rq = NULL;
2288 if (rq->cmd_flags & REQ_DONTPREP)
2291 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2293 * make sure space for the drain appears we
2294 * know we can do this because max_hw_segments
2295 * has been adjusted to be one fewer than the
2298 rq->nr_phys_segments++;
2304 ret = q->prep_rq_fn(q, rq);
2305 if (ret == BLKPREP_OK) {
2307 } else if (ret == BLKPREP_DEFER) {
2309 * the request may have been (partially) prepped.
2310 * we need to keep this request in the front to
2311 * avoid resource deadlock. REQ_STARTED will
2312 * prevent other fs requests from passing this one.
2314 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2315 !(rq->cmd_flags & REQ_DONTPREP)) {
2317 * remove the space for the drain we added
2318 * so that we don't add it again
2320 --rq->nr_phys_segments;
2325 } else if (ret == BLKPREP_KILL) {
2326 rq->cmd_flags |= REQ_QUIET;
2328 * Mark this request as started so we don't trigger
2329 * any debug logic in the end I/O path.
2331 blk_start_request(rq);
2332 __blk_end_request_all(rq, -EIO);
2334 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2341 EXPORT_SYMBOL(blk_peek_request);
2343 void blk_dequeue_request(struct request *rq)
2345 struct request_queue *q = rq->q;
2347 BUG_ON(list_empty(&rq->queuelist));
2348 BUG_ON(ELV_ON_HASH(rq));
2350 list_del_init(&rq->queuelist);
2353 * the time frame between a request being removed from the lists
2354 * and to it is freed is accounted as io that is in progress at
2357 if (blk_account_rq(rq)) {
2358 q->in_flight[rq_is_sync(rq)]++;
2359 set_io_start_time_ns(rq);
2364 * blk_start_request - start request processing on the driver
2365 * @req: request to dequeue
2368 * Dequeue @req and start timeout timer on it. This hands off the
2369 * request to the driver.
2371 * Block internal functions which don't want to start timer should
2372 * call blk_dequeue_request().
2375 * queue_lock must be held.
2377 void blk_start_request(struct request *req)
2379 blk_dequeue_request(req);
2382 * We are now handing the request to the hardware, initialize
2383 * resid_len to full count and add the timeout handler.
2385 req->resid_len = blk_rq_bytes(req);
2386 if (unlikely(blk_bidi_rq(req)))
2387 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2389 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2392 EXPORT_SYMBOL(blk_start_request);
2395 * blk_fetch_request - fetch a request from a request queue
2396 * @q: request queue to fetch a request from
2399 * Return the request at the top of @q. The request is started on
2400 * return and LLD can start processing it immediately.
2403 * Pointer to the request at the top of @q if available. Null
2407 * queue_lock must be held.
2409 struct request *blk_fetch_request(struct request_queue *q)
2413 rq = blk_peek_request(q);
2415 blk_start_request(rq);
2418 EXPORT_SYMBOL(blk_fetch_request);
2421 * blk_update_request - Special helper function for request stacking drivers
2422 * @req: the request being processed
2423 * @error: %0 for success, < %0 for error
2424 * @nr_bytes: number of bytes to complete @req
2427 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2428 * the request structure even if @req doesn't have leftover.
2429 * If @req has leftover, sets it up for the next range of segments.
2431 * This special helper function is only for request stacking drivers
2432 * (e.g. request-based dm) so that they can handle partial completion.
2433 * Actual device drivers should use blk_end_request instead.
2435 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2436 * %false return from this function.
2439 * %false - this request doesn't have any more data
2440 * %true - this request has more data
2442 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2446 trace_block_rq_complete(req->q, req, nr_bytes);
2452 * For fs requests, rq is just carrier of independent bio's
2453 * and each partial completion should be handled separately.
2454 * Reset per-request error on each partial completion.
2456 * TODO: tj: This is too subtle. It would be better to let
2457 * low level drivers do what they see fit.
2459 if (req->cmd_type == REQ_TYPE_FS)
2462 if (error && req->cmd_type == REQ_TYPE_FS &&
2463 !(req->cmd_flags & REQ_QUIET)) {
2468 error_type = "recoverable transport";
2471 error_type = "critical target";
2474 error_type = "critical nexus";
2477 error_type = "timeout";
2480 error_type = "critical space allocation";
2483 error_type = "critical medium";
2490 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2491 __func__, error_type, req->rq_disk ?
2492 req->rq_disk->disk_name : "?",
2493 (unsigned long long)blk_rq_pos(req));
2497 blk_account_io_completion(req, nr_bytes);
2501 struct bio *bio = req->bio;
2502 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2504 if (bio_bytes == bio->bi_iter.bi_size)
2505 req->bio = bio->bi_next;
2507 req_bio_endio(req, bio, bio_bytes, error);
2509 total_bytes += bio_bytes;
2510 nr_bytes -= bio_bytes;
2521 * Reset counters so that the request stacking driver
2522 * can find how many bytes remain in the request
2525 req->__data_len = 0;
2529 req->__data_len -= total_bytes;
2531 /* update sector only for requests with clear definition of sector */
2532 if (req->cmd_type == REQ_TYPE_FS)
2533 req->__sector += total_bytes >> 9;
2535 /* mixed attributes always follow the first bio */
2536 if (req->cmd_flags & REQ_MIXED_MERGE) {
2537 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2538 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2542 * If total number of sectors is less than the first segment
2543 * size, something has gone terribly wrong.
2545 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2546 blk_dump_rq_flags(req, "request botched");
2547 req->__data_len = blk_rq_cur_bytes(req);
2550 /* recalculate the number of segments */
2551 blk_recalc_rq_segments(req);
2555 EXPORT_SYMBOL_GPL(blk_update_request);
2557 static bool blk_update_bidi_request(struct request *rq, int error,
2558 unsigned int nr_bytes,
2559 unsigned int bidi_bytes)
2561 if (blk_update_request(rq, error, nr_bytes))
2564 /* Bidi request must be completed as a whole */
2565 if (unlikely(blk_bidi_rq(rq)) &&
2566 blk_update_request(rq->next_rq, error, bidi_bytes))
2569 if (blk_queue_add_random(rq->q))
2570 add_disk_randomness(rq->rq_disk);
2576 * blk_unprep_request - unprepare a request
2579 * This function makes a request ready for complete resubmission (or
2580 * completion). It happens only after all error handling is complete,
2581 * so represents the appropriate moment to deallocate any resources
2582 * that were allocated to the request in the prep_rq_fn. The queue
2583 * lock is held when calling this.
2585 void blk_unprep_request(struct request *req)
2587 struct request_queue *q = req->q;
2589 req->cmd_flags &= ~REQ_DONTPREP;
2590 if (q->unprep_rq_fn)
2591 q->unprep_rq_fn(q, req);
2593 EXPORT_SYMBOL_GPL(blk_unprep_request);
2596 * queue lock must be held
2598 void blk_finish_request(struct request *req, int error)
2600 if (req->cmd_flags & REQ_QUEUED)
2601 blk_queue_end_tag(req->q, req);
2603 BUG_ON(blk_queued_rq(req));
2605 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2606 laptop_io_completion(&req->q->backing_dev_info);
2608 blk_delete_timer(req);
2610 if (req->cmd_flags & REQ_DONTPREP)
2611 blk_unprep_request(req);
2613 blk_account_io_done(req);
2616 req->end_io(req, error);
2618 if (blk_bidi_rq(req))
2619 __blk_put_request(req->next_rq->q, req->next_rq);
2621 __blk_put_request(req->q, req);
2624 EXPORT_SYMBOL(blk_finish_request);
2627 * blk_end_bidi_request - Complete a bidi request
2628 * @rq: the request to complete
2629 * @error: %0 for success, < %0 for error
2630 * @nr_bytes: number of bytes to complete @rq
2631 * @bidi_bytes: number of bytes to complete @rq->next_rq
2634 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2635 * Drivers that supports bidi can safely call this member for any
2636 * type of request, bidi or uni. In the later case @bidi_bytes is
2640 * %false - we are done with this request
2641 * %true - still buffers pending for this request
2643 static bool blk_end_bidi_request(struct request *rq, int error,
2644 unsigned int nr_bytes, unsigned int bidi_bytes)
2646 struct request_queue *q = rq->q;
2647 unsigned long flags;
2649 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2652 spin_lock_irqsave(q->queue_lock, flags);
2653 blk_finish_request(rq, error);
2654 spin_unlock_irqrestore(q->queue_lock, flags);
2660 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2661 * @rq: the request to complete
2662 * @error: %0 for success, < %0 for error
2663 * @nr_bytes: number of bytes to complete @rq
2664 * @bidi_bytes: number of bytes to complete @rq->next_rq
2667 * Identical to blk_end_bidi_request() except that queue lock is
2668 * assumed to be locked on entry and remains so on return.
2671 * %false - we are done with this request
2672 * %true - still buffers pending for this request
2674 bool __blk_end_bidi_request(struct request *rq, int error,
2675 unsigned int nr_bytes, unsigned int bidi_bytes)
2677 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2680 blk_finish_request(rq, error);
2686 * blk_end_request - Helper function for drivers to complete the request.
2687 * @rq: the request being processed
2688 * @error: %0 for success, < %0 for error
2689 * @nr_bytes: number of bytes to complete
2692 * Ends I/O on a number of bytes attached to @rq.
2693 * If @rq has leftover, sets it up for the next range of segments.
2696 * %false - we are done with this request
2697 * %true - still buffers pending for this request
2699 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2701 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2703 EXPORT_SYMBOL(blk_end_request);
2706 * blk_end_request_all - Helper function for drives to finish the request.
2707 * @rq: the request to finish
2708 * @error: %0 for success, < %0 for error
2711 * Completely finish @rq.
2713 void blk_end_request_all(struct request *rq, int error)
2716 unsigned int bidi_bytes = 0;
2718 if (unlikely(blk_bidi_rq(rq)))
2719 bidi_bytes = blk_rq_bytes(rq->next_rq);
2721 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2724 EXPORT_SYMBOL(blk_end_request_all);
2727 * blk_end_request_cur - Helper function to finish the current request chunk.
2728 * @rq: the request to finish the current chunk for
2729 * @error: %0 for success, < %0 for error
2732 * Complete the current consecutively mapped chunk from @rq.
2735 * %false - we are done with this request
2736 * %true - still buffers pending for this request
2738 bool blk_end_request_cur(struct request *rq, int error)
2740 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2742 EXPORT_SYMBOL(blk_end_request_cur);
2745 * blk_end_request_err - Finish a request till the next failure boundary.
2746 * @rq: the request to finish till the next failure boundary for
2747 * @error: must be negative errno
2750 * Complete @rq till the next failure boundary.
2753 * %false - we are done with this request
2754 * %true - still buffers pending for this request
2756 bool blk_end_request_err(struct request *rq, int error)
2758 WARN_ON(error >= 0);
2759 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2761 EXPORT_SYMBOL_GPL(blk_end_request_err);
2764 * __blk_end_request - Helper function for drivers to complete the request.
2765 * @rq: the request being processed
2766 * @error: %0 for success, < %0 for error
2767 * @nr_bytes: number of bytes to complete
2770 * Must be called with queue lock held unlike blk_end_request().
2773 * %false - we are done with this request
2774 * %true - still buffers pending for this request
2776 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2778 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2780 EXPORT_SYMBOL(__blk_end_request);
2783 * __blk_end_request_all - Helper function for drives to finish the request.
2784 * @rq: the request to finish
2785 * @error: %0 for success, < %0 for error
2788 * Completely finish @rq. Must be called with queue lock held.
2790 void __blk_end_request_all(struct request *rq, int error)
2793 unsigned int bidi_bytes = 0;
2795 if (unlikely(blk_bidi_rq(rq)))
2796 bidi_bytes = blk_rq_bytes(rq->next_rq);
2798 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2801 EXPORT_SYMBOL(__blk_end_request_all);
2804 * __blk_end_request_cur - Helper function to finish the current request chunk.
2805 * @rq: the request to finish the current chunk for
2806 * @error: %0 for success, < %0 for error
2809 * Complete the current consecutively mapped chunk from @rq. Must
2810 * be called with queue lock held.
2813 * %false - we are done with this request
2814 * %true - still buffers pending for this request
2816 bool __blk_end_request_cur(struct request *rq, int error)
2818 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2820 EXPORT_SYMBOL(__blk_end_request_cur);
2823 * __blk_end_request_err - Finish a request till the next failure boundary.
2824 * @rq: the request to finish till the next failure boundary for
2825 * @error: must be negative errno
2828 * Complete @rq till the next failure boundary. Must be called
2829 * with queue lock held.
2832 * %false - we are done with this request
2833 * %true - still buffers pending for this request
2835 bool __blk_end_request_err(struct request *rq, int error)
2837 WARN_ON(error >= 0);
2838 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2840 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2842 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2845 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2846 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2848 if (bio_has_data(bio))
2849 rq->nr_phys_segments = bio_phys_segments(q, bio);
2851 rq->__data_len = bio->bi_iter.bi_size;
2852 rq->bio = rq->biotail = bio;
2855 rq->rq_disk = bio->bi_bdev->bd_disk;
2858 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2860 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2861 * @rq: the request to be flushed
2864 * Flush all pages in @rq.
2866 void rq_flush_dcache_pages(struct request *rq)
2868 struct req_iterator iter;
2869 struct bio_vec bvec;
2871 rq_for_each_segment(bvec, rq, iter)
2872 flush_dcache_page(bvec.bv_page);
2874 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2878 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2879 * @q : the queue of the device being checked
2882 * Check if underlying low-level drivers of a device are busy.
2883 * If the drivers want to export their busy state, they must set own
2884 * exporting function using blk_queue_lld_busy() first.
2886 * Basically, this function is used only by request stacking drivers
2887 * to stop dispatching requests to underlying devices when underlying
2888 * devices are busy. This behavior helps more I/O merging on the queue
2889 * of the request stacking driver and prevents I/O throughput regression
2890 * on burst I/O load.
2893 * 0 - Not busy (The request stacking driver should dispatch request)
2894 * 1 - Busy (The request stacking driver should stop dispatching request)
2896 int blk_lld_busy(struct request_queue *q)
2899 return q->lld_busy_fn(q);
2903 EXPORT_SYMBOL_GPL(blk_lld_busy);
2906 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2907 * @rq: the clone request to be cleaned up
2910 * Free all bios in @rq for a cloned request.
2912 void blk_rq_unprep_clone(struct request *rq)
2916 while ((bio = rq->bio) != NULL) {
2917 rq->bio = bio->bi_next;
2922 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2925 * Copy attributes of the original request to the clone request.
2926 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
2928 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2930 dst->cpu = src->cpu;
2931 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2932 dst->cmd_type = src->cmd_type;
2933 dst->__sector = blk_rq_pos(src);
2934 dst->__data_len = blk_rq_bytes(src);
2935 dst->nr_phys_segments = src->nr_phys_segments;
2936 dst->ioprio = src->ioprio;
2937 dst->extra_len = src->extra_len;
2941 * blk_rq_prep_clone - Helper function to setup clone request
2942 * @rq: the request to be setup
2943 * @rq_src: original request to be cloned
2944 * @bs: bio_set that bios for clone are allocated from
2945 * @gfp_mask: memory allocation mask for bio
2946 * @bio_ctr: setup function to be called for each clone bio.
2947 * Returns %0 for success, non %0 for failure.
2948 * @data: private data to be passed to @bio_ctr
2951 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2952 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
2953 * are not copied, and copying such parts is the caller's responsibility.
2954 * Also, pages which the original bios are pointing to are not copied
2955 * and the cloned bios just point same pages.
2956 * So cloned bios must be completed before original bios, which means
2957 * the caller must complete @rq before @rq_src.
2959 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2960 struct bio_set *bs, gfp_t gfp_mask,
2961 int (*bio_ctr)(struct bio *, struct bio *, void *),
2964 struct bio *bio, *bio_src;
2969 __rq_for_each_bio(bio_src, rq_src) {
2970 bio = bio_clone_fast(bio_src, gfp_mask, bs);
2974 if (bio_ctr && bio_ctr(bio, bio_src, data))
2978 rq->biotail->bi_next = bio;
2981 rq->bio = rq->biotail = bio;
2984 __blk_rq_prep_clone(rq, rq_src);
2991 blk_rq_unprep_clone(rq);
2995 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2997 int kblockd_schedule_work(struct work_struct *work)
2999 return queue_work(kblockd_workqueue, work);
3001 EXPORT_SYMBOL(kblockd_schedule_work);
3003 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3004 unsigned long delay)
3006 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3008 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3010 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3011 unsigned long delay)
3013 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3015 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3018 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3019 * @plug: The &struct blk_plug that needs to be initialized
3022 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3023 * pending I/O should the task end up blocking between blk_start_plug() and
3024 * blk_finish_plug(). This is important from a performance perspective, but
3025 * also ensures that we don't deadlock. For instance, if the task is blocking
3026 * for a memory allocation, memory reclaim could end up wanting to free a
3027 * page belonging to that request that is currently residing in our private
3028 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3029 * this kind of deadlock.
3031 void blk_start_plug(struct blk_plug *plug)
3033 struct task_struct *tsk = current;
3035 INIT_LIST_HEAD(&plug->list);
3036 INIT_LIST_HEAD(&plug->mq_list);
3037 INIT_LIST_HEAD(&plug->cb_list);
3040 * If this is a nested plug, don't actually assign it. It will be
3041 * flushed on its own.
3045 * Store ordering should not be needed here, since a potential
3046 * preempt will imply a full memory barrier
3051 EXPORT_SYMBOL(blk_start_plug);
3053 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3055 struct request *rqa = container_of(a, struct request, queuelist);
3056 struct request *rqb = container_of(b, struct request, queuelist);
3058 return !(rqa->q < rqb->q ||
3059 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3063 * If 'from_schedule' is true, then postpone the dispatch of requests
3064 * until a safe kblockd context. We due this to avoid accidental big
3065 * additional stack usage in driver dispatch, in places where the originally
3066 * plugger did not intend it.
3068 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3070 __releases(q->queue_lock)
3072 trace_block_unplug(q, depth, !from_schedule);
3075 blk_run_queue_async(q);
3078 spin_unlock(q->queue_lock);
3081 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3083 LIST_HEAD(callbacks);
3085 while (!list_empty(&plug->cb_list)) {
3086 list_splice_init(&plug->cb_list, &callbacks);
3088 while (!list_empty(&callbacks)) {
3089 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3092 list_del(&cb->list);
3093 cb->callback(cb, from_schedule);
3098 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3101 struct blk_plug *plug = current->plug;
3102 struct blk_plug_cb *cb;
3107 list_for_each_entry(cb, &plug->cb_list, list)
3108 if (cb->callback == unplug && cb->data == data)
3111 /* Not currently on the callback list */
3112 BUG_ON(size < sizeof(*cb));
3113 cb = kzalloc(size, GFP_ATOMIC);
3116 cb->callback = unplug;
3117 list_add(&cb->list, &plug->cb_list);
3121 EXPORT_SYMBOL(blk_check_plugged);
3123 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3125 struct request_queue *q;
3126 unsigned long flags;
3131 flush_plug_callbacks(plug, from_schedule);
3133 if (!list_empty(&plug->mq_list))
3134 blk_mq_flush_plug_list(plug, from_schedule);
3136 if (list_empty(&plug->list))
3139 list_splice_init(&plug->list, &list);
3141 list_sort(NULL, &list, plug_rq_cmp);
3147 * Save and disable interrupts here, to avoid doing it for every
3148 * queue lock we have to take.
3150 local_irq_save(flags);
3151 while (!list_empty(&list)) {
3152 rq = list_entry_rq(list.next);
3153 list_del_init(&rq->queuelist);
3157 * This drops the queue lock
3160 queue_unplugged(q, depth, from_schedule);
3163 spin_lock(q->queue_lock);
3167 * Short-circuit if @q is dead
3169 if (unlikely(blk_queue_dying(q))) {
3170 __blk_end_request_all(rq, -ENODEV);
3175 * rq is already accounted, so use raw insert
3177 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3178 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3180 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3186 * This drops the queue lock
3189 queue_unplugged(q, depth, from_schedule);
3191 local_irq_restore(flags);
3194 void blk_finish_plug(struct blk_plug *plug)
3196 blk_flush_plug_list(plug, false);
3198 if (plug == current->plug)
3199 current->plug = NULL;
3201 EXPORT_SYMBOL(blk_finish_plug);
3205 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3206 * @q: the queue of the device
3207 * @dev: the device the queue belongs to
3210 * Initialize runtime-PM-related fields for @q and start auto suspend for
3211 * @dev. Drivers that want to take advantage of request-based runtime PM
3212 * should call this function after @dev has been initialized, and its
3213 * request queue @q has been allocated, and runtime PM for it can not happen
3214 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3215 * cases, driver should call this function before any I/O has taken place.
3217 * This function takes care of setting up using auto suspend for the device,
3218 * the autosuspend delay is set to -1 to make runtime suspend impossible
3219 * until an updated value is either set by user or by driver. Drivers do
3220 * not need to touch other autosuspend settings.
3222 * The block layer runtime PM is request based, so only works for drivers
3223 * that use request as their IO unit instead of those directly use bio's.
3225 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3228 q->rpm_status = RPM_ACTIVE;
3229 pm_runtime_set_autosuspend_delay(q->dev, -1);
3230 pm_runtime_use_autosuspend(q->dev);
3232 EXPORT_SYMBOL(blk_pm_runtime_init);
3235 * blk_pre_runtime_suspend - Pre runtime suspend check
3236 * @q: the queue of the device
3239 * This function will check if runtime suspend is allowed for the device
3240 * by examining if there are any requests pending in the queue. If there
3241 * are requests pending, the device can not be runtime suspended; otherwise,
3242 * the queue's status will be updated to SUSPENDING and the driver can
3243 * proceed to suspend the device.
3245 * For the not allowed case, we mark last busy for the device so that
3246 * runtime PM core will try to autosuspend it some time later.
3248 * This function should be called near the start of the device's
3249 * runtime_suspend callback.
3252 * 0 - OK to runtime suspend the device
3253 * -EBUSY - Device should not be runtime suspended
3255 int blk_pre_runtime_suspend(struct request_queue *q)
3259 spin_lock_irq(q->queue_lock);
3260 if (q->nr_pending) {
3262 pm_runtime_mark_last_busy(q->dev);
3264 q->rpm_status = RPM_SUSPENDING;
3266 spin_unlock_irq(q->queue_lock);
3269 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3272 * blk_post_runtime_suspend - Post runtime suspend processing
3273 * @q: the queue of the device
3274 * @err: return value of the device's runtime_suspend function
3277 * Update the queue's runtime status according to the return value of the
3278 * device's runtime suspend function and mark last busy for the device so
3279 * that PM core will try to auto suspend the device at a later time.
3281 * This function should be called near the end of the device's
3282 * runtime_suspend callback.
3284 void blk_post_runtime_suspend(struct request_queue *q, int err)
3286 spin_lock_irq(q->queue_lock);
3288 q->rpm_status = RPM_SUSPENDED;
3290 q->rpm_status = RPM_ACTIVE;
3291 pm_runtime_mark_last_busy(q->dev);
3293 spin_unlock_irq(q->queue_lock);
3295 EXPORT_SYMBOL(blk_post_runtime_suspend);
3298 * blk_pre_runtime_resume - Pre runtime resume processing
3299 * @q: the queue of the device
3302 * Update the queue's runtime status to RESUMING in preparation for the
3303 * runtime resume of the device.
3305 * This function should be called near the start of the device's
3306 * runtime_resume callback.
3308 void blk_pre_runtime_resume(struct request_queue *q)
3310 spin_lock_irq(q->queue_lock);
3311 q->rpm_status = RPM_RESUMING;
3312 spin_unlock_irq(q->queue_lock);
3314 EXPORT_SYMBOL(blk_pre_runtime_resume);
3317 * blk_post_runtime_resume - Post runtime resume processing
3318 * @q: the queue of the device
3319 * @err: return value of the device's runtime_resume function
3322 * Update the queue's runtime status according to the return value of the
3323 * device's runtime_resume function. If it is successfully resumed, process
3324 * the requests that are queued into the device's queue when it is resuming
3325 * and then mark last busy and initiate autosuspend for it.
3327 * This function should be called near the end of the device's
3328 * runtime_resume callback.
3330 void blk_post_runtime_resume(struct request_queue *q, int err)
3332 spin_lock_irq(q->queue_lock);
3334 q->rpm_status = RPM_ACTIVE;
3336 pm_runtime_mark_last_busy(q->dev);
3337 pm_request_autosuspend(q->dev);
3339 q->rpm_status = RPM_SUSPENDED;
3341 spin_unlock_irq(q->queue_lock);
3343 EXPORT_SYMBOL(blk_post_runtime_resume);
3346 int __init blk_dev_init(void)
3348 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3349 sizeof(((struct request *)0)->cmd_flags));
3351 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3352 kblockd_workqueue = alloc_workqueue("kblockd",
3353 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3354 if (!kblockd_workqueue)
3355 panic("Failed to create kblockd\n");
3357 request_cachep = kmem_cache_create("blkdev_requests",
3358 sizeof(struct request), 0, SLAB_PANIC, NULL);
3360 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3361 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);