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, buffer %p, len %u\n",
150 rq->bio, rq->biotail, rq->buffer, 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->delayed_work);
257 cancel_delayed_work_sync(&q->delay_work);
260 EXPORT_SYMBOL(blk_sync_queue);
263 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
264 * @q: The queue to run
267 * Invoke request handling on a queue if there are any pending requests.
268 * May be used to restart request handling after a request has completed.
269 * This variant runs the queue whether or not the queue has been
270 * stopped. Must be called with the queue lock held and interrupts
271 * disabled. See also @blk_run_queue.
273 inline void __blk_run_queue_uncond(struct request_queue *q)
275 if (unlikely(blk_queue_dead(q)))
279 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
280 * the queue lock internally. As a result multiple threads may be
281 * running such a request function concurrently. Keep track of the
282 * number of active request_fn invocations such that blk_drain_queue()
283 * can wait until all these request_fn calls have finished.
285 q->request_fn_active++;
287 q->request_fn_active--;
291 * __blk_run_queue - run a single device queue
292 * @q: The queue to run
295 * See @blk_run_queue. This variant must be called with the queue lock
296 * held and interrupts disabled.
298 void __blk_run_queue(struct request_queue *q)
300 if (unlikely(blk_queue_stopped(q)))
303 __blk_run_queue_uncond(q);
305 EXPORT_SYMBOL(__blk_run_queue);
308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
313 * of us. The caller must hold the queue lock.
315 void blk_run_queue_async(struct request_queue *q)
317 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
318 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
320 EXPORT_SYMBOL(blk_run_queue_async);
323 * blk_run_queue - run a single device queue
324 * @q: The queue to run
327 * Invoke request handling on this queue, if it has pending work to do.
328 * May be used to restart queueing when a request has completed.
330 void blk_run_queue(struct request_queue *q)
334 spin_lock_irqsave(q->queue_lock, flags);
336 spin_unlock_irqrestore(q->queue_lock, flags);
338 EXPORT_SYMBOL(blk_run_queue);
340 void blk_put_queue(struct request_queue *q)
342 kobject_put(&q->kobj);
344 EXPORT_SYMBOL(blk_put_queue);
347 * __blk_drain_queue - drain requests from request_queue
349 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
351 * Drain requests from @q. If @drain_all is set, all requests are drained.
352 * If not, only ELVPRIV requests are drained. The caller is responsible
353 * for ensuring that no new requests which need to be drained are queued.
355 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
356 __releases(q->queue_lock)
357 __acquires(q->queue_lock)
361 lockdep_assert_held(q->queue_lock);
367 * The caller might be trying to drain @q before its
368 * elevator is initialized.
371 elv_drain_elevator(q);
373 blkcg_drain_queue(q);
376 * This function might be called on a queue which failed
377 * driver init after queue creation or is not yet fully
378 * active yet. Some drivers (e.g. fd and loop) get unhappy
379 * in such cases. Kick queue iff dispatch queue has
380 * something on it and @q has request_fn set.
382 if (!list_empty(&q->queue_head) && q->request_fn)
385 drain |= q->nr_rqs_elvpriv;
386 drain |= q->request_fn_active;
389 * Unfortunately, requests are queued at and tracked from
390 * multiple places and there's no single counter which can
391 * be drained. Check all the queues and counters.
394 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];
398 drain |= !list_empty(&q->flush_queue[i]);
405 spin_unlock_irq(q->queue_lock);
409 spin_lock_irq(q->queue_lock);
413 * With queue marked dead, any woken up waiter will fail the
414 * allocation path, so the wakeup chaining is lost and we're
415 * left with hung waiters. We need to wake up those waiters.
418 struct request_list *rl;
420 blk_queue_for_each_rl(rl, q)
421 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
422 wake_up_all(&rl->wait[i]);
427 * blk_queue_bypass_start - enter queue bypass mode
428 * @q: queue of interest
430 * In bypass mode, only the dispatch FIFO queue of @q is used. This
431 * function makes @q enter bypass mode and drains all requests which were
432 * throttled or issued before. On return, it's guaranteed that no request
433 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
434 * inside queue or RCU read lock.
436 void blk_queue_bypass_start(struct request_queue *q)
440 spin_lock_irq(q->queue_lock);
441 drain = !q->bypass_depth++;
442 queue_flag_set(QUEUE_FLAG_BYPASS, q);
443 spin_unlock_irq(q->queue_lock);
446 spin_lock_irq(q->queue_lock);
447 __blk_drain_queue(q, false);
448 spin_unlock_irq(q->queue_lock);
450 /* ensure blk_queue_bypass() is %true inside RCU read lock */
454 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
457 * blk_queue_bypass_end - leave queue bypass mode
458 * @q: queue of interest
460 * Leave bypass mode and restore the normal queueing behavior.
462 void blk_queue_bypass_end(struct request_queue *q)
464 spin_lock_irq(q->queue_lock);
465 if (!--q->bypass_depth)
466 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
467 WARN_ON_ONCE(q->bypass_depth < 0);
468 spin_unlock_irq(q->queue_lock);
470 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
473 * blk_cleanup_queue - shutdown a request queue
474 * @q: request queue to shutdown
476 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
477 * put it. All future requests will be failed immediately with -ENODEV.
479 void blk_cleanup_queue(struct request_queue *q)
481 spinlock_t *lock = q->queue_lock;
483 /* mark @q DYING, no new request or merges will be allowed afterwards */
484 mutex_lock(&q->sysfs_lock);
485 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
489 * A dying queue is permanently in bypass mode till released. Note
490 * that, unlike blk_queue_bypass_start(), we aren't performing
491 * synchronize_rcu() after entering bypass mode to avoid the delay
492 * as some drivers create and destroy a lot of queues while
493 * probing. This is still safe because blk_release_queue() will be
494 * called only after the queue refcnt drops to zero and nothing,
495 * RCU or not, would be traversing the queue by then.
498 queue_flag_set(QUEUE_FLAG_BYPASS, q);
500 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
501 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
502 queue_flag_set(QUEUE_FLAG_DYING, q);
503 spin_unlock_irq(lock);
504 mutex_unlock(&q->sysfs_lock);
507 * Drain all requests queued before DYING marking. Set DEAD flag to
508 * prevent that q->request_fn() gets invoked after draining finished.
511 blk_mq_drain_queue(q);
515 __blk_drain_queue(q, true);
517 queue_flag_set(QUEUE_FLAG_DEAD, q);
518 spin_unlock_irq(lock);
520 /* @q won't process any more request, flush async actions */
521 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
525 if (q->queue_lock != &q->__queue_lock)
526 q->queue_lock = &q->__queue_lock;
527 spin_unlock_irq(lock);
529 /* @q is and will stay empty, shutdown and put */
532 EXPORT_SYMBOL(blk_cleanup_queue);
534 int blk_init_rl(struct request_list *rl, struct request_queue *q,
537 if (unlikely(rl->rq_pool))
541 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
542 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
543 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
544 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
546 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
547 mempool_free_slab, request_cachep,
555 void blk_exit_rl(struct request_list *rl)
558 mempool_destroy(rl->rq_pool);
561 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
563 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
565 EXPORT_SYMBOL(blk_alloc_queue);
567 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
569 struct request_queue *q;
572 q = kmem_cache_alloc_node(blk_requestq_cachep,
573 gfp_mask | __GFP_ZERO, node_id);
577 if (percpu_counter_init(&q->mq_usage_counter, 0))
580 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
584 q->backing_dev_info.ra_pages =
585 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
586 q->backing_dev_info.state = 0;
587 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
588 q->backing_dev_info.name = "block";
591 err = bdi_init(&q->backing_dev_info);
595 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
596 laptop_mode_timer_fn, (unsigned long) q);
597 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
598 INIT_LIST_HEAD(&q->queue_head);
599 INIT_LIST_HEAD(&q->timeout_list);
600 INIT_LIST_HEAD(&q->icq_list);
601 #ifdef CONFIG_BLK_CGROUP
602 INIT_LIST_HEAD(&q->blkg_list);
604 INIT_LIST_HEAD(&q->flush_queue[0]);
605 INIT_LIST_HEAD(&q->flush_queue[1]);
606 INIT_LIST_HEAD(&q->flush_data_in_flight);
607 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
609 kobject_init(&q->kobj, &blk_queue_ktype);
611 mutex_init(&q->sysfs_lock);
612 spin_lock_init(&q->__queue_lock);
615 * By default initialize queue_lock to internal lock and driver can
616 * override it later if need be.
618 q->queue_lock = &q->__queue_lock;
621 * A queue starts its life with bypass turned on to avoid
622 * unnecessary bypass on/off overhead and nasty surprises during
623 * init. The initial bypass will be finished when the queue is
624 * registered by blk_register_queue().
627 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
629 init_waitqueue_head(&q->mq_freeze_wq);
631 if (blkcg_init_queue(q))
637 bdi_destroy(&q->backing_dev_info);
639 ida_simple_remove(&blk_queue_ida, q->id);
641 percpu_counter_destroy(&q->mq_usage_counter);
643 kmem_cache_free(blk_requestq_cachep, q);
646 EXPORT_SYMBOL(blk_alloc_queue_node);
649 * blk_init_queue - prepare a request queue for use with a block device
650 * @rfn: The function to be called to process requests that have been
651 * placed on the queue.
652 * @lock: Request queue spin lock
655 * If a block device wishes to use the standard request handling procedures,
656 * which sorts requests and coalesces adjacent requests, then it must
657 * call blk_init_queue(). The function @rfn will be called when there
658 * are requests on the queue that need to be processed. If the device
659 * supports plugging, then @rfn may not be called immediately when requests
660 * are available on the queue, but may be called at some time later instead.
661 * Plugged queues are generally unplugged when a buffer belonging to one
662 * of the requests on the queue is needed, or due to memory pressure.
664 * @rfn is not required, or even expected, to remove all requests off the
665 * queue, but only as many as it can handle at a time. If it does leave
666 * requests on the queue, it is responsible for arranging that the requests
667 * get dealt with eventually.
669 * The queue spin lock must be held while manipulating the requests on the
670 * request queue; this lock will be taken also from interrupt context, so irq
671 * disabling is needed for it.
673 * Function returns a pointer to the initialized request queue, or %NULL if
677 * blk_init_queue() must be paired with a blk_cleanup_queue() call
678 * when the block device is deactivated (such as at module unload).
681 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
683 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
685 EXPORT_SYMBOL(blk_init_queue);
687 struct request_queue *
688 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
690 struct request_queue *uninit_q, *q;
692 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
696 q = blk_init_allocated_queue(uninit_q, rfn, lock);
698 blk_cleanup_queue(uninit_q);
702 EXPORT_SYMBOL(blk_init_queue_node);
704 struct request_queue *
705 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
711 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
715 q->prep_rq_fn = NULL;
716 q->unprep_rq_fn = NULL;
717 q->queue_flags |= QUEUE_FLAG_DEFAULT;
719 /* Override internal queue lock with supplied lock pointer */
721 q->queue_lock = lock;
724 * This also sets hw/phys segments, boundary and size
726 blk_queue_make_request(q, blk_queue_bio);
728 q->sg_reserved_size = INT_MAX;
730 /* Protect q->elevator from elevator_change */
731 mutex_lock(&q->sysfs_lock);
734 if (elevator_init(q, NULL)) {
735 mutex_unlock(&q->sysfs_lock);
739 mutex_unlock(&q->sysfs_lock);
743 EXPORT_SYMBOL(blk_init_allocated_queue);
745 bool blk_get_queue(struct request_queue *q)
747 if (likely(!blk_queue_dying(q))) {
754 EXPORT_SYMBOL(blk_get_queue);
756 static inline void blk_free_request(struct request_list *rl, struct request *rq)
758 if (rq->cmd_flags & REQ_ELVPRIV) {
759 elv_put_request(rl->q, rq);
761 put_io_context(rq->elv.icq->ioc);
764 mempool_free(rq, rl->rq_pool);
768 * ioc_batching returns true if the ioc is a valid batching request and
769 * should be given priority access to a request.
771 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
777 * Make sure the process is able to allocate at least 1 request
778 * even if the batch times out, otherwise we could theoretically
781 return ioc->nr_batch_requests == q->nr_batching ||
782 (ioc->nr_batch_requests > 0
783 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
787 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
788 * will cause the process to be a "batcher" on all queues in the system. This
789 * is the behaviour we want though - once it gets a wakeup it should be given
792 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
794 if (!ioc || ioc_batching(q, ioc))
797 ioc->nr_batch_requests = q->nr_batching;
798 ioc->last_waited = jiffies;
801 static void __freed_request(struct request_list *rl, int sync)
803 struct request_queue *q = rl->q;
806 * bdi isn't aware of blkcg yet. As all async IOs end up root
807 * blkcg anyway, just use root blkcg state.
809 if (rl == &q->root_rl &&
810 rl->count[sync] < queue_congestion_off_threshold(q))
811 blk_clear_queue_congested(q, sync);
813 if (rl->count[sync] + 1 <= q->nr_requests) {
814 if (waitqueue_active(&rl->wait[sync]))
815 wake_up(&rl->wait[sync]);
817 blk_clear_rl_full(rl, sync);
822 * A request has just been released. Account for it, update the full and
823 * congestion status, wake up any waiters. Called under q->queue_lock.
825 static void freed_request(struct request_list *rl, unsigned int flags)
827 struct request_queue *q = rl->q;
828 int sync = rw_is_sync(flags);
832 if (flags & REQ_ELVPRIV)
835 __freed_request(rl, sync);
837 if (unlikely(rl->starved[sync ^ 1]))
838 __freed_request(rl, sync ^ 1);
842 * Determine if elevator data should be initialized when allocating the
843 * request associated with @bio.
845 static bool blk_rq_should_init_elevator(struct bio *bio)
851 * Flush requests do not use the elevator so skip initialization.
852 * This allows a request to share the flush and elevator data.
854 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
861 * rq_ioc - determine io_context for request allocation
862 * @bio: request being allocated is for this bio (can be %NULL)
864 * Determine io_context to use for request allocation for @bio. May return
865 * %NULL if %current->io_context doesn't exist.
867 static struct io_context *rq_ioc(struct bio *bio)
869 #ifdef CONFIG_BLK_CGROUP
870 if (bio && bio->bi_ioc)
873 return current->io_context;
877 * __get_request - get a free request
878 * @rl: request list to allocate from
879 * @rw_flags: RW and SYNC flags
880 * @bio: bio to allocate request for (can be %NULL)
881 * @gfp_mask: allocation mask
883 * Get a free request from @q. This function may fail under memory
884 * pressure or if @q is dead.
886 * Must be callled with @q->queue_lock held and,
887 * Returns %NULL on failure, with @q->queue_lock held.
888 * Returns !%NULL on success, with @q->queue_lock *not held*.
890 static struct request *__get_request(struct request_list *rl, int rw_flags,
891 struct bio *bio, gfp_t gfp_mask)
893 struct request_queue *q = rl->q;
895 struct elevator_type *et = q->elevator->type;
896 struct io_context *ioc = rq_ioc(bio);
897 struct io_cq *icq = NULL;
898 const bool is_sync = rw_is_sync(rw_flags) != 0;
901 if (unlikely(blk_queue_dying(q)))
904 may_queue = elv_may_queue(q, rw_flags);
905 if (may_queue == ELV_MQUEUE_NO)
908 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
909 if (rl->count[is_sync]+1 >= q->nr_requests) {
911 * The queue will fill after this allocation, so set
912 * it as full, and mark this process as "batching".
913 * This process will be allowed to complete a batch of
914 * requests, others will be blocked.
916 if (!blk_rl_full(rl, is_sync)) {
917 ioc_set_batching(q, ioc);
918 blk_set_rl_full(rl, is_sync);
920 if (may_queue != ELV_MQUEUE_MUST
921 && !ioc_batching(q, ioc)) {
923 * The queue is full and the allocating
924 * process is not a "batcher", and not
925 * exempted by the IO scheduler
932 * bdi isn't aware of blkcg yet. As all async IOs end up
933 * root blkcg anyway, just use root blkcg state.
935 if (rl == &q->root_rl)
936 blk_set_queue_congested(q, is_sync);
940 * Only allow batching queuers to allocate up to 50% over the defined
941 * limit of requests, otherwise we could have thousands of requests
942 * allocated with any setting of ->nr_requests
944 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
947 q->nr_rqs[is_sync]++;
948 rl->count[is_sync]++;
949 rl->starved[is_sync] = 0;
952 * Decide whether the new request will be managed by elevator. If
953 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
954 * prevent the current elevator from being destroyed until the new
955 * request is freed. This guarantees icq's won't be destroyed and
956 * makes creating new ones safe.
958 * Also, lookup icq while holding queue_lock. If it doesn't exist,
959 * it will be created after releasing queue_lock.
961 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
962 rw_flags |= REQ_ELVPRIV;
964 if (et->icq_cache && ioc)
965 icq = ioc_lookup_icq(ioc, q);
968 if (blk_queue_io_stat(q))
969 rw_flags |= REQ_IO_STAT;
970 spin_unlock_irq(q->queue_lock);
972 /* allocate and init request */
973 rq = mempool_alloc(rl->rq_pool, gfp_mask);
978 blk_rq_set_rl(rq, rl);
979 rq->cmd_flags = rw_flags | REQ_ALLOCED;
982 if (rw_flags & REQ_ELVPRIV) {
983 if (unlikely(et->icq_cache && !icq)) {
985 icq = ioc_create_icq(ioc, q, gfp_mask);
991 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
994 /* @rq->elv.icq holds io_context until @rq is freed */
996 get_io_context(icq->ioc);
1000 * ioc may be NULL here, and ioc_batching will be false. That's
1001 * OK, if the queue is under the request limit then requests need
1002 * not count toward the nr_batch_requests limit. There will always
1003 * be some limit enforced by BLK_BATCH_TIME.
1005 if (ioc_batching(q, ioc))
1006 ioc->nr_batch_requests--;
1008 trace_block_getrq(q, bio, rw_flags & 1);
1013 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1014 * and may fail indefinitely under memory pressure and thus
1015 * shouldn't stall IO. Treat this request as !elvpriv. This will
1016 * disturb iosched and blkcg but weird is bettern than dead.
1018 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1019 dev_name(q->backing_dev_info.dev));
1021 rq->cmd_flags &= ~REQ_ELVPRIV;
1024 spin_lock_irq(q->queue_lock);
1025 q->nr_rqs_elvpriv--;
1026 spin_unlock_irq(q->queue_lock);
1031 * Allocation failed presumably due to memory. Undo anything we
1032 * might have messed up.
1034 * Allocating task should really be put onto the front of the wait
1035 * queue, but this is pretty rare.
1037 spin_lock_irq(q->queue_lock);
1038 freed_request(rl, rw_flags);
1041 * in the very unlikely event that allocation failed and no
1042 * requests for this direction was pending, mark us starved so that
1043 * freeing of a request in the other direction will notice
1044 * us. another possible fix would be to split the rq mempool into
1048 if (unlikely(rl->count[is_sync] == 0))
1049 rl->starved[is_sync] = 1;
1054 * get_request - get a free request
1055 * @q: request_queue to allocate request from
1056 * @rw_flags: RW and SYNC flags
1057 * @bio: bio to allocate request for (can be %NULL)
1058 * @gfp_mask: allocation mask
1060 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1061 * function keeps retrying under memory pressure and fails iff @q is dead.
1063 * Must be callled with @q->queue_lock held and,
1064 * Returns %NULL on failure, with @q->queue_lock held.
1065 * Returns !%NULL on success, with @q->queue_lock *not held*.
1067 static struct request *get_request(struct request_queue *q, int rw_flags,
1068 struct bio *bio, gfp_t gfp_mask)
1070 const bool is_sync = rw_is_sync(rw_flags) != 0;
1072 struct request_list *rl;
1075 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1077 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1081 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1086 /* wait on @rl and retry */
1087 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1088 TASK_UNINTERRUPTIBLE);
1090 trace_block_sleeprq(q, bio, rw_flags & 1);
1092 spin_unlock_irq(q->queue_lock);
1096 * After sleeping, we become a "batching" process and will be able
1097 * to allocate at least one request, and up to a big batch of them
1098 * for a small period time. See ioc_batching, ioc_set_batching
1100 ioc_set_batching(q, current->io_context);
1102 spin_lock_irq(q->queue_lock);
1103 finish_wait(&rl->wait[is_sync], &wait);
1108 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1113 BUG_ON(rw != READ && rw != WRITE);
1115 /* create ioc upfront */
1116 create_io_context(gfp_mask, q->node);
1118 spin_lock_irq(q->queue_lock);
1119 rq = get_request(q, rw, NULL, gfp_mask);
1121 spin_unlock_irq(q->queue_lock);
1122 /* q->queue_lock is unlocked at this point */
1127 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1130 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1132 return blk_old_get_request(q, rw, gfp_mask);
1134 EXPORT_SYMBOL(blk_get_request);
1137 * blk_make_request - given a bio, allocate a corresponding struct request.
1138 * @q: target request queue
1139 * @bio: The bio describing the memory mappings that will be submitted for IO.
1140 * It may be a chained-bio properly constructed by block/bio layer.
1141 * @gfp_mask: gfp flags to be used for memory allocation
1143 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1144 * type commands. Where the struct request needs to be farther initialized by
1145 * the caller. It is passed a &struct bio, which describes the memory info of
1148 * The caller of blk_make_request must make sure that bi_io_vec
1149 * are set to describe the memory buffers. That bio_data_dir() will return
1150 * the needed direction of the request. (And all bio's in the passed bio-chain
1151 * are properly set accordingly)
1153 * If called under none-sleepable conditions, mapped bio buffers must not
1154 * need bouncing, by calling the appropriate masked or flagged allocator,
1155 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1158 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1159 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1160 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1161 * completion of a bio that hasn't been submitted yet, thus resulting in a
1162 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1163 * of bio_alloc(), as that avoids the mempool deadlock.
1164 * If possible a big IO should be split into smaller parts when allocation
1165 * fails. Partial allocation should not be an error, or you risk a live-lock.
1167 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1170 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1173 return ERR_PTR(-ENOMEM);
1176 struct bio *bounce_bio = bio;
1179 blk_queue_bounce(q, &bounce_bio);
1180 ret = blk_rq_append_bio(q, rq, bounce_bio);
1181 if (unlikely(ret)) {
1182 blk_put_request(rq);
1183 return ERR_PTR(ret);
1189 EXPORT_SYMBOL(blk_make_request);
1192 * blk_requeue_request - put a request back on queue
1193 * @q: request queue where request should be inserted
1194 * @rq: request to be inserted
1197 * Drivers often keep queueing requests until the hardware cannot accept
1198 * more, when that condition happens we need to put the request back
1199 * on the queue. Must be called with queue lock held.
1201 void blk_requeue_request(struct request_queue *q, struct request *rq)
1203 blk_delete_timer(rq);
1204 blk_clear_rq_complete(rq);
1205 trace_block_rq_requeue(q, rq);
1207 if (blk_rq_tagged(rq))
1208 blk_queue_end_tag(q, rq);
1210 BUG_ON(blk_queued_rq(rq));
1212 elv_requeue_request(q, rq);
1214 EXPORT_SYMBOL(blk_requeue_request);
1216 static void add_acct_request(struct request_queue *q, struct request *rq,
1219 blk_account_io_start(rq, true);
1220 __elv_add_request(q, rq, where);
1223 static void part_round_stats_single(int cpu, struct hd_struct *part,
1226 if (now == part->stamp)
1229 if (part_in_flight(part)) {
1230 __part_stat_add(cpu, part, time_in_queue,
1231 part_in_flight(part) * (now - part->stamp));
1232 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1238 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1239 * @cpu: cpu number for stats access
1240 * @part: target partition
1242 * The average IO queue length and utilisation statistics are maintained
1243 * by observing the current state of the queue length and the amount of
1244 * time it has been in this state for.
1246 * Normally, that accounting is done on IO completion, but that can result
1247 * in more than a second's worth of IO being accounted for within any one
1248 * second, leading to >100% utilisation. To deal with that, we call this
1249 * function to do a round-off before returning the results when reading
1250 * /proc/diskstats. This accounts immediately for all queue usage up to
1251 * the current jiffies and restarts the counters again.
1253 void part_round_stats(int cpu, struct hd_struct *part)
1255 unsigned long now = jiffies;
1258 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1259 part_round_stats_single(cpu, part, now);
1261 EXPORT_SYMBOL_GPL(part_round_stats);
1263 #ifdef CONFIG_PM_RUNTIME
1264 static void blk_pm_put_request(struct request *rq)
1266 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1267 pm_runtime_mark_last_busy(rq->q->dev);
1270 static inline void blk_pm_put_request(struct request *rq) {}
1274 * queue lock must be held
1276 void __blk_put_request(struct request_queue *q, struct request *req)
1282 blk_mq_free_request(req);
1286 blk_pm_put_request(req);
1288 elv_completed_request(q, req);
1290 /* this is a bio leak */
1291 WARN_ON(req->bio != NULL);
1294 * Request may not have originated from ll_rw_blk. if not,
1295 * it didn't come out of our reserved rq pools
1297 if (req->cmd_flags & REQ_ALLOCED) {
1298 unsigned int flags = req->cmd_flags;
1299 struct request_list *rl = blk_rq_rl(req);
1301 BUG_ON(!list_empty(&req->queuelist));
1302 BUG_ON(!hlist_unhashed(&req->hash));
1304 blk_free_request(rl, req);
1305 freed_request(rl, flags);
1309 EXPORT_SYMBOL_GPL(__blk_put_request);
1311 void blk_put_request(struct request *req)
1313 struct request_queue *q = req->q;
1316 blk_mq_free_request(req);
1318 unsigned long flags;
1320 spin_lock_irqsave(q->queue_lock, flags);
1321 __blk_put_request(q, req);
1322 spin_unlock_irqrestore(q->queue_lock, flags);
1325 EXPORT_SYMBOL(blk_put_request);
1328 * blk_add_request_payload - add a payload to a request
1329 * @rq: request to update
1330 * @page: page backing the payload
1331 * @len: length of the payload.
1333 * This allows to later add a payload to an already submitted request by
1334 * a block driver. The driver needs to take care of freeing the payload
1337 * Note that this is a quite horrible hack and nothing but handling of
1338 * discard requests should ever use it.
1340 void blk_add_request_payload(struct request *rq, struct page *page,
1343 struct bio *bio = rq->bio;
1345 bio->bi_io_vec->bv_page = page;
1346 bio->bi_io_vec->bv_offset = 0;
1347 bio->bi_io_vec->bv_len = len;
1349 bio->bi_iter.bi_size = len;
1351 bio->bi_phys_segments = 1;
1353 rq->__data_len = rq->resid_len = len;
1354 rq->nr_phys_segments = 1;
1355 rq->buffer = bio_data(bio);
1357 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1359 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1362 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1364 if (!ll_back_merge_fn(q, req, bio))
1367 trace_block_bio_backmerge(q, req, bio);
1369 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1370 blk_rq_set_mixed_merge(req);
1372 req->biotail->bi_next = bio;
1374 req->__data_len += bio->bi_iter.bi_size;
1375 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1377 blk_account_io_start(req, false);
1381 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1384 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1386 if (!ll_front_merge_fn(q, req, bio))
1389 trace_block_bio_frontmerge(q, req, bio);
1391 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1392 blk_rq_set_mixed_merge(req);
1394 bio->bi_next = req->bio;
1398 * may not be valid. if the low level driver said
1399 * it didn't need a bounce buffer then it better
1400 * not touch req->buffer either...
1402 req->buffer = bio_data(bio);
1403 req->__sector = bio->bi_iter.bi_sector;
1404 req->__data_len += bio->bi_iter.bi_size;
1405 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1407 blk_account_io_start(req, false);
1412 * blk_attempt_plug_merge - try to merge with %current's plugged list
1413 * @q: request_queue new bio is being queued at
1414 * @bio: new bio being queued
1415 * @request_count: out parameter for number of traversed plugged requests
1417 * Determine whether @bio being queued on @q can be merged with a request
1418 * on %current's plugged list. Returns %true if merge was successful,
1421 * Plugging coalesces IOs from the same issuer for the same purpose without
1422 * going through @q->queue_lock. As such it's more of an issuing mechanism
1423 * than scheduling, and the request, while may have elvpriv data, is not
1424 * added on the elevator at this point. In addition, we don't have
1425 * reliable access to the elevator outside queue lock. Only check basic
1426 * merging parameters without querying the elevator.
1428 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1429 unsigned int *request_count)
1431 struct blk_plug *plug;
1434 struct list_head *plug_list;
1436 if (blk_queue_nomerges(q))
1439 plug = current->plug;
1445 plug_list = &plug->mq_list;
1447 plug_list = &plug->list;
1449 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1455 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1458 el_ret = blk_try_merge(rq, bio);
1459 if (el_ret == ELEVATOR_BACK_MERGE) {
1460 ret = bio_attempt_back_merge(q, rq, bio);
1463 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1464 ret = bio_attempt_front_merge(q, rq, bio);
1473 void init_request_from_bio(struct request *req, struct bio *bio)
1475 req->cmd_type = REQ_TYPE_FS;
1477 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1478 if (bio->bi_rw & REQ_RAHEAD)
1479 req->cmd_flags |= REQ_FAILFAST_MASK;
1482 req->__sector = bio->bi_iter.bi_sector;
1483 req->ioprio = bio_prio(bio);
1484 blk_rq_bio_prep(req->q, req, bio);
1487 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1489 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1490 struct blk_plug *plug;
1491 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1492 struct request *req;
1493 unsigned int request_count = 0;
1496 * low level driver can indicate that it wants pages above a
1497 * certain limit bounced to low memory (ie for highmem, or even
1498 * ISA dma in theory)
1500 blk_queue_bounce(q, &bio);
1502 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1503 bio_endio(bio, -EIO);
1507 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1508 spin_lock_irq(q->queue_lock);
1509 where = ELEVATOR_INSERT_FLUSH;
1514 * Check if we can merge with the plugged list before grabbing
1517 if (blk_attempt_plug_merge(q, bio, &request_count))
1520 spin_lock_irq(q->queue_lock);
1522 el_ret = elv_merge(q, &req, bio);
1523 if (el_ret == ELEVATOR_BACK_MERGE) {
1524 if (bio_attempt_back_merge(q, req, bio)) {
1525 elv_bio_merged(q, req, bio);
1526 if (!attempt_back_merge(q, req))
1527 elv_merged_request(q, req, el_ret);
1530 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1531 if (bio_attempt_front_merge(q, req, bio)) {
1532 elv_bio_merged(q, req, bio);
1533 if (!attempt_front_merge(q, req))
1534 elv_merged_request(q, req, el_ret);
1541 * This sync check and mask will be re-done in init_request_from_bio(),
1542 * but we need to set it earlier to expose the sync flag to the
1543 * rq allocator and io schedulers.
1545 rw_flags = bio_data_dir(bio);
1547 rw_flags |= REQ_SYNC;
1550 * Grab a free request. This is might sleep but can not fail.
1551 * Returns with the queue unlocked.
1553 req = get_request(q, rw_flags, bio, GFP_NOIO);
1554 if (unlikely(!req)) {
1555 bio_endio(bio, -ENODEV); /* @q is dead */
1560 * After dropping the lock and possibly sleeping here, our request
1561 * may now be mergeable after it had proven unmergeable (above).
1562 * We don't worry about that case for efficiency. It won't happen
1563 * often, and the elevators are able to handle it.
1565 init_request_from_bio(req, bio);
1567 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1568 req->cpu = raw_smp_processor_id();
1570 plug = current->plug;
1573 * If this is the first request added after a plug, fire
1577 trace_block_plug(q);
1579 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1580 blk_flush_plug_list(plug, false);
1581 trace_block_plug(q);
1584 list_add_tail(&req->queuelist, &plug->list);
1585 blk_account_io_start(req, true);
1587 spin_lock_irq(q->queue_lock);
1588 add_acct_request(q, req, where);
1591 spin_unlock_irq(q->queue_lock);
1594 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1597 * If bio->bi_dev is a partition, remap the location
1599 static inline void blk_partition_remap(struct bio *bio)
1601 struct block_device *bdev = bio->bi_bdev;
1603 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1604 struct hd_struct *p = bdev->bd_part;
1606 bio->bi_iter.bi_sector += p->start_sect;
1607 bio->bi_bdev = bdev->bd_contains;
1609 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1611 bio->bi_iter.bi_sector - p->start_sect);
1615 static void handle_bad_sector(struct bio *bio)
1617 char b[BDEVNAME_SIZE];
1619 printk(KERN_INFO "attempt to access beyond end of device\n");
1620 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1621 bdevname(bio->bi_bdev, b),
1623 (unsigned long long)bio_end_sector(bio),
1624 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1626 set_bit(BIO_EOF, &bio->bi_flags);
1629 #ifdef CONFIG_FAIL_MAKE_REQUEST
1631 static DECLARE_FAULT_ATTR(fail_make_request);
1633 static int __init setup_fail_make_request(char *str)
1635 return setup_fault_attr(&fail_make_request, str);
1637 __setup("fail_make_request=", setup_fail_make_request);
1639 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1641 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1644 static int __init fail_make_request_debugfs(void)
1646 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1647 NULL, &fail_make_request);
1649 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1652 late_initcall(fail_make_request_debugfs);
1654 #else /* CONFIG_FAIL_MAKE_REQUEST */
1656 static inline bool should_fail_request(struct hd_struct *part,
1662 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1665 * Check whether this bio extends beyond the end of the device.
1667 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1674 /* Test device or partition size, when known. */
1675 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1677 sector_t sector = bio->bi_iter.bi_sector;
1679 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1681 * This may well happen - the kernel calls bread()
1682 * without checking the size of the device, e.g., when
1683 * mounting a device.
1685 handle_bad_sector(bio);
1693 static noinline_for_stack bool
1694 generic_make_request_checks(struct bio *bio)
1696 struct request_queue *q;
1697 int nr_sectors = bio_sectors(bio);
1699 char b[BDEVNAME_SIZE];
1700 struct hd_struct *part;
1704 if (bio_check_eod(bio, nr_sectors))
1707 q = bdev_get_queue(bio->bi_bdev);
1710 "generic_make_request: Trying to access "
1711 "nonexistent block-device %s (%Lu)\n",
1712 bdevname(bio->bi_bdev, b),
1713 (long long) bio->bi_iter.bi_sector);
1717 if (likely(bio_is_rw(bio) &&
1718 nr_sectors > queue_max_hw_sectors(q))) {
1719 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1720 bdevname(bio->bi_bdev, b),
1722 queue_max_hw_sectors(q));
1726 part = bio->bi_bdev->bd_part;
1727 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1728 should_fail_request(&part_to_disk(part)->part0,
1729 bio->bi_iter.bi_size))
1733 * If this device has partitions, remap block n
1734 * of partition p to block n+start(p) of the disk.
1736 blk_partition_remap(bio);
1738 if (bio_check_eod(bio, nr_sectors))
1742 * Filter flush bio's early so that make_request based
1743 * drivers without flush support don't have to worry
1746 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1747 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1754 if ((bio->bi_rw & REQ_DISCARD) &&
1755 (!blk_queue_discard(q) ||
1756 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1761 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1767 * Various block parts want %current->io_context and lazy ioc
1768 * allocation ends up trading a lot of pain for a small amount of
1769 * memory. Just allocate it upfront. This may fail and block
1770 * layer knows how to live with it.
1772 create_io_context(GFP_ATOMIC, q->node);
1774 if (blk_throtl_bio(q, bio))
1775 return false; /* throttled, will be resubmitted later */
1777 trace_block_bio_queue(q, bio);
1781 bio_endio(bio, err);
1786 * generic_make_request - hand a buffer to its device driver for I/O
1787 * @bio: The bio describing the location in memory and on the device.
1789 * generic_make_request() is used to make I/O requests of block
1790 * devices. It is passed a &struct bio, which describes the I/O that needs
1793 * generic_make_request() does not return any status. The
1794 * success/failure status of the request, along with notification of
1795 * completion, is delivered asynchronously through the bio->bi_end_io
1796 * function described (one day) else where.
1798 * The caller of generic_make_request must make sure that bi_io_vec
1799 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1800 * set to describe the device address, and the
1801 * bi_end_io and optionally bi_private are set to describe how
1802 * completion notification should be signaled.
1804 * generic_make_request and the drivers it calls may use bi_next if this
1805 * bio happens to be merged with someone else, and may resubmit the bio to
1806 * a lower device by calling into generic_make_request recursively, which
1807 * means the bio should NOT be touched after the call to ->make_request_fn.
1809 void generic_make_request(struct bio *bio)
1811 struct bio_list bio_list_on_stack;
1813 if (!generic_make_request_checks(bio))
1817 * We only want one ->make_request_fn to be active at a time, else
1818 * stack usage with stacked devices could be a problem. So use
1819 * current->bio_list to keep a list of requests submited by a
1820 * make_request_fn function. current->bio_list is also used as a
1821 * flag to say if generic_make_request is currently active in this
1822 * task or not. If it is NULL, then no make_request is active. If
1823 * it is non-NULL, then a make_request is active, and new requests
1824 * should be added at the tail
1826 if (current->bio_list) {
1827 bio_list_add(current->bio_list, bio);
1831 /* following loop may be a bit non-obvious, and so deserves some
1833 * Before entering the loop, bio->bi_next is NULL (as all callers
1834 * ensure that) so we have a list with a single bio.
1835 * We pretend that we have just taken it off a longer list, so
1836 * we assign bio_list to a pointer to the bio_list_on_stack,
1837 * thus initialising the bio_list of new bios to be
1838 * added. ->make_request() may indeed add some more bios
1839 * through a recursive call to generic_make_request. If it
1840 * did, we find a non-NULL value in bio_list and re-enter the loop
1841 * from the top. In this case we really did just take the bio
1842 * of the top of the list (no pretending) and so remove it from
1843 * bio_list, and call into ->make_request() again.
1845 BUG_ON(bio->bi_next);
1846 bio_list_init(&bio_list_on_stack);
1847 current->bio_list = &bio_list_on_stack;
1849 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1851 q->make_request_fn(q, bio);
1853 bio = bio_list_pop(current->bio_list);
1855 current->bio_list = NULL; /* deactivate */
1857 EXPORT_SYMBOL(generic_make_request);
1860 * submit_bio - submit a bio to the block device layer for I/O
1861 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1862 * @bio: The &struct bio which describes the I/O
1864 * submit_bio() is very similar in purpose to generic_make_request(), and
1865 * uses that function to do most of the work. Both are fairly rough
1866 * interfaces; @bio must be presetup and ready for I/O.
1869 void submit_bio(int rw, struct bio *bio)
1874 * If it's a regular read/write or a barrier with data attached,
1875 * go through the normal accounting stuff before submission.
1877 if (bio_has_data(bio)) {
1880 if (unlikely(rw & REQ_WRITE_SAME))
1881 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1883 count = bio_sectors(bio);
1886 count_vm_events(PGPGOUT, count);
1888 task_io_account_read(bio->bi_iter.bi_size);
1889 count_vm_events(PGPGIN, count);
1892 if (unlikely(block_dump)) {
1893 char b[BDEVNAME_SIZE];
1894 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1895 current->comm, task_pid_nr(current),
1896 (rw & WRITE) ? "WRITE" : "READ",
1897 (unsigned long long)bio->bi_iter.bi_sector,
1898 bdevname(bio->bi_bdev, b),
1903 generic_make_request(bio);
1905 EXPORT_SYMBOL(submit_bio);
1908 * blk_rq_check_limits - Helper function to check a request for the queue limit
1910 * @rq: the request being checked
1913 * @rq may have been made based on weaker limitations of upper-level queues
1914 * in request stacking drivers, and it may violate the limitation of @q.
1915 * Since the block layer and the underlying device driver trust @rq
1916 * after it is inserted to @q, it should be checked against @q before
1917 * the insertion using this generic function.
1919 * This function should also be useful for request stacking drivers
1920 * in some cases below, so export this function.
1921 * Request stacking drivers like request-based dm may change the queue
1922 * limits while requests are in the queue (e.g. dm's table swapping).
1923 * Such request stacking drivers should check those requests agaist
1924 * the new queue limits again when they dispatch those requests,
1925 * although such checkings are also done against the old queue limits
1926 * when submitting requests.
1928 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1930 if (!rq_mergeable(rq))
1933 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1934 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1939 * queue's settings related to segment counting like q->bounce_pfn
1940 * may differ from that of other stacking queues.
1941 * Recalculate it to check the request correctly on this queue's
1944 blk_recalc_rq_segments(rq);
1945 if (rq->nr_phys_segments > queue_max_segments(q)) {
1946 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1952 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1955 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1956 * @q: the queue to submit the request
1957 * @rq: the request being queued
1959 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1961 unsigned long flags;
1962 int where = ELEVATOR_INSERT_BACK;
1964 if (blk_rq_check_limits(q, rq))
1968 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1971 spin_lock_irqsave(q->queue_lock, flags);
1972 if (unlikely(blk_queue_dying(q))) {
1973 spin_unlock_irqrestore(q->queue_lock, flags);
1978 * Submitting request must be dequeued before calling this function
1979 * because it will be linked to another request_queue
1981 BUG_ON(blk_queued_rq(rq));
1983 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1984 where = ELEVATOR_INSERT_FLUSH;
1986 add_acct_request(q, rq, where);
1987 if (where == ELEVATOR_INSERT_FLUSH)
1989 spin_unlock_irqrestore(q->queue_lock, flags);
1993 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1996 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1997 * @rq: request to examine
2000 * A request could be merge of IOs which require different failure
2001 * handling. This function determines the number of bytes which
2002 * can be failed from the beginning of the request without
2003 * crossing into area which need to be retried further.
2006 * The number of bytes to fail.
2009 * queue_lock must be held.
2011 unsigned int blk_rq_err_bytes(const struct request *rq)
2013 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2014 unsigned int bytes = 0;
2017 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2018 return blk_rq_bytes(rq);
2021 * Currently the only 'mixing' which can happen is between
2022 * different fastfail types. We can safely fail portions
2023 * which have all the failfast bits that the first one has -
2024 * the ones which are at least as eager to fail as the first
2027 for (bio = rq->bio; bio; bio = bio->bi_next) {
2028 if ((bio->bi_rw & ff) != ff)
2030 bytes += bio->bi_iter.bi_size;
2033 /* this could lead to infinite loop */
2034 BUG_ON(blk_rq_bytes(rq) && !bytes);
2037 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2039 void blk_account_io_completion(struct request *req, unsigned int bytes)
2041 if (blk_do_io_stat(req)) {
2042 const int rw = rq_data_dir(req);
2043 struct hd_struct *part;
2046 cpu = part_stat_lock();
2048 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2053 void blk_account_io_done(struct request *req)
2056 * Account IO completion. flush_rq isn't accounted as a
2057 * normal IO on queueing nor completion. Accounting the
2058 * containing request is enough.
2060 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2061 unsigned long duration = jiffies - req->start_time;
2062 const int rw = rq_data_dir(req);
2063 struct hd_struct *part;
2066 cpu = part_stat_lock();
2069 part_stat_inc(cpu, part, ios[rw]);
2070 part_stat_add(cpu, part, ticks[rw], duration);
2071 part_round_stats(cpu, part);
2072 part_dec_in_flight(part, rw);
2074 hd_struct_put(part);
2079 #ifdef CONFIG_PM_RUNTIME
2081 * Don't process normal requests when queue is suspended
2082 * or in the process of suspending/resuming
2084 static struct request *blk_pm_peek_request(struct request_queue *q,
2087 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2088 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2094 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2101 void blk_account_io_start(struct request *rq, bool new_io)
2103 struct hd_struct *part;
2104 int rw = rq_data_dir(rq);
2107 if (!blk_do_io_stat(rq))
2110 cpu = part_stat_lock();
2114 part_stat_inc(cpu, part, merges[rw]);
2116 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2117 if (!hd_struct_try_get(part)) {
2119 * The partition is already being removed,
2120 * the request will be accounted on the disk only
2122 * We take a reference on disk->part0 although that
2123 * partition will never be deleted, so we can treat
2124 * it as any other partition.
2126 part = &rq->rq_disk->part0;
2127 hd_struct_get(part);
2129 part_round_stats(cpu, part);
2130 part_inc_in_flight(part, rw);
2138 * blk_peek_request - peek at the top of a request queue
2139 * @q: request queue to peek at
2142 * Return the request at the top of @q. The returned request
2143 * should be started using blk_start_request() before LLD starts
2147 * Pointer to the request at the top of @q if available. Null
2151 * queue_lock must be held.
2153 struct request *blk_peek_request(struct request_queue *q)
2158 while ((rq = __elv_next_request(q)) != NULL) {
2160 rq = blk_pm_peek_request(q, rq);
2164 if (!(rq->cmd_flags & REQ_STARTED)) {
2166 * This is the first time the device driver
2167 * sees this request (possibly after
2168 * requeueing). Notify IO scheduler.
2170 if (rq->cmd_flags & REQ_SORTED)
2171 elv_activate_rq(q, rq);
2174 * just mark as started even if we don't start
2175 * it, a request that has been delayed should
2176 * not be passed by new incoming requests
2178 rq->cmd_flags |= REQ_STARTED;
2179 trace_block_rq_issue(q, rq);
2182 if (!q->boundary_rq || q->boundary_rq == rq) {
2183 q->end_sector = rq_end_sector(rq);
2184 q->boundary_rq = NULL;
2187 if (rq->cmd_flags & REQ_DONTPREP)
2190 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2192 * make sure space for the drain appears we
2193 * know we can do this because max_hw_segments
2194 * has been adjusted to be one fewer than the
2197 rq->nr_phys_segments++;
2203 ret = q->prep_rq_fn(q, rq);
2204 if (ret == BLKPREP_OK) {
2206 } else if (ret == BLKPREP_DEFER) {
2208 * the request may have been (partially) prepped.
2209 * we need to keep this request in the front to
2210 * avoid resource deadlock. REQ_STARTED will
2211 * prevent other fs requests from passing this one.
2213 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2214 !(rq->cmd_flags & REQ_DONTPREP)) {
2216 * remove the space for the drain we added
2217 * so that we don't add it again
2219 --rq->nr_phys_segments;
2224 } else if (ret == BLKPREP_KILL) {
2225 rq->cmd_flags |= REQ_QUIET;
2227 * Mark this request as started so we don't trigger
2228 * any debug logic in the end I/O path.
2230 blk_start_request(rq);
2231 __blk_end_request_all(rq, -EIO);
2233 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2240 EXPORT_SYMBOL(blk_peek_request);
2242 void blk_dequeue_request(struct request *rq)
2244 struct request_queue *q = rq->q;
2246 BUG_ON(list_empty(&rq->queuelist));
2247 BUG_ON(ELV_ON_HASH(rq));
2249 list_del_init(&rq->queuelist);
2252 * the time frame between a request being removed from the lists
2253 * and to it is freed is accounted as io that is in progress at
2256 if (blk_account_rq(rq)) {
2257 q->in_flight[rq_is_sync(rq)]++;
2258 set_io_start_time_ns(rq);
2263 * blk_start_request - start request processing on the driver
2264 * @req: request to dequeue
2267 * Dequeue @req and start timeout timer on it. This hands off the
2268 * request to the driver.
2270 * Block internal functions which don't want to start timer should
2271 * call blk_dequeue_request().
2274 * queue_lock must be held.
2276 void blk_start_request(struct request *req)
2278 blk_dequeue_request(req);
2281 * We are now handing the request to the hardware, initialize
2282 * resid_len to full count and add the timeout handler.
2284 req->resid_len = blk_rq_bytes(req);
2285 if (unlikely(blk_bidi_rq(req)))
2286 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2288 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2291 EXPORT_SYMBOL(blk_start_request);
2294 * blk_fetch_request - fetch a request from a request queue
2295 * @q: request queue to fetch a request from
2298 * Return the request at the top of @q. The request is started on
2299 * return and LLD can start processing it immediately.
2302 * Pointer to the request at the top of @q if available. Null
2306 * queue_lock must be held.
2308 struct request *blk_fetch_request(struct request_queue *q)
2312 rq = blk_peek_request(q);
2314 blk_start_request(rq);
2317 EXPORT_SYMBOL(blk_fetch_request);
2320 * blk_update_request - Special helper function for request stacking drivers
2321 * @req: the request being processed
2322 * @error: %0 for success, < %0 for error
2323 * @nr_bytes: number of bytes to complete @req
2326 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2327 * the request structure even if @req doesn't have leftover.
2328 * If @req has leftover, sets it up for the next range of segments.
2330 * This special helper function is only for request stacking drivers
2331 * (e.g. request-based dm) so that they can handle partial completion.
2332 * Actual device drivers should use blk_end_request instead.
2334 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2335 * %false return from this function.
2338 * %false - this request doesn't have any more data
2339 * %true - this request has more data
2341 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2348 trace_block_rq_complete(req->q, req);
2351 * For fs requests, rq is just carrier of independent bio's
2352 * and each partial completion should be handled separately.
2353 * Reset per-request error on each partial completion.
2355 * TODO: tj: This is too subtle. It would be better to let
2356 * low level drivers do what they see fit.
2358 if (req->cmd_type == REQ_TYPE_FS)
2361 if (error && req->cmd_type == REQ_TYPE_FS &&
2362 !(req->cmd_flags & REQ_QUIET)) {
2367 error_type = "recoverable transport";
2370 error_type = "critical target";
2373 error_type = "critical nexus";
2376 error_type = "timeout";
2379 error_type = "critical space allocation";
2382 error_type = "critical medium";
2389 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2390 error_type, req->rq_disk ?
2391 req->rq_disk->disk_name : "?",
2392 (unsigned long long)blk_rq_pos(req));
2396 blk_account_io_completion(req, nr_bytes);
2400 struct bio *bio = req->bio;
2401 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2403 if (bio_bytes == bio->bi_iter.bi_size)
2404 req->bio = bio->bi_next;
2406 req_bio_endio(req, bio, bio_bytes, error);
2408 total_bytes += bio_bytes;
2409 nr_bytes -= bio_bytes;
2420 * Reset counters so that the request stacking driver
2421 * can find how many bytes remain in the request
2424 req->__data_len = 0;
2428 req->__data_len -= total_bytes;
2429 req->buffer = bio_data(req->bio);
2431 /* update sector only for requests with clear definition of sector */
2432 if (req->cmd_type == REQ_TYPE_FS)
2433 req->__sector += total_bytes >> 9;
2435 /* mixed attributes always follow the first bio */
2436 if (req->cmd_flags & REQ_MIXED_MERGE) {
2437 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2438 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2442 * If total number of sectors is less than the first segment
2443 * size, something has gone terribly wrong.
2445 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2446 blk_dump_rq_flags(req, "request botched");
2447 req->__data_len = blk_rq_cur_bytes(req);
2450 /* recalculate the number of segments */
2451 blk_recalc_rq_segments(req);
2455 EXPORT_SYMBOL_GPL(blk_update_request);
2457 static bool blk_update_bidi_request(struct request *rq, int error,
2458 unsigned int nr_bytes,
2459 unsigned int bidi_bytes)
2461 if (blk_update_request(rq, error, nr_bytes))
2464 /* Bidi request must be completed as a whole */
2465 if (unlikely(blk_bidi_rq(rq)) &&
2466 blk_update_request(rq->next_rq, error, bidi_bytes))
2469 if (blk_queue_add_random(rq->q))
2470 add_disk_randomness(rq->rq_disk);
2476 * blk_unprep_request - unprepare a request
2479 * This function makes a request ready for complete resubmission (or
2480 * completion). It happens only after all error handling is complete,
2481 * so represents the appropriate moment to deallocate any resources
2482 * that were allocated to the request in the prep_rq_fn. The queue
2483 * lock is held when calling this.
2485 void blk_unprep_request(struct request *req)
2487 struct request_queue *q = req->q;
2489 req->cmd_flags &= ~REQ_DONTPREP;
2490 if (q->unprep_rq_fn)
2491 q->unprep_rq_fn(q, req);
2493 EXPORT_SYMBOL_GPL(blk_unprep_request);
2496 * queue lock must be held
2498 static void blk_finish_request(struct request *req, int error)
2500 if (blk_rq_tagged(req))
2501 blk_queue_end_tag(req->q, req);
2503 BUG_ON(blk_queued_rq(req));
2505 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2506 laptop_io_completion(&req->q->backing_dev_info);
2508 blk_delete_timer(req);
2510 if (req->cmd_flags & REQ_DONTPREP)
2511 blk_unprep_request(req);
2513 blk_account_io_done(req);
2516 req->end_io(req, error);
2518 if (blk_bidi_rq(req))
2519 __blk_put_request(req->next_rq->q, req->next_rq);
2521 __blk_put_request(req->q, req);
2526 * blk_end_bidi_request - Complete a bidi request
2527 * @rq: the request to complete
2528 * @error: %0 for success, < %0 for error
2529 * @nr_bytes: number of bytes to complete @rq
2530 * @bidi_bytes: number of bytes to complete @rq->next_rq
2533 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2534 * Drivers that supports bidi can safely call this member for any
2535 * type of request, bidi or uni. In the later case @bidi_bytes is
2539 * %false - we are done with this request
2540 * %true - still buffers pending for this request
2542 static bool blk_end_bidi_request(struct request *rq, int error,
2543 unsigned int nr_bytes, unsigned int bidi_bytes)
2545 struct request_queue *q = rq->q;
2546 unsigned long flags;
2548 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2551 spin_lock_irqsave(q->queue_lock, flags);
2552 blk_finish_request(rq, error);
2553 spin_unlock_irqrestore(q->queue_lock, flags);
2559 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2560 * @rq: the request to complete
2561 * @error: %0 for success, < %0 for error
2562 * @nr_bytes: number of bytes to complete @rq
2563 * @bidi_bytes: number of bytes to complete @rq->next_rq
2566 * Identical to blk_end_bidi_request() except that queue lock is
2567 * assumed to be locked on entry and remains so on return.
2570 * %false - we are done with this request
2571 * %true - still buffers pending for this request
2573 bool __blk_end_bidi_request(struct request *rq, int error,
2574 unsigned int nr_bytes, unsigned int bidi_bytes)
2576 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2579 blk_finish_request(rq, error);
2585 * blk_end_request - Helper function for drivers to complete the request.
2586 * @rq: the request being processed
2587 * @error: %0 for success, < %0 for error
2588 * @nr_bytes: number of bytes to complete
2591 * Ends I/O on a number of bytes attached to @rq.
2592 * If @rq has leftover, sets it up for the next range of segments.
2595 * %false - we are done with this request
2596 * %true - still buffers pending for this request
2598 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2600 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2602 EXPORT_SYMBOL(blk_end_request);
2605 * blk_end_request_all - Helper function for drives to finish the request.
2606 * @rq: the request to finish
2607 * @error: %0 for success, < %0 for error
2610 * Completely finish @rq.
2612 void blk_end_request_all(struct request *rq, int error)
2615 unsigned int bidi_bytes = 0;
2617 if (unlikely(blk_bidi_rq(rq)))
2618 bidi_bytes = blk_rq_bytes(rq->next_rq);
2620 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2623 EXPORT_SYMBOL(blk_end_request_all);
2626 * blk_end_request_cur - Helper function to finish the current request chunk.
2627 * @rq: the request to finish the current chunk for
2628 * @error: %0 for success, < %0 for error
2631 * Complete the current consecutively mapped chunk from @rq.
2634 * %false - we are done with this request
2635 * %true - still buffers pending for this request
2637 bool blk_end_request_cur(struct request *rq, int error)
2639 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2641 EXPORT_SYMBOL(blk_end_request_cur);
2644 * blk_end_request_err - Finish a request till the next failure boundary.
2645 * @rq: the request to finish till the next failure boundary for
2646 * @error: must be negative errno
2649 * Complete @rq till the next failure boundary.
2652 * %false - we are done with this request
2653 * %true - still buffers pending for this request
2655 bool blk_end_request_err(struct request *rq, int error)
2657 WARN_ON(error >= 0);
2658 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2660 EXPORT_SYMBOL_GPL(blk_end_request_err);
2663 * __blk_end_request - Helper function for drivers to complete the request.
2664 * @rq: the request being processed
2665 * @error: %0 for success, < %0 for error
2666 * @nr_bytes: number of bytes to complete
2669 * Must be called with queue lock held unlike blk_end_request().
2672 * %false - we are done with this request
2673 * %true - still buffers pending for this request
2675 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2677 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2679 EXPORT_SYMBOL(__blk_end_request);
2682 * __blk_end_request_all - Helper function for drives to finish the request.
2683 * @rq: the request to finish
2684 * @error: %0 for success, < %0 for error
2687 * Completely finish @rq. Must be called with queue lock held.
2689 void __blk_end_request_all(struct request *rq, int error)
2692 unsigned int bidi_bytes = 0;
2694 if (unlikely(blk_bidi_rq(rq)))
2695 bidi_bytes = blk_rq_bytes(rq->next_rq);
2697 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2700 EXPORT_SYMBOL(__blk_end_request_all);
2703 * __blk_end_request_cur - Helper function to finish the current request chunk.
2704 * @rq: the request to finish the current chunk for
2705 * @error: %0 for success, < %0 for error
2708 * Complete the current consecutively mapped chunk from @rq. Must
2709 * be called with queue lock held.
2712 * %false - we are done with this request
2713 * %true - still buffers pending for this request
2715 bool __blk_end_request_cur(struct request *rq, int error)
2717 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2719 EXPORT_SYMBOL(__blk_end_request_cur);
2722 * __blk_end_request_err - Finish a request till the next failure boundary.
2723 * @rq: the request to finish till the next failure boundary for
2724 * @error: must be negative errno
2727 * Complete @rq till the next failure boundary. Must be called
2728 * with queue lock held.
2731 * %false - we are done with this request
2732 * %true - still buffers pending for this request
2734 bool __blk_end_request_err(struct request *rq, int error)
2736 WARN_ON(error >= 0);
2737 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2739 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2741 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2744 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2745 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2747 if (bio_has_data(bio)) {
2748 rq->nr_phys_segments = bio_phys_segments(q, bio);
2749 rq->buffer = bio_data(bio);
2751 rq->__data_len = bio->bi_iter.bi_size;
2752 rq->bio = rq->biotail = bio;
2755 rq->rq_disk = bio->bi_bdev->bd_disk;
2758 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2760 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2761 * @rq: the request to be flushed
2764 * Flush all pages in @rq.
2766 void rq_flush_dcache_pages(struct request *rq)
2768 struct req_iterator iter;
2769 struct bio_vec bvec;
2771 rq_for_each_segment(bvec, rq, iter)
2772 flush_dcache_page(bvec.bv_page);
2774 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2778 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2779 * @q : the queue of the device being checked
2782 * Check if underlying low-level drivers of a device are busy.
2783 * If the drivers want to export their busy state, they must set own
2784 * exporting function using blk_queue_lld_busy() first.
2786 * Basically, this function is used only by request stacking drivers
2787 * to stop dispatching requests to underlying devices when underlying
2788 * devices are busy. This behavior helps more I/O merging on the queue
2789 * of the request stacking driver and prevents I/O throughput regression
2790 * on burst I/O load.
2793 * 0 - Not busy (The request stacking driver should dispatch request)
2794 * 1 - Busy (The request stacking driver should stop dispatching request)
2796 int blk_lld_busy(struct request_queue *q)
2799 return q->lld_busy_fn(q);
2803 EXPORT_SYMBOL_GPL(blk_lld_busy);
2806 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2807 * @rq: the clone request to be cleaned up
2810 * Free all bios in @rq for a cloned request.
2812 void blk_rq_unprep_clone(struct request *rq)
2816 while ((bio = rq->bio) != NULL) {
2817 rq->bio = bio->bi_next;
2822 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2825 * Copy attributes of the original request to the clone request.
2826 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2828 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2830 dst->cpu = src->cpu;
2831 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2832 dst->cmd_type = src->cmd_type;
2833 dst->__sector = blk_rq_pos(src);
2834 dst->__data_len = blk_rq_bytes(src);
2835 dst->nr_phys_segments = src->nr_phys_segments;
2836 dst->ioprio = src->ioprio;
2837 dst->extra_len = src->extra_len;
2841 * blk_rq_prep_clone - Helper function to setup clone request
2842 * @rq: the request to be setup
2843 * @rq_src: original request to be cloned
2844 * @bs: bio_set that bios for clone are allocated from
2845 * @gfp_mask: memory allocation mask for bio
2846 * @bio_ctr: setup function to be called for each clone bio.
2847 * Returns %0 for success, non %0 for failure.
2848 * @data: private data to be passed to @bio_ctr
2851 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2852 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2853 * are not copied, and copying such parts is the caller's responsibility.
2854 * Also, pages which the original bios are pointing to are not copied
2855 * and the cloned bios just point same pages.
2856 * So cloned bios must be completed before original bios, which means
2857 * the caller must complete @rq before @rq_src.
2859 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2860 struct bio_set *bs, gfp_t gfp_mask,
2861 int (*bio_ctr)(struct bio *, struct bio *, void *),
2864 struct bio *bio, *bio_src;
2869 blk_rq_init(NULL, rq);
2871 __rq_for_each_bio(bio_src, rq_src) {
2872 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2876 if (bio_ctr && bio_ctr(bio, bio_src, data))
2880 rq->biotail->bi_next = bio;
2883 rq->bio = rq->biotail = bio;
2886 __blk_rq_prep_clone(rq, rq_src);
2893 blk_rq_unprep_clone(rq);
2897 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2899 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2901 return queue_work(kblockd_workqueue, work);
2903 EXPORT_SYMBOL(kblockd_schedule_work);
2905 int kblockd_schedule_delayed_work(struct request_queue *q,
2906 struct delayed_work *dwork, unsigned long delay)
2908 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2910 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2912 #define PLUG_MAGIC 0x91827364
2915 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2916 * @plug: The &struct blk_plug that needs to be initialized
2919 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2920 * pending I/O should the task end up blocking between blk_start_plug() and
2921 * blk_finish_plug(). This is important from a performance perspective, but
2922 * also ensures that we don't deadlock. For instance, if the task is blocking
2923 * for a memory allocation, memory reclaim could end up wanting to free a
2924 * page belonging to that request that is currently residing in our private
2925 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2926 * this kind of deadlock.
2928 void blk_start_plug(struct blk_plug *plug)
2930 struct task_struct *tsk = current;
2932 plug->magic = PLUG_MAGIC;
2933 INIT_LIST_HEAD(&plug->list);
2934 INIT_LIST_HEAD(&plug->mq_list);
2935 INIT_LIST_HEAD(&plug->cb_list);
2938 * If this is a nested plug, don't actually assign it. It will be
2939 * flushed on its own.
2943 * Store ordering should not be needed here, since a potential
2944 * preempt will imply a full memory barrier
2949 EXPORT_SYMBOL(blk_start_plug);
2951 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2953 struct request *rqa = container_of(a, struct request, queuelist);
2954 struct request *rqb = container_of(b, struct request, queuelist);
2956 return !(rqa->q < rqb->q ||
2957 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2961 * If 'from_schedule' is true, then postpone the dispatch of requests
2962 * until a safe kblockd context. We due this to avoid accidental big
2963 * additional stack usage in driver dispatch, in places where the originally
2964 * plugger did not intend it.
2966 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2968 __releases(q->queue_lock)
2970 trace_block_unplug(q, depth, !from_schedule);
2973 blk_run_queue_async(q);
2976 spin_unlock(q->queue_lock);
2979 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2981 LIST_HEAD(callbacks);
2983 while (!list_empty(&plug->cb_list)) {
2984 list_splice_init(&plug->cb_list, &callbacks);
2986 while (!list_empty(&callbacks)) {
2987 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2990 list_del(&cb->list);
2991 cb->callback(cb, from_schedule);
2996 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2999 struct blk_plug *plug = current->plug;
3000 struct blk_plug_cb *cb;
3005 list_for_each_entry(cb, &plug->cb_list, list)
3006 if (cb->callback == unplug && cb->data == data)
3009 /* Not currently on the callback list */
3010 BUG_ON(size < sizeof(*cb));
3011 cb = kzalloc(size, GFP_ATOMIC);
3014 cb->callback = unplug;
3015 list_add(&cb->list, &plug->cb_list);
3019 EXPORT_SYMBOL(blk_check_plugged);
3021 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3023 struct request_queue *q;
3024 unsigned long flags;
3029 BUG_ON(plug->magic != PLUG_MAGIC);
3031 flush_plug_callbacks(plug, from_schedule);
3033 if (!list_empty(&plug->mq_list))
3034 blk_mq_flush_plug_list(plug, from_schedule);
3036 if (list_empty(&plug->list))
3039 list_splice_init(&plug->list, &list);
3041 list_sort(NULL, &list, plug_rq_cmp);
3047 * Save and disable interrupts here, to avoid doing it for every
3048 * queue lock we have to take.
3050 local_irq_save(flags);
3051 while (!list_empty(&list)) {
3052 rq = list_entry_rq(list.next);
3053 list_del_init(&rq->queuelist);
3057 * This drops the queue lock
3060 queue_unplugged(q, depth, from_schedule);
3063 spin_lock(q->queue_lock);
3067 * Short-circuit if @q is dead
3069 if (unlikely(blk_queue_dying(q))) {
3070 __blk_end_request_all(rq, -ENODEV);
3075 * rq is already accounted, so use raw insert
3077 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3078 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3080 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3086 * This drops the queue lock
3089 queue_unplugged(q, depth, from_schedule);
3091 local_irq_restore(flags);
3094 void blk_finish_plug(struct blk_plug *plug)
3096 blk_flush_plug_list(plug, false);
3098 if (plug == current->plug)
3099 current->plug = NULL;
3101 EXPORT_SYMBOL(blk_finish_plug);
3103 #ifdef CONFIG_PM_RUNTIME
3105 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3106 * @q: the queue of the device
3107 * @dev: the device the queue belongs to
3110 * Initialize runtime-PM-related fields for @q and start auto suspend for
3111 * @dev. Drivers that want to take advantage of request-based runtime PM
3112 * should call this function after @dev has been initialized, and its
3113 * request queue @q has been allocated, and runtime PM for it can not happen
3114 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3115 * cases, driver should call this function before any I/O has taken place.
3117 * This function takes care of setting up using auto suspend for the device,
3118 * the autosuspend delay is set to -1 to make runtime suspend impossible
3119 * until an updated value is either set by user or by driver. Drivers do
3120 * not need to touch other autosuspend settings.
3122 * The block layer runtime PM is request based, so only works for drivers
3123 * that use request as their IO unit instead of those directly use bio's.
3125 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3128 q->rpm_status = RPM_ACTIVE;
3129 pm_runtime_set_autosuspend_delay(q->dev, -1);
3130 pm_runtime_use_autosuspend(q->dev);
3132 EXPORT_SYMBOL(blk_pm_runtime_init);
3135 * blk_pre_runtime_suspend - Pre runtime suspend check
3136 * @q: the queue of the device
3139 * This function will check if runtime suspend is allowed for the device
3140 * by examining if there are any requests pending in the queue. If there
3141 * are requests pending, the device can not be runtime suspended; otherwise,
3142 * the queue's status will be updated to SUSPENDING and the driver can
3143 * proceed to suspend the device.
3145 * For the not allowed case, we mark last busy for the device so that
3146 * runtime PM core will try to autosuspend it some time later.
3148 * This function should be called near the start of the device's
3149 * runtime_suspend callback.
3152 * 0 - OK to runtime suspend the device
3153 * -EBUSY - Device should not be runtime suspended
3155 int blk_pre_runtime_suspend(struct request_queue *q)
3159 spin_lock_irq(q->queue_lock);
3160 if (q->nr_pending) {
3162 pm_runtime_mark_last_busy(q->dev);
3164 q->rpm_status = RPM_SUSPENDING;
3166 spin_unlock_irq(q->queue_lock);
3169 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3172 * blk_post_runtime_suspend - Post runtime suspend processing
3173 * @q: the queue of the device
3174 * @err: return value of the device's runtime_suspend function
3177 * Update the queue's runtime status according to the return value of the
3178 * device's runtime suspend function and mark last busy for the device so
3179 * that PM core will try to auto suspend the device at a later time.
3181 * This function should be called near the end of the device's
3182 * runtime_suspend callback.
3184 void blk_post_runtime_suspend(struct request_queue *q, int err)
3186 spin_lock_irq(q->queue_lock);
3188 q->rpm_status = RPM_SUSPENDED;
3190 q->rpm_status = RPM_ACTIVE;
3191 pm_runtime_mark_last_busy(q->dev);
3193 spin_unlock_irq(q->queue_lock);
3195 EXPORT_SYMBOL(blk_post_runtime_suspend);
3198 * blk_pre_runtime_resume - Pre runtime resume processing
3199 * @q: the queue of the device
3202 * Update the queue's runtime status to RESUMING in preparation for the
3203 * runtime resume of the device.
3205 * This function should be called near the start of the device's
3206 * runtime_resume callback.
3208 void blk_pre_runtime_resume(struct request_queue *q)
3210 spin_lock_irq(q->queue_lock);
3211 q->rpm_status = RPM_RESUMING;
3212 spin_unlock_irq(q->queue_lock);
3214 EXPORT_SYMBOL(blk_pre_runtime_resume);
3217 * blk_post_runtime_resume - Post runtime resume processing
3218 * @q: the queue of the device
3219 * @err: return value of the device's runtime_resume function
3222 * Update the queue's runtime status according to the return value of the
3223 * device's runtime_resume function. If it is successfully resumed, process
3224 * the requests that are queued into the device's queue when it is resuming
3225 * and then mark last busy and initiate autosuspend for it.
3227 * This function should be called near the end of the device's
3228 * runtime_resume callback.
3230 void blk_post_runtime_resume(struct request_queue *q, int err)
3232 spin_lock_irq(q->queue_lock);
3234 q->rpm_status = RPM_ACTIVE;
3236 pm_runtime_mark_last_busy(q->dev);
3237 pm_request_autosuspend(q->dev);
3239 q->rpm_status = RPM_SUSPENDED;
3241 spin_unlock_irq(q->queue_lock);
3243 EXPORT_SYMBOL(blk_post_runtime_resume);
3246 int __init blk_dev_init(void)
3248 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3249 sizeof(((struct request *)0)->cmd_flags));
3251 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3252 kblockd_workqueue = alloc_workqueue("kblockd",
3253 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3254 WQ_POWER_EFFICIENT, 0);
3255 if (!kblockd_workqueue)
3256 panic("Failed to create kblockd\n");
3258 request_cachep = kmem_cache_create("blkdev_requests",
3259 sizeof(struct request), 0, SLAB_PANIC, NULL);
3261 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3262 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);