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/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/blktrace_api.h>
30 #include <linux/fault-inject.h>
31 #include <trace/block.h>
35 static int __make_request(struct request_queue *q, struct bio *bio);
38 * For the allocated request tables
40 static struct kmem_cache *request_cachep;
43 * For queue allocation
45 struct kmem_cache *blk_requestq_cachep;
48 * Controlling structure to kblockd
50 static struct workqueue_struct *kblockd_workqueue;
52 static void drive_stat_acct(struct request *rq, int new_io)
54 struct hd_struct *part;
55 int rw = rq_data_dir(rq);
58 if (!blk_fs_request(rq) || !rq->rq_disk)
61 cpu = part_stat_lock();
62 part = disk_map_sector_rcu(rq->rq_disk, rq->sector);
65 part_stat_inc(cpu, part, merges[rw]);
67 part_round_stats(cpu, part);
68 part_inc_in_flight(part);
74 void blk_queue_congestion_threshold(struct request_queue *q)
78 nr = q->nr_requests - (q->nr_requests / 8) + 1;
79 if (nr > q->nr_requests)
81 q->nr_congestion_on = nr;
83 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
86 q->nr_congestion_off = nr;
90 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
93 * Locates the passed device's request queue and returns the address of its
96 * Will return NULL if the request queue cannot be located.
98 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
100 struct backing_dev_info *ret = NULL;
101 struct request_queue *q = bdev_get_queue(bdev);
104 ret = &q->backing_dev_info;
107 EXPORT_SYMBOL(blk_get_backing_dev_info);
109 void blk_rq_init(struct request_queue *q, struct request *rq)
111 memset(rq, 0, sizeof(*rq));
113 INIT_LIST_HEAD(&rq->queuelist);
114 INIT_LIST_HEAD(&rq->timeout_list);
117 rq->sector = rq->hard_sector = (sector_t) -1;
118 INIT_HLIST_NODE(&rq->hash);
119 RB_CLEAR_NODE(&rq->rb_node);
124 EXPORT_SYMBOL(blk_rq_init);
126 static void req_bio_endio(struct request *rq, struct bio *bio,
127 unsigned int nbytes, int error)
129 struct request_queue *q = rq->q;
131 if (&q->bar_rq != rq) {
133 clear_bit(BIO_UPTODATE, &bio->bi_flags);
134 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
137 if (unlikely(nbytes > bio->bi_size)) {
138 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
139 __func__, nbytes, bio->bi_size);
140 nbytes = bio->bi_size;
143 bio->bi_size -= nbytes;
144 bio->bi_sector += (nbytes >> 9);
146 if (bio_integrity(bio))
147 bio_integrity_advance(bio, nbytes);
149 if (bio->bi_size == 0)
150 bio_endio(bio, error);
154 * Okay, this is the barrier request in progress, just
157 if (error && !q->orderr)
162 void blk_dump_rq_flags(struct request *rq, char *msg)
166 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
167 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
170 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
171 (unsigned long long)rq->sector,
173 rq->current_nr_sectors);
174 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
175 rq->bio, rq->biotail,
176 rq->buffer, rq->data,
179 if (blk_pc_request(rq)) {
180 printk(KERN_INFO " cdb: ");
181 for (bit = 0; bit < BLK_MAX_CDB; bit++)
182 printk("%02x ", rq->cmd[bit]);
186 EXPORT_SYMBOL(blk_dump_rq_flags);
189 * "plug" the device if there are no outstanding requests: this will
190 * force the transfer to start only after we have put all the requests
193 * This is called with interrupts off and no requests on the queue and
194 * with the queue lock held.
196 void blk_plug_device(struct request_queue *q)
198 WARN_ON(!irqs_disabled());
201 * don't plug a stopped queue, it must be paired with blk_start_queue()
202 * which will restart the queueing
204 if (blk_queue_stopped(q))
207 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
208 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
212 EXPORT_SYMBOL(blk_plug_device);
215 * blk_plug_device_unlocked - plug a device without queue lock held
216 * @q: The &struct request_queue to plug
219 * Like @blk_plug_device(), but grabs the queue lock and disables
222 void blk_plug_device_unlocked(struct request_queue *q)
226 spin_lock_irqsave(q->queue_lock, flags);
228 spin_unlock_irqrestore(q->queue_lock, flags);
230 EXPORT_SYMBOL(blk_plug_device_unlocked);
233 * remove the queue from the plugged list, if present. called with
234 * queue lock held and interrupts disabled.
236 int blk_remove_plug(struct request_queue *q)
238 WARN_ON(!irqs_disabled());
240 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
243 del_timer(&q->unplug_timer);
246 EXPORT_SYMBOL(blk_remove_plug);
249 * remove the plug and let it rip..
251 void __generic_unplug_device(struct request_queue *q)
253 if (unlikely(blk_queue_stopped(q)))
256 if (!blk_remove_plug(q))
263 * generic_unplug_device - fire a request queue
264 * @q: The &struct request_queue in question
267 * Linux uses plugging to build bigger requests queues before letting
268 * the device have at them. If a queue is plugged, the I/O scheduler
269 * is still adding and merging requests on the queue. Once the queue
270 * gets unplugged, the request_fn defined for the queue is invoked and
273 void generic_unplug_device(struct request_queue *q)
275 if (blk_queue_plugged(q)) {
276 spin_lock_irq(q->queue_lock);
277 __generic_unplug_device(q);
278 spin_unlock_irq(q->queue_lock);
281 EXPORT_SYMBOL(generic_unplug_device);
283 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
286 struct request_queue *q = bdi->unplug_io_data;
291 void blk_unplug_work(struct work_struct *work)
293 struct request_queue *q =
294 container_of(work, struct request_queue, unplug_work);
296 trace_block_unplug_io(q);
300 void blk_unplug_timeout(unsigned long data)
302 struct request_queue *q = (struct request_queue *)data;
304 trace_block_unplug_timer(q);
305 kblockd_schedule_work(q, &q->unplug_work);
308 void blk_unplug(struct request_queue *q)
311 * devices don't necessarily have an ->unplug_fn defined
314 trace_block_unplug_io(q);
318 EXPORT_SYMBOL(blk_unplug);
320 static void blk_invoke_request_fn(struct request_queue *q)
322 if (unlikely(blk_queue_stopped(q)))
326 * one level of recursion is ok and is much faster than kicking
327 * the unplug handling
329 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
331 queue_flag_clear(QUEUE_FLAG_REENTER, q);
333 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
334 kblockd_schedule_work(q, &q->unplug_work);
339 * blk_start_queue - restart a previously stopped queue
340 * @q: The &struct request_queue in question
343 * blk_start_queue() will clear the stop flag on the queue, and call
344 * the request_fn for the queue if it was in a stopped state when
345 * entered. Also see blk_stop_queue(). Queue lock must be held.
347 void blk_start_queue(struct request_queue *q)
349 WARN_ON(!irqs_disabled());
351 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
352 blk_invoke_request_fn(q);
354 EXPORT_SYMBOL(blk_start_queue);
357 * blk_stop_queue - stop a queue
358 * @q: The &struct request_queue in question
361 * The Linux block layer assumes that a block driver will consume all
362 * entries on the request queue when the request_fn strategy is called.
363 * Often this will not happen, because of hardware limitations (queue
364 * depth settings). If a device driver gets a 'queue full' response,
365 * or if it simply chooses not to queue more I/O at one point, it can
366 * call this function to prevent the request_fn from being called until
367 * the driver has signalled it's ready to go again. This happens by calling
368 * blk_start_queue() to restart queue operations. Queue lock must be held.
370 void blk_stop_queue(struct request_queue *q)
373 queue_flag_set(QUEUE_FLAG_STOPPED, q);
375 EXPORT_SYMBOL(blk_stop_queue);
378 * blk_sync_queue - cancel any pending callbacks on a queue
382 * The block layer may perform asynchronous callback activity
383 * on a queue, such as calling the unplug function after a timeout.
384 * A block device may call blk_sync_queue to ensure that any
385 * such activity is cancelled, thus allowing it to release resources
386 * that the callbacks might use. The caller must already have made sure
387 * that its ->make_request_fn will not re-add plugging prior to calling
391 void blk_sync_queue(struct request_queue *q)
393 del_timer_sync(&q->unplug_timer);
394 kblockd_flush_work(&q->unplug_work);
396 EXPORT_SYMBOL(blk_sync_queue);
399 * __blk_run_queue - run a single device queue
400 * @q: The queue to run
403 * See @blk_run_queue. This variant must be called with the queue lock
404 * held and interrupts disabled.
407 void __blk_run_queue(struct request_queue *q)
412 * Only recurse once to avoid overrunning the stack, let the unplug
413 * handling reinvoke the handler shortly if we already got there.
415 if (!elv_queue_empty(q))
416 blk_invoke_request_fn(q);
418 EXPORT_SYMBOL(__blk_run_queue);
421 * blk_run_queue - run a single device queue
422 * @q: The queue to run
425 * Invoke request handling on this queue, if it has pending work to do.
426 * May be used to restart queueing when a request has completed. Also
427 * See @blk_start_queueing.
430 void blk_run_queue(struct request_queue *q)
434 spin_lock_irqsave(q->queue_lock, flags);
436 spin_unlock_irqrestore(q->queue_lock, flags);
438 EXPORT_SYMBOL(blk_run_queue);
440 void blk_put_queue(struct request_queue *q)
442 kobject_put(&q->kobj);
445 void blk_cleanup_queue(struct request_queue *q)
448 * We know we have process context here, so we can be a little
449 * cautious and ensure that pending block actions on this device
450 * are done before moving on. Going into this function, we should
451 * not have processes doing IO to this device.
455 mutex_lock(&q->sysfs_lock);
456 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
457 mutex_unlock(&q->sysfs_lock);
460 elevator_exit(q->elevator);
464 EXPORT_SYMBOL(blk_cleanup_queue);
466 static int blk_init_free_list(struct request_queue *q)
468 struct request_list *rl = &q->rq;
470 rl->count[READ] = rl->count[WRITE] = 0;
471 rl->starved[READ] = rl->starved[WRITE] = 0;
473 init_waitqueue_head(&rl->wait[READ]);
474 init_waitqueue_head(&rl->wait[WRITE]);
476 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
477 mempool_free_slab, request_cachep, q->node);
485 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
487 return blk_alloc_queue_node(gfp_mask, -1);
489 EXPORT_SYMBOL(blk_alloc_queue);
491 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
493 struct request_queue *q;
496 q = kmem_cache_alloc_node(blk_requestq_cachep,
497 gfp_mask | __GFP_ZERO, node_id);
501 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
502 q->backing_dev_info.unplug_io_data = q;
503 err = bdi_init(&q->backing_dev_info);
505 kmem_cache_free(blk_requestq_cachep, q);
509 init_timer(&q->unplug_timer);
510 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
511 INIT_LIST_HEAD(&q->timeout_list);
512 INIT_WORK(&q->unplug_work, blk_unplug_work);
514 kobject_init(&q->kobj, &blk_queue_ktype);
516 mutex_init(&q->sysfs_lock);
517 spin_lock_init(&q->__queue_lock);
521 EXPORT_SYMBOL(blk_alloc_queue_node);
524 * blk_init_queue - prepare a request queue for use with a block device
525 * @rfn: The function to be called to process requests that have been
526 * placed on the queue.
527 * @lock: Request queue spin lock
530 * If a block device wishes to use the standard request handling procedures,
531 * which sorts requests and coalesces adjacent requests, then it must
532 * call blk_init_queue(). The function @rfn will be called when there
533 * are requests on the queue that need to be processed. If the device
534 * supports plugging, then @rfn may not be called immediately when requests
535 * are available on the queue, but may be called at some time later instead.
536 * Plugged queues are generally unplugged when a buffer belonging to one
537 * of the requests on the queue is needed, or due to memory pressure.
539 * @rfn is not required, or even expected, to remove all requests off the
540 * queue, but only as many as it can handle at a time. If it does leave
541 * requests on the queue, it is responsible for arranging that the requests
542 * get dealt with eventually.
544 * The queue spin lock must be held while manipulating the requests on the
545 * request queue; this lock will be taken also from interrupt context, so irq
546 * disabling is needed for it.
548 * Function returns a pointer to the initialized request queue, or %NULL if
552 * blk_init_queue() must be paired with a blk_cleanup_queue() call
553 * when the block device is deactivated (such as at module unload).
556 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
558 return blk_init_queue_node(rfn, lock, -1);
560 EXPORT_SYMBOL(blk_init_queue);
562 struct request_queue *
563 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
565 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
571 if (blk_init_free_list(q)) {
572 kmem_cache_free(blk_requestq_cachep, q);
577 * if caller didn't supply a lock, they get per-queue locking with
581 lock = &q->__queue_lock;
584 q->prep_rq_fn = NULL;
585 q->unplug_fn = generic_unplug_device;
586 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER |
587 1 << QUEUE_FLAG_STACKABLE);
588 q->queue_lock = lock;
590 blk_queue_segment_boundary(q, 0xffffffff);
592 blk_queue_make_request(q, __make_request);
593 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
595 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
596 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
598 q->sg_reserved_size = INT_MAX;
600 blk_set_cmd_filter_defaults(&q->cmd_filter);
605 if (!elevator_init(q, NULL)) {
606 blk_queue_congestion_threshold(q);
613 EXPORT_SYMBOL(blk_init_queue_node);
615 int blk_get_queue(struct request_queue *q)
617 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
618 kobject_get(&q->kobj);
625 static inline void blk_free_request(struct request_queue *q, struct request *rq)
627 if (rq->cmd_flags & REQ_ELVPRIV)
628 elv_put_request(q, rq);
629 mempool_free(rq, q->rq.rq_pool);
632 static struct request *
633 blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
635 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
642 rq->cmd_flags = rw | REQ_ALLOCED;
645 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
646 mempool_free(rq, q->rq.rq_pool);
649 rq->cmd_flags |= REQ_ELVPRIV;
656 * ioc_batching returns true if the ioc is a valid batching request and
657 * should be given priority access to a request.
659 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
665 * Make sure the process is able to allocate at least 1 request
666 * even if the batch times out, otherwise we could theoretically
669 return ioc->nr_batch_requests == q->nr_batching ||
670 (ioc->nr_batch_requests > 0
671 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
675 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
676 * will cause the process to be a "batcher" on all queues in the system. This
677 * is the behaviour we want though - once it gets a wakeup it should be given
680 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
682 if (!ioc || ioc_batching(q, ioc))
685 ioc->nr_batch_requests = q->nr_batching;
686 ioc->last_waited = jiffies;
689 static void __freed_request(struct request_queue *q, int rw)
691 struct request_list *rl = &q->rq;
693 if (rl->count[rw] < queue_congestion_off_threshold(q))
694 blk_clear_queue_congested(q, rw);
696 if (rl->count[rw] + 1 <= q->nr_requests) {
697 if (waitqueue_active(&rl->wait[rw]))
698 wake_up(&rl->wait[rw]);
700 blk_clear_queue_full(q, rw);
705 * A request has just been released. Account for it, update the full and
706 * congestion status, wake up any waiters. Called under q->queue_lock.
708 static void freed_request(struct request_queue *q, int rw, int priv)
710 struct request_list *rl = &q->rq;
716 __freed_request(q, rw);
718 if (unlikely(rl->starved[rw ^ 1]))
719 __freed_request(q, rw ^ 1);
722 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
724 * Get a free request, queue_lock must be held.
725 * Returns NULL on failure, with queue_lock held.
726 * Returns !NULL on success, with queue_lock *not held*.
728 static struct request *get_request(struct request_queue *q, int rw_flags,
729 struct bio *bio, gfp_t gfp_mask)
731 struct request *rq = NULL;
732 struct request_list *rl = &q->rq;
733 struct io_context *ioc = NULL;
734 const int rw = rw_flags & 0x01;
737 may_queue = elv_may_queue(q, rw_flags);
738 if (may_queue == ELV_MQUEUE_NO)
741 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
742 if (rl->count[rw]+1 >= q->nr_requests) {
743 ioc = current_io_context(GFP_ATOMIC, q->node);
745 * The queue will fill after this allocation, so set
746 * it as full, and mark this process as "batching".
747 * This process will be allowed to complete a batch of
748 * requests, others will be blocked.
750 if (!blk_queue_full(q, rw)) {
751 ioc_set_batching(q, ioc);
752 blk_set_queue_full(q, rw);
754 if (may_queue != ELV_MQUEUE_MUST
755 && !ioc_batching(q, ioc)) {
757 * The queue is full and the allocating
758 * process is not a "batcher", and not
759 * exempted by the IO scheduler
765 blk_set_queue_congested(q, rw);
769 * Only allow batching queuers to allocate up to 50% over the defined
770 * limit of requests, otherwise we could have thousands of requests
771 * allocated with any setting of ->nr_requests
773 if (rl->count[rw] >= (3 * q->nr_requests / 2))
779 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
783 spin_unlock_irq(q->queue_lock);
785 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
788 * Allocation failed presumably due to memory. Undo anything
789 * we might have messed up.
791 * Allocating task should really be put onto the front of the
792 * wait queue, but this is pretty rare.
794 spin_lock_irq(q->queue_lock);
795 freed_request(q, rw, priv);
798 * in the very unlikely event that allocation failed and no
799 * requests for this direction was pending, mark us starved
800 * so that freeing of a request in the other direction will
801 * notice us. another possible fix would be to split the
802 * rq mempool into READ and WRITE
805 if (unlikely(rl->count[rw] == 0))
812 * ioc may be NULL here, and ioc_batching will be false. That's
813 * OK, if the queue is under the request limit then requests need
814 * not count toward the nr_batch_requests limit. There will always
815 * be some limit enforced by BLK_BATCH_TIME.
817 if (ioc_batching(q, ioc))
818 ioc->nr_batch_requests--;
820 trace_block_getrq(q, bio, rw);
826 * No available requests for this queue, unplug the device and wait for some
827 * requests to become available.
829 * Called with q->queue_lock held, and returns with it unlocked.
831 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
834 const int rw = rw_flags & 0x01;
837 rq = get_request(q, rw_flags, bio, GFP_NOIO);
840 struct io_context *ioc;
841 struct request_list *rl = &q->rq;
843 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
844 TASK_UNINTERRUPTIBLE);
846 trace_block_sleeprq(q, bio, rw);
848 __generic_unplug_device(q);
849 spin_unlock_irq(q->queue_lock);
853 * After sleeping, we become a "batching" process and
854 * will be able to allocate at least one request, and
855 * up to a big batch of them for a small period time.
856 * See ioc_batching, ioc_set_batching
858 ioc = current_io_context(GFP_NOIO, q->node);
859 ioc_set_batching(q, ioc);
861 spin_lock_irq(q->queue_lock);
862 finish_wait(&rl->wait[rw], &wait);
864 rq = get_request(q, rw_flags, bio, GFP_NOIO);
870 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
874 BUG_ON(rw != READ && rw != WRITE);
876 spin_lock_irq(q->queue_lock);
877 if (gfp_mask & __GFP_WAIT) {
878 rq = get_request_wait(q, rw, NULL);
880 rq = get_request(q, rw, NULL, gfp_mask);
882 spin_unlock_irq(q->queue_lock);
884 /* q->queue_lock is unlocked at this point */
888 EXPORT_SYMBOL(blk_get_request);
891 * blk_start_queueing - initiate dispatch of requests to device
892 * @q: request queue to kick into gear
894 * This is basically a helper to remove the need to know whether a queue
895 * is plugged or not if someone just wants to initiate dispatch of requests
896 * for this queue. Should be used to start queueing on a device outside
897 * of ->request_fn() context. Also see @blk_run_queue.
899 * The queue lock must be held with interrupts disabled.
901 void blk_start_queueing(struct request_queue *q)
903 if (!blk_queue_plugged(q)) {
904 if (unlikely(blk_queue_stopped(q)))
908 __generic_unplug_device(q);
910 EXPORT_SYMBOL(blk_start_queueing);
913 * blk_requeue_request - put a request back on queue
914 * @q: request queue where request should be inserted
915 * @rq: request to be inserted
918 * Drivers often keep queueing requests until the hardware cannot accept
919 * more, when that condition happens we need to put the request back
920 * on the queue. Must be called with queue lock held.
922 void blk_requeue_request(struct request_queue *q, struct request *rq)
924 blk_delete_timer(rq);
925 blk_clear_rq_complete(rq);
926 trace_block_rq_requeue(q, rq);
928 if (blk_rq_tagged(rq))
929 blk_queue_end_tag(q, rq);
931 elv_requeue_request(q, rq);
933 EXPORT_SYMBOL(blk_requeue_request);
936 * blk_insert_request - insert a special request into a request queue
937 * @q: request queue where request should be inserted
938 * @rq: request to be inserted
939 * @at_head: insert request at head or tail of queue
940 * @data: private data
943 * Many block devices need to execute commands asynchronously, so they don't
944 * block the whole kernel from preemption during request execution. This is
945 * accomplished normally by inserting aritficial requests tagged as
946 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
947 * be scheduled for actual execution by the request queue.
949 * We have the option of inserting the head or the tail of the queue.
950 * Typically we use the tail for new ioctls and so forth. We use the head
951 * of the queue for things like a QUEUE_FULL message from a device, or a
952 * host that is unable to accept a particular command.
954 void blk_insert_request(struct request_queue *q, struct request *rq,
955 int at_head, void *data)
957 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
961 * tell I/O scheduler that this isn't a regular read/write (ie it
962 * must not attempt merges on this) and that it acts as a soft
965 rq->cmd_type = REQ_TYPE_SPECIAL;
966 rq->cmd_flags |= REQ_SOFTBARRIER;
970 spin_lock_irqsave(q->queue_lock, flags);
973 * If command is tagged, release the tag
975 if (blk_rq_tagged(rq))
976 blk_queue_end_tag(q, rq);
978 drive_stat_acct(rq, 1);
979 __elv_add_request(q, rq, where, 0);
980 blk_start_queueing(q);
981 spin_unlock_irqrestore(q->queue_lock, flags);
983 EXPORT_SYMBOL(blk_insert_request);
986 * add-request adds a request to the linked list.
987 * queue lock is held and interrupts disabled, as we muck with the
988 * request queue list.
990 static inline void add_request(struct request_queue *q, struct request *req)
992 drive_stat_acct(req, 1);
995 * elevator indicated where it wants this request to be
996 * inserted at elevator_merge time
998 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1001 static void part_round_stats_single(int cpu, struct hd_struct *part,
1004 if (now == part->stamp)
1007 if (part->in_flight) {
1008 __part_stat_add(cpu, part, time_in_queue,
1009 part->in_flight * (now - part->stamp));
1010 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1016 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1017 * @cpu: cpu number for stats access
1018 * @part: target partition
1020 * The average IO queue length and utilisation statistics are maintained
1021 * by observing the current state of the queue length and the amount of
1022 * time it has been in this state for.
1024 * Normally, that accounting is done on IO completion, but that can result
1025 * in more than a second's worth of IO being accounted for within any one
1026 * second, leading to >100% utilisation. To deal with that, we call this
1027 * function to do a round-off before returning the results when reading
1028 * /proc/diskstats. This accounts immediately for all queue usage up to
1029 * the current jiffies and restarts the counters again.
1031 void part_round_stats(int cpu, struct hd_struct *part)
1033 unsigned long now = jiffies;
1036 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1037 part_round_stats_single(cpu, part, now);
1039 EXPORT_SYMBOL_GPL(part_round_stats);
1042 * queue lock must be held
1044 void __blk_put_request(struct request_queue *q, struct request *req)
1048 if (unlikely(--req->ref_count))
1051 elv_completed_request(q, req);
1054 * Request may not have originated from ll_rw_blk. if not,
1055 * it didn't come out of our reserved rq pools
1057 if (req->cmd_flags & REQ_ALLOCED) {
1058 int rw = rq_data_dir(req);
1059 int priv = req->cmd_flags & REQ_ELVPRIV;
1061 BUG_ON(!list_empty(&req->queuelist));
1062 BUG_ON(!hlist_unhashed(&req->hash));
1064 blk_free_request(q, req);
1065 freed_request(q, rw, priv);
1068 EXPORT_SYMBOL_GPL(__blk_put_request);
1070 void blk_put_request(struct request *req)
1072 unsigned long flags;
1073 struct request_queue *q = req->q;
1075 spin_lock_irqsave(q->queue_lock, flags);
1076 __blk_put_request(q, req);
1077 spin_unlock_irqrestore(q->queue_lock, flags);
1079 EXPORT_SYMBOL(blk_put_request);
1081 void init_request_from_bio(struct request *req, struct bio *bio)
1083 req->cpu = bio->bi_comp_cpu;
1084 req->cmd_type = REQ_TYPE_FS;
1087 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1089 if (bio_rw_ahead(bio))
1090 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1091 REQ_FAILFAST_DRIVER);
1092 if (bio_failfast_dev(bio))
1093 req->cmd_flags |= REQ_FAILFAST_DEV;
1094 if (bio_failfast_transport(bio))
1095 req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1096 if (bio_failfast_driver(bio))
1097 req->cmd_flags |= REQ_FAILFAST_DRIVER;
1100 * REQ_BARRIER implies no merging, but lets make it explicit
1102 if (unlikely(bio_discard(bio))) {
1103 req->cmd_flags |= REQ_DISCARD;
1104 if (bio_barrier(bio))
1105 req->cmd_flags |= REQ_SOFTBARRIER;
1106 req->q->prepare_discard_fn(req->q, req);
1107 } else if (unlikely(bio_barrier(bio)))
1108 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
1111 req->cmd_flags |= REQ_RW_SYNC;
1112 if (bio_rw_meta(bio))
1113 req->cmd_flags |= REQ_RW_META;
1116 req->hard_sector = req->sector = bio->bi_sector;
1117 req->ioprio = bio_prio(bio);
1118 req->start_time = jiffies;
1119 blk_rq_bio_prep(req->q, req, bio);
1122 static int __make_request(struct request_queue *q, struct bio *bio)
1124 struct request *req;
1125 int el_ret, nr_sectors, barrier, discard, err;
1126 const unsigned short prio = bio_prio(bio);
1127 const int sync = bio_sync(bio);
1130 nr_sectors = bio_sectors(bio);
1133 * low level driver can indicate that it wants pages above a
1134 * certain limit bounced to low memory (ie for highmem, or even
1135 * ISA dma in theory)
1137 blk_queue_bounce(q, &bio);
1139 barrier = bio_barrier(bio);
1140 if (unlikely(barrier) && bio_has_data(bio) &&
1141 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1146 discard = bio_discard(bio);
1147 if (unlikely(discard) && !q->prepare_discard_fn) {
1152 spin_lock_irq(q->queue_lock);
1154 if (unlikely(barrier) || elv_queue_empty(q))
1157 el_ret = elv_merge(q, &req, bio);
1159 case ELEVATOR_BACK_MERGE:
1160 BUG_ON(!rq_mergeable(req));
1162 if (!ll_back_merge_fn(q, req, bio))
1165 trace_block_bio_backmerge(q, bio);
1167 req->biotail->bi_next = bio;
1169 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1170 req->ioprio = ioprio_best(req->ioprio, prio);
1171 if (!blk_rq_cpu_valid(req))
1172 req->cpu = bio->bi_comp_cpu;
1173 drive_stat_acct(req, 0);
1174 if (!attempt_back_merge(q, req))
1175 elv_merged_request(q, req, el_ret);
1178 case ELEVATOR_FRONT_MERGE:
1179 BUG_ON(!rq_mergeable(req));
1181 if (!ll_front_merge_fn(q, req, bio))
1184 trace_block_bio_frontmerge(q, bio);
1186 bio->bi_next = req->bio;
1190 * may not be valid. if the low level driver said
1191 * it didn't need a bounce buffer then it better
1192 * not touch req->buffer either...
1194 req->buffer = bio_data(bio);
1195 req->current_nr_sectors = bio_cur_sectors(bio);
1196 req->hard_cur_sectors = req->current_nr_sectors;
1197 req->sector = req->hard_sector = bio->bi_sector;
1198 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1199 req->ioprio = ioprio_best(req->ioprio, prio);
1200 if (!blk_rq_cpu_valid(req))
1201 req->cpu = bio->bi_comp_cpu;
1202 drive_stat_acct(req, 0);
1203 if (!attempt_front_merge(q, req))
1204 elv_merged_request(q, req, el_ret);
1207 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1214 * This sync check and mask will be re-done in init_request_from_bio(),
1215 * but we need to set it earlier to expose the sync flag to the
1216 * rq allocator and io schedulers.
1218 rw_flags = bio_data_dir(bio);
1220 rw_flags |= REQ_RW_SYNC;
1223 * Grab a free request. This is might sleep but can not fail.
1224 * Returns with the queue unlocked.
1226 req = get_request_wait(q, rw_flags, bio);
1229 * After dropping the lock and possibly sleeping here, our request
1230 * may now be mergeable after it had proven unmergeable (above).
1231 * We don't worry about that case for efficiency. It won't happen
1232 * often, and the elevators are able to handle it.
1234 init_request_from_bio(req, bio);
1236 spin_lock_irq(q->queue_lock);
1237 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1238 bio_flagged(bio, BIO_CPU_AFFINE))
1239 req->cpu = blk_cpu_to_group(smp_processor_id());
1240 if (elv_queue_empty(q))
1242 add_request(q, req);
1245 __generic_unplug_device(q);
1246 spin_unlock_irq(q->queue_lock);
1250 bio_endio(bio, err);
1255 * If bio->bi_dev is a partition, remap the location
1257 static inline void blk_partition_remap(struct bio *bio)
1259 struct block_device *bdev = bio->bi_bdev;
1261 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1262 struct hd_struct *p = bdev->bd_part;
1264 bio->bi_sector += p->start_sect;
1265 bio->bi_bdev = bdev->bd_contains;
1267 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1268 bdev->bd_dev, bio->bi_sector,
1269 bio->bi_sector - p->start_sect);
1273 static void handle_bad_sector(struct bio *bio)
1275 char b[BDEVNAME_SIZE];
1277 printk(KERN_INFO "attempt to access beyond end of device\n");
1278 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1279 bdevname(bio->bi_bdev, b),
1281 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1282 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1284 set_bit(BIO_EOF, &bio->bi_flags);
1287 #ifdef CONFIG_FAIL_MAKE_REQUEST
1289 static DECLARE_FAULT_ATTR(fail_make_request);
1291 static int __init setup_fail_make_request(char *str)
1293 return setup_fault_attr(&fail_make_request, str);
1295 __setup("fail_make_request=", setup_fail_make_request);
1297 static int should_fail_request(struct bio *bio)
1299 struct hd_struct *part = bio->bi_bdev->bd_part;
1301 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1302 return should_fail(&fail_make_request, bio->bi_size);
1307 static int __init fail_make_request_debugfs(void)
1309 return init_fault_attr_dentries(&fail_make_request,
1310 "fail_make_request");
1313 late_initcall(fail_make_request_debugfs);
1315 #else /* CONFIG_FAIL_MAKE_REQUEST */
1317 static inline int should_fail_request(struct bio *bio)
1322 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1325 * Check whether this bio extends beyond the end of the device.
1327 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1334 /* Test device or partition size, when known. */
1335 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1337 sector_t sector = bio->bi_sector;
1339 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1341 * This may well happen - the kernel calls bread()
1342 * without checking the size of the device, e.g., when
1343 * mounting a device.
1345 handle_bad_sector(bio);
1354 * generic_make_request - hand a buffer to its device driver for I/O
1355 * @bio: The bio describing the location in memory and on the device.
1357 * generic_make_request() is used to make I/O requests of block
1358 * devices. It is passed a &struct bio, which describes the I/O that needs
1361 * generic_make_request() does not return any status. The
1362 * success/failure status of the request, along with notification of
1363 * completion, is delivered asynchronously through the bio->bi_end_io
1364 * function described (one day) else where.
1366 * The caller of generic_make_request must make sure that bi_io_vec
1367 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1368 * set to describe the device address, and the
1369 * bi_end_io and optionally bi_private are set to describe how
1370 * completion notification should be signaled.
1372 * generic_make_request and the drivers it calls may use bi_next if this
1373 * bio happens to be merged with someone else, and may change bi_dev and
1374 * bi_sector for remaps as it sees fit. So the values of these fields
1375 * should NOT be depended on after the call to generic_make_request.
1377 static inline void __generic_make_request(struct bio *bio)
1379 struct request_queue *q;
1380 sector_t old_sector;
1381 int ret, nr_sectors = bio_sectors(bio);
1387 if (bio_check_eod(bio, nr_sectors))
1391 * Resolve the mapping until finished. (drivers are
1392 * still free to implement/resolve their own stacking
1393 * by explicitly returning 0)
1395 * NOTE: we don't repeat the blk_size check for each new device.
1396 * Stacking drivers are expected to know what they are doing.
1401 char b[BDEVNAME_SIZE];
1403 q = bdev_get_queue(bio->bi_bdev);
1406 "generic_make_request: Trying to access "
1407 "nonexistent block-device %s (%Lu)\n",
1408 bdevname(bio->bi_bdev, b),
1409 (long long) bio->bi_sector);
1411 bio_endio(bio, err);
1415 if (unlikely(nr_sectors > q->max_hw_sectors)) {
1416 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1417 bdevname(bio->bi_bdev, b),
1423 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1426 if (should_fail_request(bio))
1430 * If this device has partitions, remap block n
1431 * of partition p to block n+start(p) of the disk.
1433 blk_partition_remap(bio);
1435 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1438 if (old_sector != -1)
1439 trace_block_remap(q, bio, old_dev, bio->bi_sector,
1442 trace_block_bio_queue(q, bio);
1444 old_sector = bio->bi_sector;
1445 old_dev = bio->bi_bdev->bd_dev;
1447 if (bio_check_eod(bio, nr_sectors))
1449 if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) ||
1450 (bio_discard(bio) && !q->prepare_discard_fn)) {
1455 ret = q->make_request_fn(q, bio);
1460 * We only want one ->make_request_fn to be active at a time,
1461 * else stack usage with stacked devices could be a problem.
1462 * So use current->bio_{list,tail} to keep a list of requests
1463 * submited by a make_request_fn function.
1464 * current->bio_tail is also used as a flag to say if
1465 * generic_make_request is currently active in this task or not.
1466 * If it is NULL, then no make_request is active. If it is non-NULL,
1467 * then a make_request is active, and new requests should be added
1470 void generic_make_request(struct bio *bio)
1472 if (current->bio_tail) {
1473 /* make_request is active */
1474 *(current->bio_tail) = bio;
1475 bio->bi_next = NULL;
1476 current->bio_tail = &bio->bi_next;
1479 /* following loop may be a bit non-obvious, and so deserves some
1481 * Before entering the loop, bio->bi_next is NULL (as all callers
1482 * ensure that) so we have a list with a single bio.
1483 * We pretend that we have just taken it off a longer list, so
1484 * we assign bio_list to the next (which is NULL) and bio_tail
1485 * to &bio_list, thus initialising the bio_list of new bios to be
1486 * added. __generic_make_request may indeed add some more bios
1487 * through a recursive call to generic_make_request. If it
1488 * did, we find a non-NULL value in bio_list and re-enter the loop
1489 * from the top. In this case we really did just take the bio
1490 * of the top of the list (no pretending) and so fixup bio_list and
1491 * bio_tail or bi_next, and call into __generic_make_request again.
1493 * The loop was structured like this to make only one call to
1494 * __generic_make_request (which is important as it is large and
1495 * inlined) and to keep the structure simple.
1497 BUG_ON(bio->bi_next);
1499 current->bio_list = bio->bi_next;
1500 if (bio->bi_next == NULL)
1501 current->bio_tail = ¤t->bio_list;
1503 bio->bi_next = NULL;
1504 __generic_make_request(bio);
1505 bio = current->bio_list;
1507 current->bio_tail = NULL; /* deactivate */
1509 EXPORT_SYMBOL(generic_make_request);
1512 * submit_bio - submit a bio to the block device layer for I/O
1513 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1514 * @bio: The &struct bio which describes the I/O
1516 * submit_bio() is very similar in purpose to generic_make_request(), and
1517 * uses that function to do most of the work. Both are fairly rough
1518 * interfaces; @bio must be presetup and ready for I/O.
1521 void submit_bio(int rw, struct bio *bio)
1523 int count = bio_sectors(bio);
1528 * If it's a regular read/write or a barrier with data attached,
1529 * go through the normal accounting stuff before submission.
1531 if (bio_has_data(bio)) {
1533 count_vm_events(PGPGOUT, count);
1535 task_io_account_read(bio->bi_size);
1536 count_vm_events(PGPGIN, count);
1539 if (unlikely(block_dump)) {
1540 char b[BDEVNAME_SIZE];
1541 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1542 current->comm, task_pid_nr(current),
1543 (rw & WRITE) ? "WRITE" : "READ",
1544 (unsigned long long)bio->bi_sector,
1545 bdevname(bio->bi_bdev, b));
1549 generic_make_request(bio);
1551 EXPORT_SYMBOL(submit_bio);
1554 * blk_rq_check_limits - Helper function to check a request for the queue limit
1556 * @rq: the request being checked
1559 * @rq may have been made based on weaker limitations of upper-level queues
1560 * in request stacking drivers, and it may violate the limitation of @q.
1561 * Since the block layer and the underlying device driver trust @rq
1562 * after it is inserted to @q, it should be checked against @q before
1563 * the insertion using this generic function.
1565 * This function should also be useful for request stacking drivers
1566 * in some cases below, so export this fuction.
1567 * Request stacking drivers like request-based dm may change the queue
1568 * limits while requests are in the queue (e.g. dm's table swapping).
1569 * Such request stacking drivers should check those requests agaist
1570 * the new queue limits again when they dispatch those requests,
1571 * although such checkings are also done against the old queue limits
1572 * when submitting requests.
1574 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1576 if (rq->nr_sectors > q->max_sectors ||
1577 rq->data_len > q->max_hw_sectors << 9) {
1578 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1583 * queue's settings related to segment counting like q->bounce_pfn
1584 * may differ from that of other stacking queues.
1585 * Recalculate it to check the request correctly on this queue's
1588 blk_recalc_rq_segments(rq);
1589 if (rq->nr_phys_segments > q->max_phys_segments ||
1590 rq->nr_phys_segments > q->max_hw_segments) {
1591 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1597 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1600 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1601 * @q: the queue to submit the request
1602 * @rq: the request being queued
1604 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1606 unsigned long flags;
1608 if (blk_rq_check_limits(q, rq))
1611 #ifdef CONFIG_FAIL_MAKE_REQUEST
1612 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1613 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1617 spin_lock_irqsave(q->queue_lock, flags);
1620 * Submitting request must be dequeued before calling this function
1621 * because it will be linked to another request_queue
1623 BUG_ON(blk_queued_rq(rq));
1625 drive_stat_acct(rq, 1);
1626 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1628 spin_unlock_irqrestore(q->queue_lock, flags);
1632 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1635 * __end_that_request_first - end I/O on a request
1636 * @req: the request being processed
1637 * @error: %0 for success, < %0 for error
1638 * @nr_bytes: number of bytes to complete
1641 * Ends I/O on a number of bytes attached to @req, and sets it up
1642 * for the next range of segments (if any) in the cluster.
1645 * %0 - we are done with this request, call end_that_request_last()
1646 * %1 - still buffers pending for this request
1648 static int __end_that_request_first(struct request *req, int error,
1651 int total_bytes, bio_nbytes, next_idx = 0;
1654 trace_block_rq_complete(req->q, req);
1657 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1658 * sense key with us all the way through
1660 if (!blk_pc_request(req))
1663 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1664 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1665 req->rq_disk ? req->rq_disk->disk_name : "?",
1666 (unsigned long long)req->sector);
1669 if (blk_fs_request(req) && req->rq_disk) {
1670 const int rw = rq_data_dir(req);
1671 struct hd_struct *part;
1674 cpu = part_stat_lock();
1675 part = disk_map_sector_rcu(req->rq_disk, req->sector);
1676 part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9);
1680 total_bytes = bio_nbytes = 0;
1681 while ((bio = req->bio) != NULL) {
1685 * For an empty barrier request, the low level driver must
1686 * store a potential error location in ->sector. We pass
1687 * that back up in ->bi_sector.
1689 if (blk_empty_barrier(req))
1690 bio->bi_sector = req->sector;
1692 if (nr_bytes >= bio->bi_size) {
1693 req->bio = bio->bi_next;
1694 nbytes = bio->bi_size;
1695 req_bio_endio(req, bio, nbytes, error);
1699 int idx = bio->bi_idx + next_idx;
1701 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1702 blk_dump_rq_flags(req, "__end_that");
1703 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1704 __func__, bio->bi_idx, bio->bi_vcnt);
1708 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1709 BIO_BUG_ON(nbytes > bio->bi_size);
1712 * not a complete bvec done
1714 if (unlikely(nbytes > nr_bytes)) {
1715 bio_nbytes += nr_bytes;
1716 total_bytes += nr_bytes;
1721 * advance to the next vector
1724 bio_nbytes += nbytes;
1727 total_bytes += nbytes;
1733 * end more in this run, or just return 'not-done'
1735 if (unlikely(nr_bytes <= 0))
1747 * if the request wasn't completed, update state
1750 req_bio_endio(req, bio, bio_nbytes, error);
1751 bio->bi_idx += next_idx;
1752 bio_iovec(bio)->bv_offset += nr_bytes;
1753 bio_iovec(bio)->bv_len -= nr_bytes;
1756 blk_recalc_rq_sectors(req, total_bytes >> 9);
1757 blk_recalc_rq_segments(req);
1762 * queue lock must be held
1764 static void end_that_request_last(struct request *req, int error)
1766 struct gendisk *disk = req->rq_disk;
1768 if (blk_rq_tagged(req))
1769 blk_queue_end_tag(req->q, req);
1771 if (blk_queued_rq(req))
1772 blkdev_dequeue_request(req);
1774 if (unlikely(laptop_mode) && blk_fs_request(req))
1775 laptop_io_completion();
1777 blk_delete_timer(req);
1780 * Account IO completion. bar_rq isn't accounted as a normal
1781 * IO on queueing nor completion. Accounting the containing
1782 * request is enough.
1784 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1785 unsigned long duration = jiffies - req->start_time;
1786 const int rw = rq_data_dir(req);
1787 struct hd_struct *part;
1790 cpu = part_stat_lock();
1791 part = disk_map_sector_rcu(disk, req->sector);
1793 part_stat_inc(cpu, part, ios[rw]);
1794 part_stat_add(cpu, part, ticks[rw], duration);
1795 part_round_stats(cpu, part);
1796 part_dec_in_flight(part);
1802 req->end_io(req, error);
1804 if (blk_bidi_rq(req))
1805 __blk_put_request(req->next_rq->q, req->next_rq);
1807 __blk_put_request(req->q, req);
1812 * blk_rq_bytes - Returns bytes left to complete in the entire request
1813 * @rq: the request being processed
1815 unsigned int blk_rq_bytes(struct request *rq)
1817 if (blk_fs_request(rq))
1818 return rq->hard_nr_sectors << 9;
1820 return rq->data_len;
1822 EXPORT_SYMBOL_GPL(blk_rq_bytes);
1825 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1826 * @rq: the request being processed
1828 unsigned int blk_rq_cur_bytes(struct request *rq)
1830 if (blk_fs_request(rq))
1831 return rq->current_nr_sectors << 9;
1834 return rq->bio->bi_size;
1836 return rq->data_len;
1838 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1841 * end_request - end I/O on the current segment of the request
1842 * @req: the request being processed
1843 * @uptodate: error value or %0/%1 uptodate flag
1846 * Ends I/O on the current segment of a request. If that is the only
1847 * remaining segment, the request is also completed and freed.
1849 * This is a remnant of how older block drivers handled I/O completions.
1850 * Modern drivers typically end I/O on the full request in one go, unless
1851 * they have a residual value to account for. For that case this function
1852 * isn't really useful, unless the residual just happens to be the
1853 * full current segment. In other words, don't use this function in new
1854 * code. Use blk_end_request() or __blk_end_request() to end a request.
1856 void end_request(struct request *req, int uptodate)
1861 error = uptodate ? uptodate : -EIO;
1863 __blk_end_request(req, error, req->hard_cur_sectors << 9);
1865 EXPORT_SYMBOL(end_request);
1867 static int end_that_request_data(struct request *rq, int error,
1868 unsigned int nr_bytes, unsigned int bidi_bytes)
1871 if (__end_that_request_first(rq, error, nr_bytes))
1874 /* Bidi request must be completed as a whole */
1875 if (blk_bidi_rq(rq) &&
1876 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1884 * blk_end_io - Generic end_io function to complete a request.
1885 * @rq: the request being processed
1886 * @error: %0 for success, < %0 for error
1887 * @nr_bytes: number of bytes to complete @rq
1888 * @bidi_bytes: number of bytes to complete @rq->next_rq
1889 * @drv_callback: function called between completion of bios in the request
1890 * and completion of the request.
1891 * If the callback returns non %0, this helper returns without
1892 * completion of the request.
1895 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1896 * If @rq has leftover, sets it up for the next range of segments.
1899 * %0 - we are done with this request
1900 * %1 - this request is not freed yet, it still has pending buffers.
1902 static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1903 unsigned int bidi_bytes,
1904 int (drv_callback)(struct request *))
1906 struct request_queue *q = rq->q;
1907 unsigned long flags = 0UL;
1909 if (end_that_request_data(rq, error, nr_bytes, bidi_bytes))
1912 /* Special feature for tricky drivers */
1913 if (drv_callback && drv_callback(rq))
1916 add_disk_randomness(rq->rq_disk);
1918 spin_lock_irqsave(q->queue_lock, flags);
1919 end_that_request_last(rq, error);
1920 spin_unlock_irqrestore(q->queue_lock, flags);
1926 * blk_end_request - Helper function for drivers to complete the request.
1927 * @rq: the request being processed
1928 * @error: %0 for success, < %0 for error
1929 * @nr_bytes: number of bytes to complete
1932 * Ends I/O on a number of bytes attached to @rq.
1933 * If @rq has leftover, sets it up for the next range of segments.
1936 * %0 - we are done with this request
1937 * %1 - still buffers pending for this request
1939 int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1941 return blk_end_io(rq, error, nr_bytes, 0, NULL);
1943 EXPORT_SYMBOL_GPL(blk_end_request);
1946 * __blk_end_request - Helper function for drivers to complete the request.
1947 * @rq: the request being processed
1948 * @error: %0 for success, < %0 for error
1949 * @nr_bytes: number of bytes to complete
1952 * Must be called with queue lock held unlike blk_end_request().
1955 * %0 - we are done with this request
1956 * %1 - still buffers pending for this request
1958 int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1960 if (rq->bio && __end_that_request_first(rq, error, nr_bytes))
1963 add_disk_randomness(rq->rq_disk);
1965 end_that_request_last(rq, error);
1969 EXPORT_SYMBOL_GPL(__blk_end_request);
1972 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1973 * @rq: the bidi request being processed
1974 * @error: %0 for success, < %0 for error
1975 * @nr_bytes: number of bytes to complete @rq
1976 * @bidi_bytes: number of bytes to complete @rq->next_rq
1979 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1982 * %0 - we are done with this request
1983 * %1 - still buffers pending for this request
1985 int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1986 unsigned int bidi_bytes)
1988 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1990 EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1993 * blk_update_request - Special helper function for request stacking drivers
1994 * @rq: the request being processed
1995 * @error: %0 for success, < %0 for error
1996 * @nr_bytes: number of bytes to complete @rq
1999 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2000 * the request structure even if @rq doesn't have leftover.
2001 * If @rq has leftover, sets it up for the next range of segments.
2003 * This special helper function is only for request stacking drivers
2004 * (e.g. request-based dm) so that they can handle partial completion.
2005 * Actual device drivers should use blk_end_request instead.
2007 void blk_update_request(struct request *rq, int error, unsigned int nr_bytes)
2009 if (!end_that_request_data(rq, error, nr_bytes, 0)) {
2011 * These members are not updated in end_that_request_data()
2012 * when all bios are completed.
2013 * Update them so that the request stacking driver can find
2014 * how many bytes remain in the request later.
2016 rq->nr_sectors = rq->hard_nr_sectors = 0;
2017 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
2020 EXPORT_SYMBOL_GPL(blk_update_request);
2023 * blk_end_request_callback - Special helper function for tricky drivers
2024 * @rq: the request being processed
2025 * @error: %0 for success, < %0 for error
2026 * @nr_bytes: number of bytes to complete
2027 * @drv_callback: function called between completion of bios in the request
2028 * and completion of the request.
2029 * If the callback returns non %0, this helper returns without
2030 * completion of the request.
2033 * Ends I/O on a number of bytes attached to @rq.
2034 * If @rq has leftover, sets it up for the next range of segments.
2036 * This special helper function is used only for existing tricky drivers.
2037 * (e.g. cdrom_newpc_intr() of ide-cd)
2038 * This interface will be removed when such drivers are rewritten.
2039 * Don't use this interface in other places anymore.
2042 * %0 - we are done with this request
2043 * %1 - this request is not freed yet.
2044 * this request still has pending buffers or
2045 * the driver doesn't want to finish this request yet.
2047 int blk_end_request_callback(struct request *rq, int error,
2048 unsigned int nr_bytes,
2049 int (drv_callback)(struct request *))
2051 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
2053 EXPORT_SYMBOL_GPL(blk_end_request_callback);
2055 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2058 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2059 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2060 rq->cmd_flags |= (bio->bi_rw & 3);
2062 if (bio_has_data(bio)) {
2063 rq->nr_phys_segments = bio_phys_segments(q, bio);
2064 rq->buffer = bio_data(bio);
2066 rq->current_nr_sectors = bio_cur_sectors(bio);
2067 rq->hard_cur_sectors = rq->current_nr_sectors;
2068 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2069 rq->data_len = bio->bi_size;
2071 rq->bio = rq->biotail = bio;
2074 rq->rq_disk = bio->bi_bdev->bd_disk;
2078 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2079 * @q : the queue of the device being checked
2082 * Check if underlying low-level drivers of a device are busy.
2083 * If the drivers want to export their busy state, they must set own
2084 * exporting function using blk_queue_lld_busy() first.
2086 * Basically, this function is used only by request stacking drivers
2087 * to stop dispatching requests to underlying devices when underlying
2088 * devices are busy. This behavior helps more I/O merging on the queue
2089 * of the request stacking driver and prevents I/O throughput regression
2090 * on burst I/O load.
2093 * 0 - Not busy (The request stacking driver should dispatch request)
2094 * 1 - Busy (The request stacking driver should stop dispatching request)
2096 int blk_lld_busy(struct request_queue *q)
2099 return q->lld_busy_fn(q);
2103 EXPORT_SYMBOL_GPL(blk_lld_busy);
2105 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2107 return queue_work(kblockd_workqueue, work);
2109 EXPORT_SYMBOL(kblockd_schedule_work);
2111 void kblockd_flush_work(struct work_struct *work)
2113 cancel_work_sync(work);
2115 EXPORT_SYMBOL(kblockd_flush_work);
2117 int __init blk_dev_init(void)
2119 kblockd_workqueue = create_workqueue("kblockd");
2120 if (!kblockd_workqueue)
2121 panic("Failed to create kblockd\n");
2123 request_cachep = kmem_cache_create("blkdev_requests",
2124 sizeof(struct request), 0, SLAB_PANIC, NULL);
2126 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2127 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);