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/interrupt.h>
30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h>
32 #include <linux/fault-inject.h>
36 static int __make_request(struct request_queue *q, struct bio *bio);
39 * For the allocated request tables
41 static struct kmem_cache *request_cachep;
44 * For queue allocation
46 struct kmem_cache *blk_requestq_cachep;
49 * Controlling structure to kblockd
51 static struct workqueue_struct *kblockd_workqueue;
53 static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
55 static void drive_stat_acct(struct request *rq, int new_io)
57 struct hd_struct *part;
58 int rw = rq_data_dir(rq);
60 if (!blk_fs_request(rq) || !rq->rq_disk)
63 part = get_part(rq->rq_disk, rq->sector);
65 __all_stat_inc(rq->rq_disk, part, merges[rw], rq->sector);
67 disk_round_stats(rq->rq_disk);
68 rq->rq_disk->in_flight++;
70 part_round_stats(part);
76 void blk_queue_congestion_threshold(struct request_queue *q)
80 nr = q->nr_requests - (q->nr_requests / 8) + 1;
81 if (nr > q->nr_requests)
83 q->nr_congestion_on = nr;
85 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
88 q->nr_congestion_off = nr;
92 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
95 * Locates the passed device's request queue and returns the address of its
98 * Will return NULL if the request queue cannot be located.
100 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
102 struct backing_dev_info *ret = NULL;
103 struct request_queue *q = bdev_get_queue(bdev);
106 ret = &q->backing_dev_info;
109 EXPORT_SYMBOL(blk_get_backing_dev_info);
111 void blk_rq_init(struct request_queue *q, struct request *rq)
113 memset(rq, 0, sizeof(*rq));
115 INIT_LIST_HEAD(&rq->queuelist);
116 INIT_LIST_HEAD(&rq->donelist);
118 rq->sector = rq->hard_sector = (sector_t) -1;
119 INIT_HLIST_NODE(&rq->hash);
120 RB_CLEAR_NODE(&rq->rb_node);
125 EXPORT_SYMBOL(blk_rq_init);
127 static void req_bio_endio(struct request *rq, struct bio *bio,
128 unsigned int nbytes, int error)
130 struct request_queue *q = rq->q;
132 if (&q->bar_rq != rq) {
134 clear_bit(BIO_UPTODATE, &bio->bi_flags);
135 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
138 if (unlikely(nbytes > bio->bi_size)) {
139 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
140 __func__, nbytes, bio->bi_size);
141 nbytes = bio->bi_size;
144 bio->bi_size -= nbytes;
145 bio->bi_sector += (nbytes >> 9);
147 if (bio_integrity(bio))
148 bio_integrity_advance(bio, nbytes);
150 if (bio->bi_size == 0)
151 bio_endio(bio, error);
155 * Okay, this is the barrier request in progress, just
158 if (error && !q->orderr)
163 void blk_dump_rq_flags(struct request *rq, char *msg)
167 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
168 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
171 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
172 (unsigned long long)rq->sector,
174 rq->current_nr_sectors);
175 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
176 rq->bio, rq->biotail,
177 rq->buffer, rq->data,
180 if (blk_pc_request(rq)) {
181 printk(KERN_INFO " cdb: ");
182 for (bit = 0; bit < BLK_MAX_CDB; bit++)
183 printk("%02x ", rq->cmd[bit]);
187 EXPORT_SYMBOL(blk_dump_rq_flags);
190 * "plug" the device if there are no outstanding requests: this will
191 * force the transfer to start only after we have put all the requests
194 * This is called with interrupts off and no requests on the queue and
195 * with the queue lock held.
197 void blk_plug_device(struct request_queue *q)
199 WARN_ON(!irqs_disabled());
202 * don't plug a stopped queue, it must be paired with blk_start_queue()
203 * which will restart the queueing
205 if (blk_queue_stopped(q))
208 if (!test_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
209 __set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
210 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
211 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
214 EXPORT_SYMBOL(blk_plug_device);
217 * remove the queue from the plugged list, if present. called with
218 * queue lock held and interrupts disabled.
220 int blk_remove_plug(struct request_queue *q)
222 WARN_ON(!irqs_disabled());
224 if (!test_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
227 queue_flag_clear(QUEUE_FLAG_PLUGGED, q);
228 del_timer(&q->unplug_timer);
231 EXPORT_SYMBOL(blk_remove_plug);
234 * remove the plug and let it rip..
236 void __generic_unplug_device(struct request_queue *q)
238 if (unlikely(blk_queue_stopped(q)))
241 if (!blk_remove_plug(q))
246 EXPORT_SYMBOL(__generic_unplug_device);
249 * generic_unplug_device - fire a request queue
250 * @q: The &struct request_queue in question
253 * Linux uses plugging to build bigger requests queues before letting
254 * the device have at them. If a queue is plugged, the I/O scheduler
255 * is still adding and merging requests on the queue. Once the queue
256 * gets unplugged, the request_fn defined for the queue is invoked and
259 void generic_unplug_device(struct request_queue *q)
261 if (blk_queue_plugged(q)) {
262 spin_lock_irq(q->queue_lock);
263 __generic_unplug_device(q);
264 spin_unlock_irq(q->queue_lock);
267 EXPORT_SYMBOL(generic_unplug_device);
269 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
272 struct request_queue *q = bdi->unplug_io_data;
277 void blk_unplug_work(struct work_struct *work)
279 struct request_queue *q =
280 container_of(work, struct request_queue, unplug_work);
282 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
283 q->rq.count[READ] + q->rq.count[WRITE]);
288 void blk_unplug_timeout(unsigned long data)
290 struct request_queue *q = (struct request_queue *)data;
292 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
293 q->rq.count[READ] + q->rq.count[WRITE]);
295 kblockd_schedule_work(&q->unplug_work);
298 void blk_unplug(struct request_queue *q)
301 * devices don't necessarily have an ->unplug_fn defined
304 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
305 q->rq.count[READ] + q->rq.count[WRITE]);
310 EXPORT_SYMBOL(blk_unplug);
313 * blk_start_queue - restart a previously stopped queue
314 * @q: The &struct request_queue in question
317 * blk_start_queue() will clear the stop flag on the queue, and call
318 * the request_fn for the queue if it was in a stopped state when
319 * entered. Also see blk_stop_queue(). Queue lock must be held.
321 void blk_start_queue(struct request_queue *q)
323 WARN_ON(!irqs_disabled());
325 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
328 * one level of recursion is ok and is much faster than kicking
329 * the unplug handling
331 if (!test_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
332 queue_flag_set(QUEUE_FLAG_REENTER, q);
334 queue_flag_clear(QUEUE_FLAG_REENTER, q);
337 kblockd_schedule_work(&q->unplug_work);
340 EXPORT_SYMBOL(blk_start_queue);
343 * blk_stop_queue - stop a queue
344 * @q: The &struct request_queue in question
347 * The Linux block layer assumes that a block driver will consume all
348 * entries on the request queue when the request_fn strategy is called.
349 * Often this will not happen, because of hardware limitations (queue
350 * depth settings). If a device driver gets a 'queue full' response,
351 * or if it simply chooses not to queue more I/O at one point, it can
352 * call this function to prevent the request_fn from being called until
353 * the driver has signalled it's ready to go again. This happens by calling
354 * blk_start_queue() to restart queue operations. Queue lock must be held.
356 void blk_stop_queue(struct request_queue *q)
359 queue_flag_set(QUEUE_FLAG_STOPPED, q);
361 EXPORT_SYMBOL(blk_stop_queue);
364 * blk_sync_queue - cancel any pending callbacks on a queue
368 * The block layer may perform asynchronous callback activity
369 * on a queue, such as calling the unplug function after a timeout.
370 * A block device may call blk_sync_queue to ensure that any
371 * such activity is cancelled, thus allowing it to release resources
372 * that the callbacks might use. The caller must already have made sure
373 * that its ->make_request_fn will not re-add plugging prior to calling
377 void blk_sync_queue(struct request_queue *q)
379 del_timer_sync(&q->unplug_timer);
380 kblockd_flush_work(&q->unplug_work);
382 EXPORT_SYMBOL(blk_sync_queue);
385 * blk_run_queue - run a single device queue
386 * @q: The queue to run
388 void __blk_run_queue(struct request_queue *q)
393 * Only recurse once to avoid overrunning the stack, let the unplug
394 * handling reinvoke the handler shortly if we already got there.
396 if (!elv_queue_empty(q)) {
397 if (!test_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
398 queue_flag_set(QUEUE_FLAG_REENTER, q);
400 queue_flag_clear(QUEUE_FLAG_REENTER, q);
403 kblockd_schedule_work(&q->unplug_work);
407 EXPORT_SYMBOL(__blk_run_queue);
410 * blk_run_queue - run a single device queue
411 * @q: The queue to run
413 void blk_run_queue(struct request_queue *q)
417 spin_lock_irqsave(q->queue_lock, flags);
419 spin_unlock_irqrestore(q->queue_lock, flags);
421 EXPORT_SYMBOL(blk_run_queue);
423 void blk_put_queue(struct request_queue *q)
425 kobject_put(&q->kobj);
428 void blk_cleanup_queue(struct request_queue *q)
430 mutex_lock(&q->sysfs_lock);
431 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
432 mutex_unlock(&q->sysfs_lock);
435 elevator_exit(q->elevator);
439 EXPORT_SYMBOL(blk_cleanup_queue);
441 static int blk_init_free_list(struct request_queue *q)
443 struct request_list *rl = &q->rq;
445 rl->count[READ] = rl->count[WRITE] = 0;
446 rl->starved[READ] = rl->starved[WRITE] = 0;
448 init_waitqueue_head(&rl->wait[READ]);
449 init_waitqueue_head(&rl->wait[WRITE]);
451 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
452 mempool_free_slab, request_cachep, q->node);
460 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
462 return blk_alloc_queue_node(gfp_mask, -1);
464 EXPORT_SYMBOL(blk_alloc_queue);
466 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
468 struct request_queue *q;
471 q = kmem_cache_alloc_node(blk_requestq_cachep,
472 gfp_mask | __GFP_ZERO, node_id);
476 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
477 q->backing_dev_info.unplug_io_data = q;
478 err = bdi_init(&q->backing_dev_info);
480 kmem_cache_free(blk_requestq_cachep, q);
484 init_timer(&q->unplug_timer);
486 kobject_init(&q->kobj, &blk_queue_ktype);
488 mutex_init(&q->sysfs_lock);
489 spin_lock_init(&q->__queue_lock);
493 EXPORT_SYMBOL(blk_alloc_queue_node);
496 * blk_init_queue - prepare a request queue for use with a block device
497 * @rfn: The function to be called to process requests that have been
498 * placed on the queue.
499 * @lock: Request queue spin lock
502 * If a block device wishes to use the standard request handling procedures,
503 * which sorts requests and coalesces adjacent requests, then it must
504 * call blk_init_queue(). The function @rfn will be called when there
505 * are requests on the queue that need to be processed. If the device
506 * supports plugging, then @rfn may not be called immediately when requests
507 * are available on the queue, but may be called at some time later instead.
508 * Plugged queues are generally unplugged when a buffer belonging to one
509 * of the requests on the queue is needed, or due to memory pressure.
511 * @rfn is not required, or even expected, to remove all requests off the
512 * queue, but only as many as it can handle at a time. If it does leave
513 * requests on the queue, it is responsible for arranging that the requests
514 * get dealt with eventually.
516 * The queue spin lock must be held while manipulating the requests on the
517 * request queue; this lock will be taken also from interrupt context, so irq
518 * disabling is needed for it.
520 * Function returns a pointer to the initialized request queue, or NULL if
524 * blk_init_queue() must be paired with a blk_cleanup_queue() call
525 * when the block device is deactivated (such as at module unload).
528 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
530 return blk_init_queue_node(rfn, lock, -1);
532 EXPORT_SYMBOL(blk_init_queue);
534 struct request_queue *
535 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
537 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
543 if (blk_init_free_list(q)) {
544 kmem_cache_free(blk_requestq_cachep, q);
549 * if caller didn't supply a lock, they get per-queue locking with
553 lock = &q->__queue_lock;
556 q->prep_rq_fn = NULL;
557 q->unplug_fn = generic_unplug_device;
558 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
559 q->queue_lock = lock;
561 blk_queue_segment_boundary(q, 0xffffffff);
563 blk_queue_make_request(q, __make_request);
564 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
566 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
567 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
569 q->sg_reserved_size = INT_MAX;
574 if (!elevator_init(q, NULL)) {
575 blk_queue_congestion_threshold(q);
582 EXPORT_SYMBOL(blk_init_queue_node);
584 int blk_get_queue(struct request_queue *q)
586 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
587 kobject_get(&q->kobj);
594 static inline void blk_free_request(struct request_queue *q, struct request *rq)
596 if (rq->cmd_flags & REQ_ELVPRIV)
597 elv_put_request(q, rq);
598 mempool_free(rq, q->rq.rq_pool);
601 static struct request *
602 blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
604 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
612 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
613 * see bio.h and blkdev.h
615 rq->cmd_flags = rw | REQ_ALLOCED;
618 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
619 mempool_free(rq, q->rq.rq_pool);
622 rq->cmd_flags |= REQ_ELVPRIV;
629 * ioc_batching returns true if the ioc is a valid batching request and
630 * should be given priority access to a request.
632 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
638 * Make sure the process is able to allocate at least 1 request
639 * even if the batch times out, otherwise we could theoretically
642 return ioc->nr_batch_requests == q->nr_batching ||
643 (ioc->nr_batch_requests > 0
644 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
648 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
649 * will cause the process to be a "batcher" on all queues in the system. This
650 * is the behaviour we want though - once it gets a wakeup it should be given
653 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
655 if (!ioc || ioc_batching(q, ioc))
658 ioc->nr_batch_requests = q->nr_batching;
659 ioc->last_waited = jiffies;
662 static void __freed_request(struct request_queue *q, int rw)
664 struct request_list *rl = &q->rq;
666 if (rl->count[rw] < queue_congestion_off_threshold(q))
667 blk_clear_queue_congested(q, rw);
669 if (rl->count[rw] + 1 <= q->nr_requests) {
670 if (waitqueue_active(&rl->wait[rw]))
671 wake_up(&rl->wait[rw]);
673 blk_clear_queue_full(q, rw);
678 * A request has just been released. Account for it, update the full and
679 * congestion status, wake up any waiters. Called under q->queue_lock.
681 static void freed_request(struct request_queue *q, int rw, int priv)
683 struct request_list *rl = &q->rq;
689 __freed_request(q, rw);
691 if (unlikely(rl->starved[rw ^ 1]))
692 __freed_request(q, rw ^ 1);
695 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
697 * Get a free request, queue_lock must be held.
698 * Returns NULL on failure, with queue_lock held.
699 * Returns !NULL on success, with queue_lock *not held*.
701 static struct request *get_request(struct request_queue *q, int rw_flags,
702 struct bio *bio, gfp_t gfp_mask)
704 struct request *rq = NULL;
705 struct request_list *rl = &q->rq;
706 struct io_context *ioc = NULL;
707 const int rw = rw_flags & 0x01;
710 may_queue = elv_may_queue(q, rw_flags);
711 if (may_queue == ELV_MQUEUE_NO)
714 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
715 if (rl->count[rw]+1 >= q->nr_requests) {
716 ioc = current_io_context(GFP_ATOMIC, q->node);
718 * The queue will fill after this allocation, so set
719 * it as full, and mark this process as "batching".
720 * This process will be allowed to complete a batch of
721 * requests, others will be blocked.
723 if (!blk_queue_full(q, rw)) {
724 ioc_set_batching(q, ioc);
725 blk_set_queue_full(q, rw);
727 if (may_queue != ELV_MQUEUE_MUST
728 && !ioc_batching(q, ioc)) {
730 * The queue is full and the allocating
731 * process is not a "batcher", and not
732 * exempted by the IO scheduler
738 blk_set_queue_congested(q, rw);
742 * Only allow batching queuers to allocate up to 50% over the defined
743 * limit of requests, otherwise we could have thousands of requests
744 * allocated with any setting of ->nr_requests
746 if (rl->count[rw] >= (3 * q->nr_requests / 2))
752 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
756 spin_unlock_irq(q->queue_lock);
758 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
761 * Allocation failed presumably due to memory. Undo anything
762 * we might have messed up.
764 * Allocating task should really be put onto the front of the
765 * wait queue, but this is pretty rare.
767 spin_lock_irq(q->queue_lock);
768 freed_request(q, rw, priv);
771 * in the very unlikely event that allocation failed and no
772 * requests for this direction was pending, mark us starved
773 * so that freeing of a request in the other direction will
774 * notice us. another possible fix would be to split the
775 * rq mempool into READ and WRITE
778 if (unlikely(rl->count[rw] == 0))
785 * ioc may be NULL here, and ioc_batching will be false. That's
786 * OK, if the queue is under the request limit then requests need
787 * not count toward the nr_batch_requests limit. There will always
788 * be some limit enforced by BLK_BATCH_TIME.
790 if (ioc_batching(q, ioc))
791 ioc->nr_batch_requests--;
793 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
799 * No available requests for this queue, unplug the device and wait for some
800 * requests to become available.
802 * Called with q->queue_lock held, and returns with it unlocked.
804 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
807 const int rw = rw_flags & 0x01;
810 rq = get_request(q, rw_flags, bio, GFP_NOIO);
813 struct io_context *ioc;
814 struct request_list *rl = &q->rq;
816 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
817 TASK_UNINTERRUPTIBLE);
819 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
821 __generic_unplug_device(q);
822 spin_unlock_irq(q->queue_lock);
826 * After sleeping, we become a "batching" process and
827 * will be able to allocate at least one request, and
828 * up to a big batch of them for a small period time.
829 * See ioc_batching, ioc_set_batching
831 ioc = current_io_context(GFP_NOIO, q->node);
832 ioc_set_batching(q, ioc);
834 spin_lock_irq(q->queue_lock);
835 finish_wait(&rl->wait[rw], &wait);
837 rq = get_request(q, rw_flags, bio, GFP_NOIO);
843 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
847 BUG_ON(rw != READ && rw != WRITE);
849 spin_lock_irq(q->queue_lock);
850 if (gfp_mask & __GFP_WAIT) {
851 rq = get_request_wait(q, rw, NULL);
853 rq = get_request(q, rw, NULL, gfp_mask);
855 spin_unlock_irq(q->queue_lock);
857 /* q->queue_lock is unlocked at this point */
861 EXPORT_SYMBOL(blk_get_request);
864 * blk_start_queueing - initiate dispatch of requests to device
865 * @q: request queue to kick into gear
867 * This is basically a helper to remove the need to know whether a queue
868 * is plugged or not if someone just wants to initiate dispatch of requests
871 * The queue lock must be held with interrupts disabled.
873 void blk_start_queueing(struct request_queue *q)
875 if (!blk_queue_plugged(q))
878 __generic_unplug_device(q);
880 EXPORT_SYMBOL(blk_start_queueing);
883 * blk_requeue_request - put a request back on queue
884 * @q: request queue where request should be inserted
885 * @rq: request to be inserted
888 * Drivers often keep queueing requests until the hardware cannot accept
889 * more, when that condition happens we need to put the request back
890 * on the queue. Must be called with queue lock held.
892 void blk_requeue_request(struct request_queue *q, struct request *rq)
894 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
896 if (blk_rq_tagged(rq))
897 blk_queue_end_tag(q, rq);
899 elv_requeue_request(q, rq);
901 EXPORT_SYMBOL(blk_requeue_request);
904 * blk_insert_request - insert a special request in to a request queue
905 * @q: request queue where request should be inserted
906 * @rq: request to be inserted
907 * @at_head: insert request at head or tail of queue
908 * @data: private data
911 * Many block devices need to execute commands asynchronously, so they don't
912 * block the whole kernel from preemption during request execution. This is
913 * accomplished normally by inserting aritficial requests tagged as
914 * REQ_SPECIAL in to the corresponding request queue, and letting them be
915 * scheduled for actual execution by the request queue.
917 * We have the option of inserting the head or the tail of the queue.
918 * Typically we use the tail for new ioctls and so forth. We use the head
919 * of the queue for things like a QUEUE_FULL message from a device, or a
920 * host that is unable to accept a particular command.
922 void blk_insert_request(struct request_queue *q, struct request *rq,
923 int at_head, void *data)
925 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
929 * tell I/O scheduler that this isn't a regular read/write (ie it
930 * must not attempt merges on this) and that it acts as a soft
933 rq->cmd_type = REQ_TYPE_SPECIAL;
934 rq->cmd_flags |= REQ_SOFTBARRIER;
938 spin_lock_irqsave(q->queue_lock, flags);
941 * If command is tagged, release the tag
943 if (blk_rq_tagged(rq))
944 blk_queue_end_tag(q, rq);
946 drive_stat_acct(rq, 1);
947 __elv_add_request(q, rq, where, 0);
948 blk_start_queueing(q);
949 spin_unlock_irqrestore(q->queue_lock, flags);
951 EXPORT_SYMBOL(blk_insert_request);
954 * add-request adds a request to the linked list.
955 * queue lock is held and interrupts disabled, as we muck with the
956 * request queue list.
958 static inline void add_request(struct request_queue *q, struct request *req)
960 drive_stat_acct(req, 1);
963 * elevator indicated where it wants this request to be
964 * inserted at elevator_merge time
966 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
970 * disk_round_stats() - Round off the performance stats on a struct
973 * The average IO queue length and utilisation statistics are maintained
974 * by observing the current state of the queue length and the amount of
975 * time it has been in this state for.
977 * Normally, that accounting is done on IO completion, but that can result
978 * in more than a second's worth of IO being accounted for within any one
979 * second, leading to >100% utilisation. To deal with that, we call this
980 * function to do a round-off before returning the results when reading
981 * /proc/diskstats. This accounts immediately for all queue usage up to
982 * the current jiffies and restarts the counters again.
984 void disk_round_stats(struct gendisk *disk)
986 unsigned long now = jiffies;
988 if (now == disk->stamp)
991 if (disk->in_flight) {
992 __disk_stat_add(disk, time_in_queue,
993 disk->in_flight * (now - disk->stamp));
994 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
998 EXPORT_SYMBOL_GPL(disk_round_stats);
1000 void part_round_stats(struct hd_struct *part)
1002 unsigned long now = jiffies;
1004 if (now == part->stamp)
1007 if (part->in_flight) {
1008 __part_stat_add(part, time_in_queue,
1009 part->in_flight * (now - part->stamp));
1010 __part_stat_add(part, io_ticks, (now - part->stamp));
1016 * queue lock must be held
1018 void __blk_put_request(struct request_queue *q, struct request *req)
1022 if (unlikely(--req->ref_count))
1025 elv_completed_request(q, req);
1028 * Request may not have originated from ll_rw_blk. if not,
1029 * it didn't come out of our reserved rq pools
1031 if (req->cmd_flags & REQ_ALLOCED) {
1032 int rw = rq_data_dir(req);
1033 int priv = req->cmd_flags & REQ_ELVPRIV;
1035 BUG_ON(!list_empty(&req->queuelist));
1036 BUG_ON(!hlist_unhashed(&req->hash));
1038 blk_free_request(q, req);
1039 freed_request(q, rw, priv);
1042 EXPORT_SYMBOL_GPL(__blk_put_request);
1044 void blk_put_request(struct request *req)
1046 unsigned long flags;
1047 struct request_queue *q = req->q;
1050 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
1051 * following if (q) test.
1054 spin_lock_irqsave(q->queue_lock, flags);
1055 __blk_put_request(q, req);
1056 spin_unlock_irqrestore(q->queue_lock, flags);
1059 EXPORT_SYMBOL(blk_put_request);
1061 void init_request_from_bio(struct request *req, struct bio *bio)
1063 req->cmd_type = REQ_TYPE_FS;
1066 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1068 if (bio_rw_ahead(bio) || bio_failfast(bio))
1069 req->cmd_flags |= REQ_FAILFAST;
1072 * REQ_BARRIER implies no merging, but lets make it explicit
1074 if (unlikely(bio_barrier(bio)))
1075 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
1078 req->cmd_flags |= REQ_RW_SYNC;
1079 if (bio_rw_meta(bio))
1080 req->cmd_flags |= REQ_RW_META;
1083 req->hard_sector = req->sector = bio->bi_sector;
1084 req->ioprio = bio_prio(bio);
1085 req->start_time = jiffies;
1086 blk_rq_bio_prep(req->q, req, bio);
1089 static int __make_request(struct request_queue *q, struct bio *bio)
1091 struct request *req;
1092 int el_ret, nr_sectors, barrier, err;
1093 const unsigned short prio = bio_prio(bio);
1094 const int sync = bio_sync(bio);
1097 nr_sectors = bio_sectors(bio);
1100 * low level driver can indicate that it wants pages above a
1101 * certain limit bounced to low memory (ie for highmem, or even
1102 * ISA dma in theory)
1104 blk_queue_bounce(q, &bio);
1106 barrier = bio_barrier(bio);
1107 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
1112 spin_lock_irq(q->queue_lock);
1114 if (unlikely(barrier) || elv_queue_empty(q))
1117 el_ret = elv_merge(q, &req, bio);
1119 case ELEVATOR_BACK_MERGE:
1120 BUG_ON(!rq_mergeable(req));
1122 if (!ll_back_merge_fn(q, req, bio))
1125 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
1127 req->biotail->bi_next = bio;
1129 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1130 req->ioprio = ioprio_best(req->ioprio, prio);
1131 drive_stat_acct(req, 0);
1132 if (!attempt_back_merge(q, req))
1133 elv_merged_request(q, req, el_ret);
1136 case ELEVATOR_FRONT_MERGE:
1137 BUG_ON(!rq_mergeable(req));
1139 if (!ll_front_merge_fn(q, req, bio))
1142 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
1144 bio->bi_next = req->bio;
1148 * may not be valid. if the low level driver said
1149 * it didn't need a bounce buffer then it better
1150 * not touch req->buffer either...
1152 req->buffer = bio_data(bio);
1153 req->current_nr_sectors = bio_cur_sectors(bio);
1154 req->hard_cur_sectors = req->current_nr_sectors;
1155 req->sector = req->hard_sector = bio->bi_sector;
1156 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1157 req->ioprio = ioprio_best(req->ioprio, prio);
1158 drive_stat_acct(req, 0);
1159 if (!attempt_front_merge(q, req))
1160 elv_merged_request(q, req, el_ret);
1163 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1170 * This sync check and mask will be re-done in init_request_from_bio(),
1171 * but we need to set it earlier to expose the sync flag to the
1172 * rq allocator and io schedulers.
1174 rw_flags = bio_data_dir(bio);
1176 rw_flags |= REQ_RW_SYNC;
1179 * Grab a free request. This is might sleep but can not fail.
1180 * Returns with the queue unlocked.
1182 req = get_request_wait(q, rw_flags, bio);
1185 * After dropping the lock and possibly sleeping here, our request
1186 * may now be mergeable after it had proven unmergeable (above).
1187 * We don't worry about that case for efficiency. It won't happen
1188 * often, and the elevators are able to handle it.
1190 init_request_from_bio(req, bio);
1192 spin_lock_irq(q->queue_lock);
1193 if (elv_queue_empty(q))
1195 add_request(q, req);
1198 __generic_unplug_device(q);
1200 spin_unlock_irq(q->queue_lock);
1204 bio_endio(bio, err);
1209 * If bio->bi_dev is a partition, remap the location
1211 static inline void blk_partition_remap(struct bio *bio)
1213 struct block_device *bdev = bio->bi_bdev;
1215 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1216 struct hd_struct *p = bdev->bd_part;
1218 bio->bi_sector += p->start_sect;
1219 bio->bi_bdev = bdev->bd_contains;
1221 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1222 bdev->bd_dev, bio->bi_sector,
1223 bio->bi_sector - p->start_sect);
1227 static void handle_bad_sector(struct bio *bio)
1229 char b[BDEVNAME_SIZE];
1231 printk(KERN_INFO "attempt to access beyond end of device\n");
1232 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1233 bdevname(bio->bi_bdev, b),
1235 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1236 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1238 set_bit(BIO_EOF, &bio->bi_flags);
1241 #ifdef CONFIG_FAIL_MAKE_REQUEST
1243 static DECLARE_FAULT_ATTR(fail_make_request);
1245 static int __init setup_fail_make_request(char *str)
1247 return setup_fault_attr(&fail_make_request, str);
1249 __setup("fail_make_request=", setup_fail_make_request);
1251 static int should_fail_request(struct bio *bio)
1253 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
1254 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
1255 return should_fail(&fail_make_request, bio->bi_size);
1260 static int __init fail_make_request_debugfs(void)
1262 return init_fault_attr_dentries(&fail_make_request,
1263 "fail_make_request");
1266 late_initcall(fail_make_request_debugfs);
1268 #else /* CONFIG_FAIL_MAKE_REQUEST */
1270 static inline int should_fail_request(struct bio *bio)
1275 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1278 * Check whether this bio extends beyond the end of the device.
1280 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1287 /* Test device or partition size, when known. */
1288 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1290 sector_t sector = bio->bi_sector;
1292 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1294 * This may well happen - the kernel calls bread()
1295 * without checking the size of the device, e.g., when
1296 * mounting a device.
1298 handle_bad_sector(bio);
1307 * generic_make_request: hand a buffer to its device driver for I/O
1308 * @bio: The bio describing the location in memory and on the device.
1310 * generic_make_request() is used to make I/O requests of block
1311 * devices. It is passed a &struct bio, which describes the I/O that needs
1314 * generic_make_request() does not return any status. The
1315 * success/failure status of the request, along with notification of
1316 * completion, is delivered asynchronously through the bio->bi_end_io
1317 * function described (one day) else where.
1319 * The caller of generic_make_request must make sure that bi_io_vec
1320 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1321 * set to describe the device address, and the
1322 * bi_end_io and optionally bi_private are set to describe how
1323 * completion notification should be signaled.
1325 * generic_make_request and the drivers it calls may use bi_next if this
1326 * bio happens to be merged with someone else, and may change bi_dev and
1327 * bi_sector for remaps as it sees fit. So the values of these fields
1328 * should NOT be depended on after the call to generic_make_request.
1330 static inline void __generic_make_request(struct bio *bio)
1332 struct request_queue *q;
1333 sector_t old_sector;
1334 int ret, nr_sectors = bio_sectors(bio);
1340 if (bio_check_eod(bio, nr_sectors))
1344 * Resolve the mapping until finished. (drivers are
1345 * still free to implement/resolve their own stacking
1346 * by explicitly returning 0)
1348 * NOTE: we don't repeat the blk_size check for each new device.
1349 * Stacking drivers are expected to know what they are doing.
1354 char b[BDEVNAME_SIZE];
1356 q = bdev_get_queue(bio->bi_bdev);
1359 "generic_make_request: Trying to access "
1360 "nonexistent block-device %s (%Lu)\n",
1361 bdevname(bio->bi_bdev, b),
1362 (long long) bio->bi_sector);
1364 bio_endio(bio, err);
1368 if (unlikely(nr_sectors > q->max_hw_sectors)) {
1369 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1370 bdevname(bio->bi_bdev, b),
1376 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1379 if (should_fail_request(bio))
1383 * If this device has partitions, remap block n
1384 * of partition p to block n+start(p) of the disk.
1386 blk_partition_remap(bio);
1388 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1391 if (old_sector != -1)
1392 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
1395 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1397 old_sector = bio->bi_sector;
1398 old_dev = bio->bi_bdev->bd_dev;
1400 if (bio_check_eod(bio, nr_sectors))
1402 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
1407 ret = q->make_request_fn(q, bio);
1412 * We only want one ->make_request_fn to be active at a time,
1413 * else stack usage with stacked devices could be a problem.
1414 * So use current->bio_{list,tail} to keep a list of requests
1415 * submited by a make_request_fn function.
1416 * current->bio_tail is also used as a flag to say if
1417 * generic_make_request is currently active in this task or not.
1418 * If it is NULL, then no make_request is active. If it is non-NULL,
1419 * then a make_request is active, and new requests should be added
1422 void generic_make_request(struct bio *bio)
1424 if (current->bio_tail) {
1425 /* make_request is active */
1426 *(current->bio_tail) = bio;
1427 bio->bi_next = NULL;
1428 current->bio_tail = &bio->bi_next;
1431 /* following loop may be a bit non-obvious, and so deserves some
1433 * Before entering the loop, bio->bi_next is NULL (as all callers
1434 * ensure that) so we have a list with a single bio.
1435 * We pretend that we have just taken it off a longer list, so
1436 * we assign bio_list to the next (which is NULL) and bio_tail
1437 * to &bio_list, thus initialising the bio_list of new bios to be
1438 * added. __generic_make_request may indeed add some more bios
1439 * through a recursive call to generic_make_request. If it
1440 * did, we find a non-NULL value in bio_list and re-enter the loop
1441 * from the top. In this case we really did just take the bio
1442 * of the top of the list (no pretending) and so fixup bio_list and
1443 * bio_tail or bi_next, and call into __generic_make_request again.
1445 * The loop was structured like this to make only one call to
1446 * __generic_make_request (which is important as it is large and
1447 * inlined) and to keep the structure simple.
1449 BUG_ON(bio->bi_next);
1451 current->bio_list = bio->bi_next;
1452 if (bio->bi_next == NULL)
1453 current->bio_tail = ¤t->bio_list;
1455 bio->bi_next = NULL;
1456 __generic_make_request(bio);
1457 bio = current->bio_list;
1459 current->bio_tail = NULL; /* deactivate */
1461 EXPORT_SYMBOL(generic_make_request);
1464 * submit_bio: submit a bio to the block device layer for I/O
1465 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1466 * @bio: The &struct bio which describes the I/O
1468 * submit_bio() is very similar in purpose to generic_make_request(), and
1469 * uses that function to do most of the work. Both are fairly rough
1470 * interfaces, @bio must be presetup and ready for I/O.
1473 void submit_bio(int rw, struct bio *bio)
1475 int count = bio_sectors(bio);
1480 * If it's a regular read/write or a barrier with data attached,
1481 * go through the normal accounting stuff before submission.
1483 if (!bio_empty_barrier(bio)) {
1485 BIO_BUG_ON(!bio->bi_size);
1486 BIO_BUG_ON(!bio->bi_io_vec);
1489 count_vm_events(PGPGOUT, count);
1491 task_io_account_read(bio->bi_size);
1492 count_vm_events(PGPGIN, count);
1495 if (unlikely(block_dump)) {
1496 char b[BDEVNAME_SIZE];
1497 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1498 current->comm, task_pid_nr(current),
1499 (rw & WRITE) ? "WRITE" : "READ",
1500 (unsigned long long)bio->bi_sector,
1501 bdevname(bio->bi_bdev, b));
1505 generic_make_request(bio);
1507 EXPORT_SYMBOL(submit_bio);
1510 * __end_that_request_first - end I/O on a request
1511 * @req: the request being processed
1512 * @error: 0 for success, < 0 for error
1513 * @nr_bytes: number of bytes to complete
1516 * Ends I/O on a number of bytes attached to @req, and sets it up
1517 * for the next range of segments (if any) in the cluster.
1520 * 0 - we are done with this request, call end_that_request_last()
1521 * 1 - still buffers pending for this request
1523 static int __end_that_request_first(struct request *req, int error,
1526 int total_bytes, bio_nbytes, next_idx = 0;
1529 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1532 * for a REQ_BLOCK_PC request, we want to carry any eventual
1533 * sense key with us all the way through
1535 if (!blk_pc_request(req))
1538 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1539 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1540 req->rq_disk ? req->rq_disk->disk_name : "?",
1541 (unsigned long long)req->sector);
1544 if (blk_fs_request(req) && req->rq_disk) {
1545 struct hd_struct *part = get_part(req->rq_disk, req->sector);
1546 const int rw = rq_data_dir(req);
1548 all_stat_add(req->rq_disk, part, sectors[rw],
1549 nr_bytes >> 9, req->sector);
1552 total_bytes = bio_nbytes = 0;
1553 while ((bio = req->bio) != NULL) {
1557 * For an empty barrier request, the low level driver must
1558 * store a potential error location in ->sector. We pass
1559 * that back up in ->bi_sector.
1561 if (blk_empty_barrier(req))
1562 bio->bi_sector = req->sector;
1564 if (nr_bytes >= bio->bi_size) {
1565 req->bio = bio->bi_next;
1566 nbytes = bio->bi_size;
1567 req_bio_endio(req, bio, nbytes, error);
1571 int idx = bio->bi_idx + next_idx;
1573 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1574 blk_dump_rq_flags(req, "__end_that");
1575 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1576 __func__, bio->bi_idx, bio->bi_vcnt);
1580 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1581 BIO_BUG_ON(nbytes > bio->bi_size);
1584 * not a complete bvec done
1586 if (unlikely(nbytes > nr_bytes)) {
1587 bio_nbytes += nr_bytes;
1588 total_bytes += nr_bytes;
1593 * advance to the next vector
1596 bio_nbytes += nbytes;
1599 total_bytes += nbytes;
1605 * end more in this run, or just return 'not-done'
1607 if (unlikely(nr_bytes <= 0))
1619 * if the request wasn't completed, update state
1622 req_bio_endio(req, bio, bio_nbytes, error);
1623 bio->bi_idx += next_idx;
1624 bio_iovec(bio)->bv_offset += nr_bytes;
1625 bio_iovec(bio)->bv_len -= nr_bytes;
1628 blk_recalc_rq_sectors(req, total_bytes >> 9);
1629 blk_recalc_rq_segments(req);
1634 * splice the completion data to a local structure and hand off to
1635 * process_completion_queue() to complete the requests
1637 static void blk_done_softirq(struct softirq_action *h)
1639 struct list_head *cpu_list, local_list;
1641 local_irq_disable();
1642 cpu_list = &__get_cpu_var(blk_cpu_done);
1643 list_replace_init(cpu_list, &local_list);
1646 while (!list_empty(&local_list)) {
1649 rq = list_entry(local_list.next, struct request, donelist);
1650 list_del_init(&rq->donelist);
1651 rq->q->softirq_done_fn(rq);
1655 static int __cpuinit blk_cpu_notify(struct notifier_block *self,
1656 unsigned long action, void *hcpu)
1659 * If a CPU goes away, splice its entries to the current CPU
1660 * and trigger a run of the softirq
1662 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1663 int cpu = (unsigned long) hcpu;
1665 local_irq_disable();
1666 list_splice_init(&per_cpu(blk_cpu_done, cpu),
1667 &__get_cpu_var(blk_cpu_done));
1668 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1676 static struct notifier_block blk_cpu_notifier __cpuinitdata = {
1677 .notifier_call = blk_cpu_notify,
1681 * blk_complete_request - end I/O on a request
1682 * @req: the request being processed
1685 * Ends all I/O on a request. It does not handle partial completions,
1686 * unless the driver actually implements this in its completion callback
1687 * through requeueing. The actual completion happens out-of-order,
1688 * through a softirq handler. The user must have registered a completion
1689 * callback through blk_queue_softirq_done().
1692 void blk_complete_request(struct request *req)
1694 struct list_head *cpu_list;
1695 unsigned long flags;
1697 BUG_ON(!req->q->softirq_done_fn);
1699 local_irq_save(flags);
1701 cpu_list = &__get_cpu_var(blk_cpu_done);
1702 list_add_tail(&req->donelist, cpu_list);
1703 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1705 local_irq_restore(flags);
1707 EXPORT_SYMBOL(blk_complete_request);
1710 * queue lock must be held
1712 static void end_that_request_last(struct request *req, int error)
1714 struct gendisk *disk = req->rq_disk;
1716 if (blk_rq_tagged(req))
1717 blk_queue_end_tag(req->q, req);
1719 if (blk_queued_rq(req))
1720 blkdev_dequeue_request(req);
1722 if (unlikely(laptop_mode) && blk_fs_request(req))
1723 laptop_io_completion();
1726 * Account IO completion. bar_rq isn't accounted as a normal
1727 * IO on queueing nor completion. Accounting the containing
1728 * request is enough.
1730 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1731 unsigned long duration = jiffies - req->start_time;
1732 const int rw = rq_data_dir(req);
1733 struct hd_struct *part = get_part(disk, req->sector);
1735 __all_stat_inc(disk, part, ios[rw], req->sector);
1736 __all_stat_add(disk, part, ticks[rw], duration, req->sector);
1737 disk_round_stats(disk);
1740 part_round_stats(part);
1746 req->end_io(req, error);
1748 if (blk_bidi_rq(req))
1749 __blk_put_request(req->next_rq->q, req->next_rq);
1751 __blk_put_request(req->q, req);
1755 static inline void __end_request(struct request *rq, int uptodate,
1756 unsigned int nr_bytes)
1761 error = uptodate ? uptodate : -EIO;
1763 __blk_end_request(rq, error, nr_bytes);
1767 * blk_rq_bytes - Returns bytes left to complete in the entire request
1768 * @rq: the request being processed
1770 unsigned int blk_rq_bytes(struct request *rq)
1772 if (blk_fs_request(rq))
1773 return rq->hard_nr_sectors << 9;
1775 return rq->data_len;
1777 EXPORT_SYMBOL_GPL(blk_rq_bytes);
1780 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1781 * @rq: the request being processed
1783 unsigned int blk_rq_cur_bytes(struct request *rq)
1785 if (blk_fs_request(rq))
1786 return rq->current_nr_sectors << 9;
1789 return rq->bio->bi_size;
1791 return rq->data_len;
1793 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1796 * end_queued_request - end all I/O on a queued request
1797 * @rq: the request being processed
1798 * @uptodate: error value or 0/1 uptodate flag
1801 * Ends all I/O on a request, and removes it from the block layer queues.
1802 * Not suitable for normal IO completion, unless the driver still has
1803 * the request attached to the block layer.
1806 void end_queued_request(struct request *rq, int uptodate)
1808 __end_request(rq, uptodate, blk_rq_bytes(rq));
1810 EXPORT_SYMBOL(end_queued_request);
1813 * end_dequeued_request - end all I/O on a dequeued request
1814 * @rq: the request being processed
1815 * @uptodate: error value or 0/1 uptodate flag
1818 * Ends all I/O on a request. The request must already have been
1819 * dequeued using blkdev_dequeue_request(), as is normally the case
1823 void end_dequeued_request(struct request *rq, int uptodate)
1825 __end_request(rq, uptodate, blk_rq_bytes(rq));
1827 EXPORT_SYMBOL(end_dequeued_request);
1831 * end_request - end I/O on the current segment of the request
1832 * @req: the request being processed
1833 * @uptodate: error value or 0/1 uptodate flag
1836 * Ends I/O on the current segment of a request. If that is the only
1837 * remaining segment, the request is also completed and freed.
1839 * This is a remnant of how older block drivers handled IO completions.
1840 * Modern drivers typically end IO on the full request in one go, unless
1841 * they have a residual value to account for. For that case this function
1842 * isn't really useful, unless the residual just happens to be the
1843 * full current segment. In other words, don't use this function in new
1844 * code. Either use end_request_completely(), or the
1845 * end_that_request_chunk() (along with end_that_request_last()) for
1846 * partial completions.
1849 void end_request(struct request *req, int uptodate)
1851 __end_request(req, uptodate, req->hard_cur_sectors << 9);
1853 EXPORT_SYMBOL(end_request);
1856 * blk_end_io - Generic end_io function to complete a request.
1857 * @rq: the request being processed
1858 * @error: 0 for success, < 0 for error
1859 * @nr_bytes: number of bytes to complete @rq
1860 * @bidi_bytes: number of bytes to complete @rq->next_rq
1861 * @drv_callback: function called between completion of bios in the request
1862 * and completion of the request.
1863 * If the callback returns non 0, this helper returns without
1864 * completion of the request.
1867 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1868 * If @rq has leftover, sets it up for the next range of segments.
1871 * 0 - we are done with this request
1872 * 1 - this request is not freed yet, it still has pending buffers.
1874 static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1875 unsigned int bidi_bytes,
1876 int (drv_callback)(struct request *))
1878 struct request_queue *q = rq->q;
1879 unsigned long flags = 0UL;
1881 if (blk_fs_request(rq) || blk_pc_request(rq)) {
1882 if (__end_that_request_first(rq, error, nr_bytes))
1885 /* Bidi request must be completed as a whole */
1886 if (blk_bidi_rq(rq) &&
1887 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1891 /* Special feature for tricky drivers */
1892 if (drv_callback && drv_callback(rq))
1895 add_disk_randomness(rq->rq_disk);
1897 spin_lock_irqsave(q->queue_lock, flags);
1898 end_that_request_last(rq, error);
1899 spin_unlock_irqrestore(q->queue_lock, flags);
1905 * blk_end_request - Helper function for drivers to complete the request.
1906 * @rq: the request being processed
1907 * @error: 0 for success, < 0 for error
1908 * @nr_bytes: number of bytes to complete
1911 * Ends I/O on a number of bytes attached to @rq.
1912 * If @rq has leftover, sets it up for the next range of segments.
1915 * 0 - we are done with this request
1916 * 1 - still buffers pending for this request
1918 int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1920 return blk_end_io(rq, error, nr_bytes, 0, NULL);
1922 EXPORT_SYMBOL_GPL(blk_end_request);
1925 * __blk_end_request - Helper function for drivers to complete the request.
1926 * @rq: the request being processed
1927 * @error: 0 for success, < 0 for error
1928 * @nr_bytes: number of bytes to complete
1931 * Must be called with queue lock held unlike blk_end_request().
1934 * 0 - we are done with this request
1935 * 1 - still buffers pending for this request
1937 int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1939 if (blk_fs_request(rq) || blk_pc_request(rq)) {
1940 if (__end_that_request_first(rq, error, nr_bytes))
1944 add_disk_randomness(rq->rq_disk);
1946 end_that_request_last(rq, error);
1950 EXPORT_SYMBOL_GPL(__blk_end_request);
1953 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1954 * @rq: the bidi request being processed
1955 * @error: 0 for success, < 0 for error
1956 * @nr_bytes: number of bytes to complete @rq
1957 * @bidi_bytes: number of bytes to complete @rq->next_rq
1960 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1963 * 0 - we are done with this request
1964 * 1 - still buffers pending for this request
1966 int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1967 unsigned int bidi_bytes)
1969 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1971 EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1974 * blk_end_request_callback - Special helper function for tricky drivers
1975 * @rq: the request being processed
1976 * @error: 0 for success, < 0 for error
1977 * @nr_bytes: number of bytes to complete
1978 * @drv_callback: function called between completion of bios in the request
1979 * and completion of the request.
1980 * If the callback returns non 0, this helper returns without
1981 * completion of the request.
1984 * Ends I/O on a number of bytes attached to @rq.
1985 * If @rq has leftover, sets it up for the next range of segments.
1987 * This special helper function is used only for existing tricky drivers.
1988 * (e.g. cdrom_newpc_intr() of ide-cd)
1989 * This interface will be removed when such drivers are rewritten.
1990 * Don't use this interface in other places anymore.
1993 * 0 - we are done with this request
1994 * 1 - this request is not freed yet.
1995 * this request still has pending buffers or
1996 * the driver doesn't want to finish this request yet.
1998 int blk_end_request_callback(struct request *rq, int error,
1999 unsigned int nr_bytes,
2000 int (drv_callback)(struct request *))
2002 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
2004 EXPORT_SYMBOL_GPL(blk_end_request_callback);
2006 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2009 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2010 rq->cmd_flags |= (bio->bi_rw & 3);
2012 rq->nr_phys_segments = bio_phys_segments(q, bio);
2013 rq->nr_hw_segments = bio_hw_segments(q, bio);
2014 rq->current_nr_sectors = bio_cur_sectors(bio);
2015 rq->hard_cur_sectors = rq->current_nr_sectors;
2016 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2017 rq->buffer = bio_data(bio);
2018 rq->data_len = bio->bi_size;
2020 rq->bio = rq->biotail = bio;
2023 rq->rq_disk = bio->bi_bdev->bd_disk;
2026 int kblockd_schedule_work(struct work_struct *work)
2028 return queue_work(kblockd_workqueue, work);
2030 EXPORT_SYMBOL(kblockd_schedule_work);
2032 void kblockd_flush_work(struct work_struct *work)
2034 cancel_work_sync(work);
2036 EXPORT_SYMBOL(kblockd_flush_work);
2038 int __init blk_dev_init(void)
2042 kblockd_workqueue = create_workqueue("kblockd");
2043 if (!kblockd_workqueue)
2044 panic("Failed to create kblockd\n");
2046 request_cachep = kmem_cache_create("blkdev_requests",
2047 sizeof(struct request), 0, SLAB_PANIC, NULL);
2049 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2050 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2052 for_each_possible_cpu(i)
2053 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2055 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
2056 register_hotcpu_notifier(&blk_cpu_notifier);