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 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 int rw = rq_data_dir(rq);
59 if (!blk_fs_request(rq) || !rq->rq_disk)
63 __disk_stat_inc(rq->rq_disk, merges[rw]);
65 disk_round_stats(rq->rq_disk);
66 rq->rq_disk->in_flight++;
70 void blk_queue_congestion_threshold(struct request_queue *q)
74 nr = q->nr_requests - (q->nr_requests / 8) + 1;
75 if (nr > q->nr_requests)
77 q->nr_congestion_on = nr;
79 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
82 q->nr_congestion_off = nr;
86 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
89 * Locates the passed device's request queue and returns the address of its
92 * Will return NULL if the request queue cannot be located.
94 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
96 struct backing_dev_info *ret = NULL;
97 struct request_queue *q = bdev_get_queue(bdev);
100 ret = &q->backing_dev_info;
103 EXPORT_SYMBOL(blk_get_backing_dev_info);
106 * We can't just memset() the structure, since the allocation path
107 * already stored some information in the request.
109 void rq_init(struct request_queue *q, struct request *rq)
111 INIT_LIST_HEAD(&rq->queuelist);
112 INIT_LIST_HEAD(&rq->donelist);
114 rq->sector = rq->hard_sector = (sector_t) -1;
115 rq->nr_sectors = rq->hard_nr_sectors = 0;
116 rq->current_nr_sectors = rq->hard_cur_sectors = 0;
117 rq->bio = rq->biotail = NULL;
118 INIT_HLIST_NODE(&rq->hash);
119 RB_CLEAR_NODE(&rq->rb_node);
121 rq->nr_phys_segments = 0;
122 rq->nr_hw_segments = 0;
130 memset(rq->cmd, 0, sizeof(rq->cmd));
136 rq->end_io_data = NULL;
140 static void req_bio_endio(struct request *rq, struct bio *bio,
141 unsigned int nbytes, int error)
143 struct request_queue *q = rq->q;
145 if (&q->bar_rq != rq) {
147 clear_bit(BIO_UPTODATE, &bio->bi_flags);
148 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
151 if (unlikely(nbytes > bio->bi_size)) {
152 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
153 __FUNCTION__, nbytes, bio->bi_size);
154 nbytes = bio->bi_size;
157 bio->bi_size -= nbytes;
158 bio->bi_sector += (nbytes >> 9);
159 if (bio->bi_size == 0)
160 bio_endio(bio, error);
164 * Okay, this is the barrier request in progress, just
167 if (error && !q->orderr)
172 void blk_dump_rq_flags(struct request *rq, char *msg)
176 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
177 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
180 printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n",
181 (unsigned long long)rq->sector,
183 rq->current_nr_sectors);
184 printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n",
185 rq->bio, rq->biotail,
186 rq->buffer, rq->data,
189 if (blk_pc_request(rq)) {
190 printk(KERN_INFO " cdb: ");
191 for (bit = 0; bit < sizeof(rq->cmd); bit++)
192 printk("%02x ", rq->cmd[bit]);
196 EXPORT_SYMBOL(blk_dump_rq_flags);
199 * "plug" the device if there are no outstanding requests: this will
200 * force the transfer to start only after we have put all the requests
203 * This is called with interrupts off and no requests on the queue and
204 * with the queue lock held.
206 void blk_plug_device(struct request_queue *q)
208 WARN_ON(!irqs_disabled());
211 * don't plug a stopped queue, it must be paired with blk_start_queue()
212 * which will restart the queueing
214 if (blk_queue_stopped(q))
217 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
218 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
219 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
222 EXPORT_SYMBOL(blk_plug_device);
225 * remove the queue from the plugged list, if present. called with
226 * queue lock held and interrupts disabled.
228 int blk_remove_plug(struct request_queue *q)
230 WARN_ON(!irqs_disabled());
232 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
235 del_timer(&q->unplug_timer);
238 EXPORT_SYMBOL(blk_remove_plug);
241 * remove the plug and let it rip..
243 void __generic_unplug_device(struct request_queue *q)
245 if (unlikely(blk_queue_stopped(q)))
248 if (!blk_remove_plug(q))
253 EXPORT_SYMBOL(__generic_unplug_device);
256 * generic_unplug_device - fire a request queue
257 * @q: The &struct request_queue in question
260 * Linux uses plugging to build bigger requests queues before letting
261 * the device have at them. If a queue is plugged, the I/O scheduler
262 * is still adding and merging requests on the queue. Once the queue
263 * gets unplugged, the request_fn defined for the queue is invoked and
266 void generic_unplug_device(struct request_queue *q)
268 spin_lock_irq(q->queue_lock);
269 __generic_unplug_device(q);
270 spin_unlock_irq(q->queue_lock);
272 EXPORT_SYMBOL(generic_unplug_device);
274 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
277 struct request_queue *q = bdi->unplug_io_data;
282 void blk_unplug_work(struct work_struct *work)
284 struct request_queue *q =
285 container_of(work, struct request_queue, unplug_work);
287 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
288 q->rq.count[READ] + q->rq.count[WRITE]);
293 void blk_unplug_timeout(unsigned long data)
295 struct request_queue *q = (struct request_queue *)data;
297 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
298 q->rq.count[READ] + q->rq.count[WRITE]);
300 kblockd_schedule_work(&q->unplug_work);
303 void blk_unplug(struct request_queue *q)
306 * devices don't necessarily have an ->unplug_fn defined
309 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
310 q->rq.count[READ] + q->rq.count[WRITE]);
315 EXPORT_SYMBOL(blk_unplug);
318 * blk_start_queue - restart a previously stopped queue
319 * @q: The &struct request_queue in question
322 * blk_start_queue() will clear the stop flag on the queue, and call
323 * the request_fn for the queue if it was in a stopped state when
324 * entered. Also see blk_stop_queue(). Queue lock must be held.
326 void blk_start_queue(struct request_queue *q)
328 WARN_ON(!irqs_disabled());
330 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
333 * one level of recursion is ok and is much faster than kicking
334 * the unplug handling
336 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
338 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
341 kblockd_schedule_work(&q->unplug_work);
344 EXPORT_SYMBOL(blk_start_queue);
347 * blk_stop_queue - stop a queue
348 * @q: The &struct request_queue in question
351 * The Linux block layer assumes that a block driver will consume all
352 * entries on the request queue when the request_fn strategy is called.
353 * Often this will not happen, because of hardware limitations (queue
354 * depth settings). If a device driver gets a 'queue full' response,
355 * or if it simply chooses not to queue more I/O at one point, it can
356 * call this function to prevent the request_fn from being called until
357 * the driver has signalled it's ready to go again. This happens by calling
358 * blk_start_queue() to restart queue operations. Queue lock must be held.
360 void blk_stop_queue(struct request_queue *q)
363 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
365 EXPORT_SYMBOL(blk_stop_queue);
368 * blk_sync_queue - cancel any pending callbacks on a queue
372 * The block layer may perform asynchronous callback activity
373 * on a queue, such as calling the unplug function after a timeout.
374 * A block device may call blk_sync_queue to ensure that any
375 * such activity is cancelled, thus allowing it to release resources
376 * that the callbacks might use. The caller must already have made sure
377 * that its ->make_request_fn will not re-add plugging prior to calling
381 void blk_sync_queue(struct request_queue *q)
383 del_timer_sync(&q->unplug_timer);
384 kblockd_flush_work(&q->unplug_work);
386 EXPORT_SYMBOL(blk_sync_queue);
389 * blk_run_queue - run a single device queue
390 * @q: The queue to run
392 void blk_run_queue(struct request_queue *q)
396 spin_lock_irqsave(q->queue_lock, flags);
400 * Only recurse once to avoid overrunning the stack, let the unplug
401 * handling reinvoke the handler shortly if we already got there.
403 if (!elv_queue_empty(q)) {
404 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
406 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
409 kblockd_schedule_work(&q->unplug_work);
413 spin_unlock_irqrestore(q->queue_lock, flags);
415 EXPORT_SYMBOL(blk_run_queue);
417 void blk_put_queue(struct request_queue *q)
419 kobject_put(&q->kobj);
421 EXPORT_SYMBOL(blk_put_queue);
423 void blk_cleanup_queue(struct request_queue *q)
425 mutex_lock(&q->sysfs_lock);
426 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
427 mutex_unlock(&q->sysfs_lock);
430 elevator_exit(q->elevator);
434 EXPORT_SYMBOL(blk_cleanup_queue);
436 static int blk_init_free_list(struct request_queue *q)
438 struct request_list *rl = &q->rq;
440 rl->count[READ] = rl->count[WRITE] = 0;
441 rl->starved[READ] = rl->starved[WRITE] = 0;
443 init_waitqueue_head(&rl->wait[READ]);
444 init_waitqueue_head(&rl->wait[WRITE]);
446 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
447 mempool_free_slab, request_cachep, q->node);
455 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
457 return blk_alloc_queue_node(gfp_mask, -1);
459 EXPORT_SYMBOL(blk_alloc_queue);
461 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
463 struct request_queue *q;
466 q = kmem_cache_alloc_node(blk_requestq_cachep,
467 gfp_mask | __GFP_ZERO, node_id);
471 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
472 q->backing_dev_info.unplug_io_data = q;
473 err = bdi_init(&q->backing_dev_info);
475 kmem_cache_free(blk_requestq_cachep, q);
479 init_timer(&q->unplug_timer);
481 kobject_init(&q->kobj, &blk_queue_ktype);
483 mutex_init(&q->sysfs_lock);
487 EXPORT_SYMBOL(blk_alloc_queue_node);
490 * blk_init_queue - prepare a request queue for use with a block device
491 * @rfn: The function to be called to process requests that have been
492 * placed on the queue.
493 * @lock: Request queue spin lock
496 * If a block device wishes to use the standard request handling procedures,
497 * which sorts requests and coalesces adjacent requests, then it must
498 * call blk_init_queue(). The function @rfn will be called when there
499 * are requests on the queue that need to be processed. If the device
500 * supports plugging, then @rfn may not be called immediately when requests
501 * are available on the queue, but may be called at some time later instead.
502 * Plugged queues are generally unplugged when a buffer belonging to one
503 * of the requests on the queue is needed, or due to memory pressure.
505 * @rfn is not required, or even expected, to remove all requests off the
506 * queue, but only as many as it can handle at a time. If it does leave
507 * requests on the queue, it is responsible for arranging that the requests
508 * get dealt with eventually.
510 * The queue spin lock must be held while manipulating the requests on the
511 * request queue; this lock will be taken also from interrupt context, so irq
512 * disabling is needed for it.
514 * Function returns a pointer to the initialized request queue, or NULL if
518 * blk_init_queue() must be paired with a blk_cleanup_queue() call
519 * when the block device is deactivated (such as at module unload).
522 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
524 return blk_init_queue_node(rfn, lock, -1);
526 EXPORT_SYMBOL(blk_init_queue);
528 struct request_queue *
529 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
531 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
537 if (blk_init_free_list(q)) {
538 kmem_cache_free(blk_requestq_cachep, q);
543 * if caller didn't supply a lock, they get per-queue locking with
547 spin_lock_init(&q->__queue_lock);
548 lock = &q->__queue_lock;
552 q->prep_rq_fn = NULL;
553 q->unplug_fn = generic_unplug_device;
554 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
555 q->queue_lock = lock;
557 blk_queue_segment_boundary(q, 0xffffffff);
559 blk_queue_make_request(q, __make_request);
560 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
562 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
563 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
565 q->sg_reserved_size = INT_MAX;
570 if (!elevator_init(q, NULL)) {
571 blk_queue_congestion_threshold(q);
578 EXPORT_SYMBOL(blk_init_queue_node);
580 int blk_get_queue(struct request_queue *q)
582 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
583 kobject_get(&q->kobj);
589 EXPORT_SYMBOL(blk_get_queue);
591 static inline void blk_free_request(struct request_queue *q, struct request *rq)
593 if (rq->cmd_flags & REQ_ELVPRIV)
594 elv_put_request(q, rq);
595 mempool_free(rq, q->rq.rq_pool);
598 static struct request *
599 blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
601 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
607 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
608 * see bio.h and blkdev.h
610 rq->cmd_flags = rw | REQ_ALLOCED;
613 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
614 mempool_free(rq, q->rq.rq_pool);
617 rq->cmd_flags |= REQ_ELVPRIV;
624 * ioc_batching returns true if the ioc is a valid batching request and
625 * should be given priority access to a request.
627 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
633 * Make sure the process is able to allocate at least 1 request
634 * even if the batch times out, otherwise we could theoretically
637 return ioc->nr_batch_requests == q->nr_batching ||
638 (ioc->nr_batch_requests > 0
639 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
643 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
644 * will cause the process to be a "batcher" on all queues in the system. This
645 * is the behaviour we want though - once it gets a wakeup it should be given
648 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
650 if (!ioc || ioc_batching(q, ioc))
653 ioc->nr_batch_requests = q->nr_batching;
654 ioc->last_waited = jiffies;
657 static void __freed_request(struct request_queue *q, int rw)
659 struct request_list *rl = &q->rq;
661 if (rl->count[rw] < queue_congestion_off_threshold(q))
662 blk_clear_queue_congested(q, rw);
664 if (rl->count[rw] + 1 <= q->nr_requests) {
665 if (waitqueue_active(&rl->wait[rw]))
666 wake_up(&rl->wait[rw]);
668 blk_clear_queue_full(q, rw);
673 * A request has just been released. Account for it, update the full and
674 * congestion status, wake up any waiters. Called under q->queue_lock.
676 static void freed_request(struct request_queue *q, int rw, int priv)
678 struct request_list *rl = &q->rq;
684 __freed_request(q, rw);
686 if (unlikely(rl->starved[rw ^ 1]))
687 __freed_request(q, rw ^ 1);
690 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
692 * Get a free request, queue_lock must be held.
693 * Returns NULL on failure, with queue_lock held.
694 * Returns !NULL on success, with queue_lock *not held*.
696 static struct request *get_request(struct request_queue *q, int rw_flags,
697 struct bio *bio, gfp_t gfp_mask)
699 struct request *rq = NULL;
700 struct request_list *rl = &q->rq;
701 struct io_context *ioc = NULL;
702 const int rw = rw_flags & 0x01;
705 may_queue = elv_may_queue(q, rw_flags);
706 if (may_queue == ELV_MQUEUE_NO)
709 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
710 if (rl->count[rw]+1 >= q->nr_requests) {
711 ioc = current_io_context(GFP_ATOMIC, q->node);
713 * The queue will fill after this allocation, so set
714 * it as full, and mark this process as "batching".
715 * This process will be allowed to complete a batch of
716 * requests, others will be blocked.
718 if (!blk_queue_full(q, rw)) {
719 ioc_set_batching(q, ioc);
720 blk_set_queue_full(q, rw);
722 if (may_queue != ELV_MQUEUE_MUST
723 && !ioc_batching(q, ioc)) {
725 * The queue is full and the allocating
726 * process is not a "batcher", and not
727 * exempted by the IO scheduler
733 blk_set_queue_congested(q, rw);
737 * Only allow batching queuers to allocate up to 50% over the defined
738 * limit of requests, otherwise we could have thousands of requests
739 * allocated with any setting of ->nr_requests
741 if (rl->count[rw] >= (3 * q->nr_requests / 2))
747 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
751 spin_unlock_irq(q->queue_lock);
753 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
756 * Allocation failed presumably due to memory. Undo anything
757 * we might have messed up.
759 * Allocating task should really be put onto the front of the
760 * wait queue, but this is pretty rare.
762 spin_lock_irq(q->queue_lock);
763 freed_request(q, rw, priv);
766 * in the very unlikely event that allocation failed and no
767 * requests for this direction was pending, mark us starved
768 * so that freeing of a request in the other direction will
769 * notice us. another possible fix would be to split the
770 * rq mempool into READ and WRITE
773 if (unlikely(rl->count[rw] == 0))
780 * ioc may be NULL here, and ioc_batching will be false. That's
781 * OK, if the queue is under the request limit then requests need
782 * not count toward the nr_batch_requests limit. There will always
783 * be some limit enforced by BLK_BATCH_TIME.
785 if (ioc_batching(q, ioc))
786 ioc->nr_batch_requests--;
790 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
796 * No available requests for this queue, unplug the device and wait for some
797 * requests to become available.
799 * Called with q->queue_lock held, and returns with it unlocked.
801 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
804 const int rw = rw_flags & 0x01;
807 rq = get_request(q, rw_flags, bio, GFP_NOIO);
810 struct request_list *rl = &q->rq;
812 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
813 TASK_UNINTERRUPTIBLE);
815 rq = get_request(q, rw_flags, bio, GFP_NOIO);
818 struct io_context *ioc;
820 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
822 __generic_unplug_device(q);
823 spin_unlock_irq(q->queue_lock);
827 * After sleeping, we become a "batching" process and
828 * will be able to allocate at least one request, and
829 * up to a big batch of them for a small period time.
830 * See ioc_batching, ioc_set_batching
832 ioc = current_io_context(GFP_NOIO, q->node);
833 ioc_set_batching(q, ioc);
835 spin_lock_irq(q->queue_lock);
837 finish_wait(&rl->wait[rw], &wait);
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);
1001 * queue lock must be held
1003 void __blk_put_request(struct request_queue *q, struct request *req)
1007 if (unlikely(--req->ref_count))
1010 elv_completed_request(q, req);
1013 * Request may not have originated from ll_rw_blk. if not,
1014 * it didn't come out of our reserved rq pools
1016 if (req->cmd_flags & REQ_ALLOCED) {
1017 int rw = rq_data_dir(req);
1018 int priv = req->cmd_flags & REQ_ELVPRIV;
1020 BUG_ON(!list_empty(&req->queuelist));
1021 BUG_ON(!hlist_unhashed(&req->hash));
1023 blk_free_request(q, req);
1024 freed_request(q, rw, priv);
1027 EXPORT_SYMBOL_GPL(__blk_put_request);
1029 void blk_put_request(struct request *req)
1031 unsigned long flags;
1032 struct request_queue *q = req->q;
1035 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
1036 * following if (q) test.
1039 spin_lock_irqsave(q->queue_lock, flags);
1040 __blk_put_request(q, req);
1041 spin_unlock_irqrestore(q->queue_lock, flags);
1044 EXPORT_SYMBOL(blk_put_request);
1046 void init_request_from_bio(struct request *req, struct bio *bio)
1048 req->cmd_type = REQ_TYPE_FS;
1051 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1053 if (bio_rw_ahead(bio) || bio_failfast(bio))
1054 req->cmd_flags |= REQ_FAILFAST;
1057 * REQ_BARRIER implies no merging, but lets make it explicit
1059 if (unlikely(bio_barrier(bio)))
1060 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
1063 req->cmd_flags |= REQ_RW_SYNC;
1064 if (bio_rw_meta(bio))
1065 req->cmd_flags |= REQ_RW_META;
1068 req->hard_sector = req->sector = bio->bi_sector;
1069 req->ioprio = bio_prio(bio);
1070 req->start_time = jiffies;
1071 blk_rq_bio_prep(req->q, req, bio);
1074 static int __make_request(struct request_queue *q, struct bio *bio)
1076 struct request *req;
1077 int el_ret, nr_sectors, barrier, err;
1078 const unsigned short prio = bio_prio(bio);
1079 const int sync = bio_sync(bio);
1082 nr_sectors = bio_sectors(bio);
1085 * low level driver can indicate that it wants pages above a
1086 * certain limit bounced to low memory (ie for highmem, or even
1087 * ISA dma in theory)
1089 blk_queue_bounce(q, &bio);
1091 barrier = bio_barrier(bio);
1092 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
1097 spin_lock_irq(q->queue_lock);
1099 if (unlikely(barrier) || elv_queue_empty(q))
1102 el_ret = elv_merge(q, &req, bio);
1104 case ELEVATOR_BACK_MERGE:
1105 BUG_ON(!rq_mergeable(req));
1107 if (!ll_back_merge_fn(q, req, bio))
1110 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
1112 req->biotail->bi_next = bio;
1114 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1115 req->ioprio = ioprio_best(req->ioprio, prio);
1116 drive_stat_acct(req, 0);
1117 if (!attempt_back_merge(q, req))
1118 elv_merged_request(q, req, el_ret);
1121 case ELEVATOR_FRONT_MERGE:
1122 BUG_ON(!rq_mergeable(req));
1124 if (!ll_front_merge_fn(q, req, bio))
1127 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
1129 bio->bi_next = req->bio;
1133 * may not be valid. if the low level driver said
1134 * it didn't need a bounce buffer then it better
1135 * not touch req->buffer either...
1137 req->buffer = bio_data(bio);
1138 req->current_nr_sectors = bio_cur_sectors(bio);
1139 req->hard_cur_sectors = req->current_nr_sectors;
1140 req->sector = req->hard_sector = bio->bi_sector;
1141 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1142 req->ioprio = ioprio_best(req->ioprio, prio);
1143 drive_stat_acct(req, 0);
1144 if (!attempt_front_merge(q, req))
1145 elv_merged_request(q, req, el_ret);
1148 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1155 * This sync check and mask will be re-done in init_request_from_bio(),
1156 * but we need to set it earlier to expose the sync flag to the
1157 * rq allocator and io schedulers.
1159 rw_flags = bio_data_dir(bio);
1161 rw_flags |= REQ_RW_SYNC;
1164 * Grab a free request. This is might sleep but can not fail.
1165 * Returns with the queue unlocked.
1167 req = get_request_wait(q, rw_flags, bio);
1170 * After dropping the lock and possibly sleeping here, our request
1171 * may now be mergeable after it had proven unmergeable (above).
1172 * We don't worry about that case for efficiency. It won't happen
1173 * often, and the elevators are able to handle it.
1175 init_request_from_bio(req, bio);
1177 spin_lock_irq(q->queue_lock);
1178 if (elv_queue_empty(q))
1180 add_request(q, req);
1183 __generic_unplug_device(q);
1185 spin_unlock_irq(q->queue_lock);
1189 bio_endio(bio, err);
1194 * If bio->bi_dev is a partition, remap the location
1196 static inline void blk_partition_remap(struct bio *bio)
1198 struct block_device *bdev = bio->bi_bdev;
1200 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1201 struct hd_struct *p = bdev->bd_part;
1202 const int rw = bio_data_dir(bio);
1204 p->sectors[rw] += bio_sectors(bio);
1207 bio->bi_sector += p->start_sect;
1208 bio->bi_bdev = bdev->bd_contains;
1210 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1211 bdev->bd_dev, bio->bi_sector,
1212 bio->bi_sector - p->start_sect);
1216 static void handle_bad_sector(struct bio *bio)
1218 char b[BDEVNAME_SIZE];
1220 printk(KERN_INFO "attempt to access beyond end of device\n");
1221 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1222 bdevname(bio->bi_bdev, b),
1224 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1225 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1227 set_bit(BIO_EOF, &bio->bi_flags);
1230 #ifdef CONFIG_FAIL_MAKE_REQUEST
1232 static DECLARE_FAULT_ATTR(fail_make_request);
1234 static int __init setup_fail_make_request(char *str)
1236 return setup_fault_attr(&fail_make_request, str);
1238 __setup("fail_make_request=", setup_fail_make_request);
1240 static int should_fail_request(struct bio *bio)
1242 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
1243 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
1244 return should_fail(&fail_make_request, bio->bi_size);
1249 static int __init fail_make_request_debugfs(void)
1251 return init_fault_attr_dentries(&fail_make_request,
1252 "fail_make_request");
1255 late_initcall(fail_make_request_debugfs);
1257 #else /* CONFIG_FAIL_MAKE_REQUEST */
1259 static inline int should_fail_request(struct bio *bio)
1264 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1267 * Check whether this bio extends beyond the end of the device.
1269 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1276 /* Test device or partition size, when known. */
1277 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1279 sector_t sector = bio->bi_sector;
1281 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1283 * This may well happen - the kernel calls bread()
1284 * without checking the size of the device, e.g., when
1285 * mounting a device.
1287 handle_bad_sector(bio);
1296 * generic_make_request: hand a buffer to its device driver for I/O
1297 * @bio: The bio describing the location in memory and on the device.
1299 * generic_make_request() is used to make I/O requests of block
1300 * devices. It is passed a &struct bio, which describes the I/O that needs
1303 * generic_make_request() does not return any status. The
1304 * success/failure status of the request, along with notification of
1305 * completion, is delivered asynchronously through the bio->bi_end_io
1306 * function described (one day) else where.
1308 * The caller of generic_make_request must make sure that bi_io_vec
1309 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1310 * set to describe the device address, and the
1311 * bi_end_io and optionally bi_private are set to describe how
1312 * completion notification should be signaled.
1314 * generic_make_request and the drivers it calls may use bi_next if this
1315 * bio happens to be merged with someone else, and may change bi_dev and
1316 * bi_sector for remaps as it sees fit. So the values of these fields
1317 * should NOT be depended on after the call to generic_make_request.
1319 static inline void __generic_make_request(struct bio *bio)
1321 struct request_queue *q;
1322 sector_t old_sector;
1323 int ret, nr_sectors = bio_sectors(bio);
1329 if (bio_check_eod(bio, nr_sectors))
1333 * Resolve the mapping until finished. (drivers are
1334 * still free to implement/resolve their own stacking
1335 * by explicitly returning 0)
1337 * NOTE: we don't repeat the blk_size check for each new device.
1338 * Stacking drivers are expected to know what they are doing.
1343 char b[BDEVNAME_SIZE];
1345 q = bdev_get_queue(bio->bi_bdev);
1348 "generic_make_request: Trying to access "
1349 "nonexistent block-device %s (%Lu)\n",
1350 bdevname(bio->bi_bdev, b),
1351 (long long) bio->bi_sector);
1353 bio_endio(bio, err);
1357 if (unlikely(nr_sectors > q->max_hw_sectors)) {
1358 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1359 bdevname(bio->bi_bdev, b),
1365 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1368 if (should_fail_request(bio))
1372 * If this device has partitions, remap block n
1373 * of partition p to block n+start(p) of the disk.
1375 blk_partition_remap(bio);
1377 if (old_sector != -1)
1378 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
1381 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1383 old_sector = bio->bi_sector;
1384 old_dev = bio->bi_bdev->bd_dev;
1386 if (bio_check_eod(bio, nr_sectors))
1388 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
1393 ret = q->make_request_fn(q, bio);
1398 * We only want one ->make_request_fn to be active at a time,
1399 * else stack usage with stacked devices could be a problem.
1400 * So use current->bio_{list,tail} to keep a list of requests
1401 * submited by a make_request_fn function.
1402 * current->bio_tail is also used as a flag to say if
1403 * generic_make_request is currently active in this task or not.
1404 * If it is NULL, then no make_request is active. If it is non-NULL,
1405 * then a make_request is active, and new requests should be added
1408 void generic_make_request(struct bio *bio)
1410 if (current->bio_tail) {
1411 /* make_request is active */
1412 *(current->bio_tail) = bio;
1413 bio->bi_next = NULL;
1414 current->bio_tail = &bio->bi_next;
1417 /* following loop may be a bit non-obvious, and so deserves some
1419 * Before entering the loop, bio->bi_next is NULL (as all callers
1420 * ensure that) so we have a list with a single bio.
1421 * We pretend that we have just taken it off a longer list, so
1422 * we assign bio_list to the next (which is NULL) and bio_tail
1423 * to &bio_list, thus initialising the bio_list of new bios to be
1424 * added. __generic_make_request may indeed add some more bios
1425 * through a recursive call to generic_make_request. If it
1426 * did, we find a non-NULL value in bio_list and re-enter the loop
1427 * from the top. In this case we really did just take the bio
1428 * of the top of the list (no pretending) and so fixup bio_list and
1429 * bio_tail or bi_next, and call into __generic_make_request again.
1431 * The loop was structured like this to make only one call to
1432 * __generic_make_request (which is important as it is large and
1433 * inlined) and to keep the structure simple.
1435 BUG_ON(bio->bi_next);
1437 current->bio_list = bio->bi_next;
1438 if (bio->bi_next == NULL)
1439 current->bio_tail = ¤t->bio_list;
1441 bio->bi_next = NULL;
1442 __generic_make_request(bio);
1443 bio = current->bio_list;
1445 current->bio_tail = NULL; /* deactivate */
1447 EXPORT_SYMBOL(generic_make_request);
1450 * submit_bio: submit a bio to the block device layer for I/O
1451 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1452 * @bio: The &struct bio which describes the I/O
1454 * submit_bio() is very similar in purpose to generic_make_request(), and
1455 * uses that function to do most of the work. Both are fairly rough
1456 * interfaces, @bio must be presetup and ready for I/O.
1459 void submit_bio(int rw, struct bio *bio)
1461 int count = bio_sectors(bio);
1466 * If it's a regular read/write or a barrier with data attached,
1467 * go through the normal accounting stuff before submission.
1469 if (!bio_empty_barrier(bio)) {
1471 BIO_BUG_ON(!bio->bi_size);
1472 BIO_BUG_ON(!bio->bi_io_vec);
1475 count_vm_events(PGPGOUT, count);
1477 task_io_account_read(bio->bi_size);
1478 count_vm_events(PGPGIN, count);
1481 if (unlikely(block_dump)) {
1482 char b[BDEVNAME_SIZE];
1483 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1484 current->comm, task_pid_nr(current),
1485 (rw & WRITE) ? "WRITE" : "READ",
1486 (unsigned long long)bio->bi_sector,
1487 bdevname(bio->bi_bdev, b));
1491 generic_make_request(bio);
1493 EXPORT_SYMBOL(submit_bio);
1496 * __end_that_request_first - end I/O on a request
1497 * @req: the request being processed
1498 * @error: 0 for success, < 0 for error
1499 * @nr_bytes: number of bytes to complete
1502 * Ends I/O on a number of bytes attached to @req, and sets it up
1503 * for the next range of segments (if any) in the cluster.
1506 * 0 - we are done with this request, call end_that_request_last()
1507 * 1 - still buffers pending for this request
1509 static int __end_that_request_first(struct request *req, int error,
1512 int total_bytes, bio_nbytes, next_idx = 0;
1515 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1518 * for a REQ_BLOCK_PC request, we want to carry any eventual
1519 * sense key with us all the way through
1521 if (!blk_pc_request(req))
1524 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1525 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1526 req->rq_disk ? req->rq_disk->disk_name : "?",
1527 (unsigned long long)req->sector);
1530 if (blk_fs_request(req) && req->rq_disk) {
1531 const int rw = rq_data_dir(req);
1533 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
1536 total_bytes = bio_nbytes = 0;
1537 while ((bio = req->bio) != NULL) {
1541 * For an empty barrier request, the low level driver must
1542 * store a potential error location in ->sector. We pass
1543 * that back up in ->bi_sector.
1545 if (blk_empty_barrier(req))
1546 bio->bi_sector = req->sector;
1548 if (nr_bytes >= bio->bi_size) {
1549 req->bio = bio->bi_next;
1550 nbytes = bio->bi_size;
1551 req_bio_endio(req, bio, nbytes, error);
1555 int idx = bio->bi_idx + next_idx;
1557 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1558 blk_dump_rq_flags(req, "__end_that");
1559 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1560 __FUNCTION__, bio->bi_idx,
1565 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1566 BIO_BUG_ON(nbytes > bio->bi_size);
1569 * not a complete bvec done
1571 if (unlikely(nbytes > nr_bytes)) {
1572 bio_nbytes += nr_bytes;
1573 total_bytes += nr_bytes;
1578 * advance to the next vector
1581 bio_nbytes += nbytes;
1584 total_bytes += nbytes;
1590 * end more in this run, or just return 'not-done'
1592 if (unlikely(nr_bytes <= 0))
1604 * if the request wasn't completed, update state
1607 req_bio_endio(req, bio, bio_nbytes, error);
1608 bio->bi_idx += next_idx;
1609 bio_iovec(bio)->bv_offset += nr_bytes;
1610 bio_iovec(bio)->bv_len -= nr_bytes;
1613 blk_recalc_rq_sectors(req, total_bytes >> 9);
1614 blk_recalc_rq_segments(req);
1619 * splice the completion data to a local structure and hand off to
1620 * process_completion_queue() to complete the requests
1622 static void blk_done_softirq(struct softirq_action *h)
1624 struct list_head *cpu_list, local_list;
1626 local_irq_disable();
1627 cpu_list = &__get_cpu_var(blk_cpu_done);
1628 list_replace_init(cpu_list, &local_list);
1631 while (!list_empty(&local_list)) {
1634 rq = list_entry(local_list.next, struct request, donelist);
1635 list_del_init(&rq->donelist);
1636 rq->q->softirq_done_fn(rq);
1640 static int __cpuinit blk_cpu_notify(struct notifier_block *self,
1641 unsigned long action, void *hcpu)
1644 * If a CPU goes away, splice its entries to the current CPU
1645 * and trigger a run of the softirq
1647 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1648 int cpu = (unsigned long) hcpu;
1650 local_irq_disable();
1651 list_splice_init(&per_cpu(blk_cpu_done, cpu),
1652 &__get_cpu_var(blk_cpu_done));
1653 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1661 static struct notifier_block blk_cpu_notifier __cpuinitdata = {
1662 .notifier_call = blk_cpu_notify,
1666 * blk_complete_request - end I/O on a request
1667 * @req: the request being processed
1670 * Ends all I/O on a request. It does not handle partial completions,
1671 * unless the driver actually implements this in its completion callback
1672 * through requeueing. The actual completion happens out-of-order,
1673 * through a softirq handler. The user must have registered a completion
1674 * callback through blk_queue_softirq_done().
1677 void blk_complete_request(struct request *req)
1679 struct list_head *cpu_list;
1680 unsigned long flags;
1682 BUG_ON(!req->q->softirq_done_fn);
1684 local_irq_save(flags);
1686 cpu_list = &__get_cpu_var(blk_cpu_done);
1687 list_add_tail(&req->donelist, cpu_list);
1688 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1690 local_irq_restore(flags);
1692 EXPORT_SYMBOL(blk_complete_request);
1695 * queue lock must be held
1697 static void end_that_request_last(struct request *req, int error)
1699 struct gendisk *disk = req->rq_disk;
1701 if (blk_rq_tagged(req))
1702 blk_queue_end_tag(req->q, req);
1704 if (blk_queued_rq(req))
1705 blkdev_dequeue_request(req);
1707 if (unlikely(laptop_mode) && blk_fs_request(req))
1708 laptop_io_completion();
1711 * Account IO completion. bar_rq isn't accounted as a normal
1712 * IO on queueing nor completion. Accounting the containing
1713 * request is enough.
1715 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1716 unsigned long duration = jiffies - req->start_time;
1717 const int rw = rq_data_dir(req);
1719 __disk_stat_inc(disk, ios[rw]);
1720 __disk_stat_add(disk, ticks[rw], duration);
1721 disk_round_stats(disk);
1726 req->end_io(req, error);
1728 if (blk_bidi_rq(req))
1729 __blk_put_request(req->next_rq->q, req->next_rq);
1731 __blk_put_request(req->q, req);
1735 static inline void __end_request(struct request *rq, int uptodate,
1736 unsigned int nr_bytes)
1741 error = uptodate ? uptodate : -EIO;
1743 __blk_end_request(rq, error, nr_bytes);
1747 * blk_rq_bytes - Returns bytes left to complete in the entire request
1749 unsigned int blk_rq_bytes(struct request *rq)
1751 if (blk_fs_request(rq))
1752 return rq->hard_nr_sectors << 9;
1754 return rq->data_len;
1756 EXPORT_SYMBOL_GPL(blk_rq_bytes);
1759 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1761 unsigned int blk_rq_cur_bytes(struct request *rq)
1763 if (blk_fs_request(rq))
1764 return rq->current_nr_sectors << 9;
1767 return rq->bio->bi_size;
1769 return rq->data_len;
1771 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1774 * end_queued_request - end all I/O on a queued request
1775 * @rq: the request being processed
1776 * @uptodate: error value or 0/1 uptodate flag
1779 * Ends all I/O on a request, and removes it from the block layer queues.
1780 * Not suitable for normal IO completion, unless the driver still has
1781 * the request attached to the block layer.
1784 void end_queued_request(struct request *rq, int uptodate)
1786 __end_request(rq, uptodate, blk_rq_bytes(rq));
1788 EXPORT_SYMBOL(end_queued_request);
1791 * end_dequeued_request - end all I/O on a dequeued request
1792 * @rq: the request being processed
1793 * @uptodate: error value or 0/1 uptodate flag
1796 * Ends all I/O on a request. The request must already have been
1797 * dequeued using blkdev_dequeue_request(), as is normally the case
1801 void end_dequeued_request(struct request *rq, int uptodate)
1803 __end_request(rq, uptodate, blk_rq_bytes(rq));
1805 EXPORT_SYMBOL(end_dequeued_request);
1809 * end_request - end I/O on the current segment of the request
1810 * @req: the request being processed
1811 * @uptodate: error value or 0/1 uptodate flag
1814 * Ends I/O on the current segment of a request. If that is the only
1815 * remaining segment, the request is also completed and freed.
1817 * This is a remnant of how older block drivers handled IO completions.
1818 * Modern drivers typically end IO on the full request in one go, unless
1819 * they have a residual value to account for. For that case this function
1820 * isn't really useful, unless the residual just happens to be the
1821 * full current segment. In other words, don't use this function in new
1822 * code. Either use end_request_completely(), or the
1823 * end_that_request_chunk() (along with end_that_request_last()) for
1824 * partial completions.
1827 void end_request(struct request *req, int uptodate)
1829 __end_request(req, uptodate, req->hard_cur_sectors << 9);
1831 EXPORT_SYMBOL(end_request);
1834 * blk_end_io - Generic end_io function to complete a request.
1835 * @rq: the request being processed
1836 * @error: 0 for success, < 0 for error
1837 * @nr_bytes: number of bytes to complete @rq
1838 * @bidi_bytes: number of bytes to complete @rq->next_rq
1839 * @drv_callback: function called between completion of bios in the request
1840 * and completion of the request.
1841 * If the callback returns non 0, this helper returns without
1842 * completion of the request.
1845 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1846 * If @rq has leftover, sets it up for the next range of segments.
1849 * 0 - we are done with this request
1850 * 1 - this request is not freed yet, it still has pending buffers.
1852 static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1853 unsigned int bidi_bytes,
1854 int (drv_callback)(struct request *))
1856 struct request_queue *q = rq->q;
1857 unsigned long flags = 0UL;
1859 if (blk_fs_request(rq) || blk_pc_request(rq)) {
1860 if (__end_that_request_first(rq, error, nr_bytes))
1863 /* Bidi request must be completed as a whole */
1864 if (blk_bidi_rq(rq) &&
1865 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1869 /* Special feature for tricky drivers */
1870 if (drv_callback && drv_callback(rq))
1873 add_disk_randomness(rq->rq_disk);
1875 spin_lock_irqsave(q->queue_lock, flags);
1876 end_that_request_last(rq, error);
1877 spin_unlock_irqrestore(q->queue_lock, flags);
1883 * blk_end_request - Helper function for drivers to complete the request.
1884 * @rq: the request being processed
1885 * @error: 0 for success, < 0 for error
1886 * @nr_bytes: number of bytes to complete
1889 * Ends I/O on a number of bytes attached to @rq.
1890 * If @rq has leftover, sets it up for the next range of segments.
1893 * 0 - we are done with this request
1894 * 1 - still buffers pending for this request
1896 int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1898 return blk_end_io(rq, error, nr_bytes, 0, NULL);
1900 EXPORT_SYMBOL_GPL(blk_end_request);
1903 * __blk_end_request - Helper function for drivers to complete the request.
1904 * @rq: the request being processed
1905 * @error: 0 for success, < 0 for error
1906 * @nr_bytes: number of bytes to complete
1909 * Must be called with queue lock held unlike blk_end_request().
1912 * 0 - we are done with this request
1913 * 1 - still buffers pending for this request
1915 int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1917 if (blk_fs_request(rq) || blk_pc_request(rq)) {
1918 if (__end_that_request_first(rq, error, nr_bytes))
1922 add_disk_randomness(rq->rq_disk);
1924 end_that_request_last(rq, error);
1928 EXPORT_SYMBOL_GPL(__blk_end_request);
1931 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1932 * @rq: the bidi request being processed
1933 * @error: 0 for success, < 0 for error
1934 * @nr_bytes: number of bytes to complete @rq
1935 * @bidi_bytes: number of bytes to complete @rq->next_rq
1938 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1941 * 0 - we are done with this request
1942 * 1 - still buffers pending for this request
1944 int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1945 unsigned int bidi_bytes)
1947 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1949 EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1952 * blk_end_request_callback - Special helper function for tricky drivers
1953 * @rq: the request being processed
1954 * @error: 0 for success, < 0 for error
1955 * @nr_bytes: number of bytes to complete
1956 * @drv_callback: function called between completion of bios in the request
1957 * and completion of the request.
1958 * If the callback returns non 0, this helper returns without
1959 * completion of the request.
1962 * Ends I/O on a number of bytes attached to @rq.
1963 * If @rq has leftover, sets it up for the next range of segments.
1965 * This special helper function is used only for existing tricky drivers.
1966 * (e.g. cdrom_newpc_intr() of ide-cd)
1967 * This interface will be removed when such drivers are rewritten.
1968 * Don't use this interface in other places anymore.
1971 * 0 - we are done with this request
1972 * 1 - this request is not freed yet.
1973 * this request still has pending buffers or
1974 * the driver doesn't want to finish this request yet.
1976 int blk_end_request_callback(struct request *rq, int error,
1977 unsigned int nr_bytes,
1978 int (drv_callback)(struct request *))
1980 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
1982 EXPORT_SYMBOL_GPL(blk_end_request_callback);
1984 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1987 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
1988 rq->cmd_flags |= (bio->bi_rw & 3);
1990 rq->nr_phys_segments = bio_phys_segments(q, bio);
1991 rq->nr_hw_segments = bio_hw_segments(q, bio);
1992 rq->current_nr_sectors = bio_cur_sectors(bio);
1993 rq->hard_cur_sectors = rq->current_nr_sectors;
1994 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
1995 rq->buffer = bio_data(bio);
1996 rq->data_len = bio->bi_size;
1998 rq->bio = rq->biotail = bio;
2001 rq->rq_disk = bio->bi_bdev->bd_disk;
2004 int kblockd_schedule_work(struct work_struct *work)
2006 return queue_work(kblockd_workqueue, work);
2008 EXPORT_SYMBOL(kblockd_schedule_work);
2010 void kblockd_flush_work(struct work_struct *work)
2012 cancel_work_sync(work);
2014 EXPORT_SYMBOL(kblockd_flush_work);
2016 int __init blk_dev_init(void)
2020 kblockd_workqueue = create_workqueue("kblockd");
2021 if (!kblockd_workqueue)
2022 panic("Failed to create kblockd\n");
2024 request_cachep = kmem_cache_create("blkdev_requests",
2025 sizeof(struct request), 0, SLAB_PANIC, NULL);
2027 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2028 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2030 for_each_possible_cpu(i)
2031 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2033 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
2034 register_hotcpu_notifier(&blk_cpu_notifier);