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 (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
209 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
210 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
213 EXPORT_SYMBOL(blk_plug_device);
216 * remove the queue from the plugged list, if present. called with
217 * queue lock held and interrupts disabled.
219 int blk_remove_plug(struct request_queue *q)
221 WARN_ON(!irqs_disabled());
223 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
226 del_timer(&q->unplug_timer);
229 EXPORT_SYMBOL(blk_remove_plug);
232 * remove the plug and let it rip..
234 void __generic_unplug_device(struct request_queue *q)
236 if (unlikely(blk_queue_stopped(q)))
239 if (!blk_remove_plug(q))
244 EXPORT_SYMBOL(__generic_unplug_device);
247 * generic_unplug_device - fire a request queue
248 * @q: The &struct request_queue in question
251 * Linux uses plugging to build bigger requests queues before letting
252 * the device have at them. If a queue is plugged, the I/O scheduler
253 * is still adding and merging requests on the queue. Once the queue
254 * gets unplugged, the request_fn defined for the queue is invoked and
257 void generic_unplug_device(struct request_queue *q)
259 if (blk_queue_plugged(q)) {
260 spin_lock_irq(q->queue_lock);
261 __generic_unplug_device(q);
262 spin_unlock_irq(q->queue_lock);
265 EXPORT_SYMBOL(generic_unplug_device);
267 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
270 struct request_queue *q = bdi->unplug_io_data;
275 void blk_unplug_work(struct work_struct *work)
277 struct request_queue *q =
278 container_of(work, struct request_queue, unplug_work);
280 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
281 q->rq.count[READ] + q->rq.count[WRITE]);
286 void blk_unplug_timeout(unsigned long data)
288 struct request_queue *q = (struct request_queue *)data;
290 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
291 q->rq.count[READ] + q->rq.count[WRITE]);
293 kblockd_schedule_work(&q->unplug_work);
296 void blk_unplug(struct request_queue *q)
299 * devices don't necessarily have an ->unplug_fn defined
302 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
303 q->rq.count[READ] + q->rq.count[WRITE]);
308 EXPORT_SYMBOL(blk_unplug);
311 * blk_start_queue - restart a previously stopped queue
312 * @q: The &struct request_queue in question
315 * blk_start_queue() will clear the stop flag on the queue, and call
316 * the request_fn for the queue if it was in a stopped state when
317 * entered. Also see blk_stop_queue(). Queue lock must be held.
319 void blk_start_queue(struct request_queue *q)
321 WARN_ON(!irqs_disabled());
323 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
326 * one level of recursion is ok and is much faster than kicking
327 * the unplug handling
329 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
331 queue_flag_clear(QUEUE_FLAG_REENTER, q);
334 kblockd_schedule_work(&q->unplug_work);
337 EXPORT_SYMBOL(blk_start_queue);
340 * blk_stop_queue - stop a queue
341 * @q: The &struct request_queue in question
344 * The Linux block layer assumes that a block driver will consume all
345 * entries on the request queue when the request_fn strategy is called.
346 * Often this will not happen, because of hardware limitations (queue
347 * depth settings). If a device driver gets a 'queue full' response,
348 * or if it simply chooses not to queue more I/O at one point, it can
349 * call this function to prevent the request_fn from being called until
350 * the driver has signalled it's ready to go again. This happens by calling
351 * blk_start_queue() to restart queue operations. Queue lock must be held.
353 void blk_stop_queue(struct request_queue *q)
356 queue_flag_set(QUEUE_FLAG_STOPPED, q);
358 EXPORT_SYMBOL(blk_stop_queue);
361 * blk_sync_queue - cancel any pending callbacks on a queue
365 * The block layer may perform asynchronous callback activity
366 * on a queue, such as calling the unplug function after a timeout.
367 * A block device may call blk_sync_queue to ensure that any
368 * such activity is cancelled, thus allowing it to release resources
369 * that the callbacks might use. The caller must already have made sure
370 * that its ->make_request_fn will not re-add plugging prior to calling
374 void blk_sync_queue(struct request_queue *q)
376 del_timer_sync(&q->unplug_timer);
377 kblockd_flush_work(&q->unplug_work);
379 EXPORT_SYMBOL(blk_sync_queue);
382 * blk_run_queue - run a single device queue
383 * @q: The queue to run
385 void __blk_run_queue(struct request_queue *q)
390 * Only recurse once to avoid overrunning the stack, let the unplug
391 * handling reinvoke the handler shortly if we already got there.
393 if (!elv_queue_empty(q)) {
394 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
396 queue_flag_clear(QUEUE_FLAG_REENTER, q);
399 kblockd_schedule_work(&q->unplug_work);
403 EXPORT_SYMBOL(__blk_run_queue);
406 * blk_run_queue - run a single device queue
407 * @q: The queue to run
409 void blk_run_queue(struct request_queue *q)
413 spin_lock_irqsave(q->queue_lock, flags);
415 spin_unlock_irqrestore(q->queue_lock, flags);
417 EXPORT_SYMBOL(blk_run_queue);
419 void blk_put_queue(struct request_queue *q)
421 kobject_put(&q->kobj);
424 void blk_cleanup_queue(struct request_queue *q)
426 mutex_lock(&q->sysfs_lock);
427 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
428 mutex_unlock(&q->sysfs_lock);
431 elevator_exit(q->elevator);
435 EXPORT_SYMBOL(blk_cleanup_queue);
437 static int blk_init_free_list(struct request_queue *q)
439 struct request_list *rl = &q->rq;
441 rl->count[READ] = rl->count[WRITE] = 0;
442 rl->starved[READ] = rl->starved[WRITE] = 0;
444 init_waitqueue_head(&rl->wait[READ]);
445 init_waitqueue_head(&rl->wait[WRITE]);
447 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
448 mempool_free_slab, request_cachep, q->node);
456 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
458 return blk_alloc_queue_node(gfp_mask, -1);
460 EXPORT_SYMBOL(blk_alloc_queue);
462 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
464 struct request_queue *q;
467 q = kmem_cache_alloc_node(blk_requestq_cachep,
468 gfp_mask | __GFP_ZERO, node_id);
472 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
473 q->backing_dev_info.unplug_io_data = q;
474 err = bdi_init(&q->backing_dev_info);
476 kmem_cache_free(blk_requestq_cachep, q);
480 init_timer(&q->unplug_timer);
482 kobject_init(&q->kobj, &blk_queue_ktype);
484 mutex_init(&q->sysfs_lock);
485 spin_lock_init(&q->__queue_lock);
489 EXPORT_SYMBOL(blk_alloc_queue_node);
492 * blk_init_queue - prepare a request queue for use with a block device
493 * @rfn: The function to be called to process requests that have been
494 * placed on the queue.
495 * @lock: Request queue spin lock
498 * If a block device wishes to use the standard request handling procedures,
499 * which sorts requests and coalesces adjacent requests, then it must
500 * call blk_init_queue(). The function @rfn will be called when there
501 * are requests on the queue that need to be processed. If the device
502 * supports plugging, then @rfn may not be called immediately when requests
503 * are available on the queue, but may be called at some time later instead.
504 * Plugged queues are generally unplugged when a buffer belonging to one
505 * of the requests on the queue is needed, or due to memory pressure.
507 * @rfn is not required, or even expected, to remove all requests off the
508 * queue, but only as many as it can handle at a time. If it does leave
509 * requests on the queue, it is responsible for arranging that the requests
510 * get dealt with eventually.
512 * The queue spin lock must be held while manipulating the requests on the
513 * request queue; this lock will be taken also from interrupt context, so irq
514 * disabling is needed for it.
516 * Function returns a pointer to the initialized request queue, or NULL if
520 * blk_init_queue() must be paired with a blk_cleanup_queue() call
521 * when the block device is deactivated (such as at module unload).
524 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
526 return blk_init_queue_node(rfn, lock, -1);
528 EXPORT_SYMBOL(blk_init_queue);
530 struct request_queue *
531 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
533 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
539 if (blk_init_free_list(q)) {
540 kmem_cache_free(blk_requestq_cachep, q);
545 * if caller didn't supply a lock, they get per-queue locking with
549 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);
590 static inline void blk_free_request(struct request_queue *q, struct request *rq)
592 if (rq->cmd_flags & REQ_ELVPRIV)
593 elv_put_request(q, rq);
594 mempool_free(rq, q->rq.rq_pool);
597 static struct request *
598 blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
600 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
608 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
609 * see bio.h and blkdev.h
611 rq->cmd_flags = rw | REQ_ALLOCED;
614 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
615 mempool_free(rq, q->rq.rq_pool);
618 rq->cmd_flags |= REQ_ELVPRIV;
625 * ioc_batching returns true if the ioc is a valid batching request and
626 * should be given priority access to a request.
628 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
634 * Make sure the process is able to allocate at least 1 request
635 * even if the batch times out, otherwise we could theoretically
638 return ioc->nr_batch_requests == q->nr_batching ||
639 (ioc->nr_batch_requests > 0
640 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
644 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
645 * will cause the process to be a "batcher" on all queues in the system. This
646 * is the behaviour we want though - once it gets a wakeup it should be given
649 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
651 if (!ioc || ioc_batching(q, ioc))
654 ioc->nr_batch_requests = q->nr_batching;
655 ioc->last_waited = jiffies;
658 static void __freed_request(struct request_queue *q, int rw)
660 struct request_list *rl = &q->rq;
662 if (rl->count[rw] < queue_congestion_off_threshold(q))
663 blk_clear_queue_congested(q, rw);
665 if (rl->count[rw] + 1 <= q->nr_requests) {
666 if (waitqueue_active(&rl->wait[rw]))
667 wake_up(&rl->wait[rw]);
669 blk_clear_queue_full(q, rw);
674 * A request has just been released. Account for it, update the full and
675 * congestion status, wake up any waiters. Called under q->queue_lock.
677 static void freed_request(struct request_queue *q, int rw, int priv)
679 struct request_list *rl = &q->rq;
685 __freed_request(q, rw);
687 if (unlikely(rl->starved[rw ^ 1]))
688 __freed_request(q, rw ^ 1);
691 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
693 * Get a free request, queue_lock must be held.
694 * Returns NULL on failure, with queue_lock held.
695 * Returns !NULL on success, with queue_lock *not held*.
697 static struct request *get_request(struct request_queue *q, int rw_flags,
698 struct bio *bio, gfp_t gfp_mask)
700 struct request *rq = NULL;
701 struct request_list *rl = &q->rq;
702 struct io_context *ioc = NULL;
703 const int rw = rw_flags & 0x01;
706 may_queue = elv_may_queue(q, rw_flags);
707 if (may_queue == ELV_MQUEUE_NO)
710 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
711 if (rl->count[rw]+1 >= q->nr_requests) {
712 ioc = current_io_context(GFP_ATOMIC, q->node);
714 * The queue will fill after this allocation, so set
715 * it as full, and mark this process as "batching".
716 * This process will be allowed to complete a batch of
717 * requests, others will be blocked.
719 if (!blk_queue_full(q, rw)) {
720 ioc_set_batching(q, ioc);
721 blk_set_queue_full(q, rw);
723 if (may_queue != ELV_MQUEUE_MUST
724 && !ioc_batching(q, ioc)) {
726 * The queue is full and the allocating
727 * process is not a "batcher", and not
728 * exempted by the IO scheduler
734 blk_set_queue_congested(q, rw);
738 * Only allow batching queuers to allocate up to 50% over the defined
739 * limit of requests, otherwise we could have thousands of requests
740 * allocated with any setting of ->nr_requests
742 if (rl->count[rw] >= (3 * q->nr_requests / 2))
748 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
752 spin_unlock_irq(q->queue_lock);
754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
763 spin_lock_irq(q->queue_lock);
764 freed_request(q, rw, priv);
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
774 if (unlikely(rl->count[rw] == 0))
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
789 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
795 * No available requests for this queue, unplug the device and wait for some
796 * requests to become available.
798 * Called with q->queue_lock held, and returns with it unlocked.
800 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
803 const int rw = rw_flags & 0x01;
806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
809 struct io_context *ioc;
810 struct request_list *rl = &q->rq;
812 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
813 TASK_UNINTERRUPTIBLE);
815 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
817 __generic_unplug_device(q);
818 spin_unlock_irq(q->queue_lock);
822 * After sleeping, we become a "batching" process and
823 * will be able to allocate at least one request, and
824 * up to a big batch of them for a small period time.
825 * See ioc_batching, ioc_set_batching
827 ioc = current_io_context(GFP_NOIO, q->node);
828 ioc_set_batching(q, ioc);
830 spin_lock_irq(q->queue_lock);
831 finish_wait(&rl->wait[rw], &wait);
833 rq = get_request(q, rw_flags, bio, GFP_NOIO);
839 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
843 BUG_ON(rw != READ && rw != WRITE);
845 spin_lock_irq(q->queue_lock);
846 if (gfp_mask & __GFP_WAIT) {
847 rq = get_request_wait(q, rw, NULL);
849 rq = get_request(q, rw, NULL, gfp_mask);
851 spin_unlock_irq(q->queue_lock);
853 /* q->queue_lock is unlocked at this point */
857 EXPORT_SYMBOL(blk_get_request);
860 * blk_start_queueing - initiate dispatch of requests to device
861 * @q: request queue to kick into gear
863 * This is basically a helper to remove the need to know whether a queue
864 * is plugged or not if someone just wants to initiate dispatch of requests
867 * The queue lock must be held with interrupts disabled.
869 void blk_start_queueing(struct request_queue *q)
871 if (!blk_queue_plugged(q))
874 __generic_unplug_device(q);
876 EXPORT_SYMBOL(blk_start_queueing);
879 * blk_requeue_request - put a request back on queue
880 * @q: request queue where request should be inserted
881 * @rq: request to be inserted
884 * Drivers often keep queueing requests until the hardware cannot accept
885 * more, when that condition happens we need to put the request back
886 * on the queue. Must be called with queue lock held.
888 void blk_requeue_request(struct request_queue *q, struct request *rq)
890 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
892 if (blk_rq_tagged(rq))
893 blk_queue_end_tag(q, rq);
895 elv_requeue_request(q, rq);
897 EXPORT_SYMBOL(blk_requeue_request);
900 * blk_insert_request - insert a special request in to a request queue
901 * @q: request queue where request should be inserted
902 * @rq: request to be inserted
903 * @at_head: insert request at head or tail of queue
904 * @data: private data
907 * Many block devices need to execute commands asynchronously, so they don't
908 * block the whole kernel from preemption during request execution. This is
909 * accomplished normally by inserting aritficial requests tagged as
910 * REQ_SPECIAL in to the corresponding request queue, and letting them be
911 * scheduled for actual execution by the request queue.
913 * We have the option of inserting the head or the tail of the queue.
914 * Typically we use the tail for new ioctls and so forth. We use the head
915 * of the queue for things like a QUEUE_FULL message from a device, or a
916 * host that is unable to accept a particular command.
918 void blk_insert_request(struct request_queue *q, struct request *rq,
919 int at_head, void *data)
921 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
925 * tell I/O scheduler that this isn't a regular read/write (ie it
926 * must not attempt merges on this) and that it acts as a soft
929 rq->cmd_type = REQ_TYPE_SPECIAL;
930 rq->cmd_flags |= REQ_SOFTBARRIER;
934 spin_lock_irqsave(q->queue_lock, flags);
937 * If command is tagged, release the tag
939 if (blk_rq_tagged(rq))
940 blk_queue_end_tag(q, rq);
942 drive_stat_acct(rq, 1);
943 __elv_add_request(q, rq, where, 0);
944 blk_start_queueing(q);
945 spin_unlock_irqrestore(q->queue_lock, flags);
947 EXPORT_SYMBOL(blk_insert_request);
950 * add-request adds a request to the linked list.
951 * queue lock is held and interrupts disabled, as we muck with the
952 * request queue list.
954 static inline void add_request(struct request_queue *q, struct request *req)
956 drive_stat_acct(req, 1);
959 * elevator indicated where it wants this request to be
960 * inserted at elevator_merge time
962 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
966 * disk_round_stats() - Round off the performance stats on a struct
969 * The average IO queue length and utilisation statistics are maintained
970 * by observing the current state of the queue length and the amount of
971 * time it has been in this state for.
973 * Normally, that accounting is done on IO completion, but that can result
974 * in more than a second's worth of IO being accounted for within any one
975 * second, leading to >100% utilisation. To deal with that, we call this
976 * function to do a round-off before returning the results when reading
977 * /proc/diskstats. This accounts immediately for all queue usage up to
978 * the current jiffies and restarts the counters again.
980 void disk_round_stats(struct gendisk *disk)
982 unsigned long now = jiffies;
984 if (now == disk->stamp)
987 if (disk->in_flight) {
988 __disk_stat_add(disk, time_in_queue,
989 disk->in_flight * (now - disk->stamp));
990 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
994 EXPORT_SYMBOL_GPL(disk_round_stats);
996 void part_round_stats(struct hd_struct *part)
998 unsigned long now = jiffies;
1000 if (now == part->stamp)
1003 if (part->in_flight) {
1004 __part_stat_add(part, time_in_queue,
1005 part->in_flight * (now - part->stamp));
1006 __part_stat_add(part, io_ticks, (now - part->stamp));
1012 * queue lock must be held
1014 void __blk_put_request(struct request_queue *q, struct request *req)
1018 if (unlikely(--req->ref_count))
1021 elv_completed_request(q, req);
1024 * Request may not have originated from ll_rw_blk. if not,
1025 * it didn't come out of our reserved rq pools
1027 if (req->cmd_flags & REQ_ALLOCED) {
1028 int rw = rq_data_dir(req);
1029 int priv = req->cmd_flags & REQ_ELVPRIV;
1031 BUG_ON(!list_empty(&req->queuelist));
1032 BUG_ON(!hlist_unhashed(&req->hash));
1034 blk_free_request(q, req);
1035 freed_request(q, rw, priv);
1038 EXPORT_SYMBOL_GPL(__blk_put_request);
1040 void blk_put_request(struct request *req)
1042 unsigned long flags;
1043 struct request_queue *q = req->q;
1045 spin_lock_irqsave(q->queue_lock, flags);
1046 __blk_put_request(q, req);
1047 spin_unlock_irqrestore(q->queue_lock, flags);
1049 EXPORT_SYMBOL(blk_put_request);
1051 void init_request_from_bio(struct request *req, struct bio *bio)
1053 req->cmd_type = REQ_TYPE_FS;
1056 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1058 if (bio_rw_ahead(bio) || bio_failfast(bio))
1059 req->cmd_flags |= REQ_FAILFAST;
1062 * REQ_BARRIER implies no merging, but lets make it explicit
1064 if (unlikely(bio_barrier(bio)))
1065 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
1068 req->cmd_flags |= REQ_RW_SYNC;
1069 if (bio_rw_meta(bio))
1070 req->cmd_flags |= REQ_RW_META;
1073 req->hard_sector = req->sector = bio->bi_sector;
1074 req->ioprio = bio_prio(bio);
1075 req->start_time = jiffies;
1076 blk_rq_bio_prep(req->q, req, bio);
1079 static int __make_request(struct request_queue *q, struct bio *bio)
1081 struct request *req;
1082 int el_ret, nr_sectors, barrier, err;
1083 const unsigned short prio = bio_prio(bio);
1084 const int sync = bio_sync(bio);
1087 nr_sectors = bio_sectors(bio);
1090 * low level driver can indicate that it wants pages above a
1091 * certain limit bounced to low memory (ie for highmem, or even
1092 * ISA dma in theory)
1094 blk_queue_bounce(q, &bio);
1096 barrier = bio_barrier(bio);
1097 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
1102 spin_lock_irq(q->queue_lock);
1104 if (unlikely(barrier) || elv_queue_empty(q))
1107 el_ret = elv_merge(q, &req, bio);
1109 case ELEVATOR_BACK_MERGE:
1110 BUG_ON(!rq_mergeable(req));
1112 if (!ll_back_merge_fn(q, req, bio))
1115 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
1117 req->biotail->bi_next = bio;
1119 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1120 req->ioprio = ioprio_best(req->ioprio, prio);
1121 drive_stat_acct(req, 0);
1122 if (!attempt_back_merge(q, req))
1123 elv_merged_request(q, req, el_ret);
1126 case ELEVATOR_FRONT_MERGE:
1127 BUG_ON(!rq_mergeable(req));
1129 if (!ll_front_merge_fn(q, req, bio))
1132 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
1134 bio->bi_next = req->bio;
1138 * may not be valid. if the low level driver said
1139 * it didn't need a bounce buffer then it better
1140 * not touch req->buffer either...
1142 req->buffer = bio_data(bio);
1143 req->current_nr_sectors = bio_cur_sectors(bio);
1144 req->hard_cur_sectors = req->current_nr_sectors;
1145 req->sector = req->hard_sector = bio->bi_sector;
1146 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
1147 req->ioprio = ioprio_best(req->ioprio, prio);
1148 drive_stat_acct(req, 0);
1149 if (!attempt_front_merge(q, req))
1150 elv_merged_request(q, req, el_ret);
1153 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1160 * This sync check and mask will be re-done in init_request_from_bio(),
1161 * but we need to set it earlier to expose the sync flag to the
1162 * rq allocator and io schedulers.
1164 rw_flags = bio_data_dir(bio);
1166 rw_flags |= REQ_RW_SYNC;
1169 * Grab a free request. This is might sleep but can not fail.
1170 * Returns with the queue unlocked.
1172 req = get_request_wait(q, rw_flags, bio);
1175 * After dropping the lock and possibly sleeping here, our request
1176 * may now be mergeable after it had proven unmergeable (above).
1177 * We don't worry about that case for efficiency. It won't happen
1178 * often, and the elevators are able to handle it.
1180 init_request_from_bio(req, bio);
1182 spin_lock_irq(q->queue_lock);
1183 if (elv_queue_empty(q))
1185 add_request(q, req);
1188 __generic_unplug_device(q);
1190 spin_unlock_irq(q->queue_lock);
1194 bio_endio(bio, err);
1199 * If bio->bi_dev is a partition, remap the location
1201 static inline void blk_partition_remap(struct bio *bio)
1203 struct block_device *bdev = bio->bi_bdev;
1205 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1206 struct hd_struct *p = bdev->bd_part;
1208 bio->bi_sector += p->start_sect;
1209 bio->bi_bdev = bdev->bd_contains;
1211 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
1212 bdev->bd_dev, bio->bi_sector,
1213 bio->bi_sector - p->start_sect);
1217 static void handle_bad_sector(struct bio *bio)
1219 char b[BDEVNAME_SIZE];
1221 printk(KERN_INFO "attempt to access beyond end of device\n");
1222 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1223 bdevname(bio->bi_bdev, b),
1225 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1226 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1228 set_bit(BIO_EOF, &bio->bi_flags);
1231 #ifdef CONFIG_FAIL_MAKE_REQUEST
1233 static DECLARE_FAULT_ATTR(fail_make_request);
1235 static int __init setup_fail_make_request(char *str)
1237 return setup_fault_attr(&fail_make_request, str);
1239 __setup("fail_make_request=", setup_fail_make_request);
1241 static int should_fail_request(struct bio *bio)
1243 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
1244 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
1245 return should_fail(&fail_make_request, bio->bi_size);
1250 static int __init fail_make_request_debugfs(void)
1252 return init_fault_attr_dentries(&fail_make_request,
1253 "fail_make_request");
1256 late_initcall(fail_make_request_debugfs);
1258 #else /* CONFIG_FAIL_MAKE_REQUEST */
1260 static inline int should_fail_request(struct bio *bio)
1265 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1268 * Check whether this bio extends beyond the end of the device.
1270 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1277 /* Test device or partition size, when known. */
1278 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1280 sector_t sector = bio->bi_sector;
1282 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1284 * This may well happen - the kernel calls bread()
1285 * without checking the size of the device, e.g., when
1286 * mounting a device.
1288 handle_bad_sector(bio);
1297 * generic_make_request: hand a buffer to its device driver for I/O
1298 * @bio: The bio describing the location in memory and on the device.
1300 * generic_make_request() is used to make I/O requests of block
1301 * devices. It is passed a &struct bio, which describes the I/O that needs
1304 * generic_make_request() does not return any status. The
1305 * success/failure status of the request, along with notification of
1306 * completion, is delivered asynchronously through the bio->bi_end_io
1307 * function described (one day) else where.
1309 * The caller of generic_make_request must make sure that bi_io_vec
1310 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1311 * set to describe the device address, and the
1312 * bi_end_io and optionally bi_private are set to describe how
1313 * completion notification should be signaled.
1315 * generic_make_request and the drivers it calls may use bi_next if this
1316 * bio happens to be merged with someone else, and may change bi_dev and
1317 * bi_sector for remaps as it sees fit. So the values of these fields
1318 * should NOT be depended on after the call to generic_make_request.
1320 static inline void __generic_make_request(struct bio *bio)
1322 struct request_queue *q;
1323 sector_t old_sector;
1324 int ret, nr_sectors = bio_sectors(bio);
1330 if (bio_check_eod(bio, nr_sectors))
1334 * Resolve the mapping until finished. (drivers are
1335 * still free to implement/resolve their own stacking
1336 * by explicitly returning 0)
1338 * NOTE: we don't repeat the blk_size check for each new device.
1339 * Stacking drivers are expected to know what they are doing.
1344 char b[BDEVNAME_SIZE];
1346 q = bdev_get_queue(bio->bi_bdev);
1349 "generic_make_request: Trying to access "
1350 "nonexistent block-device %s (%Lu)\n",
1351 bdevname(bio->bi_bdev, b),
1352 (long long) bio->bi_sector);
1354 bio_endio(bio, err);
1358 if (unlikely(nr_sectors > q->max_hw_sectors)) {
1359 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1360 bdevname(bio->bi_bdev, b),
1366 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1369 if (should_fail_request(bio))
1373 * If this device has partitions, remap block n
1374 * of partition p to block n+start(p) of the disk.
1376 blk_partition_remap(bio);
1378 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1381 if (old_sector != -1)
1382 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
1385 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
1387 old_sector = bio->bi_sector;
1388 old_dev = bio->bi_bdev->bd_dev;
1390 if (bio_check_eod(bio, nr_sectors))
1392 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
1397 ret = q->make_request_fn(q, bio);
1402 * We only want one ->make_request_fn to be active at a time,
1403 * else stack usage with stacked devices could be a problem.
1404 * So use current->bio_{list,tail} to keep a list of requests
1405 * submited by a make_request_fn function.
1406 * current->bio_tail is also used as a flag to say if
1407 * generic_make_request is currently active in this task or not.
1408 * If it is NULL, then no make_request is active. If it is non-NULL,
1409 * then a make_request is active, and new requests should be added
1412 void generic_make_request(struct bio *bio)
1414 if (current->bio_tail) {
1415 /* make_request is active */
1416 *(current->bio_tail) = bio;
1417 bio->bi_next = NULL;
1418 current->bio_tail = &bio->bi_next;
1421 /* following loop may be a bit non-obvious, and so deserves some
1423 * Before entering the loop, bio->bi_next is NULL (as all callers
1424 * ensure that) so we have a list with a single bio.
1425 * We pretend that we have just taken it off a longer list, so
1426 * we assign bio_list to the next (which is NULL) and bio_tail
1427 * to &bio_list, thus initialising the bio_list of new bios to be
1428 * added. __generic_make_request may indeed add some more bios
1429 * through a recursive call to generic_make_request. If it
1430 * did, we find a non-NULL value in bio_list and re-enter the loop
1431 * from the top. In this case we really did just take the bio
1432 * of the top of the list (no pretending) and so fixup bio_list and
1433 * bio_tail or bi_next, and call into __generic_make_request again.
1435 * The loop was structured like this to make only one call to
1436 * __generic_make_request (which is important as it is large and
1437 * inlined) and to keep the structure simple.
1439 BUG_ON(bio->bi_next);
1441 current->bio_list = bio->bi_next;
1442 if (bio->bi_next == NULL)
1443 current->bio_tail = ¤t->bio_list;
1445 bio->bi_next = NULL;
1446 __generic_make_request(bio);
1447 bio = current->bio_list;
1449 current->bio_tail = NULL; /* deactivate */
1451 EXPORT_SYMBOL(generic_make_request);
1454 * submit_bio: submit a bio to the block device layer for I/O
1455 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1456 * @bio: The &struct bio which describes the I/O
1458 * submit_bio() is very similar in purpose to generic_make_request(), and
1459 * uses that function to do most of the work. Both are fairly rough
1460 * interfaces, @bio must be presetup and ready for I/O.
1463 void submit_bio(int rw, struct bio *bio)
1465 int count = bio_sectors(bio);
1470 * If it's a regular read/write or a barrier with data attached,
1471 * go through the normal accounting stuff before submission.
1473 if (!bio_empty_barrier(bio)) {
1475 BIO_BUG_ON(!bio->bi_size);
1476 BIO_BUG_ON(!bio->bi_io_vec);
1479 count_vm_events(PGPGOUT, count);
1481 task_io_account_read(bio->bi_size);
1482 count_vm_events(PGPGIN, count);
1485 if (unlikely(block_dump)) {
1486 char b[BDEVNAME_SIZE];
1487 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1488 current->comm, task_pid_nr(current),
1489 (rw & WRITE) ? "WRITE" : "READ",
1490 (unsigned long long)bio->bi_sector,
1491 bdevname(bio->bi_bdev, b));
1495 generic_make_request(bio);
1497 EXPORT_SYMBOL(submit_bio);
1500 * __end_that_request_first - end I/O on a request
1501 * @req: the request being processed
1502 * @error: 0 for success, < 0 for error
1503 * @nr_bytes: number of bytes to complete
1506 * Ends I/O on a number of bytes attached to @req, and sets it up
1507 * for the next range of segments (if any) in the cluster.
1510 * 0 - we are done with this request, call end_that_request_last()
1511 * 1 - still buffers pending for this request
1513 static int __end_that_request_first(struct request *req, int error,
1516 int total_bytes, bio_nbytes, next_idx = 0;
1519 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
1522 * for a REQ_BLOCK_PC request, we want to carry any eventual
1523 * sense key with us all the way through
1525 if (!blk_pc_request(req))
1528 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1529 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1530 req->rq_disk ? req->rq_disk->disk_name : "?",
1531 (unsigned long long)req->sector);
1534 if (blk_fs_request(req) && req->rq_disk) {
1535 struct hd_struct *part = get_part(req->rq_disk, req->sector);
1536 const int rw = rq_data_dir(req);
1538 all_stat_add(req->rq_disk, part, sectors[rw],
1539 nr_bytes >> 9, req->sector);
1542 total_bytes = bio_nbytes = 0;
1543 while ((bio = req->bio) != NULL) {
1547 * For an empty barrier request, the low level driver must
1548 * store a potential error location in ->sector. We pass
1549 * that back up in ->bi_sector.
1551 if (blk_empty_barrier(req))
1552 bio->bi_sector = req->sector;
1554 if (nr_bytes >= bio->bi_size) {
1555 req->bio = bio->bi_next;
1556 nbytes = bio->bi_size;
1557 req_bio_endio(req, bio, nbytes, error);
1561 int idx = bio->bi_idx + next_idx;
1563 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
1564 blk_dump_rq_flags(req, "__end_that");
1565 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1566 __func__, bio->bi_idx, bio->bi_vcnt);
1570 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1571 BIO_BUG_ON(nbytes > bio->bi_size);
1574 * not a complete bvec done
1576 if (unlikely(nbytes > nr_bytes)) {
1577 bio_nbytes += nr_bytes;
1578 total_bytes += nr_bytes;
1583 * advance to the next vector
1586 bio_nbytes += nbytes;
1589 total_bytes += nbytes;
1595 * end more in this run, or just return 'not-done'
1597 if (unlikely(nr_bytes <= 0))
1609 * if the request wasn't completed, update state
1612 req_bio_endio(req, bio, bio_nbytes, error);
1613 bio->bi_idx += next_idx;
1614 bio_iovec(bio)->bv_offset += nr_bytes;
1615 bio_iovec(bio)->bv_len -= nr_bytes;
1618 blk_recalc_rq_sectors(req, total_bytes >> 9);
1619 blk_recalc_rq_segments(req);
1624 * splice the completion data to a local structure and hand off to
1625 * process_completion_queue() to complete the requests
1627 static void blk_done_softirq(struct softirq_action *h)
1629 struct list_head *cpu_list, local_list;
1631 local_irq_disable();
1632 cpu_list = &__get_cpu_var(blk_cpu_done);
1633 list_replace_init(cpu_list, &local_list);
1636 while (!list_empty(&local_list)) {
1639 rq = list_entry(local_list.next, struct request, donelist);
1640 list_del_init(&rq->donelist);
1641 rq->q->softirq_done_fn(rq);
1645 static int __cpuinit blk_cpu_notify(struct notifier_block *self,
1646 unsigned long action, void *hcpu)
1649 * If a CPU goes away, splice its entries to the current CPU
1650 * and trigger a run of the softirq
1652 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1653 int cpu = (unsigned long) hcpu;
1655 local_irq_disable();
1656 list_splice_init(&per_cpu(blk_cpu_done, cpu),
1657 &__get_cpu_var(blk_cpu_done));
1658 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1666 static struct notifier_block blk_cpu_notifier __cpuinitdata = {
1667 .notifier_call = blk_cpu_notify,
1671 * blk_complete_request - end I/O on a request
1672 * @req: the request being processed
1675 * Ends all I/O on a request. It does not handle partial completions,
1676 * unless the driver actually implements this in its completion callback
1677 * through requeueing. The actual completion happens out-of-order,
1678 * through a softirq handler. The user must have registered a completion
1679 * callback through blk_queue_softirq_done().
1682 void blk_complete_request(struct request *req)
1684 struct list_head *cpu_list;
1685 unsigned long flags;
1687 BUG_ON(!req->q->softirq_done_fn);
1689 local_irq_save(flags);
1691 cpu_list = &__get_cpu_var(blk_cpu_done);
1692 list_add_tail(&req->donelist, cpu_list);
1693 raise_softirq_irqoff(BLOCK_SOFTIRQ);
1695 local_irq_restore(flags);
1697 EXPORT_SYMBOL(blk_complete_request);
1700 * queue lock must be held
1702 static void end_that_request_last(struct request *req, int error)
1704 struct gendisk *disk = req->rq_disk;
1706 if (blk_rq_tagged(req))
1707 blk_queue_end_tag(req->q, req);
1709 if (blk_queued_rq(req))
1710 blkdev_dequeue_request(req);
1712 if (unlikely(laptop_mode) && blk_fs_request(req))
1713 laptop_io_completion();
1716 * Account IO completion. bar_rq isn't accounted as a normal
1717 * IO on queueing nor completion. Accounting the containing
1718 * request is enough.
1720 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
1721 unsigned long duration = jiffies - req->start_time;
1722 const int rw = rq_data_dir(req);
1723 struct hd_struct *part = get_part(disk, req->sector);
1725 __all_stat_inc(disk, part, ios[rw], req->sector);
1726 __all_stat_add(disk, part, ticks[rw], duration, req->sector);
1727 disk_round_stats(disk);
1730 part_round_stats(part);
1736 req->end_io(req, error);
1738 if (blk_bidi_rq(req))
1739 __blk_put_request(req->next_rq->q, req->next_rq);
1741 __blk_put_request(req->q, req);
1745 static inline void __end_request(struct request *rq, int uptodate,
1746 unsigned int nr_bytes)
1751 error = uptodate ? uptodate : -EIO;
1753 __blk_end_request(rq, error, nr_bytes);
1757 * blk_rq_bytes - Returns bytes left to complete in the entire request
1758 * @rq: the request being processed
1760 unsigned int blk_rq_bytes(struct request *rq)
1762 if (blk_fs_request(rq))
1763 return rq->hard_nr_sectors << 9;
1765 return rq->data_len;
1767 EXPORT_SYMBOL_GPL(blk_rq_bytes);
1770 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1771 * @rq: the request being processed
1773 unsigned int blk_rq_cur_bytes(struct request *rq)
1775 if (blk_fs_request(rq))
1776 return rq->current_nr_sectors << 9;
1779 return rq->bio->bi_size;
1781 return rq->data_len;
1783 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes);
1786 * end_queued_request - end all I/O on a queued request
1787 * @rq: the request being processed
1788 * @uptodate: error value or 0/1 uptodate flag
1791 * Ends all I/O on a request, and removes it from the block layer queues.
1792 * Not suitable for normal IO completion, unless the driver still has
1793 * the request attached to the block layer.
1796 void end_queued_request(struct request *rq, int uptodate)
1798 __end_request(rq, uptodate, blk_rq_bytes(rq));
1800 EXPORT_SYMBOL(end_queued_request);
1803 * end_dequeued_request - end all I/O on a dequeued request
1804 * @rq: the request being processed
1805 * @uptodate: error value or 0/1 uptodate flag
1808 * Ends all I/O on a request. The request must already have been
1809 * dequeued using blkdev_dequeue_request(), as is normally the case
1813 void end_dequeued_request(struct request *rq, int uptodate)
1815 __end_request(rq, uptodate, blk_rq_bytes(rq));
1817 EXPORT_SYMBOL(end_dequeued_request);
1821 * end_request - end I/O on the current segment of the request
1822 * @req: the request being processed
1823 * @uptodate: error value or 0/1 uptodate flag
1826 * Ends I/O on the current segment of a request. If that is the only
1827 * remaining segment, the request is also completed and freed.
1829 * This is a remnant of how older block drivers handled IO completions.
1830 * Modern drivers typically end IO on the full request in one go, unless
1831 * they have a residual value to account for. For that case this function
1832 * isn't really useful, unless the residual just happens to be the
1833 * full current segment. In other words, don't use this function in new
1834 * code. Either use end_request_completely(), or the
1835 * end_that_request_chunk() (along with end_that_request_last()) for
1836 * partial completions.
1839 void end_request(struct request *req, int uptodate)
1841 __end_request(req, uptodate, req->hard_cur_sectors << 9);
1843 EXPORT_SYMBOL(end_request);
1846 * blk_end_io - Generic end_io function to complete a request.
1847 * @rq: the request being processed
1848 * @error: 0 for success, < 0 for error
1849 * @nr_bytes: number of bytes to complete @rq
1850 * @bidi_bytes: number of bytes to complete @rq->next_rq
1851 * @drv_callback: function called between completion of bios in the request
1852 * and completion of the request.
1853 * If the callback returns non 0, this helper returns without
1854 * completion of the request.
1857 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1858 * If @rq has leftover, sets it up for the next range of segments.
1861 * 0 - we are done with this request
1862 * 1 - this request is not freed yet, it still has pending buffers.
1864 static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes,
1865 unsigned int bidi_bytes,
1866 int (drv_callback)(struct request *))
1868 struct request_queue *q = rq->q;
1869 unsigned long flags = 0UL;
1871 if (blk_fs_request(rq) || blk_pc_request(rq)) {
1872 if (__end_that_request_first(rq, error, nr_bytes))
1875 /* Bidi request must be completed as a whole */
1876 if (blk_bidi_rq(rq) &&
1877 __end_that_request_first(rq->next_rq, error, bidi_bytes))
1881 /* Special feature for tricky drivers */
1882 if (drv_callback && drv_callback(rq))
1885 add_disk_randomness(rq->rq_disk);
1887 spin_lock_irqsave(q->queue_lock, flags);
1888 end_that_request_last(rq, error);
1889 spin_unlock_irqrestore(q->queue_lock, flags);
1895 * blk_end_request - Helper function for drivers to complete the request.
1896 * @rq: the request being processed
1897 * @error: 0 for success, < 0 for error
1898 * @nr_bytes: number of bytes to complete
1901 * Ends I/O on a number of bytes attached to @rq.
1902 * If @rq has leftover, sets it up for the next range of segments.
1905 * 0 - we are done with this request
1906 * 1 - still buffers pending for this request
1908 int blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1910 return blk_end_io(rq, error, nr_bytes, 0, NULL);
1912 EXPORT_SYMBOL_GPL(blk_end_request);
1915 * __blk_end_request - Helper function for drivers to complete the request.
1916 * @rq: the request being processed
1917 * @error: 0 for success, < 0 for error
1918 * @nr_bytes: number of bytes to complete
1921 * Must be called with queue lock held unlike blk_end_request().
1924 * 0 - we are done with this request
1925 * 1 - still buffers pending for this request
1927 int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
1929 if (blk_fs_request(rq) || blk_pc_request(rq)) {
1930 if (__end_that_request_first(rq, error, nr_bytes))
1934 add_disk_randomness(rq->rq_disk);
1936 end_that_request_last(rq, error);
1940 EXPORT_SYMBOL_GPL(__blk_end_request);
1943 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1944 * @rq: the bidi request being processed
1945 * @error: 0 for success, < 0 for error
1946 * @nr_bytes: number of bytes to complete @rq
1947 * @bidi_bytes: number of bytes to complete @rq->next_rq
1950 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1953 * 0 - we are done with this request
1954 * 1 - still buffers pending for this request
1956 int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes,
1957 unsigned int bidi_bytes)
1959 return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL);
1961 EXPORT_SYMBOL_GPL(blk_end_bidi_request);
1964 * blk_end_request_callback - Special helper function for tricky drivers
1965 * @rq: the request being processed
1966 * @error: 0 for success, < 0 for error
1967 * @nr_bytes: number of bytes to complete
1968 * @drv_callback: function called between completion of bios in the request
1969 * and completion of the request.
1970 * If the callback returns non 0, this helper returns without
1971 * completion of the request.
1974 * Ends I/O on a number of bytes attached to @rq.
1975 * If @rq has leftover, sets it up for the next range of segments.
1977 * This special helper function is used only for existing tricky drivers.
1978 * (e.g. cdrom_newpc_intr() of ide-cd)
1979 * This interface will be removed when such drivers are rewritten.
1980 * Don't use this interface in other places anymore.
1983 * 0 - we are done with this request
1984 * 1 - this request is not freed yet.
1985 * this request still has pending buffers or
1986 * the driver doesn't want to finish this request yet.
1988 int blk_end_request_callback(struct request *rq, int error,
1989 unsigned int nr_bytes,
1990 int (drv_callback)(struct request *))
1992 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
1994 EXPORT_SYMBOL_GPL(blk_end_request_callback);
1996 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1999 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2000 rq->cmd_flags |= (bio->bi_rw & 3);
2002 rq->nr_phys_segments = bio_phys_segments(q, bio);
2003 rq->nr_hw_segments = bio_hw_segments(q, bio);
2004 rq->current_nr_sectors = bio_cur_sectors(bio);
2005 rq->hard_cur_sectors = rq->current_nr_sectors;
2006 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2007 rq->buffer = bio_data(bio);
2008 rq->data_len = bio->bi_size;
2010 rq->bio = rq->biotail = bio;
2013 rq->rq_disk = bio->bi_bdev->bd_disk;
2016 int kblockd_schedule_work(struct work_struct *work)
2018 return queue_work(kblockd_workqueue, work);
2020 EXPORT_SYMBOL(kblockd_schedule_work);
2022 void kblockd_flush_work(struct work_struct *work)
2024 cancel_work_sync(work);
2026 EXPORT_SYMBOL(kblockd_flush_work);
2028 int __init blk_dev_init(void)
2032 kblockd_workqueue = create_workqueue("kblockd");
2033 if (!kblockd_workqueue)
2034 panic("Failed to create kblockd\n");
2036 request_cachep = kmem_cache_create("blkdev_requests",
2037 sizeof(struct request), 0, SLAB_PANIC, NULL);
2039 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2040 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2042 for_each_possible_cpu(i)
2043 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2045 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq);
2046 register_hotcpu_notifier(&blk_cpu_notifier);