2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/blktrace_api.h>
30 #include <linux/fault-inject.h>
31 #include <trace/block.h>
35 DEFINE_TRACE(block_plug);
36 DEFINE_TRACE(block_unplug_io);
37 DEFINE_TRACE(block_unplug_timer);
38 DEFINE_TRACE(block_getrq);
39 DEFINE_TRACE(block_sleeprq);
40 DEFINE_TRACE(block_rq_requeue);
41 DEFINE_TRACE(block_bio_backmerge);
42 DEFINE_TRACE(block_bio_frontmerge);
43 DEFINE_TRACE(block_bio_queue);
44 DEFINE_TRACE(block_rq_complete);
45 DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
48 static int __make_request(struct request_queue *q, struct bio *bio);
51 * For the allocated request tables
53 static struct kmem_cache *request_cachep;
56 * For queue allocation
58 struct kmem_cache *blk_requestq_cachep;
61 * Controlling structure to kblockd
63 static struct workqueue_struct *kblockd_workqueue;
65 static void drive_stat_acct(struct request *rq, int new_io)
67 struct hd_struct *part;
68 int rw = rq_data_dir(rq);
71 if (!blk_do_io_stat(rq))
74 cpu = part_stat_lock();
75 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
78 part_stat_inc(cpu, part, merges[rw]);
80 part_round_stats(cpu, part);
81 part_inc_in_flight(part);
87 void blk_queue_congestion_threshold(struct request_queue *q)
91 nr = q->nr_requests - (q->nr_requests / 8) + 1;
92 if (nr > q->nr_requests)
94 q->nr_congestion_on = nr;
96 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
99 q->nr_congestion_off = nr;
103 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
106 * Locates the passed device's request queue and returns the address of its
109 * Will return NULL if the request queue cannot be located.
111 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
113 struct backing_dev_info *ret = NULL;
114 struct request_queue *q = bdev_get_queue(bdev);
117 ret = &q->backing_dev_info;
120 EXPORT_SYMBOL(blk_get_backing_dev_info);
122 void blk_rq_init(struct request_queue *q, struct request *rq)
124 memset(rq, 0, sizeof(*rq));
126 INIT_LIST_HEAD(&rq->queuelist);
127 INIT_LIST_HEAD(&rq->timeout_list);
130 rq->__sector = (sector_t) -1;
131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
134 rq->cmd_len = BLK_MAX_CDB;
137 rq->start_time = jiffies;
139 EXPORT_SYMBOL(blk_rq_init);
141 static void req_bio_endio(struct request *rq, struct bio *bio,
142 unsigned int nbytes, int error)
144 struct request_queue *q = rq->q;
146 if (&q->bar_rq != rq) {
148 clear_bit(BIO_UPTODATE, &bio->bi_flags);
149 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
152 if (unlikely(nbytes > bio->bi_size)) {
153 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
154 __func__, nbytes, bio->bi_size);
155 nbytes = bio->bi_size;
158 if (unlikely(rq->cmd_flags & REQ_QUIET))
159 set_bit(BIO_QUIET, &bio->bi_flags);
161 bio->bi_size -= nbytes;
162 bio->bi_sector += (nbytes >> 9);
164 if (bio_integrity(bio))
165 bio_integrity_advance(bio, nbytes);
167 if (bio->bi_size == 0)
168 bio_endio(bio, error);
172 * Okay, this is the barrier request in progress, just
175 if (error && !q->orderr)
180 void blk_dump_rq_flags(struct request *rq, char *msg)
184 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
185 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
188 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
189 (unsigned long long)blk_rq_pos(rq),
190 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
191 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
192 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
194 if (blk_pc_request(rq)) {
195 printk(KERN_INFO " cdb: ");
196 for (bit = 0; bit < BLK_MAX_CDB; bit++)
197 printk("%02x ", rq->cmd[bit]);
201 EXPORT_SYMBOL(blk_dump_rq_flags);
204 * "plug" the device if there are no outstanding requests: this will
205 * force the transfer to start only after we have put all the requests
208 * This is called with interrupts off and no requests on the queue and
209 * with the queue lock held.
211 void blk_plug_device(struct request_queue *q)
213 WARN_ON(!irqs_disabled());
216 * don't plug a stopped queue, it must be paired with blk_start_queue()
217 * which will restart the queueing
219 if (blk_queue_stopped(q))
222 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
223 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
227 EXPORT_SYMBOL(blk_plug_device);
230 * blk_plug_device_unlocked - plug a device without queue lock held
231 * @q: The &struct request_queue to plug
234 * Like @blk_plug_device(), but grabs the queue lock and disables
237 void blk_plug_device_unlocked(struct request_queue *q)
241 spin_lock_irqsave(q->queue_lock, flags);
243 spin_unlock_irqrestore(q->queue_lock, flags);
245 EXPORT_SYMBOL(blk_plug_device_unlocked);
248 * remove the queue from the plugged list, if present. called with
249 * queue lock held and interrupts disabled.
251 int blk_remove_plug(struct request_queue *q)
253 WARN_ON(!irqs_disabled());
255 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
258 del_timer(&q->unplug_timer);
261 EXPORT_SYMBOL(blk_remove_plug);
264 * remove the plug and let it rip..
266 void __generic_unplug_device(struct request_queue *q)
268 if (unlikely(blk_queue_stopped(q)))
270 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
277 * generic_unplug_device - fire a request queue
278 * @q: The &struct request_queue in question
281 * Linux uses plugging to build bigger requests queues before letting
282 * the device have at them. If a queue is plugged, the I/O scheduler
283 * is still adding and merging requests on the queue. Once the queue
284 * gets unplugged, the request_fn defined for the queue is invoked and
287 void generic_unplug_device(struct request_queue *q)
289 if (blk_queue_plugged(q)) {
290 spin_lock_irq(q->queue_lock);
291 __generic_unplug_device(q);
292 spin_unlock_irq(q->queue_lock);
295 EXPORT_SYMBOL(generic_unplug_device);
297 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
300 struct request_queue *q = bdi->unplug_io_data;
305 void blk_unplug_work(struct work_struct *work)
307 struct request_queue *q =
308 container_of(work, struct request_queue, unplug_work);
310 trace_block_unplug_io(q);
314 void blk_unplug_timeout(unsigned long data)
316 struct request_queue *q = (struct request_queue *)data;
318 trace_block_unplug_timer(q);
319 kblockd_schedule_work(q, &q->unplug_work);
322 void blk_unplug(struct request_queue *q)
325 * devices don't necessarily have an ->unplug_fn defined
328 trace_block_unplug_io(q);
332 EXPORT_SYMBOL(blk_unplug);
335 * blk_start_queue - restart a previously stopped queue
336 * @q: The &struct request_queue in question
339 * blk_start_queue() will clear the stop flag on the queue, and call
340 * the request_fn for the queue if it was in a stopped state when
341 * entered. Also see blk_stop_queue(). Queue lock must be held.
343 void blk_start_queue(struct request_queue *q)
345 WARN_ON(!irqs_disabled());
347 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
350 EXPORT_SYMBOL(blk_start_queue);
353 * blk_stop_queue - stop a queue
354 * @q: The &struct request_queue in question
357 * The Linux block layer assumes that a block driver will consume all
358 * entries on the request queue when the request_fn strategy is called.
359 * Often this will not happen, because of hardware limitations (queue
360 * depth settings). If a device driver gets a 'queue full' response,
361 * or if it simply chooses not to queue more I/O at one point, it can
362 * call this function to prevent the request_fn from being called until
363 * the driver has signalled it's ready to go again. This happens by calling
364 * blk_start_queue() to restart queue operations. Queue lock must be held.
366 void blk_stop_queue(struct request_queue *q)
369 queue_flag_set(QUEUE_FLAG_STOPPED, q);
371 EXPORT_SYMBOL(blk_stop_queue);
374 * blk_sync_queue - cancel any pending callbacks on a queue
378 * The block layer may perform asynchronous callback activity
379 * on a queue, such as calling the unplug function after a timeout.
380 * A block device may call blk_sync_queue to ensure that any
381 * such activity is cancelled, thus allowing it to release resources
382 * that the callbacks might use. The caller must already have made sure
383 * that its ->make_request_fn will not re-add plugging prior to calling
387 void blk_sync_queue(struct request_queue *q)
389 del_timer_sync(&q->unplug_timer);
390 del_timer_sync(&q->timeout);
391 cancel_work_sync(&q->unplug_work);
393 EXPORT_SYMBOL(blk_sync_queue);
396 * __blk_run_queue - run a single device queue
397 * @q: The queue to run
400 * See @blk_run_queue. This variant must be called with the queue lock
401 * held and interrupts disabled.
404 void __blk_run_queue(struct request_queue *q)
408 if (unlikely(blk_queue_stopped(q)))
411 if (elv_queue_empty(q))
415 * Only recurse once to avoid overrunning the stack, let the unplug
416 * handling reinvoke the handler shortly if we already got there.
418 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
420 queue_flag_clear(QUEUE_FLAG_REENTER, q);
422 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
423 kblockd_schedule_work(q, &q->unplug_work);
426 EXPORT_SYMBOL(__blk_run_queue);
429 * blk_run_queue - run a single device queue
430 * @q: The queue to run
433 * Invoke request handling on this queue, if it has pending work to do.
434 * May be used to restart queueing when a request has completed.
436 void blk_run_queue(struct request_queue *q)
440 spin_lock_irqsave(q->queue_lock, flags);
442 spin_unlock_irqrestore(q->queue_lock, flags);
444 EXPORT_SYMBOL(blk_run_queue);
446 void blk_put_queue(struct request_queue *q)
448 kobject_put(&q->kobj);
451 void blk_cleanup_queue(struct request_queue *q)
454 * We know we have process context here, so we can be a little
455 * cautious and ensure that pending block actions on this device
456 * are done before moving on. Going into this function, we should
457 * not have processes doing IO to this device.
461 mutex_lock(&q->sysfs_lock);
462 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
463 mutex_unlock(&q->sysfs_lock);
466 elevator_exit(q->elevator);
470 EXPORT_SYMBOL(blk_cleanup_queue);
472 static int blk_init_free_list(struct request_queue *q)
474 struct request_list *rl = &q->rq;
476 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
477 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
479 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
480 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
482 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
483 mempool_free_slab, request_cachep, q->node);
491 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
493 return blk_alloc_queue_node(gfp_mask, -1);
495 EXPORT_SYMBOL(blk_alloc_queue);
497 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
499 struct request_queue *q;
502 q = kmem_cache_alloc_node(blk_requestq_cachep,
503 gfp_mask | __GFP_ZERO, node_id);
507 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
508 q->backing_dev_info.unplug_io_data = q;
509 err = bdi_init(&q->backing_dev_info);
511 kmem_cache_free(blk_requestq_cachep, q);
515 init_timer(&q->unplug_timer);
516 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
517 INIT_LIST_HEAD(&q->timeout_list);
518 INIT_WORK(&q->unplug_work, blk_unplug_work);
520 kobject_init(&q->kobj, &blk_queue_ktype);
522 mutex_init(&q->sysfs_lock);
523 spin_lock_init(&q->__queue_lock);
527 EXPORT_SYMBOL(blk_alloc_queue_node);
530 * blk_init_queue - prepare a request queue for use with a block device
531 * @rfn: The function to be called to process requests that have been
532 * placed on the queue.
533 * @lock: Request queue spin lock
536 * If a block device wishes to use the standard request handling procedures,
537 * which sorts requests and coalesces adjacent requests, then it must
538 * call blk_init_queue(). The function @rfn will be called when there
539 * are requests on the queue that need to be processed. If the device
540 * supports plugging, then @rfn may not be called immediately when requests
541 * are available on the queue, but may be called at some time later instead.
542 * Plugged queues are generally unplugged when a buffer belonging to one
543 * of the requests on the queue is needed, or due to memory pressure.
545 * @rfn is not required, or even expected, to remove all requests off the
546 * queue, but only as many as it can handle at a time. If it does leave
547 * requests on the queue, it is responsible for arranging that the requests
548 * get dealt with eventually.
550 * The queue spin lock must be held while manipulating the requests on the
551 * request queue; this lock will be taken also from interrupt context, so irq
552 * disabling is needed for it.
554 * Function returns a pointer to the initialized request queue, or %NULL if
558 * blk_init_queue() must be paired with a blk_cleanup_queue() call
559 * when the block device is deactivated (such as at module unload).
562 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
564 return blk_init_queue_node(rfn, lock, -1);
566 EXPORT_SYMBOL(blk_init_queue);
568 struct request_queue *
569 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
571 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
577 if (blk_init_free_list(q)) {
578 kmem_cache_free(blk_requestq_cachep, q);
583 * if caller didn't supply a lock, they get per-queue locking with
587 lock = &q->__queue_lock;
590 q->prep_rq_fn = NULL;
591 q->unplug_fn = generic_unplug_device;
592 q->queue_flags = QUEUE_FLAG_DEFAULT;
593 q->queue_lock = lock;
596 * This also sets hw/phys segments, boundary and size
598 blk_queue_make_request(q, __make_request);
600 q->sg_reserved_size = INT_MAX;
602 blk_set_cmd_filter_defaults(&q->cmd_filter);
607 if (!elevator_init(q, NULL)) {
608 blk_queue_congestion_threshold(q);
615 EXPORT_SYMBOL(blk_init_queue_node);
617 int blk_get_queue(struct request_queue *q)
619 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
620 kobject_get(&q->kobj);
627 static inline void blk_free_request(struct request_queue *q, struct request *rq)
629 if (rq->cmd_flags & REQ_ELVPRIV)
630 elv_put_request(q, rq);
631 mempool_free(rq, q->rq.rq_pool);
634 static struct request *
635 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
637 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
644 rq->cmd_flags = flags | REQ_ALLOCED;
647 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
648 mempool_free(rq, q->rq.rq_pool);
651 rq->cmd_flags |= REQ_ELVPRIV;
658 * ioc_batching returns true if the ioc is a valid batching request and
659 * should be given priority access to a request.
661 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
667 * Make sure the process is able to allocate at least 1 request
668 * even if the batch times out, otherwise we could theoretically
671 return ioc->nr_batch_requests == q->nr_batching ||
672 (ioc->nr_batch_requests > 0
673 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
677 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
678 * will cause the process to be a "batcher" on all queues in the system. This
679 * is the behaviour we want though - once it gets a wakeup it should be given
682 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
684 if (!ioc || ioc_batching(q, ioc))
687 ioc->nr_batch_requests = q->nr_batching;
688 ioc->last_waited = jiffies;
691 static void __freed_request(struct request_queue *q, int sync)
693 struct request_list *rl = &q->rq;
695 if (rl->count[sync] < queue_congestion_off_threshold(q))
696 blk_clear_queue_congested(q, sync);
698 if (rl->count[sync] + 1 <= q->nr_requests) {
699 if (waitqueue_active(&rl->wait[sync]))
700 wake_up(&rl->wait[sync]);
702 blk_clear_queue_full(q, sync);
707 * A request has just been released. Account for it, update the full and
708 * congestion status, wake up any waiters. Called under q->queue_lock.
710 static void freed_request(struct request_queue *q, int sync, int priv)
712 struct request_list *rl = &q->rq;
718 __freed_request(q, sync);
720 if (unlikely(rl->starved[sync ^ 1]))
721 __freed_request(q, sync ^ 1);
725 * Get a free request, queue_lock must be held.
726 * Returns NULL on failure, with queue_lock held.
727 * Returns !NULL on success, with queue_lock *not held*.
729 static struct request *get_request(struct request_queue *q, int rw_flags,
730 struct bio *bio, gfp_t gfp_mask)
732 struct request *rq = NULL;
733 struct request_list *rl = &q->rq;
734 struct io_context *ioc = NULL;
735 const bool is_sync = rw_is_sync(rw_flags) != 0;
738 may_queue = elv_may_queue(q, rw_flags);
739 if (may_queue == ELV_MQUEUE_NO)
742 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
743 if (rl->count[is_sync]+1 >= q->nr_requests) {
744 ioc = current_io_context(GFP_ATOMIC, q->node);
746 * The queue will fill after this allocation, so set
747 * it as full, and mark this process as "batching".
748 * This process will be allowed to complete a batch of
749 * requests, others will be blocked.
751 if (!blk_queue_full(q, is_sync)) {
752 ioc_set_batching(q, ioc);
753 blk_set_queue_full(q, is_sync);
755 if (may_queue != ELV_MQUEUE_MUST
756 && !ioc_batching(q, ioc)) {
758 * The queue is full and the allocating
759 * process is not a "batcher", and not
760 * exempted by the IO scheduler
766 blk_set_queue_congested(q, is_sync);
770 * Only allow batching queuers to allocate up to 50% over the defined
771 * limit of requests, otherwise we could have thousands of requests
772 * allocated with any setting of ->nr_requests
774 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
777 rl->count[is_sync]++;
778 rl->starved[is_sync] = 0;
780 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
784 if (blk_queue_io_stat(q))
785 rw_flags |= REQ_IO_STAT;
786 spin_unlock_irq(q->queue_lock);
788 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
791 * Allocation failed presumably due to memory. Undo anything
792 * we might have messed up.
794 * Allocating task should really be put onto the front of the
795 * wait queue, but this is pretty rare.
797 spin_lock_irq(q->queue_lock);
798 freed_request(q, is_sync, priv);
801 * in the very unlikely event that allocation failed and no
802 * requests for this direction was pending, mark us starved
803 * so that freeing of a request in the other direction will
804 * notice us. another possible fix would be to split the
805 * rq mempool into READ and WRITE
808 if (unlikely(rl->count[is_sync] == 0))
809 rl->starved[is_sync] = 1;
815 * ioc may be NULL here, and ioc_batching will be false. That's
816 * OK, if the queue is under the request limit then requests need
817 * not count toward the nr_batch_requests limit. There will always
818 * be some limit enforced by BLK_BATCH_TIME.
820 if (ioc_batching(q, ioc))
821 ioc->nr_batch_requests--;
823 trace_block_getrq(q, bio, rw_flags & 1);
829 * No available requests for this queue, unplug the device and wait for some
830 * requests to become available.
832 * Called with q->queue_lock held, and returns with it unlocked.
834 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
837 const bool is_sync = rw_is_sync(rw_flags) != 0;
840 rq = get_request(q, rw_flags, bio, GFP_NOIO);
843 struct io_context *ioc;
844 struct request_list *rl = &q->rq;
846 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
847 TASK_UNINTERRUPTIBLE);
849 trace_block_sleeprq(q, bio, rw_flags & 1);
851 __generic_unplug_device(q);
852 spin_unlock_irq(q->queue_lock);
856 * After sleeping, we become a "batching" process and
857 * will be able to allocate at least one request, and
858 * up to a big batch of them for a small period time.
859 * See ioc_batching, ioc_set_batching
861 ioc = current_io_context(GFP_NOIO, q->node);
862 ioc_set_batching(q, ioc);
864 spin_lock_irq(q->queue_lock);
865 finish_wait(&rl->wait[is_sync], &wait);
867 rq = get_request(q, rw_flags, bio, GFP_NOIO);
873 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
877 BUG_ON(rw != READ && rw != WRITE);
879 spin_lock_irq(q->queue_lock);
880 if (gfp_mask & __GFP_WAIT) {
881 rq = get_request_wait(q, rw, NULL);
883 rq = get_request(q, rw, NULL, gfp_mask);
885 spin_unlock_irq(q->queue_lock);
887 /* q->queue_lock is unlocked at this point */
891 EXPORT_SYMBOL(blk_get_request);
894 * blk_make_request - given a bio, allocate a corresponding struct request.
896 * @bio: The bio describing the memory mappings that will be submitted for IO.
897 * It may be a chained-bio properly constructed by block/bio layer.
899 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
900 * type commands. Where the struct request needs to be farther initialized by
901 * the caller. It is passed a &struct bio, which describes the memory info of
904 * The caller of blk_make_request must make sure that bi_io_vec
905 * are set to describe the memory buffers. That bio_data_dir() will return
906 * the needed direction of the request. (And all bio's in the passed bio-chain
907 * are properly set accordingly)
909 * If called under none-sleepable conditions, mapped bio buffers must not
910 * need bouncing, by calling the appropriate masked or flagged allocator,
911 * suitable for the target device. Otherwise the call to blk_queue_bounce will
914 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
915 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
916 * anything but the first bio in the chain. Otherwise you risk waiting for IO
917 * completion of a bio that hasn't been submitted yet, thus resulting in a
918 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
919 * of bio_alloc(), as that avoids the mempool deadlock.
920 * If possible a big IO should be split into smaller parts when allocation
921 * fails. Partial allocation should not be an error, or you risk a live-lock.
923 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
926 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
929 return ERR_PTR(-ENOMEM);
932 struct bio *bounce_bio = bio;
935 blk_queue_bounce(q, &bounce_bio);
936 ret = blk_rq_append_bio(q, rq, bounce_bio);
945 EXPORT_SYMBOL(blk_make_request);
948 * blk_requeue_request - put a request back on queue
949 * @q: request queue where request should be inserted
950 * @rq: request to be inserted
953 * Drivers often keep queueing requests until the hardware cannot accept
954 * more, when that condition happens we need to put the request back
955 * on the queue. Must be called with queue lock held.
957 void blk_requeue_request(struct request_queue *q, struct request *rq)
959 blk_delete_timer(rq);
960 blk_clear_rq_complete(rq);
961 trace_block_rq_requeue(q, rq);
963 if (blk_rq_tagged(rq))
964 blk_queue_end_tag(q, rq);
966 BUG_ON(blk_queued_rq(rq));
968 elv_requeue_request(q, rq);
970 EXPORT_SYMBOL(blk_requeue_request);
973 * blk_insert_request - insert a special request into a request queue
974 * @q: request queue where request should be inserted
975 * @rq: request to be inserted
976 * @at_head: insert request at head or tail of queue
977 * @data: private data
980 * Many block devices need to execute commands asynchronously, so they don't
981 * block the whole kernel from preemption during request execution. This is
982 * accomplished normally by inserting aritficial requests tagged as
983 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
984 * be scheduled for actual execution by the request queue.
986 * We have the option of inserting the head or the tail of the queue.
987 * Typically we use the tail for new ioctls and so forth. We use the head
988 * of the queue for things like a QUEUE_FULL message from a device, or a
989 * host that is unable to accept a particular command.
991 void blk_insert_request(struct request_queue *q, struct request *rq,
992 int at_head, void *data)
994 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
998 * tell I/O scheduler that this isn't a regular read/write (ie it
999 * must not attempt merges on this) and that it acts as a soft
1002 rq->cmd_type = REQ_TYPE_SPECIAL;
1006 spin_lock_irqsave(q->queue_lock, flags);
1009 * If command is tagged, release the tag
1011 if (blk_rq_tagged(rq))
1012 blk_queue_end_tag(q, rq);
1014 drive_stat_acct(rq, 1);
1015 __elv_add_request(q, rq, where, 0);
1017 spin_unlock_irqrestore(q->queue_lock, flags);
1019 EXPORT_SYMBOL(blk_insert_request);
1022 * add-request adds a request to the linked list.
1023 * queue lock is held and interrupts disabled, as we muck with the
1024 * request queue list.
1026 static inline void add_request(struct request_queue *q, struct request *req)
1028 drive_stat_acct(req, 1);
1031 * elevator indicated where it wants this request to be
1032 * inserted at elevator_merge time
1034 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1037 static void part_round_stats_single(int cpu, struct hd_struct *part,
1040 if (now == part->stamp)
1043 if (part->in_flight) {
1044 __part_stat_add(cpu, part, time_in_queue,
1045 part->in_flight * (now - part->stamp));
1046 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1052 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1053 * @cpu: cpu number for stats access
1054 * @part: target partition
1056 * The average IO queue length and utilisation statistics are maintained
1057 * by observing the current state of the queue length and the amount of
1058 * time it has been in this state for.
1060 * Normally, that accounting is done on IO completion, but that can result
1061 * in more than a second's worth of IO being accounted for within any one
1062 * second, leading to >100% utilisation. To deal with that, we call this
1063 * function to do a round-off before returning the results when reading
1064 * /proc/diskstats. This accounts immediately for all queue usage up to
1065 * the current jiffies and restarts the counters again.
1067 void part_round_stats(int cpu, struct hd_struct *part)
1069 unsigned long now = jiffies;
1072 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1073 part_round_stats_single(cpu, part, now);
1075 EXPORT_SYMBOL_GPL(part_round_stats);
1078 * queue lock must be held
1080 void __blk_put_request(struct request_queue *q, struct request *req)
1084 if (unlikely(--req->ref_count))
1087 elv_completed_request(q, req);
1089 /* this is a bio leak */
1090 WARN_ON(req->bio != NULL);
1093 * Request may not have originated from ll_rw_blk. if not,
1094 * it didn't come out of our reserved rq pools
1096 if (req->cmd_flags & REQ_ALLOCED) {
1097 int is_sync = rq_is_sync(req) != 0;
1098 int priv = req->cmd_flags & REQ_ELVPRIV;
1100 BUG_ON(!list_empty(&req->queuelist));
1101 BUG_ON(!hlist_unhashed(&req->hash));
1103 blk_free_request(q, req);
1104 freed_request(q, is_sync, priv);
1107 EXPORT_SYMBOL_GPL(__blk_put_request);
1109 void blk_put_request(struct request *req)
1111 unsigned long flags;
1112 struct request_queue *q = req->q;
1114 spin_lock_irqsave(q->queue_lock, flags);
1115 __blk_put_request(q, req);
1116 spin_unlock_irqrestore(q->queue_lock, flags);
1118 EXPORT_SYMBOL(blk_put_request);
1120 void init_request_from_bio(struct request *req, struct bio *bio)
1122 req->cpu = bio->bi_comp_cpu;
1123 req->cmd_type = REQ_TYPE_FS;
1126 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1128 if (bio_rw_ahead(bio))
1129 req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT |
1130 REQ_FAILFAST_DRIVER);
1131 if (bio_failfast_dev(bio))
1132 req->cmd_flags |= REQ_FAILFAST_DEV;
1133 if (bio_failfast_transport(bio))
1134 req->cmd_flags |= REQ_FAILFAST_TRANSPORT;
1135 if (bio_failfast_driver(bio))
1136 req->cmd_flags |= REQ_FAILFAST_DRIVER;
1138 if (unlikely(bio_discard(bio))) {
1139 req->cmd_flags |= REQ_DISCARD;
1140 if (bio_barrier(bio))
1141 req->cmd_flags |= REQ_SOFTBARRIER;
1142 req->q->prepare_discard_fn(req->q, req);
1143 } else if (unlikely(bio_barrier(bio)))
1144 req->cmd_flags |= REQ_HARDBARRIER;
1147 req->cmd_flags |= REQ_RW_SYNC;
1148 if (bio_rw_meta(bio))
1149 req->cmd_flags |= REQ_RW_META;
1150 if (bio_noidle(bio))
1151 req->cmd_flags |= REQ_NOIDLE;
1154 req->__sector = bio->bi_sector;
1155 req->ioprio = bio_prio(bio);
1156 blk_rq_bio_prep(req->q, req, bio);
1160 * Only disabling plugging for non-rotational devices if it does tagging
1161 * as well, otherwise we do need the proper merging
1163 static inline bool queue_should_plug(struct request_queue *q)
1165 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1168 static int __make_request(struct request_queue *q, struct bio *bio)
1170 struct request *req;
1172 unsigned int bytes = bio->bi_size;
1173 const unsigned short prio = bio_prio(bio);
1174 const int sync = bio_sync(bio);
1175 const int unplug = bio_unplug(bio);
1179 * low level driver can indicate that it wants pages above a
1180 * certain limit bounced to low memory (ie for highmem, or even
1181 * ISA dma in theory)
1183 blk_queue_bounce(q, &bio);
1185 spin_lock_irq(q->queue_lock);
1187 if (unlikely(bio_barrier(bio)) || elv_queue_empty(q))
1190 el_ret = elv_merge(q, &req, bio);
1192 case ELEVATOR_BACK_MERGE:
1193 BUG_ON(!rq_mergeable(req));
1195 if (!ll_back_merge_fn(q, req, bio))
1198 trace_block_bio_backmerge(q, bio);
1200 req->biotail->bi_next = bio;
1202 req->__data_len += bytes;
1203 req->ioprio = ioprio_best(req->ioprio, prio);
1204 if (!blk_rq_cpu_valid(req))
1205 req->cpu = bio->bi_comp_cpu;
1206 drive_stat_acct(req, 0);
1207 if (!attempt_back_merge(q, req))
1208 elv_merged_request(q, req, el_ret);
1211 case ELEVATOR_FRONT_MERGE:
1212 BUG_ON(!rq_mergeable(req));
1214 if (!ll_front_merge_fn(q, req, bio))
1217 trace_block_bio_frontmerge(q, bio);
1219 bio->bi_next = req->bio;
1223 * may not be valid. if the low level driver said
1224 * it didn't need a bounce buffer then it better
1225 * not touch req->buffer either...
1227 req->buffer = bio_data(bio);
1228 req->__sector = bio->bi_sector;
1229 req->__data_len += bytes;
1230 req->ioprio = ioprio_best(req->ioprio, prio);
1231 if (!blk_rq_cpu_valid(req))
1232 req->cpu = bio->bi_comp_cpu;
1233 drive_stat_acct(req, 0);
1234 if (!attempt_front_merge(q, req))
1235 elv_merged_request(q, req, el_ret);
1238 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1245 * This sync check and mask will be re-done in init_request_from_bio(),
1246 * but we need to set it earlier to expose the sync flag to the
1247 * rq allocator and io schedulers.
1249 rw_flags = bio_data_dir(bio);
1251 rw_flags |= REQ_RW_SYNC;
1254 * Grab a free request. This is might sleep but can not fail.
1255 * Returns with the queue unlocked.
1257 req = get_request_wait(q, rw_flags, bio);
1260 * After dropping the lock and possibly sleeping here, our request
1261 * may now be mergeable after it had proven unmergeable (above).
1262 * We don't worry about that case for efficiency. It won't happen
1263 * often, and the elevators are able to handle it.
1265 init_request_from_bio(req, bio);
1267 spin_lock_irq(q->queue_lock);
1268 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1269 bio_flagged(bio, BIO_CPU_AFFINE))
1270 req->cpu = blk_cpu_to_group(smp_processor_id());
1271 if (queue_should_plug(q) && elv_queue_empty(q))
1273 add_request(q, req);
1275 if (unplug || !queue_should_plug(q))
1276 __generic_unplug_device(q);
1277 spin_unlock_irq(q->queue_lock);
1282 * If bio->bi_dev is a partition, remap the location
1284 static inline void blk_partition_remap(struct bio *bio)
1286 struct block_device *bdev = bio->bi_bdev;
1288 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1289 struct hd_struct *p = bdev->bd_part;
1291 bio->bi_sector += p->start_sect;
1292 bio->bi_bdev = bdev->bd_contains;
1294 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1295 bdev->bd_dev, bio->bi_sector,
1296 bio->bi_sector - p->start_sect);
1300 static void handle_bad_sector(struct bio *bio)
1302 char b[BDEVNAME_SIZE];
1304 printk(KERN_INFO "attempt to access beyond end of device\n");
1305 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1306 bdevname(bio->bi_bdev, b),
1308 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1309 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1311 set_bit(BIO_EOF, &bio->bi_flags);
1314 #ifdef CONFIG_FAIL_MAKE_REQUEST
1316 static DECLARE_FAULT_ATTR(fail_make_request);
1318 static int __init setup_fail_make_request(char *str)
1320 return setup_fault_attr(&fail_make_request, str);
1322 __setup("fail_make_request=", setup_fail_make_request);
1324 static int should_fail_request(struct bio *bio)
1326 struct hd_struct *part = bio->bi_bdev->bd_part;
1328 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1329 return should_fail(&fail_make_request, bio->bi_size);
1334 static int __init fail_make_request_debugfs(void)
1336 return init_fault_attr_dentries(&fail_make_request,
1337 "fail_make_request");
1340 late_initcall(fail_make_request_debugfs);
1342 #else /* CONFIG_FAIL_MAKE_REQUEST */
1344 static inline int should_fail_request(struct bio *bio)
1349 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1352 * Check whether this bio extends beyond the end of the device.
1354 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1361 /* Test device or partition size, when known. */
1362 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1364 sector_t sector = bio->bi_sector;
1366 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1368 * This may well happen - the kernel calls bread()
1369 * without checking the size of the device, e.g., when
1370 * mounting a device.
1372 handle_bad_sector(bio);
1381 * generic_make_request - hand a buffer to its device driver for I/O
1382 * @bio: The bio describing the location in memory and on the device.
1384 * generic_make_request() is used to make I/O requests of block
1385 * devices. It is passed a &struct bio, which describes the I/O that needs
1388 * generic_make_request() does not return any status. The
1389 * success/failure status of the request, along with notification of
1390 * completion, is delivered asynchronously through the bio->bi_end_io
1391 * function described (one day) else where.
1393 * The caller of generic_make_request must make sure that bi_io_vec
1394 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1395 * set to describe the device address, and the
1396 * bi_end_io and optionally bi_private are set to describe how
1397 * completion notification should be signaled.
1399 * generic_make_request and the drivers it calls may use bi_next if this
1400 * bio happens to be merged with someone else, and may change bi_dev and
1401 * bi_sector for remaps as it sees fit. So the values of these fields
1402 * should NOT be depended on after the call to generic_make_request.
1404 static inline void __generic_make_request(struct bio *bio)
1406 struct request_queue *q;
1407 sector_t old_sector;
1408 int ret, nr_sectors = bio_sectors(bio);
1414 if (bio_check_eod(bio, nr_sectors))
1418 * Resolve the mapping until finished. (drivers are
1419 * still free to implement/resolve their own stacking
1420 * by explicitly returning 0)
1422 * NOTE: we don't repeat the blk_size check for each new device.
1423 * Stacking drivers are expected to know what they are doing.
1428 char b[BDEVNAME_SIZE];
1430 q = bdev_get_queue(bio->bi_bdev);
1433 "generic_make_request: Trying to access "
1434 "nonexistent block-device %s (%Lu)\n",
1435 bdevname(bio->bi_bdev, b),
1436 (long long) bio->bi_sector);
1440 if (unlikely(nr_sectors > queue_max_hw_sectors(q))) {
1441 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1442 bdevname(bio->bi_bdev, b),
1444 queue_max_hw_sectors(q));
1448 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1451 if (should_fail_request(bio))
1455 * If this device has partitions, remap block n
1456 * of partition p to block n+start(p) of the disk.
1458 blk_partition_remap(bio);
1460 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1463 if (old_sector != -1)
1464 trace_block_remap(q, bio, old_dev, bio->bi_sector,
1467 trace_block_bio_queue(q, bio);
1469 old_sector = bio->bi_sector;
1470 old_dev = bio->bi_bdev->bd_dev;
1472 if (bio_check_eod(bio, nr_sectors))
1475 if (bio_discard(bio) && !q->prepare_discard_fn) {
1479 if (bio_barrier(bio) && bio_has_data(bio) &&
1480 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1485 ret = q->make_request_fn(q, bio);
1491 bio_endio(bio, err);
1495 * We only want one ->make_request_fn to be active at a time,
1496 * else stack usage with stacked devices could be a problem.
1497 * So use current->bio_{list,tail} to keep a list of requests
1498 * submited by a make_request_fn function.
1499 * current->bio_tail is also used as a flag to say if
1500 * generic_make_request is currently active in this task or not.
1501 * If it is NULL, then no make_request is active. If it is non-NULL,
1502 * then a make_request is active, and new requests should be added
1505 void generic_make_request(struct bio *bio)
1507 if (current->bio_tail) {
1508 /* make_request is active */
1509 *(current->bio_tail) = bio;
1510 bio->bi_next = NULL;
1511 current->bio_tail = &bio->bi_next;
1514 /* following loop may be a bit non-obvious, and so deserves some
1516 * Before entering the loop, bio->bi_next is NULL (as all callers
1517 * ensure that) so we have a list with a single bio.
1518 * We pretend that we have just taken it off a longer list, so
1519 * we assign bio_list to the next (which is NULL) and bio_tail
1520 * to &bio_list, thus initialising the bio_list of new bios to be
1521 * added. __generic_make_request may indeed add some more bios
1522 * through a recursive call to generic_make_request. If it
1523 * did, we find a non-NULL value in bio_list and re-enter the loop
1524 * from the top. In this case we really did just take the bio
1525 * of the top of the list (no pretending) and so fixup bio_list and
1526 * bio_tail or bi_next, and call into __generic_make_request again.
1528 * The loop was structured like this to make only one call to
1529 * __generic_make_request (which is important as it is large and
1530 * inlined) and to keep the structure simple.
1532 BUG_ON(bio->bi_next);
1534 current->bio_list = bio->bi_next;
1535 if (bio->bi_next == NULL)
1536 current->bio_tail = ¤t->bio_list;
1538 bio->bi_next = NULL;
1539 __generic_make_request(bio);
1540 bio = current->bio_list;
1542 current->bio_tail = NULL; /* deactivate */
1544 EXPORT_SYMBOL(generic_make_request);
1547 * submit_bio - submit a bio to the block device layer for I/O
1548 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1549 * @bio: The &struct bio which describes the I/O
1551 * submit_bio() is very similar in purpose to generic_make_request(), and
1552 * uses that function to do most of the work. Both are fairly rough
1553 * interfaces; @bio must be presetup and ready for I/O.
1556 void submit_bio(int rw, struct bio *bio)
1558 int count = bio_sectors(bio);
1563 * If it's a regular read/write or a barrier with data attached,
1564 * go through the normal accounting stuff before submission.
1566 if (bio_has_data(bio)) {
1568 count_vm_events(PGPGOUT, count);
1570 task_io_account_read(bio->bi_size);
1571 count_vm_events(PGPGIN, count);
1574 if (unlikely(block_dump)) {
1575 char b[BDEVNAME_SIZE];
1576 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1577 current->comm, task_pid_nr(current),
1578 (rw & WRITE) ? "WRITE" : "READ",
1579 (unsigned long long)bio->bi_sector,
1580 bdevname(bio->bi_bdev, b));
1584 generic_make_request(bio);
1586 EXPORT_SYMBOL(submit_bio);
1589 * blk_rq_check_limits - Helper function to check a request for the queue limit
1591 * @rq: the request being checked
1594 * @rq may have been made based on weaker limitations of upper-level queues
1595 * in request stacking drivers, and it may violate the limitation of @q.
1596 * Since the block layer and the underlying device driver trust @rq
1597 * after it is inserted to @q, it should be checked against @q before
1598 * the insertion using this generic function.
1600 * This function should also be useful for request stacking drivers
1601 * in some cases below, so export this fuction.
1602 * Request stacking drivers like request-based dm may change the queue
1603 * limits while requests are in the queue (e.g. dm's table swapping).
1604 * Such request stacking drivers should check those requests agaist
1605 * the new queue limits again when they dispatch those requests,
1606 * although such checkings are also done against the old queue limits
1607 * when submitting requests.
1609 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1611 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1612 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1613 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1618 * queue's settings related to segment counting like q->bounce_pfn
1619 * may differ from that of other stacking queues.
1620 * Recalculate it to check the request correctly on this queue's
1623 blk_recalc_rq_segments(rq);
1624 if (rq->nr_phys_segments > queue_max_phys_segments(q) ||
1625 rq->nr_phys_segments > queue_max_hw_segments(q)) {
1626 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1632 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1635 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1636 * @q: the queue to submit the request
1637 * @rq: the request being queued
1639 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1641 unsigned long flags;
1643 if (blk_rq_check_limits(q, rq))
1646 #ifdef CONFIG_FAIL_MAKE_REQUEST
1647 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1648 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1652 spin_lock_irqsave(q->queue_lock, flags);
1655 * Submitting request must be dequeued before calling this function
1656 * because it will be linked to another request_queue
1658 BUG_ON(blk_queued_rq(rq));
1660 drive_stat_acct(rq, 1);
1661 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1663 spin_unlock_irqrestore(q->queue_lock, flags);
1667 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1669 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1671 if (blk_do_io_stat(req)) {
1672 const int rw = rq_data_dir(req);
1673 struct hd_struct *part;
1676 cpu = part_stat_lock();
1677 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1678 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1683 static void blk_account_io_done(struct request *req)
1686 * Account IO completion. bar_rq isn't accounted as a normal
1687 * IO on queueing nor completion. Accounting the containing
1688 * request is enough.
1690 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
1691 unsigned long duration = jiffies - req->start_time;
1692 const int rw = rq_data_dir(req);
1693 struct hd_struct *part;
1696 cpu = part_stat_lock();
1697 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1699 part_stat_inc(cpu, part, ios[rw]);
1700 part_stat_add(cpu, part, ticks[rw], duration);
1701 part_round_stats(cpu, part);
1702 part_dec_in_flight(part);
1709 * blk_peek_request - peek at the top of a request queue
1710 * @q: request queue to peek at
1713 * Return the request at the top of @q. The returned request
1714 * should be started using blk_start_request() before LLD starts
1718 * Pointer to the request at the top of @q if available. Null
1722 * queue_lock must be held.
1724 struct request *blk_peek_request(struct request_queue *q)
1729 while ((rq = __elv_next_request(q)) != NULL) {
1730 if (!(rq->cmd_flags & REQ_STARTED)) {
1732 * This is the first time the device driver
1733 * sees this request (possibly after
1734 * requeueing). Notify IO scheduler.
1736 if (blk_sorted_rq(rq))
1737 elv_activate_rq(q, rq);
1740 * just mark as started even if we don't start
1741 * it, a request that has been delayed should
1742 * not be passed by new incoming requests
1744 rq->cmd_flags |= REQ_STARTED;
1745 trace_block_rq_issue(q, rq);
1748 if (!q->boundary_rq || q->boundary_rq == rq) {
1749 q->end_sector = rq_end_sector(rq);
1750 q->boundary_rq = NULL;
1753 if (rq->cmd_flags & REQ_DONTPREP)
1756 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1758 * make sure space for the drain appears we
1759 * know we can do this because max_hw_segments
1760 * has been adjusted to be one fewer than the
1763 rq->nr_phys_segments++;
1769 ret = q->prep_rq_fn(q, rq);
1770 if (ret == BLKPREP_OK) {
1772 } else if (ret == BLKPREP_DEFER) {
1774 * the request may have been (partially) prepped.
1775 * we need to keep this request in the front to
1776 * avoid resource deadlock. REQ_STARTED will
1777 * prevent other fs requests from passing this one.
1779 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1780 !(rq->cmd_flags & REQ_DONTPREP)) {
1782 * remove the space for the drain we added
1783 * so that we don't add it again
1785 --rq->nr_phys_segments;
1790 } else if (ret == BLKPREP_KILL) {
1791 rq->cmd_flags |= REQ_QUIET;
1793 * Mark this request as started so we don't trigger
1794 * any debug logic in the end I/O path.
1796 blk_start_request(rq);
1797 __blk_end_request_all(rq, -EIO);
1799 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1806 EXPORT_SYMBOL(blk_peek_request);
1808 void blk_dequeue_request(struct request *rq)
1810 struct request_queue *q = rq->q;
1812 BUG_ON(list_empty(&rq->queuelist));
1813 BUG_ON(ELV_ON_HASH(rq));
1815 list_del_init(&rq->queuelist);
1818 * the time frame between a request being removed from the lists
1819 * and to it is freed is accounted as io that is in progress at
1822 if (blk_account_rq(rq))
1823 q->in_flight[rq_is_sync(rq)]++;
1827 * blk_start_request - start request processing on the driver
1828 * @req: request to dequeue
1831 * Dequeue @req and start timeout timer on it. This hands off the
1832 * request to the driver.
1834 * Block internal functions which don't want to start timer should
1835 * call blk_dequeue_request().
1838 * queue_lock must be held.
1840 void blk_start_request(struct request *req)
1842 blk_dequeue_request(req);
1845 * We are now handing the request to the hardware, initialize
1846 * resid_len to full count and add the timeout handler.
1848 req->resid_len = blk_rq_bytes(req);
1849 if (unlikely(blk_bidi_rq(req)))
1850 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1854 EXPORT_SYMBOL(blk_start_request);
1857 * blk_fetch_request - fetch a request from a request queue
1858 * @q: request queue to fetch a request from
1861 * Return the request at the top of @q. The request is started on
1862 * return and LLD can start processing it immediately.
1865 * Pointer to the request at the top of @q if available. Null
1869 * queue_lock must be held.
1871 struct request *blk_fetch_request(struct request_queue *q)
1875 rq = blk_peek_request(q);
1877 blk_start_request(rq);
1880 EXPORT_SYMBOL(blk_fetch_request);
1883 * blk_update_request - Special helper function for request stacking drivers
1884 * @rq: the request being processed
1885 * @error: %0 for success, < %0 for error
1886 * @nr_bytes: number of bytes to complete @rq
1889 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
1890 * the request structure even if @rq doesn't have leftover.
1891 * If @rq has leftover, sets it up for the next range of segments.
1893 * This special helper function is only for request stacking drivers
1894 * (e.g. request-based dm) so that they can handle partial completion.
1895 * Actual device drivers should use blk_end_request instead.
1897 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1898 * %false return from this function.
1901 * %false - this request doesn't have any more data
1902 * %true - this request has more data
1904 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
1906 int total_bytes, bio_nbytes, next_idx = 0;
1912 trace_block_rq_complete(req->q, req);
1915 * For fs requests, rq is just carrier of independent bio's
1916 * and each partial completion should be handled separately.
1917 * Reset per-request error on each partial completion.
1919 * TODO: tj: This is too subtle. It would be better to let
1920 * low level drivers do what they see fit.
1922 if (blk_fs_request(req))
1925 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1926 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1927 req->rq_disk ? req->rq_disk->disk_name : "?",
1928 (unsigned long long)blk_rq_pos(req));
1931 blk_account_io_completion(req, nr_bytes);
1933 total_bytes = bio_nbytes = 0;
1934 while ((bio = req->bio) != NULL) {
1937 if (nr_bytes >= bio->bi_size) {
1938 req->bio = bio->bi_next;
1939 nbytes = bio->bi_size;
1940 req_bio_endio(req, bio, nbytes, error);
1944 int idx = bio->bi_idx + next_idx;
1946 if (unlikely(idx >= bio->bi_vcnt)) {
1947 blk_dump_rq_flags(req, "__end_that");
1948 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1949 __func__, idx, bio->bi_vcnt);
1953 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1954 BIO_BUG_ON(nbytes > bio->bi_size);
1957 * not a complete bvec done
1959 if (unlikely(nbytes > nr_bytes)) {
1960 bio_nbytes += nr_bytes;
1961 total_bytes += nr_bytes;
1966 * advance to the next vector
1969 bio_nbytes += nbytes;
1972 total_bytes += nbytes;
1978 * end more in this run, or just return 'not-done'
1980 if (unlikely(nr_bytes <= 0))
1990 * Reset counters so that the request stacking driver
1991 * can find how many bytes remain in the request
1994 req->__data_len = 0;
1999 * if the request wasn't completed, update state
2002 req_bio_endio(req, bio, bio_nbytes, error);
2003 bio->bi_idx += next_idx;
2004 bio_iovec(bio)->bv_offset += nr_bytes;
2005 bio_iovec(bio)->bv_len -= nr_bytes;
2008 req->__data_len -= total_bytes;
2009 req->buffer = bio_data(req->bio);
2011 /* update sector only for requests with clear definition of sector */
2012 if (blk_fs_request(req) || blk_discard_rq(req))
2013 req->__sector += total_bytes >> 9;
2016 * If total number of sectors is less than the first segment
2017 * size, something has gone terribly wrong.
2019 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2020 printk(KERN_ERR "blk: request botched\n");
2021 req->__data_len = blk_rq_cur_bytes(req);
2024 /* recalculate the number of segments */
2025 blk_recalc_rq_segments(req);
2029 EXPORT_SYMBOL_GPL(blk_update_request);
2031 static bool blk_update_bidi_request(struct request *rq, int error,
2032 unsigned int nr_bytes,
2033 unsigned int bidi_bytes)
2035 if (blk_update_request(rq, error, nr_bytes))
2038 /* Bidi request must be completed as a whole */
2039 if (unlikely(blk_bidi_rq(rq)) &&
2040 blk_update_request(rq->next_rq, error, bidi_bytes))
2043 add_disk_randomness(rq->rq_disk);
2049 * queue lock must be held
2051 static void blk_finish_request(struct request *req, int error)
2053 if (blk_rq_tagged(req))
2054 blk_queue_end_tag(req->q, req);
2056 BUG_ON(blk_queued_rq(req));
2058 if (unlikely(laptop_mode) && blk_fs_request(req))
2059 laptop_io_completion();
2061 blk_delete_timer(req);
2063 blk_account_io_done(req);
2066 req->end_io(req, error);
2068 if (blk_bidi_rq(req))
2069 __blk_put_request(req->next_rq->q, req->next_rq);
2071 __blk_put_request(req->q, req);
2076 * blk_end_bidi_request - Complete a bidi request
2077 * @rq: the request to complete
2078 * @error: %0 for success, < %0 for error
2079 * @nr_bytes: number of bytes to complete @rq
2080 * @bidi_bytes: number of bytes to complete @rq->next_rq
2083 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2084 * Drivers that supports bidi can safely call this member for any
2085 * type of request, bidi or uni. In the later case @bidi_bytes is
2089 * %false - we are done with this request
2090 * %true - still buffers pending for this request
2092 static bool blk_end_bidi_request(struct request *rq, int error,
2093 unsigned int nr_bytes, unsigned int bidi_bytes)
2095 struct request_queue *q = rq->q;
2096 unsigned long flags;
2098 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2101 spin_lock_irqsave(q->queue_lock, flags);
2102 blk_finish_request(rq, error);
2103 spin_unlock_irqrestore(q->queue_lock, flags);
2109 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2110 * @rq: the request to complete
2111 * @error: %0 for success, < %0 for error
2112 * @nr_bytes: number of bytes to complete @rq
2113 * @bidi_bytes: number of bytes to complete @rq->next_rq
2116 * Identical to blk_end_bidi_request() except that queue lock is
2117 * assumed to be locked on entry and remains so on return.
2120 * %false - we are done with this request
2121 * %true - still buffers pending for this request
2123 static bool __blk_end_bidi_request(struct request *rq, int error,
2124 unsigned int nr_bytes, unsigned int bidi_bytes)
2126 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2129 blk_finish_request(rq, error);
2135 * blk_end_request - Helper function for drivers to complete the request.
2136 * @rq: the request being processed
2137 * @error: %0 for success, < %0 for error
2138 * @nr_bytes: number of bytes to complete
2141 * Ends I/O on a number of bytes attached to @rq.
2142 * If @rq has leftover, sets it up for the next range of segments.
2145 * %false - we are done with this request
2146 * %true - still buffers pending for this request
2148 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2150 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2152 EXPORT_SYMBOL_GPL(blk_end_request);
2155 * blk_end_request_all - Helper function for drives to finish the request.
2156 * @rq: the request to finish
2157 * @err: %0 for success, < %0 for error
2160 * Completely finish @rq.
2162 void blk_end_request_all(struct request *rq, int error)
2165 unsigned int bidi_bytes = 0;
2167 if (unlikely(blk_bidi_rq(rq)))
2168 bidi_bytes = blk_rq_bytes(rq->next_rq);
2170 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2173 EXPORT_SYMBOL_GPL(blk_end_request_all);
2176 * blk_end_request_cur - Helper function to finish the current request chunk.
2177 * @rq: the request to finish the current chunk for
2178 * @err: %0 for success, < %0 for error
2181 * Complete the current consecutively mapped chunk from @rq.
2184 * %false - we are done with this request
2185 * %true - still buffers pending for this request
2187 bool blk_end_request_cur(struct request *rq, int error)
2189 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2191 EXPORT_SYMBOL_GPL(blk_end_request_cur);
2194 * __blk_end_request - Helper function for drivers to complete the request.
2195 * @rq: the request being processed
2196 * @error: %0 for success, < %0 for error
2197 * @nr_bytes: number of bytes to complete
2200 * Must be called with queue lock held unlike blk_end_request().
2203 * %false - we are done with this request
2204 * %true - still buffers pending for this request
2206 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2208 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2210 EXPORT_SYMBOL_GPL(__blk_end_request);
2213 * __blk_end_request_all - Helper function for drives to finish the request.
2214 * @rq: the request to finish
2215 * @err: %0 for success, < %0 for error
2218 * Completely finish @rq. Must be called with queue lock held.
2220 void __blk_end_request_all(struct request *rq, int error)
2223 unsigned int bidi_bytes = 0;
2225 if (unlikely(blk_bidi_rq(rq)))
2226 bidi_bytes = blk_rq_bytes(rq->next_rq);
2228 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2231 EXPORT_SYMBOL_GPL(__blk_end_request_all);
2234 * __blk_end_request_cur - Helper function to finish the current request chunk.
2235 * @rq: the request to finish the current chunk for
2236 * @err: %0 for success, < %0 for error
2239 * Complete the current consecutively mapped chunk from @rq. Must
2240 * be called with queue lock held.
2243 * %false - we are done with this request
2244 * %true - still buffers pending for this request
2246 bool __blk_end_request_cur(struct request *rq, int error)
2248 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2250 EXPORT_SYMBOL_GPL(__blk_end_request_cur);
2252 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2255 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2256 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2257 rq->cmd_flags |= (bio->bi_rw & 3);
2259 if (bio_has_data(bio)) {
2260 rq->nr_phys_segments = bio_phys_segments(q, bio);
2261 rq->buffer = bio_data(bio);
2263 rq->__data_len = bio->bi_size;
2264 rq->bio = rq->biotail = bio;
2267 rq->rq_disk = bio->bi_bdev->bd_disk;
2271 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2272 * @q : the queue of the device being checked
2275 * Check if underlying low-level drivers of a device are busy.
2276 * If the drivers want to export their busy state, they must set own
2277 * exporting function using blk_queue_lld_busy() first.
2279 * Basically, this function is used only by request stacking drivers
2280 * to stop dispatching requests to underlying devices when underlying
2281 * devices are busy. This behavior helps more I/O merging on the queue
2282 * of the request stacking driver and prevents I/O throughput regression
2283 * on burst I/O load.
2286 * 0 - Not busy (The request stacking driver should dispatch request)
2287 * 1 - Busy (The request stacking driver should stop dispatching request)
2289 int blk_lld_busy(struct request_queue *q)
2292 return q->lld_busy_fn(q);
2296 EXPORT_SYMBOL_GPL(blk_lld_busy);
2299 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2300 * @rq: the clone request to be cleaned up
2303 * Free all bios in @rq for a cloned request.
2305 void blk_rq_unprep_clone(struct request *rq)
2309 while ((bio = rq->bio) != NULL) {
2310 rq->bio = bio->bi_next;
2315 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2318 * Copy attributes of the original request to the clone request.
2319 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2321 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2323 dst->cpu = src->cpu;
2324 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2325 dst->cmd_type = src->cmd_type;
2326 dst->__sector = blk_rq_pos(src);
2327 dst->__data_len = blk_rq_bytes(src);
2328 dst->nr_phys_segments = src->nr_phys_segments;
2329 dst->ioprio = src->ioprio;
2330 dst->extra_len = src->extra_len;
2334 * blk_rq_prep_clone - Helper function to setup clone request
2335 * @rq: the request to be setup
2336 * @rq_src: original request to be cloned
2337 * @bs: bio_set that bios for clone are allocated from
2338 * @gfp_mask: memory allocation mask for bio
2339 * @bio_ctr: setup function to be called for each clone bio.
2340 * Returns %0 for success, non %0 for failure.
2341 * @data: private data to be passed to @bio_ctr
2344 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2345 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2346 * are not copied, and copying such parts is the caller's responsibility.
2347 * Also, pages which the original bios are pointing to are not copied
2348 * and the cloned bios just point same pages.
2349 * So cloned bios must be completed before original bios, which means
2350 * the caller must complete @rq before @rq_src.
2352 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2353 struct bio_set *bs, gfp_t gfp_mask,
2354 int (*bio_ctr)(struct bio *, struct bio *, void *),
2357 struct bio *bio, *bio_src;
2362 blk_rq_init(NULL, rq);
2364 __rq_for_each_bio(bio_src, rq_src) {
2365 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2369 __bio_clone(bio, bio_src);
2371 if (bio_integrity(bio_src) &&
2372 bio_integrity_clone(bio, bio_src, gfp_mask))
2375 if (bio_ctr && bio_ctr(bio, bio_src, data))
2379 rq->biotail->bi_next = bio;
2382 rq->bio = rq->biotail = bio;
2385 __blk_rq_prep_clone(rq, rq_src);
2392 blk_rq_unprep_clone(rq);
2396 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2398 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2400 return queue_work(kblockd_workqueue, work);
2402 EXPORT_SYMBOL(kblockd_schedule_work);
2404 int __init blk_dev_init(void)
2406 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2407 sizeof(((struct request *)0)->cmd_flags));
2409 kblockd_workqueue = create_workqueue("kblockd");
2410 if (!kblockd_workqueue)
2411 panic("Failed to create kblockd\n");
2413 request_cachep = kmem_cache_create("blkdev_requests",
2414 sizeof(struct request), 0, SLAB_PANIC, NULL);
2416 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2417 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);