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/fault-inject.h>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/block.h>
36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
40 static int __make_request(struct request_queue *q, struct bio *bio);
43 * For the allocated request tables
45 static struct kmem_cache *request_cachep;
48 * For queue allocation
50 struct kmem_cache *blk_requestq_cachep;
53 * Controlling structure to kblockd
55 static struct workqueue_struct *kblockd_workqueue;
57 static void drive_stat_acct(struct request *rq, int new_io)
59 struct hd_struct *part;
60 int rw = rq_data_dir(rq);
63 if (!blk_do_io_stat(rq))
66 cpu = part_stat_lock();
67 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
70 part_stat_inc(cpu, part, merges[rw]);
72 part_round_stats(cpu, part);
73 part_inc_in_flight(part, rw);
79 void blk_queue_congestion_threshold(struct request_queue *q)
83 nr = q->nr_requests - (q->nr_requests / 8) + 1;
84 if (nr > q->nr_requests)
86 q->nr_congestion_on = nr;
88 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
91 q->nr_congestion_off = nr;
95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
98 * Locates the passed device's request queue and returns the address of its
101 * Will return NULL if the request queue cannot be located.
103 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
105 struct backing_dev_info *ret = NULL;
106 struct request_queue *q = bdev_get_queue(bdev);
109 ret = &q->backing_dev_info;
112 EXPORT_SYMBOL(blk_get_backing_dev_info);
114 void blk_rq_init(struct request_queue *q, struct request *rq)
116 memset(rq, 0, sizeof(*rq));
118 INIT_LIST_HEAD(&rq->queuelist);
119 INIT_LIST_HEAD(&rq->timeout_list);
122 rq->__sector = (sector_t) -1;
123 INIT_HLIST_NODE(&rq->hash);
124 RB_CLEAR_NODE(&rq->rb_node);
126 rq->cmd_len = BLK_MAX_CDB;
129 rq->start_time = jiffies;
131 EXPORT_SYMBOL(blk_rq_init);
133 static void req_bio_endio(struct request *rq, struct bio *bio,
134 unsigned int nbytes, int error)
136 struct request_queue *q = rq->q;
138 if (&q->bar_rq != rq) {
140 clear_bit(BIO_UPTODATE, &bio->bi_flags);
141 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
144 if (unlikely(nbytes > bio->bi_size)) {
145 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
146 __func__, nbytes, bio->bi_size);
147 nbytes = bio->bi_size;
150 if (unlikely(rq->cmd_flags & REQ_QUIET))
151 set_bit(BIO_QUIET, &bio->bi_flags);
153 bio->bi_size -= nbytes;
154 bio->bi_sector += (nbytes >> 9);
156 if (bio_integrity(bio))
157 bio_integrity_advance(bio, nbytes);
159 if (bio->bi_size == 0)
160 bio_endio(bio, error);
164 * Okay, this is the barrier request in progress, just
167 if (error && !q->orderr)
172 void blk_dump_rq_flags(struct request *rq, char *msg)
176 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
177 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
180 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
181 (unsigned long long)blk_rq_pos(rq),
182 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
183 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
184 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
186 if (blk_pc_request(rq)) {
187 printk(KERN_INFO " cdb: ");
188 for (bit = 0; bit < BLK_MAX_CDB; bit++)
189 printk("%02x ", rq->cmd[bit]);
193 EXPORT_SYMBOL(blk_dump_rq_flags);
196 * "plug" the device if there are no outstanding requests: this will
197 * force the transfer to start only after we have put all the requests
200 * This is called with interrupts off and no requests on the queue and
201 * with the queue lock held.
203 void blk_plug_device(struct request_queue *q)
205 WARN_ON(!irqs_disabled());
208 * don't plug a stopped queue, it must be paired with blk_start_queue()
209 * which will restart the queueing
211 if (blk_queue_stopped(q))
214 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
215 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
219 EXPORT_SYMBOL(blk_plug_device);
222 * blk_plug_device_unlocked - plug a device without queue lock held
223 * @q: The &struct request_queue to plug
226 * Like @blk_plug_device(), but grabs the queue lock and disables
229 void blk_plug_device_unlocked(struct request_queue *q)
233 spin_lock_irqsave(q->queue_lock, flags);
235 spin_unlock_irqrestore(q->queue_lock, flags);
237 EXPORT_SYMBOL(blk_plug_device_unlocked);
240 * remove the queue from the plugged list, if present. called with
241 * queue lock held and interrupts disabled.
243 int blk_remove_plug(struct request_queue *q)
245 WARN_ON(!irqs_disabled());
247 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
250 del_timer(&q->unplug_timer);
253 EXPORT_SYMBOL(blk_remove_plug);
256 * remove the plug and let it rip..
258 void __generic_unplug_device(struct request_queue *q)
260 if (unlikely(blk_queue_stopped(q)))
262 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
269 * generic_unplug_device - fire a request queue
270 * @q: The &struct request_queue in question
273 * Linux uses plugging to build bigger requests queues before letting
274 * the device have at them. If a queue is plugged, the I/O scheduler
275 * is still adding and merging requests on the queue. Once the queue
276 * gets unplugged, the request_fn defined for the queue is invoked and
279 void generic_unplug_device(struct request_queue *q)
281 if (blk_queue_plugged(q)) {
282 spin_lock_irq(q->queue_lock);
283 __generic_unplug_device(q);
284 spin_unlock_irq(q->queue_lock);
287 EXPORT_SYMBOL(generic_unplug_device);
289 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
292 struct request_queue *q = bdi->unplug_io_data;
297 void blk_unplug_work(struct work_struct *work)
299 struct request_queue *q =
300 container_of(work, struct request_queue, unplug_work);
302 trace_block_unplug_io(q);
306 void blk_unplug_timeout(unsigned long data)
308 struct request_queue *q = (struct request_queue *)data;
310 trace_block_unplug_timer(q);
311 kblockd_schedule_work(q, &q->unplug_work);
314 void blk_unplug(struct request_queue *q)
317 * devices don't necessarily have an ->unplug_fn defined
320 trace_block_unplug_io(q);
324 EXPORT_SYMBOL(blk_unplug);
327 * blk_start_queue - restart a previously stopped queue
328 * @q: The &struct request_queue in question
331 * blk_start_queue() will clear the stop flag on the queue, and call
332 * the request_fn for the queue if it was in a stopped state when
333 * entered. Also see blk_stop_queue(). Queue lock must be held.
335 void blk_start_queue(struct request_queue *q)
337 WARN_ON(!irqs_disabled());
339 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
342 EXPORT_SYMBOL(blk_start_queue);
345 * blk_stop_queue - stop a queue
346 * @q: The &struct request_queue in question
349 * The Linux block layer assumes that a block driver will consume all
350 * entries on the request queue when the request_fn strategy is called.
351 * Often this will not happen, because of hardware limitations (queue
352 * depth settings). If a device driver gets a 'queue full' response,
353 * or if it simply chooses not to queue more I/O at one point, it can
354 * call this function to prevent the request_fn from being called until
355 * the driver has signalled it's ready to go again. This happens by calling
356 * blk_start_queue() to restart queue operations. Queue lock must be held.
358 void blk_stop_queue(struct request_queue *q)
361 queue_flag_set(QUEUE_FLAG_STOPPED, q);
363 EXPORT_SYMBOL(blk_stop_queue);
366 * blk_sync_queue - cancel any pending callbacks on a queue
370 * The block layer may perform asynchronous callback activity
371 * on a queue, such as calling the unplug function after a timeout.
372 * A block device may call blk_sync_queue to ensure that any
373 * such activity is cancelled, thus allowing it to release resources
374 * that the callbacks might use. The caller must already have made sure
375 * that its ->make_request_fn will not re-add plugging prior to calling
379 void blk_sync_queue(struct request_queue *q)
381 del_timer_sync(&q->unplug_timer);
382 del_timer_sync(&q->timeout);
383 cancel_work_sync(&q->unplug_work);
385 EXPORT_SYMBOL(blk_sync_queue);
388 * __blk_run_queue - run a single device queue
389 * @q: The queue to run
392 * See @blk_run_queue. This variant must be called with the queue lock
393 * held and interrupts disabled.
396 void __blk_run_queue(struct request_queue *q)
400 if (unlikely(blk_queue_stopped(q)))
403 if (elv_queue_empty(q))
407 * Only recurse once to avoid overrunning the stack, let the unplug
408 * handling reinvoke the handler shortly if we already got there.
410 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
412 queue_flag_clear(QUEUE_FLAG_REENTER, q);
414 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
415 kblockd_schedule_work(q, &q->unplug_work);
418 EXPORT_SYMBOL(__blk_run_queue);
421 * blk_run_queue - run a single device queue
422 * @q: The queue to run
425 * Invoke request handling on this queue, if it has pending work to do.
426 * May be used to restart queueing when a request has completed.
428 void blk_run_queue(struct request_queue *q)
432 spin_lock_irqsave(q->queue_lock, flags);
434 spin_unlock_irqrestore(q->queue_lock, flags);
436 EXPORT_SYMBOL(blk_run_queue);
438 void blk_put_queue(struct request_queue *q)
440 kobject_put(&q->kobj);
443 void blk_cleanup_queue(struct request_queue *q)
446 * We know we have process context here, so we can be a little
447 * cautious and ensure that pending block actions on this device
448 * are done before moving on. Going into this function, we should
449 * not have processes doing IO to this device.
453 mutex_lock(&q->sysfs_lock);
454 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
455 mutex_unlock(&q->sysfs_lock);
458 elevator_exit(q->elevator);
462 EXPORT_SYMBOL(blk_cleanup_queue);
464 static int blk_init_free_list(struct request_queue *q)
466 struct request_list *rl = &q->rq;
468 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
469 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
471 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
472 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
474 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
475 mempool_free_slab, request_cachep, q->node);
483 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
485 return blk_alloc_queue_node(gfp_mask, -1);
487 EXPORT_SYMBOL(blk_alloc_queue);
489 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
491 struct request_queue *q;
494 q = kmem_cache_alloc_node(blk_requestq_cachep,
495 gfp_mask | __GFP_ZERO, node_id);
499 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
500 q->backing_dev_info.unplug_io_data = q;
501 q->backing_dev_info.ra_pages =
502 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
503 q->backing_dev_info.state = 0;
504 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
505 q->backing_dev_info.name = "block";
507 err = bdi_init(&q->backing_dev_info);
509 kmem_cache_free(blk_requestq_cachep, q);
513 init_timer(&q->unplug_timer);
514 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
515 INIT_LIST_HEAD(&q->timeout_list);
516 INIT_WORK(&q->unplug_work, blk_unplug_work);
518 kobject_init(&q->kobj, &blk_queue_ktype);
520 mutex_init(&q->sysfs_lock);
521 spin_lock_init(&q->__queue_lock);
525 EXPORT_SYMBOL(blk_alloc_queue_node);
528 * blk_init_queue - prepare a request queue for use with a block device
529 * @rfn: The function to be called to process requests that have been
530 * placed on the queue.
531 * @lock: Request queue spin lock
534 * If a block device wishes to use the standard request handling procedures,
535 * which sorts requests and coalesces adjacent requests, then it must
536 * call blk_init_queue(). The function @rfn will be called when there
537 * are requests on the queue that need to be processed. If the device
538 * supports plugging, then @rfn may not be called immediately when requests
539 * are available on the queue, but may be called at some time later instead.
540 * Plugged queues are generally unplugged when a buffer belonging to one
541 * of the requests on the queue is needed, or due to memory pressure.
543 * @rfn is not required, or even expected, to remove all requests off the
544 * queue, but only as many as it can handle at a time. If it does leave
545 * requests on the queue, it is responsible for arranging that the requests
546 * get dealt with eventually.
548 * The queue spin lock must be held while manipulating the requests on the
549 * request queue; this lock will be taken also from interrupt context, so irq
550 * disabling is needed for it.
552 * Function returns a pointer to the initialized request queue, or %NULL if
556 * blk_init_queue() must be paired with a blk_cleanup_queue() call
557 * when the block device is deactivated (such as at module unload).
560 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
562 return blk_init_queue_node(rfn, lock, -1);
564 EXPORT_SYMBOL(blk_init_queue);
566 struct request_queue *
567 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
569 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
575 if (blk_init_free_list(q)) {
576 kmem_cache_free(blk_requestq_cachep, q);
581 q->prep_rq_fn = NULL;
582 q->unplug_fn = generic_unplug_device;
583 q->queue_flags = QUEUE_FLAG_DEFAULT;
584 q->queue_lock = lock;
587 * This also sets hw/phys segments, boundary and size
589 blk_queue_make_request(q, __make_request);
591 q->sg_reserved_size = INT_MAX;
596 if (!elevator_init(q, NULL)) {
597 blk_queue_congestion_threshold(q);
604 EXPORT_SYMBOL(blk_init_queue_node);
606 int blk_get_queue(struct request_queue *q)
608 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
609 kobject_get(&q->kobj);
616 static inline void blk_free_request(struct request_queue *q, struct request *rq)
618 if (rq->cmd_flags & REQ_ELVPRIV)
619 elv_put_request(q, rq);
620 mempool_free(rq, q->rq.rq_pool);
623 static struct request *
624 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
626 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
633 rq->cmd_flags = flags | REQ_ALLOCED;
636 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
637 mempool_free(rq, q->rq.rq_pool);
640 rq->cmd_flags |= REQ_ELVPRIV;
647 * ioc_batching returns true if the ioc is a valid batching request and
648 * should be given priority access to a request.
650 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
656 * Make sure the process is able to allocate at least 1 request
657 * even if the batch times out, otherwise we could theoretically
660 return ioc->nr_batch_requests == q->nr_batching ||
661 (ioc->nr_batch_requests > 0
662 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
666 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
667 * will cause the process to be a "batcher" on all queues in the system. This
668 * is the behaviour we want though - once it gets a wakeup it should be given
671 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
673 if (!ioc || ioc_batching(q, ioc))
676 ioc->nr_batch_requests = q->nr_batching;
677 ioc->last_waited = jiffies;
680 static void __freed_request(struct request_queue *q, int sync)
682 struct request_list *rl = &q->rq;
684 if (rl->count[sync] < queue_congestion_off_threshold(q))
685 blk_clear_queue_congested(q, sync);
687 if (rl->count[sync] + 1 <= q->nr_requests) {
688 if (waitqueue_active(&rl->wait[sync]))
689 wake_up(&rl->wait[sync]);
691 blk_clear_queue_full(q, sync);
696 * A request has just been released. Account for it, update the full and
697 * congestion status, wake up any waiters. Called under q->queue_lock.
699 static void freed_request(struct request_queue *q, int sync, int priv)
701 struct request_list *rl = &q->rq;
707 __freed_request(q, sync);
709 if (unlikely(rl->starved[sync ^ 1]))
710 __freed_request(q, sync ^ 1);
714 * Get a free request, queue_lock must be held.
715 * Returns NULL on failure, with queue_lock held.
716 * Returns !NULL on success, with queue_lock *not held*.
718 static struct request *get_request(struct request_queue *q, int rw_flags,
719 struct bio *bio, gfp_t gfp_mask)
721 struct request *rq = NULL;
722 struct request_list *rl = &q->rq;
723 struct io_context *ioc = NULL;
724 const bool is_sync = rw_is_sync(rw_flags) != 0;
727 may_queue = elv_may_queue(q, rw_flags);
728 if (may_queue == ELV_MQUEUE_NO)
731 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
732 if (rl->count[is_sync]+1 >= q->nr_requests) {
733 ioc = current_io_context(GFP_ATOMIC, q->node);
735 * The queue will fill after this allocation, so set
736 * it as full, and mark this process as "batching".
737 * This process will be allowed to complete a batch of
738 * requests, others will be blocked.
740 if (!blk_queue_full(q, is_sync)) {
741 ioc_set_batching(q, ioc);
742 blk_set_queue_full(q, is_sync);
744 if (may_queue != ELV_MQUEUE_MUST
745 && !ioc_batching(q, ioc)) {
747 * The queue is full and the allocating
748 * process is not a "batcher", and not
749 * exempted by the IO scheduler
755 blk_set_queue_congested(q, is_sync);
759 * Only allow batching queuers to allocate up to 50% over the defined
760 * limit of requests, otherwise we could have thousands of requests
761 * allocated with any setting of ->nr_requests
763 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
766 rl->count[is_sync]++;
767 rl->starved[is_sync] = 0;
769 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
773 if (blk_queue_io_stat(q))
774 rw_flags |= REQ_IO_STAT;
775 spin_unlock_irq(q->queue_lock);
777 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
780 * Allocation failed presumably due to memory. Undo anything
781 * we might have messed up.
783 * Allocating task should really be put onto the front of the
784 * wait queue, but this is pretty rare.
786 spin_lock_irq(q->queue_lock);
787 freed_request(q, is_sync, priv);
790 * in the very unlikely event that allocation failed and no
791 * requests for this direction was pending, mark us starved
792 * so that freeing of a request in the other direction will
793 * notice us. another possible fix would be to split the
794 * rq mempool into READ and WRITE
797 if (unlikely(rl->count[is_sync] == 0))
798 rl->starved[is_sync] = 1;
804 * ioc may be NULL here, and ioc_batching will be false. That's
805 * OK, if the queue is under the request limit then requests need
806 * not count toward the nr_batch_requests limit. There will always
807 * be some limit enforced by BLK_BATCH_TIME.
809 if (ioc_batching(q, ioc))
810 ioc->nr_batch_requests--;
812 trace_block_getrq(q, bio, rw_flags & 1);
818 * No available requests for this queue, unplug the device and wait for some
819 * requests to become available.
821 * Called with q->queue_lock held, and returns with it unlocked.
823 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
826 const bool is_sync = rw_is_sync(rw_flags) != 0;
829 rq = get_request(q, rw_flags, bio, GFP_NOIO);
832 struct io_context *ioc;
833 struct request_list *rl = &q->rq;
835 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
836 TASK_UNINTERRUPTIBLE);
838 trace_block_sleeprq(q, bio, rw_flags & 1);
840 __generic_unplug_device(q);
841 spin_unlock_irq(q->queue_lock);
845 * After sleeping, we become a "batching" process and
846 * will be able to allocate at least one request, and
847 * up to a big batch of them for a small period time.
848 * See ioc_batching, ioc_set_batching
850 ioc = current_io_context(GFP_NOIO, q->node);
851 ioc_set_batching(q, ioc);
853 spin_lock_irq(q->queue_lock);
854 finish_wait(&rl->wait[is_sync], &wait);
856 rq = get_request(q, rw_flags, bio, GFP_NOIO);
862 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
866 BUG_ON(rw != READ && rw != WRITE);
868 spin_lock_irq(q->queue_lock);
869 if (gfp_mask & __GFP_WAIT) {
870 rq = get_request_wait(q, rw, NULL);
872 rq = get_request(q, rw, NULL, gfp_mask);
874 spin_unlock_irq(q->queue_lock);
876 /* q->queue_lock is unlocked at this point */
880 EXPORT_SYMBOL(blk_get_request);
883 * blk_make_request - given a bio, allocate a corresponding struct request.
884 * @q: target request queue
885 * @bio: The bio describing the memory mappings that will be submitted for IO.
886 * It may be a chained-bio properly constructed by block/bio layer.
887 * @gfp_mask: gfp flags to be used for memory allocation
889 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
890 * type commands. Where the struct request needs to be farther initialized by
891 * the caller. It is passed a &struct bio, which describes the memory info of
894 * The caller of blk_make_request must make sure that bi_io_vec
895 * are set to describe the memory buffers. That bio_data_dir() will return
896 * the needed direction of the request. (And all bio's in the passed bio-chain
897 * are properly set accordingly)
899 * If called under none-sleepable conditions, mapped bio buffers must not
900 * need bouncing, by calling the appropriate masked or flagged allocator,
901 * suitable for the target device. Otherwise the call to blk_queue_bounce will
904 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
905 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
906 * anything but the first bio in the chain. Otherwise you risk waiting for IO
907 * completion of a bio that hasn't been submitted yet, thus resulting in a
908 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
909 * of bio_alloc(), as that avoids the mempool deadlock.
910 * If possible a big IO should be split into smaller parts when allocation
911 * fails. Partial allocation should not be an error, or you risk a live-lock.
913 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
916 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
919 return ERR_PTR(-ENOMEM);
922 struct bio *bounce_bio = bio;
925 blk_queue_bounce(q, &bounce_bio);
926 ret = blk_rq_append_bio(q, rq, bounce_bio);
935 EXPORT_SYMBOL(blk_make_request);
938 * blk_requeue_request - put a request back on queue
939 * @q: request queue where request should be inserted
940 * @rq: request to be inserted
943 * Drivers often keep queueing requests until the hardware cannot accept
944 * more, when that condition happens we need to put the request back
945 * on the queue. Must be called with queue lock held.
947 void blk_requeue_request(struct request_queue *q, struct request *rq)
949 blk_delete_timer(rq);
950 blk_clear_rq_complete(rq);
951 trace_block_rq_requeue(q, rq);
953 if (blk_rq_tagged(rq))
954 blk_queue_end_tag(q, rq);
956 BUG_ON(blk_queued_rq(rq));
958 elv_requeue_request(q, rq);
960 EXPORT_SYMBOL(blk_requeue_request);
963 * blk_insert_request - insert a special request into a request queue
964 * @q: request queue where request should be inserted
965 * @rq: request to be inserted
966 * @at_head: insert request at head or tail of queue
967 * @data: private data
970 * Many block devices need to execute commands asynchronously, so they don't
971 * block the whole kernel from preemption during request execution. This is
972 * accomplished normally by inserting aritficial requests tagged as
973 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
974 * be scheduled for actual execution by the request queue.
976 * We have the option of inserting the head or the tail of the queue.
977 * Typically we use the tail for new ioctls and so forth. We use the head
978 * of the queue for things like a QUEUE_FULL message from a device, or a
979 * host that is unable to accept a particular command.
981 void blk_insert_request(struct request_queue *q, struct request *rq,
982 int at_head, void *data)
984 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
988 * tell I/O scheduler that this isn't a regular read/write (ie it
989 * must not attempt merges on this) and that it acts as a soft
992 rq->cmd_type = REQ_TYPE_SPECIAL;
996 spin_lock_irqsave(q->queue_lock, flags);
999 * If command is tagged, release the tag
1001 if (blk_rq_tagged(rq))
1002 blk_queue_end_tag(q, rq);
1004 drive_stat_acct(rq, 1);
1005 __elv_add_request(q, rq, where, 0);
1007 spin_unlock_irqrestore(q->queue_lock, flags);
1009 EXPORT_SYMBOL(blk_insert_request);
1012 * add-request adds a request to the linked list.
1013 * queue lock is held and interrupts disabled, as we muck with the
1014 * request queue list.
1016 static inline void add_request(struct request_queue *q, struct request *req)
1018 drive_stat_acct(req, 1);
1021 * elevator indicated where it wants this request to be
1022 * inserted at elevator_merge time
1024 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1027 static void part_round_stats_single(int cpu, struct hd_struct *part,
1030 if (now == part->stamp)
1033 if (part_in_flight(part)) {
1034 __part_stat_add(cpu, part, time_in_queue,
1035 part_in_flight(part) * (now - part->stamp));
1036 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1042 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1043 * @cpu: cpu number for stats access
1044 * @part: target partition
1046 * The average IO queue length and utilisation statistics are maintained
1047 * by observing the current state of the queue length and the amount of
1048 * time it has been in this state for.
1050 * Normally, that accounting is done on IO completion, but that can result
1051 * in more than a second's worth of IO being accounted for within any one
1052 * second, leading to >100% utilisation. To deal with that, we call this
1053 * function to do a round-off before returning the results when reading
1054 * /proc/diskstats. This accounts immediately for all queue usage up to
1055 * the current jiffies and restarts the counters again.
1057 void part_round_stats(int cpu, struct hd_struct *part)
1059 unsigned long now = jiffies;
1062 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1063 part_round_stats_single(cpu, part, now);
1065 EXPORT_SYMBOL_GPL(part_round_stats);
1068 * queue lock must be held
1070 void __blk_put_request(struct request_queue *q, struct request *req)
1074 if (unlikely(--req->ref_count))
1077 elv_completed_request(q, req);
1079 /* this is a bio leak */
1080 WARN_ON(req->bio != NULL);
1083 * Request may not have originated from ll_rw_blk. if not,
1084 * it didn't come out of our reserved rq pools
1086 if (req->cmd_flags & REQ_ALLOCED) {
1087 int is_sync = rq_is_sync(req) != 0;
1088 int priv = req->cmd_flags & REQ_ELVPRIV;
1090 BUG_ON(!list_empty(&req->queuelist));
1091 BUG_ON(!hlist_unhashed(&req->hash));
1093 blk_free_request(q, req);
1094 freed_request(q, is_sync, priv);
1097 EXPORT_SYMBOL_GPL(__blk_put_request);
1099 void blk_put_request(struct request *req)
1101 unsigned long flags;
1102 struct request_queue *q = req->q;
1104 spin_lock_irqsave(q->queue_lock, flags);
1105 __blk_put_request(q, req);
1106 spin_unlock_irqrestore(q->queue_lock, flags);
1108 EXPORT_SYMBOL(blk_put_request);
1110 void init_request_from_bio(struct request *req, struct bio *bio)
1112 req->cpu = bio->bi_comp_cpu;
1113 req->cmd_type = REQ_TYPE_FS;
1116 * Inherit FAILFAST from bio (for read-ahead, and explicit
1117 * FAILFAST). FAILFAST flags are identical for req and bio.
1119 if (bio_rw_flagged(bio, BIO_RW_AHEAD))
1120 req->cmd_flags |= REQ_FAILFAST_MASK;
1122 req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK;
1124 if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) {
1125 req->cmd_flags |= REQ_DISCARD;
1126 if (bio_rw_flagged(bio, BIO_RW_BARRIER))
1127 req->cmd_flags |= REQ_SOFTBARRIER;
1128 } else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
1129 req->cmd_flags |= REQ_HARDBARRIER;
1131 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
1132 req->cmd_flags |= REQ_RW_SYNC;
1133 if (bio_rw_flagged(bio, BIO_RW_META))
1134 req->cmd_flags |= REQ_RW_META;
1135 if (bio_rw_flagged(bio, BIO_RW_NOIDLE))
1136 req->cmd_flags |= REQ_NOIDLE;
1139 req->__sector = bio->bi_sector;
1140 req->ioprio = bio_prio(bio);
1141 blk_rq_bio_prep(req->q, req, bio);
1145 * Only disabling plugging for non-rotational devices if it does tagging
1146 * as well, otherwise we do need the proper merging
1148 static inline bool queue_should_plug(struct request_queue *q)
1150 return !(blk_queue_nonrot(q) && blk_queue_queuing(q));
1153 static int __make_request(struct request_queue *q, struct bio *bio)
1155 struct request *req;
1157 unsigned int bytes = bio->bi_size;
1158 const unsigned short prio = bio_prio(bio);
1159 const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
1160 const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG);
1161 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1164 if (bio_rw_flagged(bio, BIO_RW_BARRIER) &&
1165 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1166 bio_endio(bio, -EOPNOTSUPP);
1170 * low level driver can indicate that it wants pages above a
1171 * certain limit bounced to low memory (ie for highmem, or even
1172 * ISA dma in theory)
1174 blk_queue_bounce(q, &bio);
1176 spin_lock_irq(q->queue_lock);
1178 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q))
1181 el_ret = elv_merge(q, &req, bio);
1183 case ELEVATOR_BACK_MERGE:
1184 BUG_ON(!rq_mergeable(req));
1186 if (!ll_back_merge_fn(q, req, bio))
1189 trace_block_bio_backmerge(q, bio);
1191 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1192 blk_rq_set_mixed_merge(req);
1194 req->biotail->bi_next = bio;
1196 req->__data_len += bytes;
1197 req->ioprio = ioprio_best(req->ioprio, prio);
1198 if (!blk_rq_cpu_valid(req))
1199 req->cpu = bio->bi_comp_cpu;
1200 drive_stat_acct(req, 0);
1201 if (!attempt_back_merge(q, req))
1202 elv_merged_request(q, req, el_ret);
1205 case ELEVATOR_FRONT_MERGE:
1206 BUG_ON(!rq_mergeable(req));
1208 if (!ll_front_merge_fn(q, req, bio))
1211 trace_block_bio_frontmerge(q, bio);
1213 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1214 blk_rq_set_mixed_merge(req);
1215 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1216 req->cmd_flags |= ff;
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,
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(!bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1441 nr_sectors > queue_max_hw_sectors(q))) {
1442 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1443 bdevname(bio->bi_bdev, b),
1445 queue_max_hw_sectors(q));
1449 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1452 if (should_fail_request(bio))
1456 * If this device has partitions, remap block n
1457 * of partition p to block n+start(p) of the disk.
1459 blk_partition_remap(bio);
1461 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1464 if (old_sector != -1)
1465 trace_block_remap(q, bio, old_dev, old_sector);
1467 old_sector = bio->bi_sector;
1468 old_dev = bio->bi_bdev->bd_dev;
1470 if (bio_check_eod(bio, nr_sectors))
1473 if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1474 !blk_queue_discard(q)) {
1479 trace_block_bio_queue(q, bio);
1481 ret = q->make_request_fn(q, bio);
1487 bio_endio(bio, err);
1491 * We only want one ->make_request_fn to be active at a time,
1492 * else stack usage with stacked devices could be a problem.
1493 * So use current->bio_{list,tail} to keep a list of requests
1494 * submited by a make_request_fn function.
1495 * current->bio_tail is also used as a flag to say if
1496 * generic_make_request is currently active in this task or not.
1497 * If it is NULL, then no make_request is active. If it is non-NULL,
1498 * then a make_request is active, and new requests should be added
1501 void generic_make_request(struct bio *bio)
1503 if (current->bio_tail) {
1504 /* make_request is active */
1505 *(current->bio_tail) = bio;
1506 bio->bi_next = NULL;
1507 current->bio_tail = &bio->bi_next;
1510 /* following loop may be a bit non-obvious, and so deserves some
1512 * Before entering the loop, bio->bi_next is NULL (as all callers
1513 * ensure that) so we have a list with a single bio.
1514 * We pretend that we have just taken it off a longer list, so
1515 * we assign bio_list to the next (which is NULL) and bio_tail
1516 * to &bio_list, thus initialising the bio_list of new bios to be
1517 * added. __generic_make_request may indeed add some more bios
1518 * through a recursive call to generic_make_request. If it
1519 * did, we find a non-NULL value in bio_list and re-enter the loop
1520 * from the top. In this case we really did just take the bio
1521 * of the top of the list (no pretending) and so fixup bio_list and
1522 * bio_tail or bi_next, and call into __generic_make_request again.
1524 * The loop was structured like this to make only one call to
1525 * __generic_make_request (which is important as it is large and
1526 * inlined) and to keep the structure simple.
1528 BUG_ON(bio->bi_next);
1530 current->bio_list = bio->bi_next;
1531 if (bio->bi_next == NULL)
1532 current->bio_tail = ¤t->bio_list;
1534 bio->bi_next = NULL;
1535 __generic_make_request(bio);
1536 bio = current->bio_list;
1538 current->bio_tail = NULL; /* deactivate */
1540 EXPORT_SYMBOL(generic_make_request);
1543 * submit_bio - submit a bio to the block device layer for I/O
1544 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1545 * @bio: The &struct bio which describes the I/O
1547 * submit_bio() is very similar in purpose to generic_make_request(), and
1548 * uses that function to do most of the work. Both are fairly rough
1549 * interfaces; @bio must be presetup and ready for I/O.
1552 void submit_bio(int rw, struct bio *bio)
1554 int count = bio_sectors(bio);
1559 * If it's a regular read/write or a barrier with data attached,
1560 * go through the normal accounting stuff before submission.
1562 if (bio_has_data(bio)) {
1564 count_vm_events(PGPGOUT, count);
1566 task_io_account_read(bio->bi_size);
1567 count_vm_events(PGPGIN, count);
1570 if (unlikely(block_dump)) {
1571 char b[BDEVNAME_SIZE];
1572 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1573 current->comm, task_pid_nr(current),
1574 (rw & WRITE) ? "WRITE" : "READ",
1575 (unsigned long long)bio->bi_sector,
1576 bdevname(bio->bi_bdev, b),
1581 generic_make_request(bio);
1583 EXPORT_SYMBOL(submit_bio);
1586 * blk_rq_check_limits - Helper function to check a request for the queue limit
1588 * @rq: the request being checked
1591 * @rq may have been made based on weaker limitations of upper-level queues
1592 * in request stacking drivers, and it may violate the limitation of @q.
1593 * Since the block layer and the underlying device driver trust @rq
1594 * after it is inserted to @q, it should be checked against @q before
1595 * the insertion using this generic function.
1597 * This function should also be useful for request stacking drivers
1598 * in some cases below, so export this fuction.
1599 * Request stacking drivers like request-based dm may change the queue
1600 * limits while requests are in the queue (e.g. dm's table swapping).
1601 * Such request stacking drivers should check those requests agaist
1602 * the new queue limits again when they dispatch those requests,
1603 * although such checkings are also done against the old queue limits
1604 * when submitting requests.
1606 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1608 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1609 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1610 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1615 * queue's settings related to segment counting like q->bounce_pfn
1616 * may differ from that of other stacking queues.
1617 * Recalculate it to check the request correctly on this queue's
1620 blk_recalc_rq_segments(rq);
1621 if (rq->nr_phys_segments > queue_max_phys_segments(q) ||
1622 rq->nr_phys_segments > queue_max_hw_segments(q)) {
1623 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1629 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1632 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1633 * @q: the queue to submit the request
1634 * @rq: the request being queued
1636 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1638 unsigned long flags;
1640 if (blk_rq_check_limits(q, rq))
1643 #ifdef CONFIG_FAIL_MAKE_REQUEST
1644 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1645 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1649 spin_lock_irqsave(q->queue_lock, flags);
1652 * Submitting request must be dequeued before calling this function
1653 * because it will be linked to another request_queue
1655 BUG_ON(blk_queued_rq(rq));
1657 drive_stat_acct(rq, 1);
1658 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1660 spin_unlock_irqrestore(q->queue_lock, flags);
1664 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1667 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1668 * @rq: request to examine
1671 * A request could be merge of IOs which require different failure
1672 * handling. This function determines the number of bytes which
1673 * can be failed from the beginning of the request without
1674 * crossing into area which need to be retried further.
1677 * The number of bytes to fail.
1680 * queue_lock must be held.
1682 unsigned int blk_rq_err_bytes(const struct request *rq)
1684 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1685 unsigned int bytes = 0;
1688 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1689 return blk_rq_bytes(rq);
1692 * Currently the only 'mixing' which can happen is between
1693 * different fastfail types. We can safely fail portions
1694 * which have all the failfast bits that the first one has -
1695 * the ones which are at least as eager to fail as the first
1698 for (bio = rq->bio; bio; bio = bio->bi_next) {
1699 if ((bio->bi_rw & ff) != ff)
1701 bytes += bio->bi_size;
1704 /* this could lead to infinite loop */
1705 BUG_ON(blk_rq_bytes(rq) && !bytes);
1708 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1710 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1712 if (blk_do_io_stat(req)) {
1713 const int rw = rq_data_dir(req);
1714 struct hd_struct *part;
1717 cpu = part_stat_lock();
1718 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1719 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1724 static void blk_account_io_done(struct request *req)
1727 * Account IO completion. bar_rq isn't accounted as a normal
1728 * IO on queueing nor completion. Accounting the containing
1729 * request is enough.
1731 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
1732 unsigned long duration = jiffies - req->start_time;
1733 const int rw = rq_data_dir(req);
1734 struct hd_struct *part;
1737 cpu = part_stat_lock();
1738 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1740 part_stat_inc(cpu, part, ios[rw]);
1741 part_stat_add(cpu, part, ticks[rw], duration);
1742 part_round_stats(cpu, part);
1743 part_dec_in_flight(part, rw);
1750 * blk_peek_request - peek at the top of a request queue
1751 * @q: request queue to peek at
1754 * Return the request at the top of @q. The returned request
1755 * should be started using blk_start_request() before LLD starts
1759 * Pointer to the request at the top of @q if available. Null
1763 * queue_lock must be held.
1765 struct request *blk_peek_request(struct request_queue *q)
1770 while ((rq = __elv_next_request(q)) != NULL) {
1771 if (!(rq->cmd_flags & REQ_STARTED)) {
1773 * This is the first time the device driver
1774 * sees this request (possibly after
1775 * requeueing). Notify IO scheduler.
1777 if (blk_sorted_rq(rq))
1778 elv_activate_rq(q, rq);
1781 * just mark as started even if we don't start
1782 * it, a request that has been delayed should
1783 * not be passed by new incoming requests
1785 rq->cmd_flags |= REQ_STARTED;
1786 trace_block_rq_issue(q, rq);
1789 if (!q->boundary_rq || q->boundary_rq == rq) {
1790 q->end_sector = rq_end_sector(rq);
1791 q->boundary_rq = NULL;
1794 if (rq->cmd_flags & REQ_DONTPREP)
1797 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1799 * make sure space for the drain appears we
1800 * know we can do this because max_hw_segments
1801 * has been adjusted to be one fewer than the
1804 rq->nr_phys_segments++;
1810 ret = q->prep_rq_fn(q, rq);
1811 if (ret == BLKPREP_OK) {
1813 } else if (ret == BLKPREP_DEFER) {
1815 * the request may have been (partially) prepped.
1816 * we need to keep this request in the front to
1817 * avoid resource deadlock. REQ_STARTED will
1818 * prevent other fs requests from passing this one.
1820 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1821 !(rq->cmd_flags & REQ_DONTPREP)) {
1823 * remove the space for the drain we added
1824 * so that we don't add it again
1826 --rq->nr_phys_segments;
1831 } else if (ret == BLKPREP_KILL) {
1832 rq->cmd_flags |= REQ_QUIET;
1834 * Mark this request as started so we don't trigger
1835 * any debug logic in the end I/O path.
1837 blk_start_request(rq);
1838 __blk_end_request_all(rq, -EIO);
1840 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1847 EXPORT_SYMBOL(blk_peek_request);
1849 void blk_dequeue_request(struct request *rq)
1851 struct request_queue *q = rq->q;
1853 BUG_ON(list_empty(&rq->queuelist));
1854 BUG_ON(ELV_ON_HASH(rq));
1856 list_del_init(&rq->queuelist);
1859 * the time frame between a request being removed from the lists
1860 * and to it is freed is accounted as io that is in progress at
1863 if (blk_account_rq(rq)) {
1864 q->in_flight[rq_is_sync(rq)]++;
1866 * Mark this device as supporting hardware queuing, if
1867 * we have more IOs in flight than 4.
1869 if (!blk_queue_queuing(q) && queue_in_flight(q) > 4)
1870 set_bit(QUEUE_FLAG_CQ, &q->queue_flags);
1875 * blk_start_request - start request processing on the driver
1876 * @req: request to dequeue
1879 * Dequeue @req and start timeout timer on it. This hands off the
1880 * request to the driver.
1882 * Block internal functions which don't want to start timer should
1883 * call blk_dequeue_request().
1886 * queue_lock must be held.
1888 void blk_start_request(struct request *req)
1890 blk_dequeue_request(req);
1893 * We are now handing the request to the hardware, initialize
1894 * resid_len to full count and add the timeout handler.
1896 req->resid_len = blk_rq_bytes(req);
1897 if (unlikely(blk_bidi_rq(req)))
1898 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1902 EXPORT_SYMBOL(blk_start_request);
1905 * blk_fetch_request - fetch a request from a request queue
1906 * @q: request queue to fetch a request from
1909 * Return the request at the top of @q. The request is started on
1910 * return and LLD can start processing it immediately.
1913 * Pointer to the request at the top of @q if available. Null
1917 * queue_lock must be held.
1919 struct request *blk_fetch_request(struct request_queue *q)
1923 rq = blk_peek_request(q);
1925 blk_start_request(rq);
1928 EXPORT_SYMBOL(blk_fetch_request);
1931 * blk_update_request - Special helper function for request stacking drivers
1932 * @req: the request being processed
1933 * @error: %0 for success, < %0 for error
1934 * @nr_bytes: number of bytes to complete @req
1937 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1938 * the request structure even if @req doesn't have leftover.
1939 * If @req has leftover, sets it up for the next range of segments.
1941 * This special helper function is only for request stacking drivers
1942 * (e.g. request-based dm) so that they can handle partial completion.
1943 * Actual device drivers should use blk_end_request instead.
1945 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1946 * %false return from this function.
1949 * %false - this request doesn't have any more data
1950 * %true - this request has more data
1952 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
1954 int total_bytes, bio_nbytes, next_idx = 0;
1960 trace_block_rq_complete(req->q, req);
1963 * For fs requests, rq is just carrier of independent bio's
1964 * and each partial completion should be handled separately.
1965 * Reset per-request error on each partial completion.
1967 * TODO: tj: This is too subtle. It would be better to let
1968 * low level drivers do what they see fit.
1970 if (blk_fs_request(req))
1973 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1974 printk(KERN_DEBUG "end_request: I/O error, dev %s, sector %llu\n",
1975 req->rq_disk ? req->rq_disk->disk_name : "?",
1976 (unsigned long long)blk_rq_pos(req));
1979 blk_account_io_completion(req, nr_bytes);
1981 total_bytes = bio_nbytes = 0;
1982 while ((bio = req->bio) != NULL) {
1985 if (nr_bytes >= bio->bi_size) {
1986 req->bio = bio->bi_next;
1987 nbytes = bio->bi_size;
1988 req_bio_endio(req, bio, nbytes, error);
1992 int idx = bio->bi_idx + next_idx;
1994 if (unlikely(idx >= bio->bi_vcnt)) {
1995 blk_dump_rq_flags(req, "__end_that");
1996 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1997 __func__, idx, bio->bi_vcnt);
2001 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2002 BIO_BUG_ON(nbytes > bio->bi_size);
2005 * not a complete bvec done
2007 if (unlikely(nbytes > nr_bytes)) {
2008 bio_nbytes += nr_bytes;
2009 total_bytes += nr_bytes;
2014 * advance to the next vector
2017 bio_nbytes += nbytes;
2020 total_bytes += nbytes;
2026 * end more in this run, or just return 'not-done'
2028 if (unlikely(nr_bytes <= 0))
2038 * Reset counters so that the request stacking driver
2039 * can find how many bytes remain in the request
2042 req->__data_len = 0;
2047 * if the request wasn't completed, update state
2050 req_bio_endio(req, bio, bio_nbytes, error);
2051 bio->bi_idx += next_idx;
2052 bio_iovec(bio)->bv_offset += nr_bytes;
2053 bio_iovec(bio)->bv_len -= nr_bytes;
2056 req->__data_len -= total_bytes;
2057 req->buffer = bio_data(req->bio);
2059 /* update sector only for requests with clear definition of sector */
2060 if (blk_fs_request(req) || blk_discard_rq(req))
2061 req->__sector += total_bytes >> 9;
2063 /* mixed attributes always follow the first bio */
2064 if (req->cmd_flags & REQ_MIXED_MERGE) {
2065 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2066 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2070 * If total number of sectors is less than the first segment
2071 * size, something has gone terribly wrong.
2073 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2074 printk(KERN_ERR "blk: request botched\n");
2075 req->__data_len = blk_rq_cur_bytes(req);
2078 /* recalculate the number of segments */
2079 blk_recalc_rq_segments(req);
2083 EXPORT_SYMBOL_GPL(blk_update_request);
2085 static bool blk_update_bidi_request(struct request *rq, int error,
2086 unsigned int nr_bytes,
2087 unsigned int bidi_bytes)
2089 if (blk_update_request(rq, error, nr_bytes))
2092 /* Bidi request must be completed as a whole */
2093 if (unlikely(blk_bidi_rq(rq)) &&
2094 blk_update_request(rq->next_rq, error, bidi_bytes))
2097 add_disk_randomness(rq->rq_disk);
2103 * queue lock must be held
2105 static void blk_finish_request(struct request *req, int error)
2107 if (blk_rq_tagged(req))
2108 blk_queue_end_tag(req->q, req);
2110 BUG_ON(blk_queued_rq(req));
2112 if (unlikely(laptop_mode) && blk_fs_request(req))
2113 laptop_io_completion();
2115 blk_delete_timer(req);
2117 blk_account_io_done(req);
2120 req->end_io(req, error);
2122 if (blk_bidi_rq(req))
2123 __blk_put_request(req->next_rq->q, req->next_rq);
2125 __blk_put_request(req->q, req);
2130 * blk_end_bidi_request - Complete a bidi request
2131 * @rq: the request to complete
2132 * @error: %0 for success, < %0 for error
2133 * @nr_bytes: number of bytes to complete @rq
2134 * @bidi_bytes: number of bytes to complete @rq->next_rq
2137 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2138 * Drivers that supports bidi can safely call this member for any
2139 * type of request, bidi or uni. In the later case @bidi_bytes is
2143 * %false - we are done with this request
2144 * %true - still buffers pending for this request
2146 static bool blk_end_bidi_request(struct request *rq, int error,
2147 unsigned int nr_bytes, unsigned int bidi_bytes)
2149 struct request_queue *q = rq->q;
2150 unsigned long flags;
2152 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2155 spin_lock_irqsave(q->queue_lock, flags);
2156 blk_finish_request(rq, error);
2157 spin_unlock_irqrestore(q->queue_lock, flags);
2163 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2164 * @rq: the request to complete
2165 * @error: %0 for success, < %0 for error
2166 * @nr_bytes: number of bytes to complete @rq
2167 * @bidi_bytes: number of bytes to complete @rq->next_rq
2170 * Identical to blk_end_bidi_request() except that queue lock is
2171 * assumed to be locked on entry and remains so on return.
2174 * %false - we are done with this request
2175 * %true - still buffers pending for this request
2177 static bool __blk_end_bidi_request(struct request *rq, int error,
2178 unsigned int nr_bytes, unsigned int bidi_bytes)
2180 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2183 blk_finish_request(rq, error);
2189 * blk_end_request - Helper function for drivers to complete the request.
2190 * @rq: the request being processed
2191 * @error: %0 for success, < %0 for error
2192 * @nr_bytes: number of bytes to complete
2195 * Ends I/O on a number of bytes attached to @rq.
2196 * If @rq has leftover, sets it up for the next range of segments.
2199 * %false - we are done with this request
2200 * %true - still buffers pending for this request
2202 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2204 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2206 EXPORT_SYMBOL(blk_end_request);
2209 * blk_end_request_all - Helper function for drives to finish the request.
2210 * @rq: the request to finish
2211 * @error: %0 for success, < %0 for error
2214 * Completely finish @rq.
2216 void blk_end_request_all(struct request *rq, int error)
2219 unsigned int bidi_bytes = 0;
2221 if (unlikely(blk_bidi_rq(rq)))
2222 bidi_bytes = blk_rq_bytes(rq->next_rq);
2224 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2227 EXPORT_SYMBOL(blk_end_request_all);
2230 * blk_end_request_cur - Helper function to finish the current request chunk.
2231 * @rq: the request to finish the current chunk for
2232 * @error: %0 for success, < %0 for error
2235 * Complete the current consecutively mapped chunk from @rq.
2238 * %false - we are done with this request
2239 * %true - still buffers pending for this request
2241 bool blk_end_request_cur(struct request *rq, int error)
2243 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2245 EXPORT_SYMBOL(blk_end_request_cur);
2248 * blk_end_request_err - Finish a request till the next failure boundary.
2249 * @rq: the request to finish till the next failure boundary for
2250 * @error: must be negative errno
2253 * Complete @rq till the next failure boundary.
2256 * %false - we are done with this request
2257 * %true - still buffers pending for this request
2259 bool blk_end_request_err(struct request *rq, int error)
2261 WARN_ON(error >= 0);
2262 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2264 EXPORT_SYMBOL_GPL(blk_end_request_err);
2267 * __blk_end_request - Helper function for drivers to complete the request.
2268 * @rq: the request being processed
2269 * @error: %0 for success, < %0 for error
2270 * @nr_bytes: number of bytes to complete
2273 * Must be called with queue lock held unlike blk_end_request().
2276 * %false - we are done with this request
2277 * %true - still buffers pending for this request
2279 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2281 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2283 EXPORT_SYMBOL(__blk_end_request);
2286 * __blk_end_request_all - Helper function for drives to finish the request.
2287 * @rq: the request to finish
2288 * @error: %0 for success, < %0 for error
2291 * Completely finish @rq. Must be called with queue lock held.
2293 void __blk_end_request_all(struct request *rq, int error)
2296 unsigned int bidi_bytes = 0;
2298 if (unlikely(blk_bidi_rq(rq)))
2299 bidi_bytes = blk_rq_bytes(rq->next_rq);
2301 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2304 EXPORT_SYMBOL(__blk_end_request_all);
2307 * __blk_end_request_cur - Helper function to finish the current request chunk.
2308 * @rq: the request to finish the current chunk for
2309 * @error: %0 for success, < %0 for error
2312 * Complete the current consecutively mapped chunk from @rq. Must
2313 * be called with queue lock held.
2316 * %false - we are done with this request
2317 * %true - still buffers pending for this request
2319 bool __blk_end_request_cur(struct request *rq, int error)
2321 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2323 EXPORT_SYMBOL(__blk_end_request_cur);
2326 * __blk_end_request_err - Finish a request till the next failure boundary.
2327 * @rq: the request to finish till the next failure boundary for
2328 * @error: must be negative errno
2331 * Complete @rq till the next failure boundary. Must be called
2332 * with queue lock held.
2335 * %false - we are done with this request
2336 * %true - still buffers pending for this request
2338 bool __blk_end_request_err(struct request *rq, int error)
2340 WARN_ON(error >= 0);
2341 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2343 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2345 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2348 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2349 rq->cmd_flags |= bio->bi_rw & REQ_RW;
2351 if (bio_has_data(bio)) {
2352 rq->nr_phys_segments = bio_phys_segments(q, bio);
2353 rq->buffer = bio_data(bio);
2355 rq->__data_len = bio->bi_size;
2356 rq->bio = rq->biotail = bio;
2359 rq->rq_disk = bio->bi_bdev->bd_disk;
2363 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2364 * @q : the queue of the device being checked
2367 * Check if underlying low-level drivers of a device are busy.
2368 * If the drivers want to export their busy state, they must set own
2369 * exporting function using blk_queue_lld_busy() first.
2371 * Basically, this function is used only by request stacking drivers
2372 * to stop dispatching requests to underlying devices when underlying
2373 * devices are busy. This behavior helps more I/O merging on the queue
2374 * of the request stacking driver and prevents I/O throughput regression
2375 * on burst I/O load.
2378 * 0 - Not busy (The request stacking driver should dispatch request)
2379 * 1 - Busy (The request stacking driver should stop dispatching request)
2381 int blk_lld_busy(struct request_queue *q)
2384 return q->lld_busy_fn(q);
2388 EXPORT_SYMBOL_GPL(blk_lld_busy);
2391 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2392 * @rq: the clone request to be cleaned up
2395 * Free all bios in @rq for a cloned request.
2397 void blk_rq_unprep_clone(struct request *rq)
2401 while ((bio = rq->bio) != NULL) {
2402 rq->bio = bio->bi_next;
2407 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2410 * Copy attributes of the original request to the clone request.
2411 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2413 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2415 dst->cpu = src->cpu;
2416 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2417 dst->cmd_type = src->cmd_type;
2418 dst->__sector = blk_rq_pos(src);
2419 dst->__data_len = blk_rq_bytes(src);
2420 dst->nr_phys_segments = src->nr_phys_segments;
2421 dst->ioprio = src->ioprio;
2422 dst->extra_len = src->extra_len;
2426 * blk_rq_prep_clone - Helper function to setup clone request
2427 * @rq: the request to be setup
2428 * @rq_src: original request to be cloned
2429 * @bs: bio_set that bios for clone are allocated from
2430 * @gfp_mask: memory allocation mask for bio
2431 * @bio_ctr: setup function to be called for each clone bio.
2432 * Returns %0 for success, non %0 for failure.
2433 * @data: private data to be passed to @bio_ctr
2436 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2437 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2438 * are not copied, and copying such parts is the caller's responsibility.
2439 * Also, pages which the original bios are pointing to are not copied
2440 * and the cloned bios just point same pages.
2441 * So cloned bios must be completed before original bios, which means
2442 * the caller must complete @rq before @rq_src.
2444 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2445 struct bio_set *bs, gfp_t gfp_mask,
2446 int (*bio_ctr)(struct bio *, struct bio *, void *),
2449 struct bio *bio, *bio_src;
2454 blk_rq_init(NULL, rq);
2456 __rq_for_each_bio(bio_src, rq_src) {
2457 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2461 __bio_clone(bio, bio_src);
2463 if (bio_integrity(bio_src) &&
2464 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2467 if (bio_ctr && bio_ctr(bio, bio_src, data))
2471 rq->biotail->bi_next = bio;
2474 rq->bio = rq->biotail = bio;
2477 __blk_rq_prep_clone(rq, rq_src);
2484 blk_rq_unprep_clone(rq);
2488 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2490 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2492 return queue_work(kblockd_workqueue, work);
2494 EXPORT_SYMBOL(kblockd_schedule_work);
2496 int __init blk_dev_init(void)
2498 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2499 sizeof(((struct request *)0)->cmd_flags));
2501 kblockd_workqueue = create_workqueue("kblockd");
2502 if (!kblockd_workqueue)
2503 panic("Failed to create kblockd\n");
2505 request_cachep = kmem_cache_create("blkdev_requests",
2506 sizeof(struct request), 0, SLAB_PANIC, NULL);
2508 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2509 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);