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>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/block.h>
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
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();
70 part_stat_inc(cpu, part, merges[rw]);
72 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
73 if (!hd_struct_try_get(part)) {
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
82 part = &rq->rq_disk->part0;
85 part_round_stats(cpu, part);
86 part_inc_in_flight(part, rw);
93 void blk_queue_congestion_threshold(struct request_queue *q)
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
100 q->nr_congestion_on = nr;
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 q->nr_congestion_off = nr;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
115 * Will return NULL if the request queue cannot be located.
117 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119 struct backing_dev_info *ret = NULL;
120 struct request_queue *q = bdev_get_queue(bdev);
123 ret = &q->backing_dev_info;
126 EXPORT_SYMBOL(blk_get_backing_dev_info);
128 void blk_rq_init(struct request_queue *q, struct request *rq)
130 memset(rq, 0, sizeof(*rq));
132 INIT_LIST_HEAD(&rq->queuelist);
133 INIT_LIST_HEAD(&rq->timeout_list);
136 rq->__sector = (sector_t) -1;
137 INIT_HLIST_NODE(&rq->hash);
138 RB_CLEAR_NODE(&rq->rb_node);
140 rq->cmd_len = BLK_MAX_CDB;
143 rq->start_time = jiffies;
144 set_start_time_ns(rq);
147 EXPORT_SYMBOL(blk_rq_init);
149 static void req_bio_endio(struct request *rq, struct bio *bio,
150 unsigned int nbytes, int error)
153 clear_bit(BIO_UPTODATE, &bio->bi_flags);
154 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
157 if (unlikely(nbytes > bio->bi_size)) {
158 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
159 __func__, nbytes, bio->bi_size);
160 nbytes = bio->bi_size;
163 if (unlikely(rq->cmd_flags & REQ_QUIET))
164 set_bit(BIO_QUIET, &bio->bi_flags);
166 bio->bi_size -= nbytes;
167 bio->bi_sector += (nbytes >> 9);
169 if (bio_integrity(bio))
170 bio_integrity_advance(bio, nbytes);
172 /* don't actually finish bio if it's part of flush sequence */
173 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
174 bio_endio(bio, error);
177 void blk_dump_rq_flags(struct request *rq, char *msg)
181 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
182 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
185 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
186 (unsigned long long)blk_rq_pos(rq),
187 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
188 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
189 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
191 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
192 printk(KERN_INFO " cdb: ");
193 for (bit = 0; bit < BLK_MAX_CDB; bit++)
194 printk("%02x ", rq->cmd[bit]);
198 EXPORT_SYMBOL(blk_dump_rq_flags);
200 static void blk_delay_work(struct work_struct *work)
202 struct request_queue *q;
204 q = container_of(work, struct request_queue, delay_work.work);
205 spin_lock_irq(q->queue_lock);
207 spin_unlock_irq(q->queue_lock);
211 * blk_delay_queue - restart queueing after defined interval
212 * @q: The &struct request_queue in question
213 * @msecs: Delay in msecs
216 * Sometimes queueing needs to be postponed for a little while, to allow
217 * resources to come back. This function will make sure that queueing is
218 * restarted around the specified time.
220 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
222 queue_delayed_work(kblockd_workqueue, &q->delay_work,
223 msecs_to_jiffies(msecs));
225 EXPORT_SYMBOL(blk_delay_queue);
228 * blk_start_queue - restart a previously stopped queue
229 * @q: The &struct request_queue in question
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 void blk_start_queue(struct request_queue *q)
238 WARN_ON(!irqs_disabled());
240 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
243 EXPORT_SYMBOL(blk_start_queue);
246 * blk_stop_queue - stop a queue
247 * @q: The &struct request_queue in question
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 void blk_stop_queue(struct request_queue *q)
261 __cancel_delayed_work(&q->delay_work);
262 queue_flag_set(QUEUE_FLAG_STOPPED, q);
264 EXPORT_SYMBOL(blk_stop_queue);
267 * blk_sync_queue - cancel any pending callbacks on a queue
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
275 * that the callbacks might use. The caller must already have made sure
276 * that its ->make_request_fn will not re-add plugging prior to calling
279 * This function does not cancel any asynchronous activity arising
280 * out of elevator or throttling code. That would require elevaotor_exit()
281 * and blk_throtl_exit() to be called with queue lock initialized.
284 void blk_sync_queue(struct request_queue *q)
286 del_timer_sync(&q->timeout);
287 cancel_delayed_work_sync(&q->delay_work);
289 EXPORT_SYMBOL(blk_sync_queue);
292 * __blk_run_queue - run a single device queue
293 * @q: The queue to run
296 * See @blk_run_queue. This variant must be called with the queue lock
297 * held and interrupts disabled.
299 void __blk_run_queue(struct request_queue *q)
301 if (unlikely(blk_queue_stopped(q)))
306 EXPORT_SYMBOL(__blk_run_queue);
309 * blk_run_queue_async - run a single device queue in workqueue context
310 * @q: The queue to run
313 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
316 void blk_run_queue_async(struct request_queue *q)
318 if (likely(!blk_queue_stopped(q))) {
319 __cancel_delayed_work(&q->delay_work);
320 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
323 EXPORT_SYMBOL(blk_run_queue_async);
326 * blk_run_queue - run a single device queue
327 * @q: The queue to run
330 * Invoke request handling on this queue, if it has pending work to do.
331 * May be used to restart queueing when a request has completed.
333 void blk_run_queue(struct request_queue *q)
337 spin_lock_irqsave(q->queue_lock, flags);
339 spin_unlock_irqrestore(q->queue_lock, flags);
341 EXPORT_SYMBOL(blk_run_queue);
343 void blk_put_queue(struct request_queue *q)
345 kobject_put(&q->kobj);
347 EXPORT_SYMBOL(blk_put_queue);
350 * blk_drain_queue - drain requests from request_queue
353 * Drain ELV_PRIV requests from @q. The caller is responsible for ensuring
354 * that no new requests which need to be drained are queued.
356 void blk_drain_queue(struct request_queue *q)
361 spin_lock_irq(q->queue_lock);
363 elv_drain_elevator(q);
366 nr_rqs = q->rq.elvpriv;
368 spin_unlock_irq(q->queue_lock);
377 * Note: If a driver supplied the queue lock, it is disconnected
378 * by this function. The actual state of the lock doesn't matter
379 * here as the request_queue isn't accessible after this point
380 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
382 void blk_cleanup_queue(struct request_queue *q)
385 * We know we have process context here, so we can be a little
386 * cautious and ensure that pending block actions on this device
387 * are done before moving on. Going into this function, we should
388 * not have processes doing IO to this device.
392 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
393 mutex_lock(&q->sysfs_lock);
394 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
395 mutex_unlock(&q->sysfs_lock);
397 if (q->queue_lock != &q->__queue_lock)
398 q->queue_lock = &q->__queue_lock;
402 EXPORT_SYMBOL(blk_cleanup_queue);
404 static int blk_init_free_list(struct request_queue *q)
406 struct request_list *rl = &q->rq;
408 if (unlikely(rl->rq_pool))
411 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
412 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
414 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
415 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
417 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
418 mempool_free_slab, request_cachep, q->node);
426 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
428 return blk_alloc_queue_node(gfp_mask, -1);
430 EXPORT_SYMBOL(blk_alloc_queue);
432 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
434 struct request_queue *q;
437 q = kmem_cache_alloc_node(blk_requestq_cachep,
438 gfp_mask | __GFP_ZERO, node_id);
442 q->backing_dev_info.ra_pages =
443 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
444 q->backing_dev_info.state = 0;
445 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
446 q->backing_dev_info.name = "block";
448 err = bdi_init(&q->backing_dev_info);
450 kmem_cache_free(blk_requestq_cachep, q);
454 if (blk_throtl_init(q)) {
455 kmem_cache_free(blk_requestq_cachep, q);
459 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
460 laptop_mode_timer_fn, (unsigned long) q);
461 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
462 INIT_LIST_HEAD(&q->timeout_list);
463 INIT_LIST_HEAD(&q->flush_queue[0]);
464 INIT_LIST_HEAD(&q->flush_queue[1]);
465 INIT_LIST_HEAD(&q->flush_data_in_flight);
466 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
468 kobject_init(&q->kobj, &blk_queue_ktype);
470 mutex_init(&q->sysfs_lock);
471 spin_lock_init(&q->__queue_lock);
474 * By default initialize queue_lock to internal lock and driver can
475 * override it later if need be.
477 q->queue_lock = &q->__queue_lock;
481 EXPORT_SYMBOL(blk_alloc_queue_node);
484 * blk_init_queue - prepare a request queue for use with a block device
485 * @rfn: The function to be called to process requests that have been
486 * placed on the queue.
487 * @lock: Request queue spin lock
490 * If a block device wishes to use the standard request handling procedures,
491 * which sorts requests and coalesces adjacent requests, then it must
492 * call blk_init_queue(). The function @rfn will be called when there
493 * are requests on the queue that need to be processed. If the device
494 * supports plugging, then @rfn may not be called immediately when requests
495 * are available on the queue, but may be called at some time later instead.
496 * Plugged queues are generally unplugged when a buffer belonging to one
497 * of the requests on the queue is needed, or due to memory pressure.
499 * @rfn is not required, or even expected, to remove all requests off the
500 * queue, but only as many as it can handle at a time. If it does leave
501 * requests on the queue, it is responsible for arranging that the requests
502 * get dealt with eventually.
504 * The queue spin lock must be held while manipulating the requests on the
505 * request queue; this lock will be taken also from interrupt context, so irq
506 * disabling is needed for it.
508 * Function returns a pointer to the initialized request queue, or %NULL if
512 * blk_init_queue() must be paired with a blk_cleanup_queue() call
513 * when the block device is deactivated (such as at module unload).
516 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
518 return blk_init_queue_node(rfn, lock, -1);
520 EXPORT_SYMBOL(blk_init_queue);
522 struct request_queue *
523 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
525 struct request_queue *uninit_q, *q;
527 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
531 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
533 blk_cleanup_queue(uninit_q);
537 EXPORT_SYMBOL(blk_init_queue_node);
539 struct request_queue *
540 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
543 return blk_init_allocated_queue_node(q, rfn, lock, -1);
545 EXPORT_SYMBOL(blk_init_allocated_queue);
547 struct request_queue *
548 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
549 spinlock_t *lock, int node_id)
555 if (blk_init_free_list(q))
559 q->prep_rq_fn = NULL;
560 q->unprep_rq_fn = NULL;
561 q->queue_flags = QUEUE_FLAG_DEFAULT;
563 /* Override internal queue lock with supplied lock pointer */
565 q->queue_lock = lock;
568 * This also sets hw/phys segments, boundary and size
570 blk_queue_make_request(q, blk_queue_bio);
572 q->sg_reserved_size = INT_MAX;
577 if (!elevator_init(q, NULL)) {
578 blk_queue_congestion_threshold(q);
584 EXPORT_SYMBOL(blk_init_allocated_queue_node);
586 int blk_get_queue(struct request_queue *q)
588 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
589 kobject_get(&q->kobj);
595 EXPORT_SYMBOL(blk_get_queue);
597 static inline void blk_free_request(struct request_queue *q, struct request *rq)
599 if (rq->cmd_flags & REQ_ELVPRIV)
600 elv_put_request(q, rq);
601 mempool_free(rq, q->rq.rq_pool);
604 static struct request *
605 blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
607 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
614 rq->cmd_flags = flags | REQ_ALLOCED;
616 if ((flags & REQ_ELVPRIV) &&
617 unlikely(elv_set_request(q, rq, gfp_mask))) {
618 mempool_free(rq, q->rq.rq_pool);
626 * ioc_batching returns true if the ioc is a valid batching request and
627 * should be given priority access to a request.
629 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
635 * Make sure the process is able to allocate at least 1 request
636 * even if the batch times out, otherwise we could theoretically
639 return ioc->nr_batch_requests == q->nr_batching ||
640 (ioc->nr_batch_requests > 0
641 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
645 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
646 * will cause the process to be a "batcher" on all queues in the system. This
647 * is the behaviour we want though - once it gets a wakeup it should be given
650 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
652 if (!ioc || ioc_batching(q, ioc))
655 ioc->nr_batch_requests = q->nr_batching;
656 ioc->last_waited = jiffies;
659 static void __freed_request(struct request_queue *q, int sync)
661 struct request_list *rl = &q->rq;
663 if (rl->count[sync] < queue_congestion_off_threshold(q))
664 blk_clear_queue_congested(q, sync);
666 if (rl->count[sync] + 1 <= q->nr_requests) {
667 if (waitqueue_active(&rl->wait[sync]))
668 wake_up(&rl->wait[sync]);
670 blk_clear_queue_full(q, sync);
675 * A request has just been released. Account for it, update the full and
676 * congestion status, wake up any waiters. Called under q->queue_lock.
678 static void freed_request(struct request_queue *q, unsigned int flags)
680 struct request_list *rl = &q->rq;
681 int sync = rw_is_sync(flags);
684 if (flags & REQ_ELVPRIV)
687 __freed_request(q, sync);
689 if (unlikely(rl->starved[sync ^ 1]))
690 __freed_request(q, sync ^ 1);
694 * Determine if elevator data should be initialized when allocating the
695 * request associated with @bio.
697 static bool blk_rq_should_init_elevator(struct bio *bio)
703 * Flush requests do not use the elevator so skip initialization.
704 * This allows a request to share the flush and elevator data.
706 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
713 * Get a free request, queue_lock must be held.
714 * Returns NULL on failure, with queue_lock held.
715 * Returns !NULL on success, with queue_lock *not held*.
717 static struct request *get_request(struct request_queue *q, int rw_flags,
718 struct bio *bio, gfp_t gfp_mask)
720 struct request *rq = NULL;
721 struct request_list *rl = &q->rq;
722 struct io_context *ioc = NULL;
723 const bool is_sync = rw_is_sync(rw_flags) != 0;
726 may_queue = elv_may_queue(q, rw_flags);
727 if (may_queue == ELV_MQUEUE_NO)
730 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
731 if (rl->count[is_sync]+1 >= q->nr_requests) {
732 ioc = current_io_context(GFP_ATOMIC, q->node);
734 * The queue will fill after this allocation, so set
735 * it as full, and mark this process as "batching".
736 * This process will be allowed to complete a batch of
737 * requests, others will be blocked.
739 if (!blk_queue_full(q, is_sync)) {
740 ioc_set_batching(q, ioc);
741 blk_set_queue_full(q, is_sync);
743 if (may_queue != ELV_MQUEUE_MUST
744 && !ioc_batching(q, ioc)) {
746 * The queue is full and the allocating
747 * process is not a "batcher", and not
748 * exempted by the IO scheduler
754 blk_set_queue_congested(q, is_sync);
758 * Only allow batching queuers to allocate up to 50% over the defined
759 * limit of requests, otherwise we could have thousands of requests
760 * allocated with any setting of ->nr_requests
762 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
765 rl->count[is_sync]++;
766 rl->starved[is_sync] = 0;
768 if (blk_rq_should_init_elevator(bio) &&
769 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
770 rw_flags |= REQ_ELVPRIV;
774 if (blk_queue_io_stat(q))
775 rw_flags |= REQ_IO_STAT;
776 spin_unlock_irq(q->queue_lock);
778 rq = blk_alloc_request(q, rw_flags, gfp_mask);
781 * Allocation failed presumably due to memory. Undo anything
782 * we might have messed up.
784 * Allocating task should really be put onto the front of the
785 * wait queue, but this is pretty rare.
787 spin_lock_irq(q->queue_lock);
788 freed_request(q, rw_flags);
791 * in the very unlikely event that allocation failed and no
792 * requests for this direction was pending, mark us starved
793 * so that freeing of a request in the other direction will
794 * notice us. another possible fix would be to split the
795 * rq mempool into READ and WRITE
798 if (unlikely(rl->count[is_sync] == 0))
799 rl->starved[is_sync] = 1;
805 * ioc may be NULL here, and ioc_batching will be false. That's
806 * OK, if the queue is under the request limit then requests need
807 * not count toward the nr_batch_requests limit. There will always
808 * be some limit enforced by BLK_BATCH_TIME.
810 if (ioc_batching(q, ioc))
811 ioc->nr_batch_requests--;
813 trace_block_getrq(q, bio, rw_flags & 1);
819 * No available requests for this queue, wait for some requests to become
822 * Called with q->queue_lock held, and returns with it unlocked.
824 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
827 const bool is_sync = rw_is_sync(rw_flags) != 0;
830 rq = get_request(q, rw_flags, bio, GFP_NOIO);
833 struct io_context *ioc;
834 struct request_list *rl = &q->rq;
836 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
837 TASK_UNINTERRUPTIBLE);
839 trace_block_sleeprq(q, bio, rw_flags & 1);
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 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
869 BUG_ON(rw != READ && rw != WRITE);
871 spin_lock_irq(q->queue_lock);
872 if (gfp_mask & __GFP_WAIT) {
873 rq = get_request_wait(q, rw, NULL);
875 rq = get_request(q, rw, NULL, gfp_mask);
877 spin_unlock_irq(q->queue_lock);
879 /* q->queue_lock is unlocked at this point */
883 EXPORT_SYMBOL(blk_get_request);
886 * blk_make_request - given a bio, allocate a corresponding struct request.
887 * @q: target request queue
888 * @bio: The bio describing the memory mappings that will be submitted for IO.
889 * It may be a chained-bio properly constructed by block/bio layer.
890 * @gfp_mask: gfp flags to be used for memory allocation
892 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
893 * type commands. Where the struct request needs to be farther initialized by
894 * the caller. It is passed a &struct bio, which describes the memory info of
897 * The caller of blk_make_request must make sure that bi_io_vec
898 * are set to describe the memory buffers. That bio_data_dir() will return
899 * the needed direction of the request. (And all bio's in the passed bio-chain
900 * are properly set accordingly)
902 * If called under none-sleepable conditions, mapped bio buffers must not
903 * need bouncing, by calling the appropriate masked or flagged allocator,
904 * suitable for the target device. Otherwise the call to blk_queue_bounce will
907 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
908 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
909 * anything but the first bio in the chain. Otherwise you risk waiting for IO
910 * completion of a bio that hasn't been submitted yet, thus resulting in a
911 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
912 * of bio_alloc(), as that avoids the mempool deadlock.
913 * If possible a big IO should be split into smaller parts when allocation
914 * fails. Partial allocation should not be an error, or you risk a live-lock.
916 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
919 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
922 return ERR_PTR(-ENOMEM);
925 struct bio *bounce_bio = bio;
928 blk_queue_bounce(q, &bounce_bio);
929 ret = blk_rq_append_bio(q, rq, bounce_bio);
938 EXPORT_SYMBOL(blk_make_request);
941 * blk_requeue_request - put a request back on queue
942 * @q: request queue where request should be inserted
943 * @rq: request to be inserted
946 * Drivers often keep queueing requests until the hardware cannot accept
947 * more, when that condition happens we need to put the request back
948 * on the queue. Must be called with queue lock held.
950 void blk_requeue_request(struct request_queue *q, struct request *rq)
952 blk_delete_timer(rq);
953 blk_clear_rq_complete(rq);
954 trace_block_rq_requeue(q, rq);
956 if (blk_rq_tagged(rq))
957 blk_queue_end_tag(q, rq);
959 BUG_ON(blk_queued_rq(rq));
961 elv_requeue_request(q, rq);
963 EXPORT_SYMBOL(blk_requeue_request);
965 static void add_acct_request(struct request_queue *q, struct request *rq,
968 drive_stat_acct(rq, 1);
969 __elv_add_request(q, rq, where);
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 add_acct_request(q, rq, where);
1016 spin_unlock_irqrestore(q->queue_lock, flags);
1018 EXPORT_SYMBOL(blk_insert_request);
1020 static void part_round_stats_single(int cpu, struct hd_struct *part,
1023 if (now == part->stamp)
1026 if (part_in_flight(part)) {
1027 __part_stat_add(cpu, part, time_in_queue,
1028 part_in_flight(part) * (now - part->stamp));
1029 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1035 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1036 * @cpu: cpu number for stats access
1037 * @part: target partition
1039 * The average IO queue length and utilisation statistics are maintained
1040 * by observing the current state of the queue length and the amount of
1041 * time it has been in this state for.
1043 * Normally, that accounting is done on IO completion, but that can result
1044 * in more than a second's worth of IO being accounted for within any one
1045 * second, leading to >100% utilisation. To deal with that, we call this
1046 * function to do a round-off before returning the results when reading
1047 * /proc/diskstats. This accounts immediately for all queue usage up to
1048 * the current jiffies and restarts the counters again.
1050 void part_round_stats(int cpu, struct hd_struct *part)
1052 unsigned long now = jiffies;
1055 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1056 part_round_stats_single(cpu, part, now);
1058 EXPORT_SYMBOL_GPL(part_round_stats);
1061 * queue lock must be held
1063 void __blk_put_request(struct request_queue *q, struct request *req)
1067 if (unlikely(--req->ref_count))
1070 elv_completed_request(q, req);
1072 /* this is a bio leak */
1073 WARN_ON(req->bio != NULL);
1076 * Request may not have originated from ll_rw_blk. if not,
1077 * it didn't come out of our reserved rq pools
1079 if (req->cmd_flags & REQ_ALLOCED) {
1080 unsigned int flags = req->cmd_flags;
1082 BUG_ON(!list_empty(&req->queuelist));
1083 BUG_ON(!hlist_unhashed(&req->hash));
1085 blk_free_request(q, req);
1086 freed_request(q, flags);
1089 EXPORT_SYMBOL_GPL(__blk_put_request);
1091 void blk_put_request(struct request *req)
1093 unsigned long flags;
1094 struct request_queue *q = req->q;
1096 spin_lock_irqsave(q->queue_lock, flags);
1097 __blk_put_request(q, req);
1098 spin_unlock_irqrestore(q->queue_lock, flags);
1100 EXPORT_SYMBOL(blk_put_request);
1103 * blk_add_request_payload - add a payload to a request
1104 * @rq: request to update
1105 * @page: page backing the payload
1106 * @len: length of the payload.
1108 * This allows to later add a payload to an already submitted request by
1109 * a block driver. The driver needs to take care of freeing the payload
1112 * Note that this is a quite horrible hack and nothing but handling of
1113 * discard requests should ever use it.
1115 void blk_add_request_payload(struct request *rq, struct page *page,
1118 struct bio *bio = rq->bio;
1120 bio->bi_io_vec->bv_page = page;
1121 bio->bi_io_vec->bv_offset = 0;
1122 bio->bi_io_vec->bv_len = len;
1126 bio->bi_phys_segments = 1;
1128 rq->__data_len = rq->resid_len = len;
1129 rq->nr_phys_segments = 1;
1130 rq->buffer = bio_data(bio);
1132 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1134 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1137 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1139 if (!ll_back_merge_fn(q, req, bio))
1142 trace_block_bio_backmerge(q, bio);
1144 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1145 blk_rq_set_mixed_merge(req);
1147 req->biotail->bi_next = bio;
1149 req->__data_len += bio->bi_size;
1150 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1152 drive_stat_acct(req, 0);
1153 elv_bio_merged(q, req, bio);
1157 static bool bio_attempt_front_merge(struct request_queue *q,
1158 struct request *req, struct bio *bio)
1160 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1162 if (!ll_front_merge_fn(q, req, bio))
1165 trace_block_bio_frontmerge(q, bio);
1167 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1168 blk_rq_set_mixed_merge(req);
1170 bio->bi_next = req->bio;
1174 * may not be valid. if the low level driver said
1175 * it didn't need a bounce buffer then it better
1176 * not touch req->buffer either...
1178 req->buffer = bio_data(bio);
1179 req->__sector = bio->bi_sector;
1180 req->__data_len += bio->bi_size;
1181 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1183 drive_stat_acct(req, 0);
1184 elv_bio_merged(q, req, bio);
1189 * Attempts to merge with the plugged list in the current process. Returns
1190 * true if merge was successful, otherwise false.
1192 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1193 struct bio *bio, unsigned int *request_count)
1195 struct blk_plug *plug;
1204 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1212 el_ret = elv_try_merge(rq, bio);
1213 if (el_ret == ELEVATOR_BACK_MERGE) {
1214 ret = bio_attempt_back_merge(q, rq, bio);
1217 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1218 ret = bio_attempt_front_merge(q, rq, bio);
1227 void init_request_from_bio(struct request *req, struct bio *bio)
1229 req->cpu = bio->bi_comp_cpu;
1230 req->cmd_type = REQ_TYPE_FS;
1232 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1233 if (bio->bi_rw & REQ_RAHEAD)
1234 req->cmd_flags |= REQ_FAILFAST_MASK;
1237 req->__sector = bio->bi_sector;
1238 req->ioprio = bio_prio(bio);
1239 blk_rq_bio_prep(req->q, req, bio);
1242 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1244 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1245 struct blk_plug *plug;
1246 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1247 struct request *req;
1248 unsigned int request_count = 0;
1251 * low level driver can indicate that it wants pages above a
1252 * certain limit bounced to low memory (ie for highmem, or even
1253 * ISA dma in theory)
1255 blk_queue_bounce(q, &bio);
1257 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1258 spin_lock_irq(q->queue_lock);
1259 where = ELEVATOR_INSERT_FLUSH;
1264 * Check if we can merge with the plugged list before grabbing
1267 if (attempt_plug_merge(current, q, bio, &request_count))
1270 spin_lock_irq(q->queue_lock);
1272 el_ret = elv_merge(q, &req, bio);
1273 if (el_ret == ELEVATOR_BACK_MERGE) {
1274 if (bio_attempt_back_merge(q, req, bio)) {
1275 if (!attempt_back_merge(q, req))
1276 elv_merged_request(q, req, el_ret);
1279 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1280 if (bio_attempt_front_merge(q, req, bio)) {
1281 if (!attempt_front_merge(q, req))
1282 elv_merged_request(q, req, el_ret);
1289 * This sync check and mask will be re-done in init_request_from_bio(),
1290 * but we need to set it earlier to expose the sync flag to the
1291 * rq allocator and io schedulers.
1293 rw_flags = bio_data_dir(bio);
1295 rw_flags |= REQ_SYNC;
1298 * Grab a free request. This is might sleep but can not fail.
1299 * Returns with the queue unlocked.
1301 req = get_request_wait(q, rw_flags, bio);
1304 * After dropping the lock and possibly sleeping here, our request
1305 * may now be mergeable after it had proven unmergeable (above).
1306 * We don't worry about that case for efficiency. It won't happen
1307 * often, and the elevators are able to handle it.
1309 init_request_from_bio(req, bio);
1311 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1312 bio_flagged(bio, BIO_CPU_AFFINE))
1313 req->cpu = raw_smp_processor_id();
1315 plug = current->plug;
1318 * If this is the first request added after a plug, fire
1319 * of a plug trace. If others have been added before, check
1320 * if we have multiple devices in this plug. If so, make a
1321 * note to sort the list before dispatch.
1323 if (list_empty(&plug->list))
1324 trace_block_plug(q);
1325 else if (!plug->should_sort) {
1326 struct request *__rq;
1328 __rq = list_entry_rq(plug->list.prev);
1330 plug->should_sort = 1;
1332 if (request_count >= BLK_MAX_REQUEST_COUNT)
1333 blk_flush_plug_list(plug, false);
1334 list_add_tail(&req->queuelist, &plug->list);
1335 drive_stat_acct(req, 1);
1337 spin_lock_irq(q->queue_lock);
1338 add_acct_request(q, req, where);
1341 spin_unlock_irq(q->queue_lock);
1344 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1347 * If bio->bi_dev is a partition, remap the location
1349 static inline void blk_partition_remap(struct bio *bio)
1351 struct block_device *bdev = bio->bi_bdev;
1353 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1354 struct hd_struct *p = bdev->bd_part;
1356 bio->bi_sector += p->start_sect;
1357 bio->bi_bdev = bdev->bd_contains;
1359 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1361 bio->bi_sector - p->start_sect);
1365 static void handle_bad_sector(struct bio *bio)
1367 char b[BDEVNAME_SIZE];
1369 printk(KERN_INFO "attempt to access beyond end of device\n");
1370 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1371 bdevname(bio->bi_bdev, b),
1373 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1374 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1376 set_bit(BIO_EOF, &bio->bi_flags);
1379 #ifdef CONFIG_FAIL_MAKE_REQUEST
1381 static DECLARE_FAULT_ATTR(fail_make_request);
1383 static int __init setup_fail_make_request(char *str)
1385 return setup_fault_attr(&fail_make_request, str);
1387 __setup("fail_make_request=", setup_fail_make_request);
1389 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1391 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1394 static int __init fail_make_request_debugfs(void)
1396 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1397 NULL, &fail_make_request);
1399 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1402 late_initcall(fail_make_request_debugfs);
1404 #else /* CONFIG_FAIL_MAKE_REQUEST */
1406 static inline bool should_fail_request(struct hd_struct *part,
1412 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1415 * Check whether this bio extends beyond the end of the device.
1417 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1424 /* Test device or partition size, when known. */
1425 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1427 sector_t sector = bio->bi_sector;
1429 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1431 * This may well happen - the kernel calls bread()
1432 * without checking the size of the device, e.g., when
1433 * mounting a device.
1435 handle_bad_sector(bio);
1443 static noinline_for_stack bool
1444 generic_make_request_checks(struct bio *bio)
1446 struct request_queue *q;
1447 int nr_sectors = bio_sectors(bio);
1449 char b[BDEVNAME_SIZE];
1450 struct hd_struct *part;
1454 if (bio_check_eod(bio, nr_sectors))
1457 q = bdev_get_queue(bio->bi_bdev);
1460 "generic_make_request: Trying to access "
1461 "nonexistent block-device %s (%Lu)\n",
1462 bdevname(bio->bi_bdev, b),
1463 (long long) bio->bi_sector);
1467 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1468 nr_sectors > queue_max_hw_sectors(q))) {
1469 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1470 bdevname(bio->bi_bdev, b),
1472 queue_max_hw_sectors(q));
1476 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1479 part = bio->bi_bdev->bd_part;
1480 if (should_fail_request(part, bio->bi_size) ||
1481 should_fail_request(&part_to_disk(part)->part0,
1486 * If this device has partitions, remap block n
1487 * of partition p to block n+start(p) of the disk.
1489 blk_partition_remap(bio);
1491 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1494 if (bio_check_eod(bio, nr_sectors))
1498 * Filter flush bio's early so that make_request based
1499 * drivers without flush support don't have to worry
1502 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1503 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1510 if ((bio->bi_rw & REQ_DISCARD) &&
1511 (!blk_queue_discard(q) ||
1512 ((bio->bi_rw & REQ_SECURE) &&
1513 !blk_queue_secdiscard(q)))) {
1518 if (blk_throtl_bio(q, &bio))
1521 /* if bio = NULL, bio has been throttled and will be submitted later. */
1525 trace_block_bio_queue(q, bio);
1529 bio_endio(bio, err);
1534 * generic_make_request - hand a buffer to its device driver for I/O
1535 * @bio: The bio describing the location in memory and on the device.
1537 * generic_make_request() is used to make I/O requests of block
1538 * devices. It is passed a &struct bio, which describes the I/O that needs
1541 * generic_make_request() does not return any status. The
1542 * success/failure status of the request, along with notification of
1543 * completion, is delivered asynchronously through the bio->bi_end_io
1544 * function described (one day) else where.
1546 * The caller of generic_make_request must make sure that bi_io_vec
1547 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1548 * set to describe the device address, and the
1549 * bi_end_io and optionally bi_private are set to describe how
1550 * completion notification should be signaled.
1552 * generic_make_request and the drivers it calls may use bi_next if this
1553 * bio happens to be merged with someone else, and may resubmit the bio to
1554 * a lower device by calling into generic_make_request recursively, which
1555 * means the bio should NOT be touched after the call to ->make_request_fn.
1557 void generic_make_request(struct bio *bio)
1559 struct bio_list bio_list_on_stack;
1561 if (!generic_make_request_checks(bio))
1565 * We only want one ->make_request_fn to be active at a time, else
1566 * stack usage with stacked devices could be a problem. So use
1567 * current->bio_list to keep a list of requests submited by a
1568 * make_request_fn function. current->bio_list is also used as a
1569 * flag to say if generic_make_request is currently active in this
1570 * task or not. If it is NULL, then no make_request is active. If
1571 * it is non-NULL, then a make_request is active, and new requests
1572 * should be added at the tail
1574 if (current->bio_list) {
1575 bio_list_add(current->bio_list, bio);
1579 /* following loop may be a bit non-obvious, and so deserves some
1581 * Before entering the loop, bio->bi_next is NULL (as all callers
1582 * ensure that) so we have a list with a single bio.
1583 * We pretend that we have just taken it off a longer list, so
1584 * we assign bio_list to a pointer to the bio_list_on_stack,
1585 * thus initialising the bio_list of new bios to be
1586 * added. ->make_request() may indeed add some more bios
1587 * through a recursive call to generic_make_request. If it
1588 * did, we find a non-NULL value in bio_list and re-enter the loop
1589 * from the top. In this case we really did just take the bio
1590 * of the top of the list (no pretending) and so remove it from
1591 * bio_list, and call into ->make_request() again.
1593 BUG_ON(bio->bi_next);
1594 bio_list_init(&bio_list_on_stack);
1595 current->bio_list = &bio_list_on_stack;
1597 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1599 q->make_request_fn(q, bio);
1601 bio = bio_list_pop(current->bio_list);
1603 current->bio_list = NULL; /* deactivate */
1605 EXPORT_SYMBOL(generic_make_request);
1608 * submit_bio - submit a bio to the block device layer for I/O
1609 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1610 * @bio: The &struct bio which describes the I/O
1612 * submit_bio() is very similar in purpose to generic_make_request(), and
1613 * uses that function to do most of the work. Both are fairly rough
1614 * interfaces; @bio must be presetup and ready for I/O.
1617 void submit_bio(int rw, struct bio *bio)
1619 int count = bio_sectors(bio);
1624 * If it's a regular read/write or a barrier with data attached,
1625 * go through the normal accounting stuff before submission.
1627 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1629 count_vm_events(PGPGOUT, count);
1631 task_io_account_read(bio->bi_size);
1632 count_vm_events(PGPGIN, count);
1635 if (unlikely(block_dump)) {
1636 char b[BDEVNAME_SIZE];
1637 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1638 current->comm, task_pid_nr(current),
1639 (rw & WRITE) ? "WRITE" : "READ",
1640 (unsigned long long)bio->bi_sector,
1641 bdevname(bio->bi_bdev, b),
1646 generic_make_request(bio);
1648 EXPORT_SYMBOL(submit_bio);
1651 * blk_rq_check_limits - Helper function to check a request for the queue limit
1653 * @rq: the request being checked
1656 * @rq may have been made based on weaker limitations of upper-level queues
1657 * in request stacking drivers, and it may violate the limitation of @q.
1658 * Since the block layer and the underlying device driver trust @rq
1659 * after it is inserted to @q, it should be checked against @q before
1660 * the insertion using this generic function.
1662 * This function should also be useful for request stacking drivers
1663 * in some cases below, so export this function.
1664 * Request stacking drivers like request-based dm may change the queue
1665 * limits while requests are in the queue (e.g. dm's table swapping).
1666 * Such request stacking drivers should check those requests agaist
1667 * the new queue limits again when they dispatch those requests,
1668 * although such checkings are also done against the old queue limits
1669 * when submitting requests.
1671 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1673 if (rq->cmd_flags & REQ_DISCARD)
1676 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1677 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1678 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1683 * queue's settings related to segment counting like q->bounce_pfn
1684 * may differ from that of other stacking queues.
1685 * Recalculate it to check the request correctly on this queue's
1688 blk_recalc_rq_segments(rq);
1689 if (rq->nr_phys_segments > queue_max_segments(q)) {
1690 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1696 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1699 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1700 * @q: the queue to submit the request
1701 * @rq: the request being queued
1703 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1705 unsigned long flags;
1706 int where = ELEVATOR_INSERT_BACK;
1708 if (blk_rq_check_limits(q, rq))
1712 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1715 spin_lock_irqsave(q->queue_lock, flags);
1718 * Submitting request must be dequeued before calling this function
1719 * because it will be linked to another request_queue
1721 BUG_ON(blk_queued_rq(rq));
1723 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1724 where = ELEVATOR_INSERT_FLUSH;
1726 add_acct_request(q, rq, where);
1727 spin_unlock_irqrestore(q->queue_lock, flags);
1731 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1734 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1735 * @rq: request to examine
1738 * A request could be merge of IOs which require different failure
1739 * handling. This function determines the number of bytes which
1740 * can be failed from the beginning of the request without
1741 * crossing into area which need to be retried further.
1744 * The number of bytes to fail.
1747 * queue_lock must be held.
1749 unsigned int blk_rq_err_bytes(const struct request *rq)
1751 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1752 unsigned int bytes = 0;
1755 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1756 return blk_rq_bytes(rq);
1759 * Currently the only 'mixing' which can happen is between
1760 * different fastfail types. We can safely fail portions
1761 * which have all the failfast bits that the first one has -
1762 * the ones which are at least as eager to fail as the first
1765 for (bio = rq->bio; bio; bio = bio->bi_next) {
1766 if ((bio->bi_rw & ff) != ff)
1768 bytes += bio->bi_size;
1771 /* this could lead to infinite loop */
1772 BUG_ON(blk_rq_bytes(rq) && !bytes);
1775 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1777 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1779 if (blk_do_io_stat(req)) {
1780 const int rw = rq_data_dir(req);
1781 struct hd_struct *part;
1784 cpu = part_stat_lock();
1786 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1791 static void blk_account_io_done(struct request *req)
1794 * Account IO completion. flush_rq isn't accounted as a
1795 * normal IO on queueing nor completion. Accounting the
1796 * containing request is enough.
1798 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1799 unsigned long duration = jiffies - req->start_time;
1800 const int rw = rq_data_dir(req);
1801 struct hd_struct *part;
1804 cpu = part_stat_lock();
1807 part_stat_inc(cpu, part, ios[rw]);
1808 part_stat_add(cpu, part, ticks[rw], duration);
1809 part_round_stats(cpu, part);
1810 part_dec_in_flight(part, rw);
1812 hd_struct_put(part);
1818 * blk_peek_request - peek at the top of a request queue
1819 * @q: request queue to peek at
1822 * Return the request at the top of @q. The returned request
1823 * should be started using blk_start_request() before LLD starts
1827 * Pointer to the request at the top of @q if available. Null
1831 * queue_lock must be held.
1833 struct request *blk_peek_request(struct request_queue *q)
1838 while ((rq = __elv_next_request(q)) != NULL) {
1839 if (!(rq->cmd_flags & REQ_STARTED)) {
1841 * This is the first time the device driver
1842 * sees this request (possibly after
1843 * requeueing). Notify IO scheduler.
1845 if (rq->cmd_flags & REQ_SORTED)
1846 elv_activate_rq(q, rq);
1849 * just mark as started even if we don't start
1850 * it, a request that has been delayed should
1851 * not be passed by new incoming requests
1853 rq->cmd_flags |= REQ_STARTED;
1854 trace_block_rq_issue(q, rq);
1857 if (!q->boundary_rq || q->boundary_rq == rq) {
1858 q->end_sector = rq_end_sector(rq);
1859 q->boundary_rq = NULL;
1862 if (rq->cmd_flags & REQ_DONTPREP)
1865 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1867 * make sure space for the drain appears we
1868 * know we can do this because max_hw_segments
1869 * has been adjusted to be one fewer than the
1872 rq->nr_phys_segments++;
1878 ret = q->prep_rq_fn(q, rq);
1879 if (ret == BLKPREP_OK) {
1881 } else if (ret == BLKPREP_DEFER) {
1883 * the request may have been (partially) prepped.
1884 * we need to keep this request in the front to
1885 * avoid resource deadlock. REQ_STARTED will
1886 * prevent other fs requests from passing this one.
1888 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1889 !(rq->cmd_flags & REQ_DONTPREP)) {
1891 * remove the space for the drain we added
1892 * so that we don't add it again
1894 --rq->nr_phys_segments;
1899 } else if (ret == BLKPREP_KILL) {
1900 rq->cmd_flags |= REQ_QUIET;
1902 * Mark this request as started so we don't trigger
1903 * any debug logic in the end I/O path.
1905 blk_start_request(rq);
1906 __blk_end_request_all(rq, -EIO);
1908 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1915 EXPORT_SYMBOL(blk_peek_request);
1917 void blk_dequeue_request(struct request *rq)
1919 struct request_queue *q = rq->q;
1921 BUG_ON(list_empty(&rq->queuelist));
1922 BUG_ON(ELV_ON_HASH(rq));
1924 list_del_init(&rq->queuelist);
1927 * the time frame between a request being removed from the lists
1928 * and to it is freed is accounted as io that is in progress at
1931 if (blk_account_rq(rq)) {
1932 q->in_flight[rq_is_sync(rq)]++;
1933 set_io_start_time_ns(rq);
1938 * blk_start_request - start request processing on the driver
1939 * @req: request to dequeue
1942 * Dequeue @req and start timeout timer on it. This hands off the
1943 * request to the driver.
1945 * Block internal functions which don't want to start timer should
1946 * call blk_dequeue_request().
1949 * queue_lock must be held.
1951 void blk_start_request(struct request *req)
1953 blk_dequeue_request(req);
1956 * We are now handing the request to the hardware, initialize
1957 * resid_len to full count and add the timeout handler.
1959 req->resid_len = blk_rq_bytes(req);
1960 if (unlikely(blk_bidi_rq(req)))
1961 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1965 EXPORT_SYMBOL(blk_start_request);
1968 * blk_fetch_request - fetch a request from a request queue
1969 * @q: request queue to fetch a request from
1972 * Return the request at the top of @q. The request is started on
1973 * return and LLD can start processing it immediately.
1976 * Pointer to the request at the top of @q if available. Null
1980 * queue_lock must be held.
1982 struct request *blk_fetch_request(struct request_queue *q)
1986 rq = blk_peek_request(q);
1988 blk_start_request(rq);
1991 EXPORT_SYMBOL(blk_fetch_request);
1994 * blk_update_request - Special helper function for request stacking drivers
1995 * @req: the request being processed
1996 * @error: %0 for success, < %0 for error
1997 * @nr_bytes: number of bytes to complete @req
2000 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2001 * the request structure even if @req doesn't have leftover.
2002 * If @req has leftover, sets it up for the next range of segments.
2004 * This special helper function is only for request stacking drivers
2005 * (e.g. request-based dm) so that they can handle partial completion.
2006 * Actual device drivers should use blk_end_request instead.
2008 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2009 * %false return from this function.
2012 * %false - this request doesn't have any more data
2013 * %true - this request has more data
2015 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2017 int total_bytes, bio_nbytes, next_idx = 0;
2023 trace_block_rq_complete(req->q, req);
2026 * For fs requests, rq is just carrier of independent bio's
2027 * and each partial completion should be handled separately.
2028 * Reset per-request error on each partial completion.
2030 * TODO: tj: This is too subtle. It would be better to let
2031 * low level drivers do what they see fit.
2033 if (req->cmd_type == REQ_TYPE_FS)
2036 if (error && req->cmd_type == REQ_TYPE_FS &&
2037 !(req->cmd_flags & REQ_QUIET)) {
2042 error_type = "recoverable transport";
2045 error_type = "critical target";
2048 error_type = "critical nexus";
2055 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2056 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2057 (unsigned long long)blk_rq_pos(req));
2060 blk_account_io_completion(req, nr_bytes);
2062 total_bytes = bio_nbytes = 0;
2063 while ((bio = req->bio) != NULL) {
2066 if (nr_bytes >= bio->bi_size) {
2067 req->bio = bio->bi_next;
2068 nbytes = bio->bi_size;
2069 req_bio_endio(req, bio, nbytes, error);
2073 int idx = bio->bi_idx + next_idx;
2075 if (unlikely(idx >= bio->bi_vcnt)) {
2076 blk_dump_rq_flags(req, "__end_that");
2077 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2078 __func__, idx, bio->bi_vcnt);
2082 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2083 BIO_BUG_ON(nbytes > bio->bi_size);
2086 * not a complete bvec done
2088 if (unlikely(nbytes > nr_bytes)) {
2089 bio_nbytes += nr_bytes;
2090 total_bytes += nr_bytes;
2095 * advance to the next vector
2098 bio_nbytes += nbytes;
2101 total_bytes += nbytes;
2107 * end more in this run, or just return 'not-done'
2109 if (unlikely(nr_bytes <= 0))
2119 * Reset counters so that the request stacking driver
2120 * can find how many bytes remain in the request
2123 req->__data_len = 0;
2128 * if the request wasn't completed, update state
2131 req_bio_endio(req, bio, bio_nbytes, error);
2132 bio->bi_idx += next_idx;
2133 bio_iovec(bio)->bv_offset += nr_bytes;
2134 bio_iovec(bio)->bv_len -= nr_bytes;
2137 req->__data_len -= total_bytes;
2138 req->buffer = bio_data(req->bio);
2140 /* update sector only for requests with clear definition of sector */
2141 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2142 req->__sector += total_bytes >> 9;
2144 /* mixed attributes always follow the first bio */
2145 if (req->cmd_flags & REQ_MIXED_MERGE) {
2146 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2147 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2151 * If total number of sectors is less than the first segment
2152 * size, something has gone terribly wrong.
2154 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2155 blk_dump_rq_flags(req, "request botched");
2156 req->__data_len = blk_rq_cur_bytes(req);
2159 /* recalculate the number of segments */
2160 blk_recalc_rq_segments(req);
2164 EXPORT_SYMBOL_GPL(blk_update_request);
2166 static bool blk_update_bidi_request(struct request *rq, int error,
2167 unsigned int nr_bytes,
2168 unsigned int bidi_bytes)
2170 if (blk_update_request(rq, error, nr_bytes))
2173 /* Bidi request must be completed as a whole */
2174 if (unlikely(blk_bidi_rq(rq)) &&
2175 blk_update_request(rq->next_rq, error, bidi_bytes))
2178 if (blk_queue_add_random(rq->q))
2179 add_disk_randomness(rq->rq_disk);
2185 * blk_unprep_request - unprepare a request
2188 * This function makes a request ready for complete resubmission (or
2189 * completion). It happens only after all error handling is complete,
2190 * so represents the appropriate moment to deallocate any resources
2191 * that were allocated to the request in the prep_rq_fn. The queue
2192 * lock is held when calling this.
2194 void blk_unprep_request(struct request *req)
2196 struct request_queue *q = req->q;
2198 req->cmd_flags &= ~REQ_DONTPREP;
2199 if (q->unprep_rq_fn)
2200 q->unprep_rq_fn(q, req);
2202 EXPORT_SYMBOL_GPL(blk_unprep_request);
2205 * queue lock must be held
2207 static void blk_finish_request(struct request *req, int error)
2209 if (blk_rq_tagged(req))
2210 blk_queue_end_tag(req->q, req);
2212 BUG_ON(blk_queued_rq(req));
2214 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2215 laptop_io_completion(&req->q->backing_dev_info);
2217 blk_delete_timer(req);
2219 if (req->cmd_flags & REQ_DONTPREP)
2220 blk_unprep_request(req);
2223 blk_account_io_done(req);
2226 req->end_io(req, error);
2228 if (blk_bidi_rq(req))
2229 __blk_put_request(req->next_rq->q, req->next_rq);
2231 __blk_put_request(req->q, req);
2236 * blk_end_bidi_request - Complete a bidi request
2237 * @rq: the request to complete
2238 * @error: %0 for success, < %0 for error
2239 * @nr_bytes: number of bytes to complete @rq
2240 * @bidi_bytes: number of bytes to complete @rq->next_rq
2243 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2244 * Drivers that supports bidi can safely call this member for any
2245 * type of request, bidi or uni. In the later case @bidi_bytes is
2249 * %false - we are done with this request
2250 * %true - still buffers pending for this request
2252 static bool blk_end_bidi_request(struct request *rq, int error,
2253 unsigned int nr_bytes, unsigned int bidi_bytes)
2255 struct request_queue *q = rq->q;
2256 unsigned long flags;
2258 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2261 spin_lock_irqsave(q->queue_lock, flags);
2262 blk_finish_request(rq, error);
2263 spin_unlock_irqrestore(q->queue_lock, flags);
2269 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2270 * @rq: the request to complete
2271 * @error: %0 for success, < %0 for error
2272 * @nr_bytes: number of bytes to complete @rq
2273 * @bidi_bytes: number of bytes to complete @rq->next_rq
2276 * Identical to blk_end_bidi_request() except that queue lock is
2277 * assumed to be locked on entry and remains so on return.
2280 * %false - we are done with this request
2281 * %true - still buffers pending for this request
2283 bool __blk_end_bidi_request(struct request *rq, int error,
2284 unsigned int nr_bytes, unsigned int bidi_bytes)
2286 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2289 blk_finish_request(rq, error);
2295 * blk_end_request - Helper function for drivers to complete the request.
2296 * @rq: the request being processed
2297 * @error: %0 for success, < %0 for error
2298 * @nr_bytes: number of bytes to complete
2301 * Ends I/O on a number of bytes attached to @rq.
2302 * If @rq has leftover, sets it up for the next range of segments.
2305 * %false - we are done with this request
2306 * %true - still buffers pending for this request
2308 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2310 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2312 EXPORT_SYMBOL(blk_end_request);
2315 * blk_end_request_all - Helper function for drives to finish the request.
2316 * @rq: the request to finish
2317 * @error: %0 for success, < %0 for error
2320 * Completely finish @rq.
2322 void blk_end_request_all(struct request *rq, int error)
2325 unsigned int bidi_bytes = 0;
2327 if (unlikely(blk_bidi_rq(rq)))
2328 bidi_bytes = blk_rq_bytes(rq->next_rq);
2330 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2333 EXPORT_SYMBOL(blk_end_request_all);
2336 * blk_end_request_cur - Helper function to finish the current request chunk.
2337 * @rq: the request to finish the current chunk for
2338 * @error: %0 for success, < %0 for error
2341 * Complete the current consecutively mapped chunk from @rq.
2344 * %false - we are done with this request
2345 * %true - still buffers pending for this request
2347 bool blk_end_request_cur(struct request *rq, int error)
2349 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2351 EXPORT_SYMBOL(blk_end_request_cur);
2354 * blk_end_request_err - Finish a request till the next failure boundary.
2355 * @rq: the request to finish till the next failure boundary for
2356 * @error: must be negative errno
2359 * Complete @rq till the next failure boundary.
2362 * %false - we are done with this request
2363 * %true - still buffers pending for this request
2365 bool blk_end_request_err(struct request *rq, int error)
2367 WARN_ON(error >= 0);
2368 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2370 EXPORT_SYMBOL_GPL(blk_end_request_err);
2373 * __blk_end_request - Helper function for drivers to complete the request.
2374 * @rq: the request being processed
2375 * @error: %0 for success, < %0 for error
2376 * @nr_bytes: number of bytes to complete
2379 * Must be called with queue lock held unlike blk_end_request().
2382 * %false - we are done with this request
2383 * %true - still buffers pending for this request
2385 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2387 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2389 EXPORT_SYMBOL(__blk_end_request);
2392 * __blk_end_request_all - Helper function for drives to finish the request.
2393 * @rq: the request to finish
2394 * @error: %0 for success, < %0 for error
2397 * Completely finish @rq. Must be called with queue lock held.
2399 void __blk_end_request_all(struct request *rq, int error)
2402 unsigned int bidi_bytes = 0;
2404 if (unlikely(blk_bidi_rq(rq)))
2405 bidi_bytes = blk_rq_bytes(rq->next_rq);
2407 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2410 EXPORT_SYMBOL(__blk_end_request_all);
2413 * __blk_end_request_cur - Helper function to finish the current request chunk.
2414 * @rq: the request to finish the current chunk for
2415 * @error: %0 for success, < %0 for error
2418 * Complete the current consecutively mapped chunk from @rq. Must
2419 * be called with queue lock held.
2422 * %false - we are done with this request
2423 * %true - still buffers pending for this request
2425 bool __blk_end_request_cur(struct request *rq, int error)
2427 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2429 EXPORT_SYMBOL(__blk_end_request_cur);
2432 * __blk_end_request_err - Finish a request till the next failure boundary.
2433 * @rq: the request to finish till the next failure boundary for
2434 * @error: must be negative errno
2437 * Complete @rq till the next failure boundary. Must be called
2438 * with queue lock held.
2441 * %false - we are done with this request
2442 * %true - still buffers pending for this request
2444 bool __blk_end_request_err(struct request *rq, int error)
2446 WARN_ON(error >= 0);
2447 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2449 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2451 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2454 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2455 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2457 if (bio_has_data(bio)) {
2458 rq->nr_phys_segments = bio_phys_segments(q, bio);
2459 rq->buffer = bio_data(bio);
2461 rq->__data_len = bio->bi_size;
2462 rq->bio = rq->biotail = bio;
2465 rq->rq_disk = bio->bi_bdev->bd_disk;
2468 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2470 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2471 * @rq: the request to be flushed
2474 * Flush all pages in @rq.
2476 void rq_flush_dcache_pages(struct request *rq)
2478 struct req_iterator iter;
2479 struct bio_vec *bvec;
2481 rq_for_each_segment(bvec, rq, iter)
2482 flush_dcache_page(bvec->bv_page);
2484 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2488 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2489 * @q : the queue of the device being checked
2492 * Check if underlying low-level drivers of a device are busy.
2493 * If the drivers want to export their busy state, they must set own
2494 * exporting function using blk_queue_lld_busy() first.
2496 * Basically, this function is used only by request stacking drivers
2497 * to stop dispatching requests to underlying devices when underlying
2498 * devices are busy. This behavior helps more I/O merging on the queue
2499 * of the request stacking driver and prevents I/O throughput regression
2500 * on burst I/O load.
2503 * 0 - Not busy (The request stacking driver should dispatch request)
2504 * 1 - Busy (The request stacking driver should stop dispatching request)
2506 int blk_lld_busy(struct request_queue *q)
2509 return q->lld_busy_fn(q);
2513 EXPORT_SYMBOL_GPL(blk_lld_busy);
2516 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2517 * @rq: the clone request to be cleaned up
2520 * Free all bios in @rq for a cloned request.
2522 void blk_rq_unprep_clone(struct request *rq)
2526 while ((bio = rq->bio) != NULL) {
2527 rq->bio = bio->bi_next;
2532 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2535 * Copy attributes of the original request to the clone request.
2536 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2538 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2540 dst->cpu = src->cpu;
2541 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2542 dst->cmd_type = src->cmd_type;
2543 dst->__sector = blk_rq_pos(src);
2544 dst->__data_len = blk_rq_bytes(src);
2545 dst->nr_phys_segments = src->nr_phys_segments;
2546 dst->ioprio = src->ioprio;
2547 dst->extra_len = src->extra_len;
2551 * blk_rq_prep_clone - Helper function to setup clone request
2552 * @rq: the request to be setup
2553 * @rq_src: original request to be cloned
2554 * @bs: bio_set that bios for clone are allocated from
2555 * @gfp_mask: memory allocation mask for bio
2556 * @bio_ctr: setup function to be called for each clone bio.
2557 * Returns %0 for success, non %0 for failure.
2558 * @data: private data to be passed to @bio_ctr
2561 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2562 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2563 * are not copied, and copying such parts is the caller's responsibility.
2564 * Also, pages which the original bios are pointing to are not copied
2565 * and the cloned bios just point same pages.
2566 * So cloned bios must be completed before original bios, which means
2567 * the caller must complete @rq before @rq_src.
2569 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2570 struct bio_set *bs, gfp_t gfp_mask,
2571 int (*bio_ctr)(struct bio *, struct bio *, void *),
2574 struct bio *bio, *bio_src;
2579 blk_rq_init(NULL, rq);
2581 __rq_for_each_bio(bio_src, rq_src) {
2582 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2586 __bio_clone(bio, bio_src);
2588 if (bio_integrity(bio_src) &&
2589 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2592 if (bio_ctr && bio_ctr(bio, bio_src, data))
2596 rq->biotail->bi_next = bio;
2599 rq->bio = rq->biotail = bio;
2602 __blk_rq_prep_clone(rq, rq_src);
2609 blk_rq_unprep_clone(rq);
2613 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2615 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2617 return queue_work(kblockd_workqueue, work);
2619 EXPORT_SYMBOL(kblockd_schedule_work);
2621 int kblockd_schedule_delayed_work(struct request_queue *q,
2622 struct delayed_work *dwork, unsigned long delay)
2624 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2626 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2628 #define PLUG_MAGIC 0x91827364
2631 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2632 * @plug: The &struct blk_plug that needs to be initialized
2635 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2636 * pending I/O should the task end up blocking between blk_start_plug() and
2637 * blk_finish_plug(). This is important from a performance perspective, but
2638 * also ensures that we don't deadlock. For instance, if the task is blocking
2639 * for a memory allocation, memory reclaim could end up wanting to free a
2640 * page belonging to that request that is currently residing in our private
2641 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2642 * this kind of deadlock.
2644 void blk_start_plug(struct blk_plug *plug)
2646 struct task_struct *tsk = current;
2648 plug->magic = PLUG_MAGIC;
2649 INIT_LIST_HEAD(&plug->list);
2650 INIT_LIST_HEAD(&plug->cb_list);
2651 plug->should_sort = 0;
2654 * If this is a nested plug, don't actually assign it. It will be
2655 * flushed on its own.
2659 * Store ordering should not be needed here, since a potential
2660 * preempt will imply a full memory barrier
2665 EXPORT_SYMBOL(blk_start_plug);
2667 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2669 struct request *rqa = container_of(a, struct request, queuelist);
2670 struct request *rqb = container_of(b, struct request, queuelist);
2672 return !(rqa->q <= rqb->q);
2676 * If 'from_schedule' is true, then postpone the dispatch of requests
2677 * until a safe kblockd context. We due this to avoid accidental big
2678 * additional stack usage in driver dispatch, in places where the originally
2679 * plugger did not intend it.
2681 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2683 __releases(q->queue_lock)
2685 trace_block_unplug(q, depth, !from_schedule);
2688 * If we are punting this to kblockd, then we can safely drop
2689 * the queue_lock before waking kblockd (which needs to take
2692 if (from_schedule) {
2693 spin_unlock(q->queue_lock);
2694 blk_run_queue_async(q);
2697 spin_unlock(q->queue_lock);
2702 static void flush_plug_callbacks(struct blk_plug *plug)
2704 LIST_HEAD(callbacks);
2706 if (list_empty(&plug->cb_list))
2709 list_splice_init(&plug->cb_list, &callbacks);
2711 while (!list_empty(&callbacks)) {
2712 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2715 list_del(&cb->list);
2720 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2722 struct request_queue *q;
2723 unsigned long flags;
2728 BUG_ON(plug->magic != PLUG_MAGIC);
2730 flush_plug_callbacks(plug);
2731 if (list_empty(&plug->list))
2734 list_splice_init(&plug->list, &list);
2736 if (plug->should_sort) {
2737 list_sort(NULL, &list, plug_rq_cmp);
2738 plug->should_sort = 0;
2745 * Save and disable interrupts here, to avoid doing it for every
2746 * queue lock we have to take.
2748 local_irq_save(flags);
2749 while (!list_empty(&list)) {
2750 rq = list_entry_rq(list.next);
2751 list_del_init(&rq->queuelist);
2755 * This drops the queue lock
2758 queue_unplugged(q, depth, from_schedule);
2761 spin_lock(q->queue_lock);
2764 * rq is already accounted, so use raw insert
2766 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2767 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2769 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2775 * This drops the queue lock
2778 queue_unplugged(q, depth, from_schedule);
2780 local_irq_restore(flags);
2783 void blk_finish_plug(struct blk_plug *plug)
2785 blk_flush_plug_list(plug, false);
2787 if (plug == current->plug)
2788 current->plug = NULL;
2790 EXPORT_SYMBOL(blk_finish_plug);
2792 int __init blk_dev_init(void)
2794 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2795 sizeof(((struct request *)0)->cmd_flags));
2797 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2798 kblockd_workqueue = alloc_workqueue("kblockd",
2799 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2800 if (!kblockd_workqueue)
2801 panic("Failed to create kblockd\n");
2803 request_cachep = kmem_cache_create("blkdev_requests",
2804 sizeof(struct request), 0, SLAB_PANIC, NULL);
2806 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2807 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);