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>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
41 static int __make_request(struct request_queue *q, struct bio *bio);
44 * For the allocated request tables
46 static struct kmem_cache *request_cachep;
49 * For queue allocation
51 struct kmem_cache *blk_requestq_cachep;
54 * Controlling structure to kblockd
56 static struct workqueue_struct *kblockd_workqueue;
58 static void drive_stat_acct(struct request *rq, int new_io)
60 struct hd_struct *part;
61 int rw = rq_data_dir(rq);
64 if (!blk_do_io_stat(rq))
67 cpu = part_stat_lock();
71 part_stat_inc(cpu, part, merges[rw]);
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
74 if (!hd_struct_try_get(part)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part = &rq->rq_disk->part0;
86 part_round_stats(cpu, part);
87 part_inc_in_flight(part, rw);
94 void blk_queue_congestion_threshold(struct request_queue *q)
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
101 q->nr_congestion_on = nr;
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
106 q->nr_congestion_off = nr;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120 struct backing_dev_info *ret = NULL;
121 struct request_queue *q = bdev_get_queue(bdev);
124 ret = &q->backing_dev_info;
127 EXPORT_SYMBOL(blk_get_backing_dev_info);
129 void blk_rq_init(struct request_queue *q, struct request *rq)
131 memset(rq, 0, sizeof(*rq));
133 INIT_LIST_HEAD(&rq->queuelist);
134 INIT_LIST_HEAD(&rq->timeout_list);
137 rq->__sector = (sector_t) -1;
138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
141 rq->cmd_len = BLK_MAX_CDB;
144 rq->start_time = jiffies;
145 set_start_time_ns(rq);
148 EXPORT_SYMBOL(blk_rq_init);
150 static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
178 void blk_dump_rq_flags(struct request *rq, char *msg)
182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
193 printk(KERN_INFO " cdb: ");
194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
195 printk("%02x ", rq->cmd[bit]);
199 EXPORT_SYMBOL(blk_dump_rq_flags);
201 static void blk_delay_work(struct work_struct *work)
203 struct request_queue *q;
205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
208 spin_unlock_irq(q->queue_lock);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
226 EXPORT_SYMBOL(blk_delay_queue);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue *q)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
244 EXPORT_SYMBOL(blk_start_queue);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue *q)
262 __cancel_delayed_work(&q->delay_work);
263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
265 EXPORT_SYMBOL(blk_stop_queue);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue *q)
287 del_timer_sync(&q->timeout);
288 cancel_delayed_work_sync(&q->delay_work);
290 EXPORT_SYMBOL(blk_sync_queue);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue *q)
302 if (unlikely(blk_queue_stopped(q)))
307 EXPORT_SYMBOL(__blk_run_queue);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
317 void blk_run_queue_async(struct request_queue *q)
319 if (likely(!blk_queue_stopped(q))) {
320 __cancel_delayed_work(&q->delay_work);
321 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
324 EXPORT_SYMBOL(blk_run_queue_async);
327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
331 * Invoke request handling on this queue, if it has pending work to do.
332 * May be used to restart queueing when a request has completed.
334 void blk_run_queue(struct request_queue *q)
338 spin_lock_irqsave(q->queue_lock, flags);
340 spin_unlock_irqrestore(q->queue_lock, flags);
342 EXPORT_SYMBOL(blk_run_queue);
344 void blk_put_queue(struct request_queue *q)
346 kobject_put(&q->kobj);
350 * Note: If a driver supplied the queue lock, it should not zap that lock
351 * unexpectedly as some queue cleanup components like elevator_exit() and
352 * blk_throtl_exit() need queue lock.
354 void blk_cleanup_queue(struct request_queue *q)
357 * We know we have process context here, so we can be a little
358 * cautious and ensure that pending block actions on this device
359 * are done before moving on. Going into this function, we should
360 * not have processes doing IO to this device.
364 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
365 mutex_lock(&q->sysfs_lock);
366 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
367 mutex_unlock(&q->sysfs_lock);
370 elevator_exit(q->elevator);
376 EXPORT_SYMBOL(blk_cleanup_queue);
378 static int blk_init_free_list(struct request_queue *q)
380 struct request_list *rl = &q->rq;
382 if (unlikely(rl->rq_pool))
385 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
386 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
388 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
389 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
391 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
392 mempool_free_slab, request_cachep, q->node);
400 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
402 return blk_alloc_queue_node(gfp_mask, -1);
404 EXPORT_SYMBOL(blk_alloc_queue);
406 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
408 struct request_queue *q;
411 q = kmem_cache_alloc_node(blk_requestq_cachep,
412 gfp_mask | __GFP_ZERO, node_id);
416 q->backing_dev_info.ra_pages =
417 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 q->backing_dev_info.state = 0;
419 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
420 q->backing_dev_info.name = "block";
422 err = bdi_init(&q->backing_dev_info);
424 kmem_cache_free(blk_requestq_cachep, q);
428 if (blk_throtl_init(q)) {
429 kmem_cache_free(blk_requestq_cachep, q);
433 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
434 laptop_mode_timer_fn, (unsigned long) q);
435 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
436 INIT_LIST_HEAD(&q->timeout_list);
437 INIT_LIST_HEAD(&q->flush_queue[0]);
438 INIT_LIST_HEAD(&q->flush_queue[1]);
439 INIT_LIST_HEAD(&q->flush_data_in_flight);
440 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
442 kobject_init(&q->kobj, &blk_queue_ktype);
444 mutex_init(&q->sysfs_lock);
445 spin_lock_init(&q->__queue_lock);
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
451 q->queue_lock = &q->__queue_lock;
455 EXPORT_SYMBOL(blk_alloc_queue_node);
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
478 * The queue spin lock must be held while manipulating the requests on the
479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
482 * Function returns a pointer to the initialized request queue, or %NULL if
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
490 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
492 return blk_init_queue_node(rfn, lock, -1);
494 EXPORT_SYMBOL(blk_init_queue);
496 struct request_queue *
497 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
499 struct request_queue *uninit_q, *q;
501 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
505 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
507 blk_cleanup_queue(uninit_q);
511 EXPORT_SYMBOL(blk_init_queue_node);
513 struct request_queue *
514 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
517 return blk_init_allocated_queue_node(q, rfn, lock, -1);
519 EXPORT_SYMBOL(blk_init_allocated_queue);
521 struct request_queue *
522 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
523 spinlock_t *lock, int node_id)
529 if (blk_init_free_list(q))
533 q->prep_rq_fn = NULL;
534 q->unprep_rq_fn = NULL;
535 q->queue_flags = QUEUE_FLAG_DEFAULT;
537 /* Override internal queue lock with supplied lock pointer */
539 q->queue_lock = lock;
542 * This also sets hw/phys segments, boundary and size
544 blk_queue_make_request(q, __make_request);
546 q->sg_reserved_size = INT_MAX;
551 if (!elevator_init(q, NULL)) {
552 blk_queue_congestion_threshold(q);
558 EXPORT_SYMBOL(blk_init_allocated_queue_node);
560 int blk_get_queue(struct request_queue *q)
562 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
563 kobject_get(&q->kobj);
570 static inline void blk_free_request(struct request_queue *q, struct request *rq)
572 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
574 if (rq->cmd_flags & REQ_ELVPRIV)
575 elv_put_request(q, rq);
576 mempool_free(rq, q->rq.rq_pool);
579 static struct request *
580 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
582 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
589 rq->cmd_flags = flags | REQ_ALLOCED;
592 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
593 mempool_free(rq, q->rq.rq_pool);
596 rq->cmd_flags |= REQ_ELVPRIV;
603 * ioc_batching returns true if the ioc is a valid batching request and
604 * should be given priority access to a request.
606 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
612 * Make sure the process is able to allocate at least 1 request
613 * even if the batch times out, otherwise we could theoretically
616 return ioc->nr_batch_requests == q->nr_batching ||
617 (ioc->nr_batch_requests > 0
618 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
622 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
623 * will cause the process to be a "batcher" on all queues in the system. This
624 * is the behaviour we want though - once it gets a wakeup it should be given
627 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
629 if (!ioc || ioc_batching(q, ioc))
632 ioc->nr_batch_requests = q->nr_batching;
633 ioc->last_waited = jiffies;
636 static void __freed_request(struct request_queue *q, int sync)
638 struct request_list *rl = &q->rq;
640 if (rl->count[sync] < queue_congestion_off_threshold(q))
641 blk_clear_queue_congested(q, sync);
643 if (rl->count[sync] + 1 <= q->nr_requests) {
644 if (waitqueue_active(&rl->wait[sync]))
645 wake_up(&rl->wait[sync]);
647 blk_clear_queue_full(q, sync);
652 * A request has just been released. Account for it, update the full and
653 * congestion status, wake up any waiters. Called under q->queue_lock.
655 static void freed_request(struct request_queue *q, int sync, int priv)
657 struct request_list *rl = &q->rq;
663 __freed_request(q, sync);
665 if (unlikely(rl->starved[sync ^ 1]))
666 __freed_request(q, sync ^ 1);
670 * Determine if elevator data should be initialized when allocating the
671 * request associated with @bio.
673 static bool blk_rq_should_init_elevator(struct bio *bio)
679 * Flush requests do not use the elevator so skip initialization.
680 * This allows a request to share the flush and elevator data.
682 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
689 * Get a free request, queue_lock must be held.
690 * Returns NULL on failure, with queue_lock held.
691 * Returns !NULL on success, with queue_lock *not held*.
693 static struct request *get_request(struct request_queue *q, int rw_flags,
694 struct bio *bio, gfp_t gfp_mask)
696 struct request *rq = NULL;
697 struct request_list *rl = &q->rq;
698 struct io_context *ioc = NULL;
699 const bool is_sync = rw_is_sync(rw_flags) != 0;
700 int may_queue, priv = 0;
702 may_queue = elv_may_queue(q, rw_flags);
703 if (may_queue == ELV_MQUEUE_NO)
706 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
707 if (rl->count[is_sync]+1 >= q->nr_requests) {
708 ioc = current_io_context(GFP_ATOMIC, q->node);
710 * The queue will fill after this allocation, so set
711 * it as full, and mark this process as "batching".
712 * This process will be allowed to complete a batch of
713 * requests, others will be blocked.
715 if (!blk_queue_full(q, is_sync)) {
716 ioc_set_batching(q, ioc);
717 blk_set_queue_full(q, is_sync);
719 if (may_queue != ELV_MQUEUE_MUST
720 && !ioc_batching(q, ioc)) {
722 * The queue is full and the allocating
723 * process is not a "batcher", and not
724 * exempted by the IO scheduler
730 blk_set_queue_congested(q, is_sync);
734 * Only allow batching queuers to allocate up to 50% over the defined
735 * limit of requests, otherwise we could have thousands of requests
736 * allocated with any setting of ->nr_requests
738 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
741 rl->count[is_sync]++;
742 rl->starved[is_sync] = 0;
744 if (blk_rq_should_init_elevator(bio)) {
745 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
750 if (blk_queue_io_stat(q))
751 rw_flags |= REQ_IO_STAT;
752 spin_unlock_irq(q->queue_lock);
754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
763 spin_lock_irq(q->queue_lock);
764 freed_request(q, is_sync, priv);
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
774 if (unlikely(rl->count[is_sync] == 0))
775 rl->starved[is_sync] = 1;
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
789 trace_block_getrq(q, bio, rw_flags & 1);
795 * No available requests for this queue, wait for some requests to become
798 * Called with q->queue_lock held, and returns with it unlocked.
800 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
803 const bool is_sync = rw_is_sync(rw_flags) != 0;
806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
809 struct io_context *ioc;
810 struct request_list *rl = &q->rq;
812 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
813 TASK_UNINTERRUPTIBLE);
815 trace_block_sleeprq(q, bio, rw_flags & 1);
817 spin_unlock_irq(q->queue_lock);
821 * After sleeping, we become a "batching" process and
822 * will be able to allocate at least one request, and
823 * up to a big batch of them for a small period time.
824 * See ioc_batching, ioc_set_batching
826 ioc = current_io_context(GFP_NOIO, q->node);
827 ioc_set_batching(q, ioc);
829 spin_lock_irq(q->queue_lock);
830 finish_wait(&rl->wait[is_sync], &wait);
832 rq = get_request(q, rw_flags, bio, GFP_NOIO);
838 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
842 BUG_ON(rw != READ && rw != WRITE);
844 spin_lock_irq(q->queue_lock);
845 if (gfp_mask & __GFP_WAIT) {
846 rq = get_request_wait(q, rw, NULL);
848 rq = get_request(q, rw, NULL, gfp_mask);
850 spin_unlock_irq(q->queue_lock);
852 /* q->queue_lock is unlocked at this point */
856 EXPORT_SYMBOL(blk_get_request);
859 * blk_make_request - given a bio, allocate a corresponding struct request.
860 * @q: target request queue
861 * @bio: The bio describing the memory mappings that will be submitted for IO.
862 * It may be a chained-bio properly constructed by block/bio layer.
863 * @gfp_mask: gfp flags to be used for memory allocation
865 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
866 * type commands. Where the struct request needs to be farther initialized by
867 * the caller. It is passed a &struct bio, which describes the memory info of
870 * The caller of blk_make_request must make sure that bi_io_vec
871 * are set to describe the memory buffers. That bio_data_dir() will return
872 * the needed direction of the request. (And all bio's in the passed bio-chain
873 * are properly set accordingly)
875 * If called under none-sleepable conditions, mapped bio buffers must not
876 * need bouncing, by calling the appropriate masked or flagged allocator,
877 * suitable for the target device. Otherwise the call to blk_queue_bounce will
880 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
881 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
882 * anything but the first bio in the chain. Otherwise you risk waiting for IO
883 * completion of a bio that hasn't been submitted yet, thus resulting in a
884 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
885 * of bio_alloc(), as that avoids the mempool deadlock.
886 * If possible a big IO should be split into smaller parts when allocation
887 * fails. Partial allocation should not be an error, or you risk a live-lock.
889 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
892 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
895 return ERR_PTR(-ENOMEM);
898 struct bio *bounce_bio = bio;
901 blk_queue_bounce(q, &bounce_bio);
902 ret = blk_rq_append_bio(q, rq, bounce_bio);
911 EXPORT_SYMBOL(blk_make_request);
914 * blk_requeue_request - put a request back on queue
915 * @q: request queue where request should be inserted
916 * @rq: request to be inserted
919 * Drivers often keep queueing requests until the hardware cannot accept
920 * more, when that condition happens we need to put the request back
921 * on the queue. Must be called with queue lock held.
923 void blk_requeue_request(struct request_queue *q, struct request *rq)
925 blk_delete_timer(rq);
926 blk_clear_rq_complete(rq);
927 trace_block_rq_requeue(q, rq);
929 if (blk_rq_tagged(rq))
930 blk_queue_end_tag(q, rq);
932 BUG_ON(blk_queued_rq(rq));
934 elv_requeue_request(q, rq);
936 EXPORT_SYMBOL(blk_requeue_request);
938 static void add_acct_request(struct request_queue *q, struct request *rq,
941 drive_stat_acct(rq, 1);
942 __elv_add_request(q, rq, where);
946 * blk_insert_request - insert a special request into a request queue
947 * @q: request queue where request should be inserted
948 * @rq: request to be inserted
949 * @at_head: insert request at head or tail of queue
950 * @data: private data
953 * Many block devices need to execute commands asynchronously, so they don't
954 * block the whole kernel from preemption during request execution. This is
955 * accomplished normally by inserting aritficial requests tagged as
956 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
957 * be scheduled for actual execution by the request queue.
959 * We have the option of inserting the head or the tail of the queue.
960 * Typically we use the tail for new ioctls and so forth. We use the head
961 * of the queue for things like a QUEUE_FULL message from a device, or a
962 * host that is unable to accept a particular command.
964 void blk_insert_request(struct request_queue *q, struct request *rq,
965 int at_head, void *data)
967 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
971 * tell I/O scheduler that this isn't a regular read/write (ie it
972 * must not attempt merges on this) and that it acts as a soft
975 rq->cmd_type = REQ_TYPE_SPECIAL;
979 spin_lock_irqsave(q->queue_lock, flags);
982 * If command is tagged, release the tag
984 if (blk_rq_tagged(rq))
985 blk_queue_end_tag(q, rq);
987 add_acct_request(q, rq, where);
989 spin_unlock_irqrestore(q->queue_lock, flags);
991 EXPORT_SYMBOL(blk_insert_request);
993 static void part_round_stats_single(int cpu, struct hd_struct *part,
996 if (now == part->stamp)
999 if (part_in_flight(part)) {
1000 __part_stat_add(cpu, part, time_in_queue,
1001 part_in_flight(part) * (now - part->stamp));
1002 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1008 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1009 * @cpu: cpu number for stats access
1010 * @part: target partition
1012 * The average IO queue length and utilisation statistics are maintained
1013 * by observing the current state of the queue length and the amount of
1014 * time it has been in this state for.
1016 * Normally, that accounting is done on IO completion, but that can result
1017 * in more than a second's worth of IO being accounted for within any one
1018 * second, leading to >100% utilisation. To deal with that, we call this
1019 * function to do a round-off before returning the results when reading
1020 * /proc/diskstats. This accounts immediately for all queue usage up to
1021 * the current jiffies and restarts the counters again.
1023 void part_round_stats(int cpu, struct hd_struct *part)
1025 unsigned long now = jiffies;
1028 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1029 part_round_stats_single(cpu, part, now);
1031 EXPORT_SYMBOL_GPL(part_round_stats);
1034 * queue lock must be held
1036 void __blk_put_request(struct request_queue *q, struct request *req)
1040 if (unlikely(--req->ref_count))
1043 elv_completed_request(q, req);
1045 /* this is a bio leak */
1046 WARN_ON(req->bio != NULL);
1049 * Request may not have originated from ll_rw_blk. if not,
1050 * it didn't come out of our reserved rq pools
1052 if (req->cmd_flags & REQ_ALLOCED) {
1053 int is_sync = rq_is_sync(req) != 0;
1054 int priv = req->cmd_flags & REQ_ELVPRIV;
1056 BUG_ON(!list_empty(&req->queuelist));
1057 BUG_ON(!hlist_unhashed(&req->hash));
1059 blk_free_request(q, req);
1060 freed_request(q, is_sync, priv);
1063 EXPORT_SYMBOL_GPL(__blk_put_request);
1065 void blk_put_request(struct request *req)
1067 unsigned long flags;
1068 struct request_queue *q = req->q;
1070 spin_lock_irqsave(q->queue_lock, flags);
1071 __blk_put_request(q, req);
1072 spin_unlock_irqrestore(q->queue_lock, flags);
1074 EXPORT_SYMBOL(blk_put_request);
1077 * blk_add_request_payload - add a payload to a request
1078 * @rq: request to update
1079 * @page: page backing the payload
1080 * @len: length of the payload.
1082 * This allows to later add a payload to an already submitted request by
1083 * a block driver. The driver needs to take care of freeing the payload
1086 * Note that this is a quite horrible hack and nothing but handling of
1087 * discard requests should ever use it.
1089 void blk_add_request_payload(struct request *rq, struct page *page,
1092 struct bio *bio = rq->bio;
1094 bio->bi_io_vec->bv_page = page;
1095 bio->bi_io_vec->bv_offset = 0;
1096 bio->bi_io_vec->bv_len = len;
1100 bio->bi_phys_segments = 1;
1102 rq->__data_len = rq->resid_len = len;
1103 rq->nr_phys_segments = 1;
1104 rq->buffer = bio_data(bio);
1106 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1108 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1111 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1114 * Debug stuff, kill later
1116 if (!rq_mergeable(req)) {
1117 blk_dump_rq_flags(req, "back");
1121 if (!ll_back_merge_fn(q, req, bio))
1124 trace_block_bio_backmerge(q, bio);
1126 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1127 blk_rq_set_mixed_merge(req);
1129 req->biotail->bi_next = bio;
1131 req->__data_len += bio->bi_size;
1132 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1134 drive_stat_acct(req, 0);
1138 static bool bio_attempt_front_merge(struct request_queue *q,
1139 struct request *req, struct bio *bio)
1141 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1145 * Debug stuff, kill later
1147 if (!rq_mergeable(req)) {
1148 blk_dump_rq_flags(req, "front");
1152 if (!ll_front_merge_fn(q, req, bio))
1155 trace_block_bio_frontmerge(q, bio);
1157 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1158 blk_rq_set_mixed_merge(req);
1160 sector = bio->bi_sector;
1162 bio->bi_next = req->bio;
1166 * may not be valid. if the low level driver said
1167 * it didn't need a bounce buffer then it better
1168 * not touch req->buffer either...
1170 req->buffer = bio_data(bio);
1171 req->__sector = bio->bi_sector;
1172 req->__data_len += bio->bi_size;
1173 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1175 drive_stat_acct(req, 0);
1180 * Attempts to merge with the plugged list in the current process. Returns
1181 * true if merge was successful, otherwise false.
1183 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1186 struct blk_plug *plug;
1194 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1200 el_ret = elv_try_merge(rq, bio);
1201 if (el_ret == ELEVATOR_BACK_MERGE) {
1202 ret = bio_attempt_back_merge(q, rq, bio);
1205 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1206 ret = bio_attempt_front_merge(q, rq, bio);
1215 void init_request_from_bio(struct request *req, struct bio *bio)
1217 req->cpu = bio->bi_comp_cpu;
1218 req->cmd_type = REQ_TYPE_FS;
1220 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1221 if (bio->bi_rw & REQ_RAHEAD)
1222 req->cmd_flags |= REQ_FAILFAST_MASK;
1225 req->__sector = bio->bi_sector;
1226 req->ioprio = bio_prio(bio);
1227 blk_rq_bio_prep(req->q, req, bio);
1230 static int __make_request(struct request_queue *q, struct bio *bio)
1232 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1233 struct blk_plug *plug;
1234 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1235 struct request *req;
1238 * low level driver can indicate that it wants pages above a
1239 * certain limit bounced to low memory (ie for highmem, or even
1240 * ISA dma in theory)
1242 blk_queue_bounce(q, &bio);
1244 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1245 spin_lock_irq(q->queue_lock);
1246 where = ELEVATOR_INSERT_FLUSH;
1251 * Check if we can merge with the plugged list before grabbing
1254 if (attempt_plug_merge(current, q, bio))
1257 spin_lock_irq(q->queue_lock);
1259 el_ret = elv_merge(q, &req, bio);
1260 if (el_ret == ELEVATOR_BACK_MERGE) {
1261 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1262 if (bio_attempt_back_merge(q, req, bio)) {
1263 if (!attempt_back_merge(q, req))
1264 elv_merged_request(q, req, el_ret);
1267 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1268 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1269 if (bio_attempt_front_merge(q, req, bio)) {
1270 if (!attempt_front_merge(q, req))
1271 elv_merged_request(q, req, el_ret);
1278 * This sync check and mask will be re-done in init_request_from_bio(),
1279 * but we need to set it earlier to expose the sync flag to the
1280 * rq allocator and io schedulers.
1282 rw_flags = bio_data_dir(bio);
1284 rw_flags |= REQ_SYNC;
1287 * Grab a free request. This is might sleep but can not fail.
1288 * Returns with the queue unlocked.
1290 req = get_request_wait(q, rw_flags, bio);
1293 * After dropping the lock and possibly sleeping here, our request
1294 * may now be mergeable after it had proven unmergeable (above).
1295 * We don't worry about that case for efficiency. It won't happen
1296 * often, and the elevators are able to handle it.
1298 init_request_from_bio(req, bio);
1300 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1301 bio_flagged(bio, BIO_CPU_AFFINE)) {
1302 req->cpu = blk_cpu_to_group(get_cpu());
1306 plug = current->plug;
1309 * If this is the first request added after a plug, fire
1310 * of a plug trace. If others have been added before, check
1311 * if we have multiple devices in this plug. If so, make a
1312 * note to sort the list before dispatch.
1314 if (list_empty(&plug->list))
1315 trace_block_plug(q);
1316 else if (!plug->should_sort) {
1317 struct request *__rq;
1319 __rq = list_entry_rq(plug->list.prev);
1321 plug->should_sort = 1;
1324 * Debug flag, kill later
1326 req->cmd_flags |= REQ_ON_PLUG;
1327 list_add_tail(&req->queuelist, &plug->list);
1328 drive_stat_acct(req, 1);
1330 spin_lock_irq(q->queue_lock);
1331 add_acct_request(q, req, where);
1334 spin_unlock_irq(q->queue_lock);
1341 * If bio->bi_dev is a partition, remap the location
1343 static inline void blk_partition_remap(struct bio *bio)
1345 struct block_device *bdev = bio->bi_bdev;
1347 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1348 struct hd_struct *p = bdev->bd_part;
1350 bio->bi_sector += p->start_sect;
1351 bio->bi_bdev = bdev->bd_contains;
1353 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1355 bio->bi_sector - p->start_sect);
1359 static void handle_bad_sector(struct bio *bio)
1361 char b[BDEVNAME_SIZE];
1363 printk(KERN_INFO "attempt to access beyond end of device\n");
1364 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1365 bdevname(bio->bi_bdev, b),
1367 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1368 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1370 set_bit(BIO_EOF, &bio->bi_flags);
1373 #ifdef CONFIG_FAIL_MAKE_REQUEST
1375 static DECLARE_FAULT_ATTR(fail_make_request);
1377 static int __init setup_fail_make_request(char *str)
1379 return setup_fault_attr(&fail_make_request, str);
1381 __setup("fail_make_request=", setup_fail_make_request);
1383 static int should_fail_request(struct bio *bio)
1385 struct hd_struct *part = bio->bi_bdev->bd_part;
1387 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1388 return should_fail(&fail_make_request, bio->bi_size);
1393 static int __init fail_make_request_debugfs(void)
1395 return init_fault_attr_dentries(&fail_make_request,
1396 "fail_make_request");
1399 late_initcall(fail_make_request_debugfs);
1401 #else /* CONFIG_FAIL_MAKE_REQUEST */
1403 static inline int should_fail_request(struct bio *bio)
1408 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1411 * Check whether this bio extends beyond the end of the device.
1413 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1420 /* Test device or partition size, when known. */
1421 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1423 sector_t sector = bio->bi_sector;
1425 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1427 * This may well happen - the kernel calls bread()
1428 * without checking the size of the device, e.g., when
1429 * mounting a device.
1431 handle_bad_sector(bio);
1440 * generic_make_request - hand a buffer to its device driver for I/O
1441 * @bio: The bio describing the location in memory and on the device.
1443 * generic_make_request() is used to make I/O requests of block
1444 * devices. It is passed a &struct bio, which describes the I/O that needs
1447 * generic_make_request() does not return any status. The
1448 * success/failure status of the request, along with notification of
1449 * completion, is delivered asynchronously through the bio->bi_end_io
1450 * function described (one day) else where.
1452 * The caller of generic_make_request must make sure that bi_io_vec
1453 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1454 * set to describe the device address, and the
1455 * bi_end_io and optionally bi_private are set to describe how
1456 * completion notification should be signaled.
1458 * generic_make_request and the drivers it calls may use bi_next if this
1459 * bio happens to be merged with someone else, and may change bi_dev and
1460 * bi_sector for remaps as it sees fit. So the values of these fields
1461 * should NOT be depended on after the call to generic_make_request.
1463 static inline void __generic_make_request(struct bio *bio)
1465 struct request_queue *q;
1466 sector_t old_sector;
1467 int ret, nr_sectors = bio_sectors(bio);
1473 if (bio_check_eod(bio, nr_sectors))
1477 * Resolve the mapping until finished. (drivers are
1478 * still free to implement/resolve their own stacking
1479 * by explicitly returning 0)
1481 * NOTE: we don't repeat the blk_size check for each new device.
1482 * Stacking drivers are expected to know what they are doing.
1487 char b[BDEVNAME_SIZE];
1489 q = bdev_get_queue(bio->bi_bdev);
1492 "generic_make_request: Trying to access "
1493 "nonexistent block-device %s (%Lu)\n",
1494 bdevname(bio->bi_bdev, b),
1495 (long long) bio->bi_sector);
1499 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1500 nr_sectors > queue_max_hw_sectors(q))) {
1501 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1502 bdevname(bio->bi_bdev, b),
1504 queue_max_hw_sectors(q));
1508 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1511 if (should_fail_request(bio))
1515 * If this device has partitions, remap block n
1516 * of partition p to block n+start(p) of the disk.
1518 blk_partition_remap(bio);
1520 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1523 if (old_sector != -1)
1524 trace_block_bio_remap(q, bio, old_dev, old_sector);
1526 old_sector = bio->bi_sector;
1527 old_dev = bio->bi_bdev->bd_dev;
1529 if (bio_check_eod(bio, nr_sectors))
1533 * Filter flush bio's early so that make_request based
1534 * drivers without flush support don't have to worry
1537 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1538 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1545 if ((bio->bi_rw & REQ_DISCARD) &&
1546 (!blk_queue_discard(q) ||
1547 ((bio->bi_rw & REQ_SECURE) &&
1548 !blk_queue_secdiscard(q)))) {
1553 if (blk_throtl_bio(q, &bio))
1557 * If bio = NULL, bio has been throttled and will be submitted
1563 trace_block_bio_queue(q, bio);
1565 ret = q->make_request_fn(q, bio);
1571 bio_endio(bio, err);
1575 * We only want one ->make_request_fn to be active at a time,
1576 * else stack usage with stacked devices could be a problem.
1577 * So use current->bio_list to keep a list of requests
1578 * submited by a make_request_fn function.
1579 * current->bio_list is also used as a flag to say if
1580 * generic_make_request is currently active in this task or not.
1581 * If it is NULL, then no make_request is active. If it is non-NULL,
1582 * then a make_request is active, and new requests should be added
1585 void generic_make_request(struct bio *bio)
1587 struct bio_list bio_list_on_stack;
1589 if (current->bio_list) {
1590 /* make_request is active */
1591 bio_list_add(current->bio_list, bio);
1594 /* following loop may be a bit non-obvious, and so deserves some
1596 * Before entering the loop, bio->bi_next is NULL (as all callers
1597 * ensure that) so we have a list with a single bio.
1598 * We pretend that we have just taken it off a longer list, so
1599 * we assign bio_list to a pointer to the bio_list_on_stack,
1600 * thus initialising the bio_list of new bios to be
1601 * added. __generic_make_request may indeed add some more bios
1602 * through a recursive call to generic_make_request. If it
1603 * did, we find a non-NULL value in bio_list and re-enter the loop
1604 * from the top. In this case we really did just take the bio
1605 * of the top of the list (no pretending) and so remove it from
1606 * bio_list, and call into __generic_make_request again.
1608 * The loop was structured like this to make only one call to
1609 * __generic_make_request (which is important as it is large and
1610 * inlined) and to keep the structure simple.
1612 BUG_ON(bio->bi_next);
1613 bio_list_init(&bio_list_on_stack);
1614 current->bio_list = &bio_list_on_stack;
1616 __generic_make_request(bio);
1617 bio = bio_list_pop(current->bio_list);
1619 current->bio_list = NULL; /* deactivate */
1621 EXPORT_SYMBOL(generic_make_request);
1624 * submit_bio - submit a bio to the block device layer for I/O
1625 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1626 * @bio: The &struct bio which describes the I/O
1628 * submit_bio() is very similar in purpose to generic_make_request(), and
1629 * uses that function to do most of the work. Both are fairly rough
1630 * interfaces; @bio must be presetup and ready for I/O.
1633 void submit_bio(int rw, struct bio *bio)
1635 int count = bio_sectors(bio);
1640 * If it's a regular read/write or a barrier with data attached,
1641 * go through the normal accounting stuff before submission.
1643 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1645 count_vm_events(PGPGOUT, count);
1647 task_io_account_read(bio->bi_size);
1648 count_vm_events(PGPGIN, count);
1651 if (unlikely(block_dump)) {
1652 char b[BDEVNAME_SIZE];
1653 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1654 current->comm, task_pid_nr(current),
1655 (rw & WRITE) ? "WRITE" : "READ",
1656 (unsigned long long)bio->bi_sector,
1657 bdevname(bio->bi_bdev, b),
1662 generic_make_request(bio);
1664 EXPORT_SYMBOL(submit_bio);
1667 * blk_rq_check_limits - Helper function to check a request for the queue limit
1669 * @rq: the request being checked
1672 * @rq may have been made based on weaker limitations of upper-level queues
1673 * in request stacking drivers, and it may violate the limitation of @q.
1674 * Since the block layer and the underlying device driver trust @rq
1675 * after it is inserted to @q, it should be checked against @q before
1676 * the insertion using this generic function.
1678 * This function should also be useful for request stacking drivers
1679 * in some cases below, so export this function.
1680 * Request stacking drivers like request-based dm may change the queue
1681 * limits while requests are in the queue (e.g. dm's table swapping).
1682 * Such request stacking drivers should check those requests agaist
1683 * the new queue limits again when they dispatch those requests,
1684 * although such checkings are also done against the old queue limits
1685 * when submitting requests.
1687 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1689 if (rq->cmd_flags & REQ_DISCARD)
1692 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1693 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1694 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1699 * queue's settings related to segment counting like q->bounce_pfn
1700 * may differ from that of other stacking queues.
1701 * Recalculate it to check the request correctly on this queue's
1704 blk_recalc_rq_segments(rq);
1705 if (rq->nr_phys_segments > queue_max_segments(q)) {
1706 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1712 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1715 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1716 * @q: the queue to submit the request
1717 * @rq: the request being queued
1719 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1721 unsigned long flags;
1723 if (blk_rq_check_limits(q, rq))
1726 #ifdef CONFIG_FAIL_MAKE_REQUEST
1727 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1728 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1732 spin_lock_irqsave(q->queue_lock, flags);
1735 * Submitting request must be dequeued before calling this function
1736 * because it will be linked to another request_queue
1738 BUG_ON(blk_queued_rq(rq));
1740 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1741 spin_unlock_irqrestore(q->queue_lock, flags);
1745 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1748 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1749 * @rq: request to examine
1752 * A request could be merge of IOs which require different failure
1753 * handling. This function determines the number of bytes which
1754 * can be failed from the beginning of the request without
1755 * crossing into area which need to be retried further.
1758 * The number of bytes to fail.
1761 * queue_lock must be held.
1763 unsigned int blk_rq_err_bytes(const struct request *rq)
1765 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1766 unsigned int bytes = 0;
1769 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1770 return blk_rq_bytes(rq);
1773 * Currently the only 'mixing' which can happen is between
1774 * different fastfail types. We can safely fail portions
1775 * which have all the failfast bits that the first one has -
1776 * the ones which are at least as eager to fail as the first
1779 for (bio = rq->bio; bio; bio = bio->bi_next) {
1780 if ((bio->bi_rw & ff) != ff)
1782 bytes += bio->bi_size;
1785 /* this could lead to infinite loop */
1786 BUG_ON(blk_rq_bytes(rq) && !bytes);
1789 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1791 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1793 if (blk_do_io_stat(req)) {
1794 const int rw = rq_data_dir(req);
1795 struct hd_struct *part;
1798 cpu = part_stat_lock();
1800 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1805 static void blk_account_io_done(struct request *req)
1808 * Account IO completion. flush_rq isn't accounted as a
1809 * normal IO on queueing nor completion. Accounting the
1810 * containing request is enough.
1812 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1813 unsigned long duration = jiffies - req->start_time;
1814 const int rw = rq_data_dir(req);
1815 struct hd_struct *part;
1818 cpu = part_stat_lock();
1821 part_stat_inc(cpu, part, ios[rw]);
1822 part_stat_add(cpu, part, ticks[rw], duration);
1823 part_round_stats(cpu, part);
1824 part_dec_in_flight(part, rw);
1826 hd_struct_put(part);
1832 * blk_peek_request - peek at the top of a request queue
1833 * @q: request queue to peek at
1836 * Return the request at the top of @q. The returned request
1837 * should be started using blk_start_request() before LLD starts
1841 * Pointer to the request at the top of @q if available. Null
1845 * queue_lock must be held.
1847 struct request *blk_peek_request(struct request_queue *q)
1852 while ((rq = __elv_next_request(q)) != NULL) {
1853 if (!(rq->cmd_flags & REQ_STARTED)) {
1855 * This is the first time the device driver
1856 * sees this request (possibly after
1857 * requeueing). Notify IO scheduler.
1859 if (rq->cmd_flags & REQ_SORTED)
1860 elv_activate_rq(q, rq);
1863 * just mark as started even if we don't start
1864 * it, a request that has been delayed should
1865 * not be passed by new incoming requests
1867 rq->cmd_flags |= REQ_STARTED;
1868 trace_block_rq_issue(q, rq);
1871 if (!q->boundary_rq || q->boundary_rq == rq) {
1872 q->end_sector = rq_end_sector(rq);
1873 q->boundary_rq = NULL;
1876 if (rq->cmd_flags & REQ_DONTPREP)
1879 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1881 * make sure space for the drain appears we
1882 * know we can do this because max_hw_segments
1883 * has been adjusted to be one fewer than the
1886 rq->nr_phys_segments++;
1892 ret = q->prep_rq_fn(q, rq);
1893 if (ret == BLKPREP_OK) {
1895 } else if (ret == BLKPREP_DEFER) {
1897 * the request may have been (partially) prepped.
1898 * we need to keep this request in the front to
1899 * avoid resource deadlock. REQ_STARTED will
1900 * prevent other fs requests from passing this one.
1902 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1903 !(rq->cmd_flags & REQ_DONTPREP)) {
1905 * remove the space for the drain we added
1906 * so that we don't add it again
1908 --rq->nr_phys_segments;
1913 } else if (ret == BLKPREP_KILL) {
1914 rq->cmd_flags |= REQ_QUIET;
1916 * Mark this request as started so we don't trigger
1917 * any debug logic in the end I/O path.
1919 blk_start_request(rq);
1920 __blk_end_request_all(rq, -EIO);
1922 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1929 EXPORT_SYMBOL(blk_peek_request);
1931 void blk_dequeue_request(struct request *rq)
1933 struct request_queue *q = rq->q;
1935 BUG_ON(list_empty(&rq->queuelist));
1936 BUG_ON(ELV_ON_HASH(rq));
1938 list_del_init(&rq->queuelist);
1941 * the time frame between a request being removed from the lists
1942 * and to it is freed is accounted as io that is in progress at
1945 if (blk_account_rq(rq)) {
1946 q->in_flight[rq_is_sync(rq)]++;
1947 set_io_start_time_ns(rq);
1952 * blk_start_request - start request processing on the driver
1953 * @req: request to dequeue
1956 * Dequeue @req and start timeout timer on it. This hands off the
1957 * request to the driver.
1959 * Block internal functions which don't want to start timer should
1960 * call blk_dequeue_request().
1963 * queue_lock must be held.
1965 void blk_start_request(struct request *req)
1967 blk_dequeue_request(req);
1970 * We are now handing the request to the hardware, initialize
1971 * resid_len to full count and add the timeout handler.
1973 req->resid_len = blk_rq_bytes(req);
1974 if (unlikely(blk_bidi_rq(req)))
1975 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1979 EXPORT_SYMBOL(blk_start_request);
1982 * blk_fetch_request - fetch a request from a request queue
1983 * @q: request queue to fetch a request from
1986 * Return the request at the top of @q. The request is started on
1987 * return and LLD can start processing it immediately.
1990 * Pointer to the request at the top of @q if available. Null
1994 * queue_lock must be held.
1996 struct request *blk_fetch_request(struct request_queue *q)
2000 rq = blk_peek_request(q);
2002 blk_start_request(rq);
2005 EXPORT_SYMBOL(blk_fetch_request);
2008 * blk_update_request - Special helper function for request stacking drivers
2009 * @req: the request being processed
2010 * @error: %0 for success, < %0 for error
2011 * @nr_bytes: number of bytes to complete @req
2014 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2015 * the request structure even if @req doesn't have leftover.
2016 * If @req has leftover, sets it up for the next range of segments.
2018 * This special helper function is only for request stacking drivers
2019 * (e.g. request-based dm) so that they can handle partial completion.
2020 * Actual device drivers should use blk_end_request instead.
2022 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2023 * %false return from this function.
2026 * %false - this request doesn't have any more data
2027 * %true - this request has more data
2029 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2031 int total_bytes, bio_nbytes, next_idx = 0;
2037 trace_block_rq_complete(req->q, req);
2040 * For fs requests, rq is just carrier of independent bio's
2041 * and each partial completion should be handled separately.
2042 * Reset per-request error on each partial completion.
2044 * TODO: tj: This is too subtle. It would be better to let
2045 * low level drivers do what they see fit.
2047 if (req->cmd_type == REQ_TYPE_FS)
2050 if (error && req->cmd_type == REQ_TYPE_FS &&
2051 !(req->cmd_flags & REQ_QUIET)) {
2056 error_type = "recoverable transport";
2059 error_type = "critical target";
2062 error_type = "critical nexus";
2069 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2070 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2071 (unsigned long long)blk_rq_pos(req));
2074 blk_account_io_completion(req, nr_bytes);
2076 total_bytes = bio_nbytes = 0;
2077 while ((bio = req->bio) != NULL) {
2080 if (nr_bytes >= bio->bi_size) {
2081 req->bio = bio->bi_next;
2082 nbytes = bio->bi_size;
2083 req_bio_endio(req, bio, nbytes, error);
2087 int idx = bio->bi_idx + next_idx;
2089 if (unlikely(idx >= bio->bi_vcnt)) {
2090 blk_dump_rq_flags(req, "__end_that");
2091 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2092 __func__, idx, bio->bi_vcnt);
2096 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2097 BIO_BUG_ON(nbytes > bio->bi_size);
2100 * not a complete bvec done
2102 if (unlikely(nbytes > nr_bytes)) {
2103 bio_nbytes += nr_bytes;
2104 total_bytes += nr_bytes;
2109 * advance to the next vector
2112 bio_nbytes += nbytes;
2115 total_bytes += nbytes;
2121 * end more in this run, or just return 'not-done'
2123 if (unlikely(nr_bytes <= 0))
2133 * Reset counters so that the request stacking driver
2134 * can find how many bytes remain in the request
2137 req->__data_len = 0;
2142 * if the request wasn't completed, update state
2145 req_bio_endio(req, bio, bio_nbytes, error);
2146 bio->bi_idx += next_idx;
2147 bio_iovec(bio)->bv_offset += nr_bytes;
2148 bio_iovec(bio)->bv_len -= nr_bytes;
2151 req->__data_len -= total_bytes;
2152 req->buffer = bio_data(req->bio);
2154 /* update sector only for requests with clear definition of sector */
2155 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2156 req->__sector += total_bytes >> 9;
2158 /* mixed attributes always follow the first bio */
2159 if (req->cmd_flags & REQ_MIXED_MERGE) {
2160 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2161 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2165 * If total number of sectors is less than the first segment
2166 * size, something has gone terribly wrong.
2168 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2169 blk_dump_rq_flags(req, "request botched");
2170 req->__data_len = blk_rq_cur_bytes(req);
2173 /* recalculate the number of segments */
2174 blk_recalc_rq_segments(req);
2178 EXPORT_SYMBOL_GPL(blk_update_request);
2180 static bool blk_update_bidi_request(struct request *rq, int error,
2181 unsigned int nr_bytes,
2182 unsigned int bidi_bytes)
2184 if (blk_update_request(rq, error, nr_bytes))
2187 /* Bidi request must be completed as a whole */
2188 if (unlikely(blk_bidi_rq(rq)) &&
2189 blk_update_request(rq->next_rq, error, bidi_bytes))
2192 if (blk_queue_add_random(rq->q))
2193 add_disk_randomness(rq->rq_disk);
2199 * blk_unprep_request - unprepare a request
2202 * This function makes a request ready for complete resubmission (or
2203 * completion). It happens only after all error handling is complete,
2204 * so represents the appropriate moment to deallocate any resources
2205 * that were allocated to the request in the prep_rq_fn. The queue
2206 * lock is held when calling this.
2208 void blk_unprep_request(struct request *req)
2210 struct request_queue *q = req->q;
2212 req->cmd_flags &= ~REQ_DONTPREP;
2213 if (q->unprep_rq_fn)
2214 q->unprep_rq_fn(q, req);
2216 EXPORT_SYMBOL_GPL(blk_unprep_request);
2219 * queue lock must be held
2221 static void blk_finish_request(struct request *req, int error)
2223 if (blk_rq_tagged(req))
2224 blk_queue_end_tag(req->q, req);
2226 BUG_ON(blk_queued_rq(req));
2228 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2229 laptop_io_completion(&req->q->backing_dev_info);
2231 blk_delete_timer(req);
2233 if (req->cmd_flags & REQ_DONTPREP)
2234 blk_unprep_request(req);
2237 blk_account_io_done(req);
2240 req->end_io(req, error);
2242 if (blk_bidi_rq(req))
2243 __blk_put_request(req->next_rq->q, req->next_rq);
2245 __blk_put_request(req->q, req);
2250 * blk_end_bidi_request - Complete a bidi request
2251 * @rq: the request to complete
2252 * @error: %0 for success, < %0 for error
2253 * @nr_bytes: number of bytes to complete @rq
2254 * @bidi_bytes: number of bytes to complete @rq->next_rq
2257 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2258 * Drivers that supports bidi can safely call this member for any
2259 * type of request, bidi or uni. In the later case @bidi_bytes is
2263 * %false - we are done with this request
2264 * %true - still buffers pending for this request
2266 static bool blk_end_bidi_request(struct request *rq, int error,
2267 unsigned int nr_bytes, unsigned int bidi_bytes)
2269 struct request_queue *q = rq->q;
2270 unsigned long flags;
2272 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2275 spin_lock_irqsave(q->queue_lock, flags);
2276 blk_finish_request(rq, error);
2277 spin_unlock_irqrestore(q->queue_lock, flags);
2283 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2284 * @rq: the request to complete
2285 * @error: %0 for success, < %0 for error
2286 * @nr_bytes: number of bytes to complete @rq
2287 * @bidi_bytes: number of bytes to complete @rq->next_rq
2290 * Identical to blk_end_bidi_request() except that queue lock is
2291 * assumed to be locked on entry and remains so on return.
2294 * %false - we are done with this request
2295 * %true - still buffers pending for this request
2297 static bool __blk_end_bidi_request(struct request *rq, int error,
2298 unsigned int nr_bytes, unsigned int bidi_bytes)
2300 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2303 blk_finish_request(rq, error);
2309 * blk_end_request - Helper function for drivers to complete the request.
2310 * @rq: the request being processed
2311 * @error: %0 for success, < %0 for error
2312 * @nr_bytes: number of bytes to complete
2315 * Ends I/O on a number of bytes attached to @rq.
2316 * If @rq has leftover, sets it up for the next range of segments.
2319 * %false - we are done with this request
2320 * %true - still buffers pending for this request
2322 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2324 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2326 EXPORT_SYMBOL(blk_end_request);
2329 * blk_end_request_all - Helper function for drives to finish the request.
2330 * @rq: the request to finish
2331 * @error: %0 for success, < %0 for error
2334 * Completely finish @rq.
2336 void blk_end_request_all(struct request *rq, int error)
2339 unsigned int bidi_bytes = 0;
2341 if (unlikely(blk_bidi_rq(rq)))
2342 bidi_bytes = blk_rq_bytes(rq->next_rq);
2344 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2347 EXPORT_SYMBOL(blk_end_request_all);
2350 * blk_end_request_cur - Helper function to finish the current request chunk.
2351 * @rq: the request to finish the current chunk for
2352 * @error: %0 for success, < %0 for error
2355 * Complete the current consecutively mapped chunk from @rq.
2358 * %false - we are done with this request
2359 * %true - still buffers pending for this request
2361 bool blk_end_request_cur(struct request *rq, int error)
2363 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2365 EXPORT_SYMBOL(blk_end_request_cur);
2368 * blk_end_request_err - Finish a request till the next failure boundary.
2369 * @rq: the request to finish till the next failure boundary for
2370 * @error: must be negative errno
2373 * Complete @rq till the next failure boundary.
2376 * %false - we are done with this request
2377 * %true - still buffers pending for this request
2379 bool blk_end_request_err(struct request *rq, int error)
2381 WARN_ON(error >= 0);
2382 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2384 EXPORT_SYMBOL_GPL(blk_end_request_err);
2387 * __blk_end_request - Helper function for drivers to complete the request.
2388 * @rq: the request being processed
2389 * @error: %0 for success, < %0 for error
2390 * @nr_bytes: number of bytes to complete
2393 * Must be called with queue lock held unlike blk_end_request().
2396 * %false - we are done with this request
2397 * %true - still buffers pending for this request
2399 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2401 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2403 EXPORT_SYMBOL(__blk_end_request);
2406 * __blk_end_request_all - Helper function for drives to finish the request.
2407 * @rq: the request to finish
2408 * @error: %0 for success, < %0 for error
2411 * Completely finish @rq. Must be called with queue lock held.
2413 void __blk_end_request_all(struct request *rq, int error)
2416 unsigned int bidi_bytes = 0;
2418 if (unlikely(blk_bidi_rq(rq)))
2419 bidi_bytes = blk_rq_bytes(rq->next_rq);
2421 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2424 EXPORT_SYMBOL(__blk_end_request_all);
2427 * __blk_end_request_cur - Helper function to finish the current request chunk.
2428 * @rq: the request to finish the current chunk for
2429 * @error: %0 for success, < %0 for error
2432 * Complete the current consecutively mapped chunk from @rq. Must
2433 * be called with queue lock held.
2436 * %false - we are done with this request
2437 * %true - still buffers pending for this request
2439 bool __blk_end_request_cur(struct request *rq, int error)
2441 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2443 EXPORT_SYMBOL(__blk_end_request_cur);
2446 * __blk_end_request_err - Finish a request till the next failure boundary.
2447 * @rq: the request to finish till the next failure boundary for
2448 * @error: must be negative errno
2451 * Complete @rq till the next failure boundary. Must be called
2452 * with queue lock held.
2455 * %false - we are done with this request
2456 * %true - still buffers pending for this request
2458 bool __blk_end_request_err(struct request *rq, int error)
2460 WARN_ON(error >= 0);
2461 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2463 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2465 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2468 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2469 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2471 if (bio_has_data(bio)) {
2472 rq->nr_phys_segments = bio_phys_segments(q, bio);
2473 rq->buffer = bio_data(bio);
2475 rq->__data_len = bio->bi_size;
2476 rq->bio = rq->biotail = bio;
2479 rq->rq_disk = bio->bi_bdev->bd_disk;
2482 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2484 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2485 * @rq: the request to be flushed
2488 * Flush all pages in @rq.
2490 void rq_flush_dcache_pages(struct request *rq)
2492 struct req_iterator iter;
2493 struct bio_vec *bvec;
2495 rq_for_each_segment(bvec, rq, iter)
2496 flush_dcache_page(bvec->bv_page);
2498 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2502 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2503 * @q : the queue of the device being checked
2506 * Check if underlying low-level drivers of a device are busy.
2507 * If the drivers want to export their busy state, they must set own
2508 * exporting function using blk_queue_lld_busy() first.
2510 * Basically, this function is used only by request stacking drivers
2511 * to stop dispatching requests to underlying devices when underlying
2512 * devices are busy. This behavior helps more I/O merging on the queue
2513 * of the request stacking driver and prevents I/O throughput regression
2514 * on burst I/O load.
2517 * 0 - Not busy (The request stacking driver should dispatch request)
2518 * 1 - Busy (The request stacking driver should stop dispatching request)
2520 int blk_lld_busy(struct request_queue *q)
2523 return q->lld_busy_fn(q);
2527 EXPORT_SYMBOL_GPL(blk_lld_busy);
2530 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2531 * @rq: the clone request to be cleaned up
2534 * Free all bios in @rq for a cloned request.
2536 void blk_rq_unprep_clone(struct request *rq)
2540 while ((bio = rq->bio) != NULL) {
2541 rq->bio = bio->bi_next;
2546 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2549 * Copy attributes of the original request to the clone request.
2550 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2552 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2554 dst->cpu = src->cpu;
2555 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2556 dst->cmd_type = src->cmd_type;
2557 dst->__sector = blk_rq_pos(src);
2558 dst->__data_len = blk_rq_bytes(src);
2559 dst->nr_phys_segments = src->nr_phys_segments;
2560 dst->ioprio = src->ioprio;
2561 dst->extra_len = src->extra_len;
2565 * blk_rq_prep_clone - Helper function to setup clone request
2566 * @rq: the request to be setup
2567 * @rq_src: original request to be cloned
2568 * @bs: bio_set that bios for clone are allocated from
2569 * @gfp_mask: memory allocation mask for bio
2570 * @bio_ctr: setup function to be called for each clone bio.
2571 * Returns %0 for success, non %0 for failure.
2572 * @data: private data to be passed to @bio_ctr
2575 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2576 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2577 * are not copied, and copying such parts is the caller's responsibility.
2578 * Also, pages which the original bios are pointing to are not copied
2579 * and the cloned bios just point same pages.
2580 * So cloned bios must be completed before original bios, which means
2581 * the caller must complete @rq before @rq_src.
2583 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2584 struct bio_set *bs, gfp_t gfp_mask,
2585 int (*bio_ctr)(struct bio *, struct bio *, void *),
2588 struct bio *bio, *bio_src;
2593 blk_rq_init(NULL, rq);
2595 __rq_for_each_bio(bio_src, rq_src) {
2596 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2600 __bio_clone(bio, bio_src);
2602 if (bio_integrity(bio_src) &&
2603 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2606 if (bio_ctr && bio_ctr(bio, bio_src, data))
2610 rq->biotail->bi_next = bio;
2613 rq->bio = rq->biotail = bio;
2616 __blk_rq_prep_clone(rq, rq_src);
2623 blk_rq_unprep_clone(rq);
2627 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2629 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2631 return queue_work(kblockd_workqueue, work);
2633 EXPORT_SYMBOL(kblockd_schedule_work);
2635 int kblockd_schedule_delayed_work(struct request_queue *q,
2636 struct delayed_work *dwork, unsigned long delay)
2638 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2640 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2642 #define PLUG_MAGIC 0x91827364
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);
2752 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2756 * This drops the queue lock
2759 queue_unplugged(q, depth, from_schedule);
2762 spin_lock(q->queue_lock);
2764 rq->cmd_flags &= ~REQ_ON_PLUG;
2767 * rq is already accounted, so use raw insert
2769 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2770 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2772 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2778 * This drops the queue lock
2781 queue_unplugged(q, depth, from_schedule);
2783 local_irq_restore(flags);
2786 void blk_finish_plug(struct blk_plug *plug)
2788 blk_flush_plug_list(plug, false);
2790 if (plug == current->plug)
2791 current->plug = NULL;
2793 EXPORT_SYMBOL(blk_finish_plug);
2795 int __init blk_dev_init(void)
2797 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2798 sizeof(((struct request *)0)->cmd_flags));
2800 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2801 kblockd_workqueue = alloc_workqueue("kblockd",
2802 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2803 if (!kblockd_workqueue)
2804 panic("Failed to create kblockd\n");
2806 request_cachep = kmem_cache_create("blkdev_requests",
2807 sizeof(struct request), 0, SLAB_PANIC, NULL);
2809 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2810 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);