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(!in_interrupt() && !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
352 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
354 * Drain requests from @q. If @drain_all is set, all requests are drained.
355 * If not, only ELVPRIV requests are drained. The caller is responsible
356 * for ensuring that no new requests which need to be drained are queued.
358 void blk_drain_queue(struct request_queue *q, bool drain_all)
363 spin_lock_irq(q->queue_lock);
365 elv_drain_elevator(q);
370 * This function might be called on a queue which failed
371 * driver init after queue creation. Some drivers
372 * (e.g. fd) get unhappy in such cases. Kick queue iff
373 * dispatch queue has something on it.
375 if (!list_empty(&q->queue_head))
379 nr_rqs = q->rq.count[0] + q->rq.count[1];
381 nr_rqs = q->rq.elvpriv;
383 spin_unlock_irq(q->queue_lock);
392 * blk_cleanup_queue - shutdown a request queue
393 * @q: request queue to shutdown
395 * Mark @q DEAD, drain all pending requests, destroy and put it. All
396 * future requests will be failed immediately with -ENODEV.
398 void blk_cleanup_queue(struct request_queue *q)
400 spinlock_t *lock = q->queue_lock;
402 /* mark @q DEAD, no new request or merges will be allowed afterwards */
403 mutex_lock(&q->sysfs_lock);
404 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
407 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
408 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
409 queue_flag_set(QUEUE_FLAG_DEAD, q);
411 if (q->queue_lock != &q->__queue_lock)
412 q->queue_lock = &q->__queue_lock;
414 spin_unlock_irq(lock);
415 mutex_unlock(&q->sysfs_lock);
418 * Drain all requests queued before DEAD marking. The caller might
419 * be trying to tear down @q before its elevator is initialized, in
420 * which case we don't want to call into draining.
423 blk_drain_queue(q, true);
425 /* @q won't process any more request, flush async actions */
426 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
429 /* @q is and will stay empty, shutdown and put */
432 EXPORT_SYMBOL(blk_cleanup_queue);
434 static int blk_init_free_list(struct request_queue *q)
436 struct request_list *rl = &q->rq;
438 if (unlikely(rl->rq_pool))
441 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
442 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
444 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
445 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
447 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
448 mempool_free_slab, request_cachep, q->node);
456 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
458 return blk_alloc_queue_node(gfp_mask, -1);
460 EXPORT_SYMBOL(blk_alloc_queue);
462 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
464 struct request_queue *q;
467 q = kmem_cache_alloc_node(blk_requestq_cachep,
468 gfp_mask | __GFP_ZERO, node_id);
472 q->backing_dev_info.ra_pages =
473 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
474 q->backing_dev_info.state = 0;
475 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
476 q->backing_dev_info.name = "block";
479 err = bdi_init(&q->backing_dev_info);
481 kmem_cache_free(blk_requestq_cachep, q);
485 if (blk_throtl_init(q)) {
486 bdi_destroy(&q->backing_dev_info);
487 kmem_cache_free(blk_requestq_cachep, q);
491 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
492 laptop_mode_timer_fn, (unsigned long) q);
493 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
494 INIT_LIST_HEAD(&q->timeout_list);
495 INIT_LIST_HEAD(&q->flush_queue[0]);
496 INIT_LIST_HEAD(&q->flush_queue[1]);
497 INIT_LIST_HEAD(&q->flush_data_in_flight);
498 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
500 kobject_init(&q->kobj, &blk_queue_ktype);
502 #ifdef CONFIG_BLK_DEV_IO_TRACE
503 mutex_init(&q->blk_trace_mutex);
505 mutex_init(&q->sysfs_lock);
506 spin_lock_init(&q->__queue_lock);
509 * By default initialize queue_lock to internal lock and driver can
510 * override it later if need be.
512 q->queue_lock = &q->__queue_lock;
516 EXPORT_SYMBOL(blk_alloc_queue_node);
519 * blk_init_queue - prepare a request queue for use with a block device
520 * @rfn: The function to be called to process requests that have been
521 * placed on the queue.
522 * @lock: Request queue spin lock
525 * If a block device wishes to use the standard request handling procedures,
526 * which sorts requests and coalesces adjacent requests, then it must
527 * call blk_init_queue(). The function @rfn will be called when there
528 * are requests on the queue that need to be processed. If the device
529 * supports plugging, then @rfn may not be called immediately when requests
530 * are available on the queue, but may be called at some time later instead.
531 * Plugged queues are generally unplugged when a buffer belonging to one
532 * of the requests on the queue is needed, or due to memory pressure.
534 * @rfn is not required, or even expected, to remove all requests off the
535 * queue, but only as many as it can handle at a time. If it does leave
536 * requests on the queue, it is responsible for arranging that the requests
537 * get dealt with eventually.
539 * The queue spin lock must be held while manipulating the requests on the
540 * request queue; this lock will be taken also from interrupt context, so irq
541 * disabling is needed for it.
543 * Function returns a pointer to the initialized request queue, or %NULL if
547 * blk_init_queue() must be paired with a blk_cleanup_queue() call
548 * when the block device is deactivated (such as at module unload).
551 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
553 return blk_init_queue_node(rfn, lock, -1);
555 EXPORT_SYMBOL(blk_init_queue);
557 struct request_queue *
558 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
560 struct request_queue *uninit_q, *q;
562 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
566 q = blk_init_allocated_queue(uninit_q, rfn, lock);
568 blk_cleanup_queue(uninit_q);
572 EXPORT_SYMBOL(blk_init_queue_node);
574 struct request_queue *
575 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
581 if (blk_init_free_list(q))
585 q->prep_rq_fn = NULL;
586 q->unprep_rq_fn = NULL;
587 q->queue_flags = QUEUE_FLAG_DEFAULT;
589 /* Override internal queue lock with supplied lock pointer */
591 q->queue_lock = lock;
594 * This also sets hw/phys segments, boundary and size
596 blk_queue_make_request(q, blk_queue_bio);
598 q->sg_reserved_size = INT_MAX;
603 if (!elevator_init(q, NULL)) {
604 blk_queue_congestion_threshold(q);
610 EXPORT_SYMBOL(blk_init_allocated_queue);
612 int blk_get_queue(struct request_queue *q)
614 if (likely(!blk_queue_dead(q))) {
615 kobject_get(&q->kobj);
621 EXPORT_SYMBOL(blk_get_queue);
623 static inline void blk_free_request(struct request_queue *q, struct request *rq)
625 if (rq->cmd_flags & REQ_ELVPRIV)
626 elv_put_request(q, rq);
627 mempool_free(rq, q->rq.rq_pool);
630 static struct request *
631 blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
633 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
640 rq->cmd_flags = flags | REQ_ALLOCED;
642 if ((flags & REQ_ELVPRIV) &&
643 unlikely(elv_set_request(q, rq, gfp_mask))) {
644 mempool_free(rq, q->rq.rq_pool);
652 * ioc_batching returns true if the ioc is a valid batching request and
653 * should be given priority access to a request.
655 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
661 * Make sure the process is able to allocate at least 1 request
662 * even if the batch times out, otherwise we could theoretically
665 return ioc->nr_batch_requests == q->nr_batching ||
666 (ioc->nr_batch_requests > 0
667 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
671 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
672 * will cause the process to be a "batcher" on all queues in the system. This
673 * is the behaviour we want though - once it gets a wakeup it should be given
676 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
678 if (!ioc || ioc_batching(q, ioc))
681 ioc->nr_batch_requests = q->nr_batching;
682 ioc->last_waited = jiffies;
685 static void __freed_request(struct request_queue *q, int sync)
687 struct request_list *rl = &q->rq;
689 if (rl->count[sync] < queue_congestion_off_threshold(q))
690 blk_clear_queue_congested(q, sync);
692 if (rl->count[sync] + 1 <= q->nr_requests) {
693 if (waitqueue_active(&rl->wait[sync]))
694 wake_up(&rl->wait[sync]);
696 blk_clear_queue_full(q, sync);
701 * A request has just been released. Account for it, update the full and
702 * congestion status, wake up any waiters. Called under q->queue_lock.
704 static void freed_request(struct request_queue *q, unsigned int flags)
706 struct request_list *rl = &q->rq;
707 int sync = rw_is_sync(flags);
710 if (flags & REQ_ELVPRIV)
713 __freed_request(q, sync);
715 if (unlikely(rl->starved[sync ^ 1]))
716 __freed_request(q, sync ^ 1);
720 * Determine if elevator data should be initialized when allocating the
721 * request associated with @bio.
723 static bool blk_rq_should_init_elevator(struct bio *bio)
729 * Flush requests do not use the elevator so skip initialization.
730 * This allows a request to share the flush and elevator data.
732 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
739 * get_request - get a free request
740 * @q: request_queue to allocate request from
741 * @rw_flags: RW and SYNC flags
742 * @bio: bio to allocate request for (can be %NULL)
743 * @gfp_mask: allocation mask
745 * Get a free request from @q. This function may fail under memory
746 * pressure or if @q is dead.
748 * Must be callled with @q->queue_lock held and,
749 * Returns %NULL on failure, with @q->queue_lock held.
750 * Returns !%NULL on success, with @q->queue_lock *not held*.
752 static struct request *get_request(struct request_queue *q, int rw_flags,
753 struct bio *bio, gfp_t gfp_mask)
755 struct request *rq = NULL;
756 struct request_list *rl = &q->rq;
757 struct io_context *ioc = NULL;
758 const bool is_sync = rw_is_sync(rw_flags) != 0;
761 if (unlikely(blk_queue_dead(q)))
764 may_queue = elv_may_queue(q, rw_flags);
765 if (may_queue == ELV_MQUEUE_NO)
768 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
769 if (rl->count[is_sync]+1 >= q->nr_requests) {
770 ioc = current_io_context(GFP_ATOMIC, q->node);
772 * The queue will fill after this allocation, so set
773 * it as full, and mark this process as "batching".
774 * This process will be allowed to complete a batch of
775 * requests, others will be blocked.
777 if (!blk_queue_full(q, is_sync)) {
778 ioc_set_batching(q, ioc);
779 blk_set_queue_full(q, is_sync);
781 if (may_queue != ELV_MQUEUE_MUST
782 && !ioc_batching(q, ioc)) {
784 * The queue is full and the allocating
785 * process is not a "batcher", and not
786 * exempted by the IO scheduler
792 blk_set_queue_congested(q, is_sync);
796 * Only allow batching queuers to allocate up to 50% over the defined
797 * limit of requests, otherwise we could have thousands of requests
798 * allocated with any setting of ->nr_requests
800 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
803 rl->count[is_sync]++;
804 rl->starved[is_sync] = 0;
806 if (blk_rq_should_init_elevator(bio) &&
807 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
808 rw_flags |= REQ_ELVPRIV;
812 if (blk_queue_io_stat(q))
813 rw_flags |= REQ_IO_STAT;
814 spin_unlock_irq(q->queue_lock);
816 rq = blk_alloc_request(q, rw_flags, gfp_mask);
819 * Allocation failed presumably due to memory. Undo anything
820 * we might have messed up.
822 * Allocating task should really be put onto the front of the
823 * wait queue, but this is pretty rare.
825 spin_lock_irq(q->queue_lock);
826 freed_request(q, rw_flags);
829 * in the very unlikely event that allocation failed and no
830 * requests for this direction was pending, mark us starved
831 * so that freeing of a request in the other direction will
832 * notice us. another possible fix would be to split the
833 * rq mempool into READ and WRITE
836 if (unlikely(rl->count[is_sync] == 0))
837 rl->starved[is_sync] = 1;
843 * ioc may be NULL here, and ioc_batching will be false. That's
844 * OK, if the queue is under the request limit then requests need
845 * not count toward the nr_batch_requests limit. There will always
846 * be some limit enforced by BLK_BATCH_TIME.
848 if (ioc_batching(q, ioc))
849 ioc->nr_batch_requests--;
851 trace_block_getrq(q, bio, rw_flags & 1);
857 * get_request_wait - get a free request with retry
858 * @q: request_queue to allocate request from
859 * @rw_flags: RW and SYNC flags
860 * @bio: bio to allocate request for (can be %NULL)
862 * Get a free request from @q. This function keeps retrying under memory
863 * pressure and fails iff @q is dead.
865 * Must be callled with @q->queue_lock held and,
866 * Returns %NULL on failure, with @q->queue_lock held.
867 * Returns !%NULL on success, with @q->queue_lock *not held*.
869 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
872 const bool is_sync = rw_is_sync(rw_flags) != 0;
875 rq = get_request(q, rw_flags, bio, GFP_NOIO);
878 struct io_context *ioc;
879 struct request_list *rl = &q->rq;
881 if (unlikely(blk_queue_dead(q)))
884 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
885 TASK_UNINTERRUPTIBLE);
887 trace_block_sleeprq(q, bio, rw_flags & 1);
889 spin_unlock_irq(q->queue_lock);
893 * After sleeping, we become a "batching" process and
894 * will be able to allocate at least one request, and
895 * up to a big batch of them for a small period time.
896 * See ioc_batching, ioc_set_batching
898 ioc = current_io_context(GFP_NOIO, q->node);
899 ioc_set_batching(q, ioc);
901 spin_lock_irq(q->queue_lock);
902 finish_wait(&rl->wait[is_sync], &wait);
904 rq = get_request(q, rw_flags, bio, GFP_NOIO);
910 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
914 BUG_ON(rw != READ && rw != WRITE);
916 spin_lock_irq(q->queue_lock);
917 if (gfp_mask & __GFP_WAIT)
918 rq = get_request_wait(q, rw, NULL);
920 rq = get_request(q, rw, NULL, gfp_mask);
922 spin_unlock_irq(q->queue_lock);
923 /* q->queue_lock is unlocked at this point */
927 EXPORT_SYMBOL(blk_get_request);
930 * blk_make_request - given a bio, allocate a corresponding struct request.
931 * @q: target request queue
932 * @bio: The bio describing the memory mappings that will be submitted for IO.
933 * It may be a chained-bio properly constructed by block/bio layer.
934 * @gfp_mask: gfp flags to be used for memory allocation
936 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
937 * type commands. Where the struct request needs to be farther initialized by
938 * the caller. It is passed a &struct bio, which describes the memory info of
941 * The caller of blk_make_request must make sure that bi_io_vec
942 * are set to describe the memory buffers. That bio_data_dir() will return
943 * the needed direction of the request. (And all bio's in the passed bio-chain
944 * are properly set accordingly)
946 * If called under none-sleepable conditions, mapped bio buffers must not
947 * need bouncing, by calling the appropriate masked or flagged allocator,
948 * suitable for the target device. Otherwise the call to blk_queue_bounce will
951 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
952 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
953 * anything but the first bio in the chain. Otherwise you risk waiting for IO
954 * completion of a bio that hasn't been submitted yet, thus resulting in a
955 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
956 * of bio_alloc(), as that avoids the mempool deadlock.
957 * If possible a big IO should be split into smaller parts when allocation
958 * fails. Partial allocation should not be an error, or you risk a live-lock.
960 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
963 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
966 return ERR_PTR(-ENOMEM);
969 struct bio *bounce_bio = bio;
972 blk_queue_bounce(q, &bounce_bio);
973 ret = blk_rq_append_bio(q, rq, bounce_bio);
982 EXPORT_SYMBOL(blk_make_request);
985 * blk_requeue_request - put a request back on queue
986 * @q: request queue where request should be inserted
987 * @rq: request to be inserted
990 * Drivers often keep queueing requests until the hardware cannot accept
991 * more, when that condition happens we need to put the request back
992 * on the queue. Must be called with queue lock held.
994 void blk_requeue_request(struct request_queue *q, struct request *rq)
996 blk_delete_timer(rq);
997 blk_clear_rq_complete(rq);
998 trace_block_rq_requeue(q, rq);
1000 if (blk_rq_tagged(rq))
1001 blk_queue_end_tag(q, rq);
1003 BUG_ON(blk_queued_rq(rq));
1005 elv_requeue_request(q, rq);
1007 EXPORT_SYMBOL(blk_requeue_request);
1009 static void add_acct_request(struct request_queue *q, struct request *rq,
1012 drive_stat_acct(rq, 1);
1013 __elv_add_request(q, rq, where);
1017 * blk_insert_request - insert a special request into a request queue
1018 * @q: request queue where request should be inserted
1019 * @rq: request to be inserted
1020 * @at_head: insert request at head or tail of queue
1021 * @data: private data
1024 * Many block devices need to execute commands asynchronously, so they don't
1025 * block the whole kernel from preemption during request execution. This is
1026 * accomplished normally by inserting aritficial requests tagged as
1027 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1028 * be scheduled for actual execution by the request queue.
1030 * We have the option of inserting the head or the tail of the queue.
1031 * Typically we use the tail for new ioctls and so forth. We use the head
1032 * of the queue for things like a QUEUE_FULL message from a device, or a
1033 * host that is unable to accept a particular command.
1035 void blk_insert_request(struct request_queue *q, struct request *rq,
1036 int at_head, void *data)
1038 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1039 unsigned long flags;
1042 * tell I/O scheduler that this isn't a regular read/write (ie it
1043 * must not attempt merges on this) and that it acts as a soft
1046 rq->cmd_type = REQ_TYPE_SPECIAL;
1050 spin_lock_irqsave(q->queue_lock, flags);
1053 * If command is tagged, release the tag
1055 if (blk_rq_tagged(rq))
1056 blk_queue_end_tag(q, rq);
1058 add_acct_request(q, rq, where);
1060 spin_unlock_irqrestore(q->queue_lock, flags);
1062 EXPORT_SYMBOL(blk_insert_request);
1064 static void part_round_stats_single(int cpu, struct hd_struct *part,
1067 if (now == part->stamp)
1070 if (part_in_flight(part)) {
1071 __part_stat_add(cpu, part, time_in_queue,
1072 part_in_flight(part) * (now - part->stamp));
1073 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1079 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1080 * @cpu: cpu number for stats access
1081 * @part: target partition
1083 * The average IO queue length and utilisation statistics are maintained
1084 * by observing the current state of the queue length and the amount of
1085 * time it has been in this state for.
1087 * Normally, that accounting is done on IO completion, but that can result
1088 * in more than a second's worth of IO being accounted for within any one
1089 * second, leading to >100% utilisation. To deal with that, we call this
1090 * function to do a round-off before returning the results when reading
1091 * /proc/diskstats. This accounts immediately for all queue usage up to
1092 * the current jiffies and restarts the counters again.
1094 void part_round_stats(int cpu, struct hd_struct *part)
1096 unsigned long now = jiffies;
1099 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1100 part_round_stats_single(cpu, part, now);
1102 EXPORT_SYMBOL_GPL(part_round_stats);
1105 * queue lock must be held
1107 void __blk_put_request(struct request_queue *q, struct request *req)
1111 if (unlikely(--req->ref_count))
1114 elv_completed_request(q, req);
1116 /* this is a bio leak */
1117 WARN_ON(req->bio != NULL);
1120 * Request may not have originated from ll_rw_blk. if not,
1121 * it didn't come out of our reserved rq pools
1123 if (req->cmd_flags & REQ_ALLOCED) {
1124 unsigned int flags = req->cmd_flags;
1126 BUG_ON(!list_empty(&req->queuelist));
1127 BUG_ON(!hlist_unhashed(&req->hash));
1129 blk_free_request(q, req);
1130 freed_request(q, flags);
1133 EXPORT_SYMBOL_GPL(__blk_put_request);
1135 void blk_put_request(struct request *req)
1137 unsigned long flags;
1138 struct request_queue *q = req->q;
1140 spin_lock_irqsave(q->queue_lock, flags);
1141 __blk_put_request(q, req);
1142 spin_unlock_irqrestore(q->queue_lock, flags);
1144 EXPORT_SYMBOL(blk_put_request);
1147 * blk_add_request_payload - add a payload to a request
1148 * @rq: request to update
1149 * @page: page backing the payload
1150 * @len: length of the payload.
1152 * This allows to later add a payload to an already submitted request by
1153 * a block driver. The driver needs to take care of freeing the payload
1156 * Note that this is a quite horrible hack and nothing but handling of
1157 * discard requests should ever use it.
1159 void blk_add_request_payload(struct request *rq, struct page *page,
1162 struct bio *bio = rq->bio;
1164 bio->bi_io_vec->bv_page = page;
1165 bio->bi_io_vec->bv_offset = 0;
1166 bio->bi_io_vec->bv_len = len;
1170 bio->bi_phys_segments = 1;
1172 rq->__data_len = rq->resid_len = len;
1173 rq->nr_phys_segments = 1;
1174 rq->buffer = bio_data(bio);
1176 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1178 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1181 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1183 if (!ll_back_merge_fn(q, req, bio))
1186 trace_block_bio_backmerge(q, bio);
1188 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1189 blk_rq_set_mixed_merge(req);
1191 req->biotail->bi_next = bio;
1193 req->__data_len += bio->bi_size;
1194 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1196 drive_stat_acct(req, 0);
1197 elv_bio_merged(q, req, bio);
1201 static bool bio_attempt_front_merge(struct request_queue *q,
1202 struct request *req, struct bio *bio)
1204 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1206 if (!ll_front_merge_fn(q, req, bio))
1209 trace_block_bio_frontmerge(q, bio);
1211 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1212 blk_rq_set_mixed_merge(req);
1214 bio->bi_next = req->bio;
1218 * may not be valid. if the low level driver said
1219 * it didn't need a bounce buffer then it better
1220 * not touch req->buffer either...
1222 req->buffer = bio_data(bio);
1223 req->__sector = bio->bi_sector;
1224 req->__data_len += bio->bi_size;
1225 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1227 drive_stat_acct(req, 0);
1228 elv_bio_merged(q, req, bio);
1233 * attempt_plug_merge - try to merge with %current's plugged list
1234 * @q: request_queue new bio is being queued at
1235 * @bio: new bio being queued
1236 * @request_count: out parameter for number of traversed plugged requests
1238 * Determine whether @bio being queued on @q can be merged with a request
1239 * on %current's plugged list. Returns %true if merge was successful,
1242 * This function is called without @q->queue_lock; however, elevator is
1243 * accessed iff there already are requests on the plugged list which in
1244 * turn guarantees validity of the elevator.
1246 * Note that, on successful merge, elevator operation
1247 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1248 * must be ready for this.
1250 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1251 unsigned int *request_count)
1253 struct blk_plug *plug;
1257 plug = current->plug;
1262 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1270 el_ret = elv_try_merge(rq, bio);
1271 if (el_ret == ELEVATOR_BACK_MERGE) {
1272 ret = bio_attempt_back_merge(q, rq, bio);
1275 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1276 ret = bio_attempt_front_merge(q, rq, bio);
1285 void init_request_from_bio(struct request *req, struct bio *bio)
1287 req->cmd_type = REQ_TYPE_FS;
1289 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1290 if (bio->bi_rw & REQ_RAHEAD)
1291 req->cmd_flags |= REQ_FAILFAST_MASK;
1294 req->__sector = bio->bi_sector;
1295 req->ioprio = bio_prio(bio);
1296 blk_rq_bio_prep(req->q, req, bio);
1299 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1301 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1302 struct blk_plug *plug;
1303 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1304 struct request *req;
1305 unsigned int request_count = 0;
1308 * low level driver can indicate that it wants pages above a
1309 * certain limit bounced to low memory (ie for highmem, or even
1310 * ISA dma in theory)
1312 blk_queue_bounce(q, &bio);
1314 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1315 spin_lock_irq(q->queue_lock);
1316 where = ELEVATOR_INSERT_FLUSH;
1321 * Check if we can merge with the plugged list before grabbing
1324 if (attempt_plug_merge(q, bio, &request_count))
1327 spin_lock_irq(q->queue_lock);
1329 el_ret = elv_merge(q, &req, bio);
1330 if (el_ret == ELEVATOR_BACK_MERGE) {
1331 if (bio_attempt_back_merge(q, req, bio)) {
1332 if (!attempt_back_merge(q, req))
1333 elv_merged_request(q, req, el_ret);
1336 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1337 if (bio_attempt_front_merge(q, req, bio)) {
1338 if (!attempt_front_merge(q, req))
1339 elv_merged_request(q, req, el_ret);
1346 * This sync check and mask will be re-done in init_request_from_bio(),
1347 * but we need to set it earlier to expose the sync flag to the
1348 * rq allocator and io schedulers.
1350 rw_flags = bio_data_dir(bio);
1352 rw_flags |= REQ_SYNC;
1355 * Grab a free request. This is might sleep but can not fail.
1356 * Returns with the queue unlocked.
1358 req = get_request_wait(q, rw_flags, bio);
1359 if (unlikely(!req)) {
1360 bio_endio(bio, -ENODEV); /* @q is dead */
1365 * After dropping the lock and possibly sleeping here, our request
1366 * may now be mergeable after it had proven unmergeable (above).
1367 * We don't worry about that case for efficiency. It won't happen
1368 * often, and the elevators are able to handle it.
1370 init_request_from_bio(req, bio);
1372 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1373 req->cpu = raw_smp_processor_id();
1375 plug = current->plug;
1378 * If this is the first request added after a plug, fire
1379 * of a plug trace. If others have been added before, check
1380 * if we have multiple devices in this plug. If so, make a
1381 * note to sort the list before dispatch.
1383 if (list_empty(&plug->list))
1384 trace_block_plug(q);
1386 if (!plug->should_sort) {
1387 struct request *__rq;
1389 __rq = list_entry_rq(plug->list.prev);
1391 plug->should_sort = 1;
1393 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1394 blk_flush_plug_list(plug, false);
1395 trace_block_plug(q);
1398 list_add_tail(&req->queuelist, &plug->list);
1399 drive_stat_acct(req, 1);
1401 spin_lock_irq(q->queue_lock);
1402 add_acct_request(q, req, where);
1405 spin_unlock_irq(q->queue_lock);
1408 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1411 * If bio->bi_dev is a partition, remap the location
1413 static inline void blk_partition_remap(struct bio *bio)
1415 struct block_device *bdev = bio->bi_bdev;
1417 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1418 struct hd_struct *p = bdev->bd_part;
1420 bio->bi_sector += p->start_sect;
1421 bio->bi_bdev = bdev->bd_contains;
1423 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1425 bio->bi_sector - p->start_sect);
1429 static void handle_bad_sector(struct bio *bio)
1431 char b[BDEVNAME_SIZE];
1433 printk(KERN_INFO "attempt to access beyond end of device\n");
1434 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1435 bdevname(bio->bi_bdev, b),
1437 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1438 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1440 set_bit(BIO_EOF, &bio->bi_flags);
1443 #ifdef CONFIG_FAIL_MAKE_REQUEST
1445 static DECLARE_FAULT_ATTR(fail_make_request);
1447 static int __init setup_fail_make_request(char *str)
1449 return setup_fault_attr(&fail_make_request, str);
1451 __setup("fail_make_request=", setup_fail_make_request);
1453 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1455 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1458 static int __init fail_make_request_debugfs(void)
1460 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1461 NULL, &fail_make_request);
1463 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1466 late_initcall(fail_make_request_debugfs);
1468 #else /* CONFIG_FAIL_MAKE_REQUEST */
1470 static inline bool should_fail_request(struct hd_struct *part,
1476 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1479 * Check whether this bio extends beyond the end of the device.
1481 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1488 /* Test device or partition size, when known. */
1489 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1491 sector_t sector = bio->bi_sector;
1493 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1495 * This may well happen - the kernel calls bread()
1496 * without checking the size of the device, e.g., when
1497 * mounting a device.
1499 handle_bad_sector(bio);
1507 static noinline_for_stack bool
1508 generic_make_request_checks(struct bio *bio)
1510 struct request_queue *q;
1511 int nr_sectors = bio_sectors(bio);
1513 char b[BDEVNAME_SIZE];
1514 struct hd_struct *part;
1518 if (bio_check_eod(bio, nr_sectors))
1521 q = bdev_get_queue(bio->bi_bdev);
1524 "generic_make_request: Trying to access "
1525 "nonexistent block-device %s (%Lu)\n",
1526 bdevname(bio->bi_bdev, b),
1527 (long long) bio->bi_sector);
1531 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1532 nr_sectors > queue_max_hw_sectors(q))) {
1533 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1534 bdevname(bio->bi_bdev, b),
1536 queue_max_hw_sectors(q));
1540 part = bio->bi_bdev->bd_part;
1541 if (should_fail_request(part, bio->bi_size) ||
1542 should_fail_request(&part_to_disk(part)->part0,
1547 * If this device has partitions, remap block n
1548 * of partition p to block n+start(p) of the disk.
1550 blk_partition_remap(bio);
1552 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1555 if (bio_check_eod(bio, nr_sectors))
1559 * Filter flush bio's early so that make_request based
1560 * drivers without flush support don't have to worry
1563 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1564 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1571 if ((bio->bi_rw & REQ_DISCARD) &&
1572 (!blk_queue_discard(q) ||
1573 ((bio->bi_rw & REQ_SECURE) &&
1574 !blk_queue_secdiscard(q)))) {
1579 if (blk_throtl_bio(q, bio))
1580 return false; /* throttled, will be resubmitted later */
1582 trace_block_bio_queue(q, bio);
1586 bio_endio(bio, err);
1591 * generic_make_request - hand a buffer to its device driver for I/O
1592 * @bio: The bio describing the location in memory and on the device.
1594 * generic_make_request() is used to make I/O requests of block
1595 * devices. It is passed a &struct bio, which describes the I/O that needs
1598 * generic_make_request() does not return any status. The
1599 * success/failure status of the request, along with notification of
1600 * completion, is delivered asynchronously through the bio->bi_end_io
1601 * function described (one day) else where.
1603 * The caller of generic_make_request must make sure that bi_io_vec
1604 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1605 * set to describe the device address, and the
1606 * bi_end_io and optionally bi_private are set to describe how
1607 * completion notification should be signaled.
1609 * generic_make_request and the drivers it calls may use bi_next if this
1610 * bio happens to be merged with someone else, and may resubmit the bio to
1611 * a lower device by calling into generic_make_request recursively, which
1612 * means the bio should NOT be touched after the call to ->make_request_fn.
1614 void generic_make_request(struct bio *bio)
1616 struct bio_list bio_list_on_stack;
1618 if (!generic_make_request_checks(bio))
1622 * We only want one ->make_request_fn to be active at a time, else
1623 * stack usage with stacked devices could be a problem. So use
1624 * current->bio_list to keep a list of requests submited by a
1625 * make_request_fn function. current->bio_list is also used as a
1626 * flag to say if generic_make_request is currently active in this
1627 * task or not. If it is NULL, then no make_request is active. If
1628 * it is non-NULL, then a make_request is active, and new requests
1629 * should be added at the tail
1631 if (current->bio_list) {
1632 bio_list_add(current->bio_list, bio);
1636 /* following loop may be a bit non-obvious, and so deserves some
1638 * Before entering the loop, bio->bi_next is NULL (as all callers
1639 * ensure that) so we have a list with a single bio.
1640 * We pretend that we have just taken it off a longer list, so
1641 * we assign bio_list to a pointer to the bio_list_on_stack,
1642 * thus initialising the bio_list of new bios to be
1643 * added. ->make_request() may indeed add some more bios
1644 * through a recursive call to generic_make_request. If it
1645 * did, we find a non-NULL value in bio_list and re-enter the loop
1646 * from the top. In this case we really did just take the bio
1647 * of the top of the list (no pretending) and so remove it from
1648 * bio_list, and call into ->make_request() again.
1650 BUG_ON(bio->bi_next);
1651 bio_list_init(&bio_list_on_stack);
1652 current->bio_list = &bio_list_on_stack;
1654 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1656 q->make_request_fn(q, bio);
1658 bio = bio_list_pop(current->bio_list);
1660 current->bio_list = NULL; /* deactivate */
1662 EXPORT_SYMBOL(generic_make_request);
1665 * submit_bio - submit a bio to the block device layer for I/O
1666 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1667 * @bio: The &struct bio which describes the I/O
1669 * submit_bio() is very similar in purpose to generic_make_request(), and
1670 * uses that function to do most of the work. Both are fairly rough
1671 * interfaces; @bio must be presetup and ready for I/O.
1674 void submit_bio(int rw, struct bio *bio)
1676 int count = bio_sectors(bio);
1681 * If it's a regular read/write or a barrier with data attached,
1682 * go through the normal accounting stuff before submission.
1684 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1686 count_vm_events(PGPGOUT, count);
1688 task_io_account_read(bio->bi_size);
1689 count_vm_events(PGPGIN, count);
1692 if (unlikely(block_dump)) {
1693 char b[BDEVNAME_SIZE];
1694 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1695 current->comm, task_pid_nr(current),
1696 (rw & WRITE) ? "WRITE" : "READ",
1697 (unsigned long long)bio->bi_sector,
1698 bdevname(bio->bi_bdev, b),
1703 generic_make_request(bio);
1705 EXPORT_SYMBOL(submit_bio);
1708 * blk_rq_check_limits - Helper function to check a request for the queue limit
1710 * @rq: the request being checked
1713 * @rq may have been made based on weaker limitations of upper-level queues
1714 * in request stacking drivers, and it may violate the limitation of @q.
1715 * Since the block layer and the underlying device driver trust @rq
1716 * after it is inserted to @q, it should be checked against @q before
1717 * the insertion using this generic function.
1719 * This function should also be useful for request stacking drivers
1720 * in some cases below, so export this function.
1721 * Request stacking drivers like request-based dm may change the queue
1722 * limits while requests are in the queue (e.g. dm's table swapping).
1723 * Such request stacking drivers should check those requests agaist
1724 * the new queue limits again when they dispatch those requests,
1725 * although such checkings are also done against the old queue limits
1726 * when submitting requests.
1728 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1730 if (rq->cmd_flags & REQ_DISCARD)
1733 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1734 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1735 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1740 * queue's settings related to segment counting like q->bounce_pfn
1741 * may differ from that of other stacking queues.
1742 * Recalculate it to check the request correctly on this queue's
1745 blk_recalc_rq_segments(rq);
1746 if (rq->nr_phys_segments > queue_max_segments(q)) {
1747 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1753 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1756 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1757 * @q: the queue to submit the request
1758 * @rq: the request being queued
1760 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1762 unsigned long flags;
1763 int where = ELEVATOR_INSERT_BACK;
1765 if (blk_rq_check_limits(q, rq))
1769 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1772 spin_lock_irqsave(q->queue_lock, flags);
1775 * Submitting request must be dequeued before calling this function
1776 * because it will be linked to another request_queue
1778 BUG_ON(blk_queued_rq(rq));
1780 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1781 where = ELEVATOR_INSERT_FLUSH;
1783 add_acct_request(q, rq, where);
1784 if (where == ELEVATOR_INSERT_FLUSH)
1786 spin_unlock_irqrestore(q->queue_lock, flags);
1790 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1793 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1794 * @rq: request to examine
1797 * A request could be merge of IOs which require different failure
1798 * handling. This function determines the number of bytes which
1799 * can be failed from the beginning of the request without
1800 * crossing into area which need to be retried further.
1803 * The number of bytes to fail.
1806 * queue_lock must be held.
1808 unsigned int blk_rq_err_bytes(const struct request *rq)
1810 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1811 unsigned int bytes = 0;
1814 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1815 return blk_rq_bytes(rq);
1818 * Currently the only 'mixing' which can happen is between
1819 * different fastfail types. We can safely fail portions
1820 * which have all the failfast bits that the first one has -
1821 * the ones which are at least as eager to fail as the first
1824 for (bio = rq->bio; bio; bio = bio->bi_next) {
1825 if ((bio->bi_rw & ff) != ff)
1827 bytes += bio->bi_size;
1830 /* this could lead to infinite loop */
1831 BUG_ON(blk_rq_bytes(rq) && !bytes);
1834 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1836 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1838 if (blk_do_io_stat(req)) {
1839 const int rw = rq_data_dir(req);
1840 struct hd_struct *part;
1843 cpu = part_stat_lock();
1845 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1850 static void blk_account_io_done(struct request *req)
1853 * Account IO completion. flush_rq isn't accounted as a
1854 * normal IO on queueing nor completion. Accounting the
1855 * containing request is enough.
1857 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1858 unsigned long duration = jiffies - req->start_time;
1859 const int rw = rq_data_dir(req);
1860 struct hd_struct *part;
1863 cpu = part_stat_lock();
1866 part_stat_inc(cpu, part, ios[rw]);
1867 part_stat_add(cpu, part, ticks[rw], duration);
1868 part_round_stats(cpu, part);
1869 part_dec_in_flight(part, rw);
1871 hd_struct_put(part);
1877 * blk_peek_request - peek at the top of a request queue
1878 * @q: request queue to peek at
1881 * Return the request at the top of @q. The returned request
1882 * should be started using blk_start_request() before LLD starts
1886 * Pointer to the request at the top of @q if available. Null
1890 * queue_lock must be held.
1892 struct request *blk_peek_request(struct request_queue *q)
1897 while ((rq = __elv_next_request(q)) != NULL) {
1898 if (!(rq->cmd_flags & REQ_STARTED)) {
1900 * This is the first time the device driver
1901 * sees this request (possibly after
1902 * requeueing). Notify IO scheduler.
1904 if (rq->cmd_flags & REQ_SORTED)
1905 elv_activate_rq(q, rq);
1908 * just mark as started even if we don't start
1909 * it, a request that has been delayed should
1910 * not be passed by new incoming requests
1912 rq->cmd_flags |= REQ_STARTED;
1913 trace_block_rq_issue(q, rq);
1916 if (!q->boundary_rq || q->boundary_rq == rq) {
1917 q->end_sector = rq_end_sector(rq);
1918 q->boundary_rq = NULL;
1921 if (rq->cmd_flags & REQ_DONTPREP)
1924 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1926 * make sure space for the drain appears we
1927 * know we can do this because max_hw_segments
1928 * has been adjusted to be one fewer than the
1931 rq->nr_phys_segments++;
1937 ret = q->prep_rq_fn(q, rq);
1938 if (ret == BLKPREP_OK) {
1940 } else if (ret == BLKPREP_DEFER) {
1942 * the request may have been (partially) prepped.
1943 * we need to keep this request in the front to
1944 * avoid resource deadlock. REQ_STARTED will
1945 * prevent other fs requests from passing this one.
1947 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1948 !(rq->cmd_flags & REQ_DONTPREP)) {
1950 * remove the space for the drain we added
1951 * so that we don't add it again
1953 --rq->nr_phys_segments;
1958 } else if (ret == BLKPREP_KILL) {
1959 rq->cmd_flags |= REQ_QUIET;
1961 * Mark this request as started so we don't trigger
1962 * any debug logic in the end I/O path.
1964 blk_start_request(rq);
1965 __blk_end_request_all(rq, -EIO);
1967 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1974 EXPORT_SYMBOL(blk_peek_request);
1976 void blk_dequeue_request(struct request *rq)
1978 struct request_queue *q = rq->q;
1980 BUG_ON(list_empty(&rq->queuelist));
1981 BUG_ON(ELV_ON_HASH(rq));
1983 list_del_init(&rq->queuelist);
1986 * the time frame between a request being removed from the lists
1987 * and to it is freed is accounted as io that is in progress at
1990 if (blk_account_rq(rq)) {
1991 q->in_flight[rq_is_sync(rq)]++;
1992 set_io_start_time_ns(rq);
1997 * blk_start_request - start request processing on the driver
1998 * @req: request to dequeue
2001 * Dequeue @req and start timeout timer on it. This hands off the
2002 * request to the driver.
2004 * Block internal functions which don't want to start timer should
2005 * call blk_dequeue_request().
2008 * queue_lock must be held.
2010 void blk_start_request(struct request *req)
2012 blk_dequeue_request(req);
2015 * We are now handing the request to the hardware, initialize
2016 * resid_len to full count and add the timeout handler.
2018 req->resid_len = blk_rq_bytes(req);
2019 if (unlikely(blk_bidi_rq(req)))
2020 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2022 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2025 EXPORT_SYMBOL(blk_start_request);
2028 * blk_fetch_request - fetch a request from a request queue
2029 * @q: request queue to fetch a request from
2032 * Return the request at the top of @q. The request is started on
2033 * return and LLD can start processing it immediately.
2036 * Pointer to the request at the top of @q if available. Null
2040 * queue_lock must be held.
2042 struct request *blk_fetch_request(struct request_queue *q)
2046 rq = blk_peek_request(q);
2048 blk_start_request(rq);
2051 EXPORT_SYMBOL(blk_fetch_request);
2054 * blk_update_request - Special helper function for request stacking drivers
2055 * @req: the request being processed
2056 * @error: %0 for success, < %0 for error
2057 * @nr_bytes: number of bytes to complete @req
2060 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2061 * the request structure even if @req doesn't have leftover.
2062 * If @req has leftover, sets it up for the next range of segments.
2064 * This special helper function is only for request stacking drivers
2065 * (e.g. request-based dm) so that they can handle partial completion.
2066 * Actual device drivers should use blk_end_request instead.
2068 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2069 * %false return from this function.
2072 * %false - this request doesn't have any more data
2073 * %true - this request has more data
2075 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2077 int total_bytes, bio_nbytes, next_idx = 0;
2083 trace_block_rq_complete(req->q, req, nr_bytes);
2086 * For fs requests, rq is just carrier of independent bio's
2087 * and each partial completion should be handled separately.
2088 * Reset per-request error on each partial completion.
2090 * TODO: tj: This is too subtle. It would be better to let
2091 * low level drivers do what they see fit.
2093 if (req->cmd_type == REQ_TYPE_FS)
2096 if (error && req->cmd_type == REQ_TYPE_FS &&
2097 !(req->cmd_flags & REQ_QUIET)) {
2102 error_type = "recoverable transport";
2105 error_type = "critical target";
2108 error_type = "critical nexus";
2115 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2116 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2117 (unsigned long long)blk_rq_pos(req));
2120 blk_account_io_completion(req, nr_bytes);
2122 total_bytes = bio_nbytes = 0;
2123 while ((bio = req->bio) != NULL) {
2126 if (nr_bytes >= bio->bi_size) {
2127 req->bio = bio->bi_next;
2128 nbytes = bio->bi_size;
2129 req_bio_endio(req, bio, nbytes, error);
2133 int idx = bio->bi_idx + next_idx;
2135 if (unlikely(idx >= bio->bi_vcnt)) {
2136 blk_dump_rq_flags(req, "__end_that");
2137 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2138 __func__, idx, bio->bi_vcnt);
2142 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2143 BIO_BUG_ON(nbytes > bio->bi_size);
2146 * not a complete bvec done
2148 if (unlikely(nbytes > nr_bytes)) {
2149 bio_nbytes += nr_bytes;
2150 total_bytes += nr_bytes;
2155 * advance to the next vector
2158 bio_nbytes += nbytes;
2161 total_bytes += nbytes;
2167 * end more in this run, or just return 'not-done'
2169 if (unlikely(nr_bytes <= 0))
2179 * Reset counters so that the request stacking driver
2180 * can find how many bytes remain in the request
2183 req->__data_len = 0;
2188 * if the request wasn't completed, update state
2191 req_bio_endio(req, bio, bio_nbytes, error);
2192 bio->bi_idx += next_idx;
2193 bio_iovec(bio)->bv_offset += nr_bytes;
2194 bio_iovec(bio)->bv_len -= nr_bytes;
2197 req->__data_len -= total_bytes;
2198 req->buffer = bio_data(req->bio);
2200 /* update sector only for requests with clear definition of sector */
2201 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2202 req->__sector += total_bytes >> 9;
2204 /* mixed attributes always follow the first bio */
2205 if (req->cmd_flags & REQ_MIXED_MERGE) {
2206 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2207 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2211 * If total number of sectors is less than the first segment
2212 * size, something has gone terribly wrong.
2214 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2215 blk_dump_rq_flags(req, "request botched");
2216 req->__data_len = blk_rq_cur_bytes(req);
2219 /* recalculate the number of segments */
2220 blk_recalc_rq_segments(req);
2224 EXPORT_SYMBOL_GPL(blk_update_request);
2226 static bool blk_update_bidi_request(struct request *rq, int error,
2227 unsigned int nr_bytes,
2228 unsigned int bidi_bytes)
2230 if (blk_update_request(rq, error, nr_bytes))
2233 /* Bidi request must be completed as a whole */
2234 if (unlikely(blk_bidi_rq(rq)) &&
2235 blk_update_request(rq->next_rq, error, bidi_bytes))
2238 if (blk_queue_add_random(rq->q))
2239 add_disk_randomness(rq->rq_disk);
2245 * blk_unprep_request - unprepare a request
2248 * This function makes a request ready for complete resubmission (or
2249 * completion). It happens only after all error handling is complete,
2250 * so represents the appropriate moment to deallocate any resources
2251 * that were allocated to the request in the prep_rq_fn. The queue
2252 * lock is held when calling this.
2254 void blk_unprep_request(struct request *req)
2256 struct request_queue *q = req->q;
2258 req->cmd_flags &= ~REQ_DONTPREP;
2259 if (q->unprep_rq_fn)
2260 q->unprep_rq_fn(q, req);
2262 EXPORT_SYMBOL_GPL(blk_unprep_request);
2265 * queue lock must be held
2267 static void blk_finish_request(struct request *req, int error)
2269 if (blk_rq_tagged(req))
2270 blk_queue_end_tag(req->q, req);
2272 BUG_ON(blk_queued_rq(req));
2274 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2275 laptop_io_completion(&req->q->backing_dev_info);
2277 blk_delete_timer(req);
2279 if (req->cmd_flags & REQ_DONTPREP)
2280 blk_unprep_request(req);
2283 blk_account_io_done(req);
2286 req->end_io(req, error);
2288 if (blk_bidi_rq(req))
2289 __blk_put_request(req->next_rq->q, req->next_rq);
2291 __blk_put_request(req->q, req);
2296 * blk_end_bidi_request - Complete a bidi request
2297 * @rq: the request to complete
2298 * @error: %0 for success, < %0 for error
2299 * @nr_bytes: number of bytes to complete @rq
2300 * @bidi_bytes: number of bytes to complete @rq->next_rq
2303 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2304 * Drivers that supports bidi can safely call this member for any
2305 * type of request, bidi or uni. In the later case @bidi_bytes is
2309 * %false - we are done with this request
2310 * %true - still buffers pending for this request
2312 static bool blk_end_bidi_request(struct request *rq, int error,
2313 unsigned int nr_bytes, unsigned int bidi_bytes)
2315 struct request_queue *q = rq->q;
2316 unsigned long flags;
2318 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2321 spin_lock_irqsave(q->queue_lock, flags);
2322 blk_finish_request(rq, error);
2323 spin_unlock_irqrestore(q->queue_lock, flags);
2329 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2330 * @rq: the request to complete
2331 * @error: %0 for success, < %0 for error
2332 * @nr_bytes: number of bytes to complete @rq
2333 * @bidi_bytes: number of bytes to complete @rq->next_rq
2336 * Identical to blk_end_bidi_request() except that queue lock is
2337 * assumed to be locked on entry and remains so on return.
2340 * %false - we are done with this request
2341 * %true - still buffers pending for this request
2343 bool __blk_end_bidi_request(struct request *rq, int error,
2344 unsigned int nr_bytes, unsigned int bidi_bytes)
2346 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2349 blk_finish_request(rq, error);
2355 * blk_end_request - Helper function for drivers to complete the request.
2356 * @rq: the request being processed
2357 * @error: %0 for success, < %0 for error
2358 * @nr_bytes: number of bytes to complete
2361 * Ends I/O on a number of bytes attached to @rq.
2362 * If @rq has leftover, sets it up for the next range of segments.
2365 * %false - we are done with this request
2366 * %true - still buffers pending for this request
2368 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2370 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2372 EXPORT_SYMBOL(blk_end_request);
2375 * blk_end_request_all - Helper function for drives to finish the request.
2376 * @rq: the request to finish
2377 * @error: %0 for success, < %0 for error
2380 * Completely finish @rq.
2382 void blk_end_request_all(struct request *rq, int error)
2385 unsigned int bidi_bytes = 0;
2387 if (unlikely(blk_bidi_rq(rq)))
2388 bidi_bytes = blk_rq_bytes(rq->next_rq);
2390 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2393 EXPORT_SYMBOL(blk_end_request_all);
2396 * blk_end_request_cur - Helper function to finish the current request chunk.
2397 * @rq: the request to finish the current chunk for
2398 * @error: %0 for success, < %0 for error
2401 * Complete the current consecutively mapped chunk from @rq.
2404 * %false - we are done with this request
2405 * %true - still buffers pending for this request
2407 bool blk_end_request_cur(struct request *rq, int error)
2409 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2411 EXPORT_SYMBOL(blk_end_request_cur);
2414 * blk_end_request_err - Finish a request till the next failure boundary.
2415 * @rq: the request to finish till the next failure boundary for
2416 * @error: must be negative errno
2419 * Complete @rq till the next failure boundary.
2422 * %false - we are done with this request
2423 * %true - still buffers pending for this request
2425 bool blk_end_request_err(struct request *rq, int error)
2427 WARN_ON(error >= 0);
2428 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2430 EXPORT_SYMBOL_GPL(blk_end_request_err);
2433 * __blk_end_request - Helper function for drivers to complete the request.
2434 * @rq: the request being processed
2435 * @error: %0 for success, < %0 for error
2436 * @nr_bytes: number of bytes to complete
2439 * Must be called with queue lock held unlike blk_end_request().
2442 * %false - we are done with this request
2443 * %true - still buffers pending for this request
2445 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2447 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2449 EXPORT_SYMBOL(__blk_end_request);
2452 * __blk_end_request_all - Helper function for drives to finish the request.
2453 * @rq: the request to finish
2454 * @error: %0 for success, < %0 for error
2457 * Completely finish @rq. Must be called with queue lock held.
2459 void __blk_end_request_all(struct request *rq, int error)
2462 unsigned int bidi_bytes = 0;
2464 if (unlikely(blk_bidi_rq(rq)))
2465 bidi_bytes = blk_rq_bytes(rq->next_rq);
2467 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2470 EXPORT_SYMBOL(__blk_end_request_all);
2473 * __blk_end_request_cur - Helper function to finish the current request chunk.
2474 * @rq: the request to finish the current chunk for
2475 * @error: %0 for success, < %0 for error
2478 * Complete the current consecutively mapped chunk from @rq. Must
2479 * be called with queue lock held.
2482 * %false - we are done with this request
2483 * %true - still buffers pending for this request
2485 bool __blk_end_request_cur(struct request *rq, int error)
2487 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2489 EXPORT_SYMBOL(__blk_end_request_cur);
2492 * __blk_end_request_err - Finish a request till the next failure boundary.
2493 * @rq: the request to finish till the next failure boundary for
2494 * @error: must be negative errno
2497 * Complete @rq till the next failure boundary. Must be called
2498 * with queue lock held.
2501 * %false - we are done with this request
2502 * %true - still buffers pending for this request
2504 bool __blk_end_request_err(struct request *rq, int error)
2506 WARN_ON(error >= 0);
2507 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2509 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2511 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2514 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2515 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2517 if (bio_has_data(bio)) {
2518 rq->nr_phys_segments = bio_phys_segments(q, bio);
2519 rq->buffer = bio_data(bio);
2521 rq->__data_len = bio->bi_size;
2522 rq->bio = rq->biotail = bio;
2525 rq->rq_disk = bio->bi_bdev->bd_disk;
2528 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2530 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2531 * @rq: the request to be flushed
2534 * Flush all pages in @rq.
2536 void rq_flush_dcache_pages(struct request *rq)
2538 struct req_iterator iter;
2539 struct bio_vec *bvec;
2541 rq_for_each_segment(bvec, rq, iter)
2542 flush_dcache_page(bvec->bv_page);
2544 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2548 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2549 * @q : the queue of the device being checked
2552 * Check if underlying low-level drivers of a device are busy.
2553 * If the drivers want to export their busy state, they must set own
2554 * exporting function using blk_queue_lld_busy() first.
2556 * Basically, this function is used only by request stacking drivers
2557 * to stop dispatching requests to underlying devices when underlying
2558 * devices are busy. This behavior helps more I/O merging on the queue
2559 * of the request stacking driver and prevents I/O throughput regression
2560 * on burst I/O load.
2563 * 0 - Not busy (The request stacking driver should dispatch request)
2564 * 1 - Busy (The request stacking driver should stop dispatching request)
2566 int blk_lld_busy(struct request_queue *q)
2569 return q->lld_busy_fn(q);
2573 EXPORT_SYMBOL_GPL(blk_lld_busy);
2576 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2577 * @rq: the clone request to be cleaned up
2580 * Free all bios in @rq for a cloned request.
2582 void blk_rq_unprep_clone(struct request *rq)
2586 while ((bio = rq->bio) != NULL) {
2587 rq->bio = bio->bi_next;
2592 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2595 * Copy attributes of the original request to the clone request.
2596 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2598 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2600 dst->cpu = src->cpu;
2601 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2602 dst->cmd_type = src->cmd_type;
2603 dst->__sector = blk_rq_pos(src);
2604 dst->__data_len = blk_rq_bytes(src);
2605 dst->nr_phys_segments = src->nr_phys_segments;
2606 dst->ioprio = src->ioprio;
2607 dst->extra_len = src->extra_len;
2611 * blk_rq_prep_clone - Helper function to setup clone request
2612 * @rq: the request to be setup
2613 * @rq_src: original request to be cloned
2614 * @bs: bio_set that bios for clone are allocated from
2615 * @gfp_mask: memory allocation mask for bio
2616 * @bio_ctr: setup function to be called for each clone bio.
2617 * Returns %0 for success, non %0 for failure.
2618 * @data: private data to be passed to @bio_ctr
2621 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2622 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2623 * are not copied, and copying such parts is the caller's responsibility.
2624 * Also, pages which the original bios are pointing to are not copied
2625 * and the cloned bios just point same pages.
2626 * So cloned bios must be completed before original bios, which means
2627 * the caller must complete @rq before @rq_src.
2629 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2630 struct bio_set *bs, gfp_t gfp_mask,
2631 int (*bio_ctr)(struct bio *, struct bio *, void *),
2634 struct bio *bio, *bio_src;
2639 blk_rq_init(NULL, rq);
2641 __rq_for_each_bio(bio_src, rq_src) {
2642 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2646 __bio_clone(bio, bio_src);
2648 if (bio_integrity(bio_src) &&
2649 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2652 if (bio_ctr && bio_ctr(bio, bio_src, data))
2656 rq->biotail->bi_next = bio;
2659 rq->bio = rq->biotail = bio;
2662 __blk_rq_prep_clone(rq, rq_src);
2669 blk_rq_unprep_clone(rq);
2673 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2675 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2677 return queue_work(kblockd_workqueue, work);
2679 EXPORT_SYMBOL(kblockd_schedule_work);
2681 int kblockd_schedule_delayed_work(struct request_queue *q,
2682 struct delayed_work *dwork, unsigned long delay)
2684 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2686 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2688 #define PLUG_MAGIC 0x91827364
2691 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2692 * @plug: The &struct blk_plug that needs to be initialized
2695 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2696 * pending I/O should the task end up blocking between blk_start_plug() and
2697 * blk_finish_plug(). This is important from a performance perspective, but
2698 * also ensures that we don't deadlock. For instance, if the task is blocking
2699 * for a memory allocation, memory reclaim could end up wanting to free a
2700 * page belonging to that request that is currently residing in our private
2701 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2702 * this kind of deadlock.
2704 void blk_start_plug(struct blk_plug *plug)
2706 struct task_struct *tsk = current;
2708 plug->magic = PLUG_MAGIC;
2709 INIT_LIST_HEAD(&plug->list);
2710 INIT_LIST_HEAD(&plug->cb_list);
2711 plug->should_sort = 0;
2714 * If this is a nested plug, don't actually assign it. It will be
2715 * flushed on its own.
2719 * Store ordering should not be needed here, since a potential
2720 * preempt will imply a full memory barrier
2725 EXPORT_SYMBOL(blk_start_plug);
2727 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2729 struct request *rqa = container_of(a, struct request, queuelist);
2730 struct request *rqb = container_of(b, struct request, queuelist);
2732 return !(rqa->q <= rqb->q);
2736 * If 'from_schedule' is true, then postpone the dispatch of requests
2737 * until a safe kblockd context. We due this to avoid accidental big
2738 * additional stack usage in driver dispatch, in places where the originally
2739 * plugger did not intend it.
2741 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2743 __releases(q->queue_lock)
2745 trace_block_unplug(q, depth, !from_schedule);
2748 * If we are punting this to kblockd, then we can safely drop
2749 * the queue_lock before waking kblockd (which needs to take
2752 if (from_schedule) {
2753 spin_unlock(q->queue_lock);
2754 blk_run_queue_async(q);
2757 spin_unlock(q->queue_lock);
2762 static void flush_plug_callbacks(struct blk_plug *plug)
2764 LIST_HEAD(callbacks);
2766 if (list_empty(&plug->cb_list))
2769 list_splice_init(&plug->cb_list, &callbacks);
2771 while (!list_empty(&callbacks)) {
2772 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2775 list_del(&cb->list);
2780 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2782 struct request_queue *q;
2783 unsigned long flags;
2788 BUG_ON(plug->magic != PLUG_MAGIC);
2790 flush_plug_callbacks(plug);
2791 if (list_empty(&plug->list))
2794 list_splice_init(&plug->list, &list);
2796 if (plug->should_sort) {
2797 list_sort(NULL, &list, plug_rq_cmp);
2798 plug->should_sort = 0;
2805 * Save and disable interrupts here, to avoid doing it for every
2806 * queue lock we have to take.
2808 local_irq_save(flags);
2809 while (!list_empty(&list)) {
2810 rq = list_entry_rq(list.next);
2811 list_del_init(&rq->queuelist);
2815 * This drops the queue lock
2818 queue_unplugged(q, depth, from_schedule);
2821 spin_lock(q->queue_lock);
2824 * rq is already accounted, so use raw insert
2826 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2827 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2829 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2835 * This drops the queue lock
2838 queue_unplugged(q, depth, from_schedule);
2840 local_irq_restore(flags);
2843 void blk_finish_plug(struct blk_plug *plug)
2845 blk_flush_plug_list(plug, false);
2847 if (plug == current->plug)
2848 current->plug = NULL;
2850 EXPORT_SYMBOL(blk_finish_plug);
2852 int __init blk_dev_init(void)
2854 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2855 sizeof(((struct request *)0)->cmd_flags));
2857 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2858 kblockd_workqueue = alloc_workqueue("kblockd",
2859 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2860 if (!kblockd_workqueue)
2861 panic("Failed to create kblockd\n");
2863 request_cachep = kmem_cache_create("blkdev_requests",
2864 sizeof(struct request), 0, SLAB_PANIC, NULL);
2866 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2867 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);