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);
42 static int __make_request(struct request_queue *q, struct bio *bio);
45 * For the allocated request tables
47 static struct kmem_cache *request_cachep;
50 * For queue allocation
52 struct kmem_cache *blk_requestq_cachep;
55 * Controlling structure to kblockd
57 static struct workqueue_struct *kblockd_workqueue;
59 static void drive_stat_acct(struct request *rq, int new_io)
61 struct hd_struct *part;
62 int rw = rq_data_dir(rq);
65 if (!blk_do_io_stat(rq))
68 cpu = part_stat_lock();
72 part_stat_inc(cpu, part, merges[rw]);
74 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
75 if (!hd_struct_try_get(part)) {
77 * The partition is already being removed,
78 * the request will be accounted on the disk only
80 * We take a reference on disk->part0 although that
81 * partition will never be deleted, so we can treat
82 * it as any other partition.
84 part = &rq->rq_disk->part0;
87 part_round_stats(cpu, part);
88 part_inc_in_flight(part, rw);
95 void blk_queue_congestion_threshold(struct request_queue *q)
99 nr = q->nr_requests - (q->nr_requests / 8) + 1;
100 if (nr > q->nr_requests)
102 q->nr_congestion_on = nr;
104 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 q->nr_congestion_off = nr;
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * Locates the passed device's request queue and returns the address of its
117 * Will return NULL if the request queue cannot be located.
119 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
121 struct backing_dev_info *ret = NULL;
122 struct request_queue *q = bdev_get_queue(bdev);
125 ret = &q->backing_dev_info;
128 EXPORT_SYMBOL(blk_get_backing_dev_info);
130 void blk_rq_init(struct request_queue *q, struct request *rq)
132 memset(rq, 0, sizeof(*rq));
134 INIT_LIST_HEAD(&rq->queuelist);
135 INIT_LIST_HEAD(&rq->timeout_list);
138 rq->__sector = (sector_t) -1;
139 INIT_HLIST_NODE(&rq->hash);
140 RB_CLEAR_NODE(&rq->rb_node);
142 rq->cmd_len = BLK_MAX_CDB;
145 rq->start_time = jiffies;
146 set_start_time_ns(rq);
149 EXPORT_SYMBOL(blk_rq_init);
151 static void req_bio_endio(struct request *rq, struct bio *bio,
152 unsigned int nbytes, int error)
155 clear_bit(BIO_UPTODATE, &bio->bi_flags);
156 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
159 if (unlikely(nbytes > bio->bi_size)) {
160 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
161 __func__, nbytes, bio->bi_size);
162 nbytes = bio->bi_size;
165 if (unlikely(rq->cmd_flags & REQ_QUIET))
166 set_bit(BIO_QUIET, &bio->bi_flags);
168 bio->bi_size -= nbytes;
169 bio->bi_sector += (nbytes >> 9);
171 if (bio_integrity(bio))
172 bio_integrity_advance(bio, nbytes);
174 /* don't actually finish bio if it's part of flush sequence */
175 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
176 bio_endio(bio, error);
179 void blk_dump_rq_flags(struct request *rq, char *msg)
183 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
184 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
187 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
188 (unsigned long long)blk_rq_pos(rq),
189 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
190 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
191 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
193 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
194 printk(KERN_INFO " cdb: ");
195 for (bit = 0; bit < BLK_MAX_CDB; bit++)
196 printk("%02x ", rq->cmd[bit]);
200 EXPORT_SYMBOL(blk_dump_rq_flags);
202 static void blk_delay_work(struct work_struct *work)
204 struct request_queue *q;
206 q = container_of(work, struct request_queue, delay_work.work);
207 spin_lock_irq(q->queue_lock);
209 spin_unlock_irq(q->queue_lock);
213 * blk_delay_queue - restart queueing after defined interval
214 * @q: The &struct request_queue in question
215 * @msecs: Delay in msecs
218 * Sometimes queueing needs to be postponed for a little while, to allow
219 * resources to come back. This function will make sure that queueing is
220 * restarted around the specified time.
222 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
224 queue_delayed_work(kblockd_workqueue, &q->delay_work,
225 msecs_to_jiffies(msecs));
227 EXPORT_SYMBOL(blk_delay_queue);
230 * blk_start_queue - restart a previously stopped queue
231 * @q: The &struct request_queue in question
234 * blk_start_queue() will clear the stop flag on the queue, and call
235 * the request_fn for the queue if it was in a stopped state when
236 * entered. Also see blk_stop_queue(). Queue lock must be held.
238 void blk_start_queue(struct request_queue *q)
240 WARN_ON(!irqs_disabled());
242 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
245 EXPORT_SYMBOL(blk_start_queue);
248 * blk_stop_queue - stop a queue
249 * @q: The &struct request_queue in question
252 * The Linux block layer assumes that a block driver will consume all
253 * entries on the request queue when the request_fn strategy is called.
254 * Often this will not happen, because of hardware limitations (queue
255 * depth settings). If a device driver gets a 'queue full' response,
256 * or if it simply chooses not to queue more I/O at one point, it can
257 * call this function to prevent the request_fn from being called until
258 * the driver has signalled it's ready to go again. This happens by calling
259 * blk_start_queue() to restart queue operations. Queue lock must be held.
261 void blk_stop_queue(struct request_queue *q)
263 __cancel_delayed_work(&q->delay_work);
264 queue_flag_set(QUEUE_FLAG_STOPPED, q);
266 EXPORT_SYMBOL(blk_stop_queue);
269 * blk_sync_queue - cancel any pending callbacks on a queue
273 * The block layer may perform asynchronous callback activity
274 * on a queue, such as calling the unplug function after a timeout.
275 * A block device may call blk_sync_queue to ensure that any
276 * such activity is cancelled, thus allowing it to release resources
277 * that the callbacks might use. The caller must already have made sure
278 * that its ->make_request_fn will not re-add plugging prior to calling
281 * This function does not cancel any asynchronous activity arising
282 * out of elevator or throttling code. That would require elevaotor_exit()
283 * and blk_throtl_exit() to be called with queue lock initialized.
286 void blk_sync_queue(struct request_queue *q)
288 del_timer_sync(&q->timeout);
289 cancel_delayed_work_sync(&q->delay_work);
291 EXPORT_SYMBOL(blk_sync_queue);
294 * __blk_run_queue - run a single device queue
295 * @q: The queue to run
298 * See @blk_run_queue. This variant must be called with the queue lock
299 * held and interrupts disabled.
301 void __blk_run_queue(struct request_queue *q)
303 if (unlikely(blk_queue_stopped(q)))
308 EXPORT_SYMBOL(__blk_run_queue);
311 * blk_run_queue_async - run a single device queue in workqueue context
312 * @q: The queue to run
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
318 void blk_run_queue_async(struct request_queue *q)
320 if (likely(!blk_queue_stopped(q))) {
321 __cancel_delayed_work(&q->delay_work);
322 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
325 EXPORT_SYMBOL(blk_run_queue_async);
328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
332 * Invoke request handling on this queue, if it has pending work to do.
333 * May be used to restart queueing when a request has completed.
335 void blk_run_queue(struct request_queue *q)
339 spin_lock_irqsave(q->queue_lock, flags);
341 spin_unlock_irqrestore(q->queue_lock, flags);
343 EXPORT_SYMBOL(blk_run_queue);
345 void blk_put_queue(struct request_queue *q)
347 kobject_put(&q->kobj);
349 EXPORT_SYMBOL(blk_put_queue);
351 static void blk_drain_queue(struct request_queue *q)
358 spin_lock_irq(q->queue_lock);
361 elv_drain_elevator(q);
364 * This function might be called on a queue which failed
365 * driver init after queue creation or is not yet fully
366 * active yet. Some drivers (e.g. fd and loop) get unhappy
367 * in such cases. Kick queue iff dispatch queue has
368 * something on it and @q has request_fn set.
370 if (!list_empty(&q->queue_head) && q->request_fn)
373 drain |= q->rq.elvpriv;
374 // drain |= q->request_fn_active;
376 * Unfortunately, requests are queued at and tracked from
377 * multiple places and there's no single counter which can
378 * be drained. Check all the queues and counters.
380 drain |= !list_empty(&q->queue_head);
381 for (i = 0; i < 2; i++) {
382 drain |= q->rq.count[i];
383 drain |= q->in_flight[i];
384 drain |= !list_empty(&q->flush_queue[i]);
387 spin_unlock_irq(q->queue_lock);
397 * Note: If a driver supplied the queue lock, it is disconnected
398 * by this function. The actual state of the lock doesn't matter
399 * here as the request_queue isn't accessible after this point
400 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
402 void blk_cleanup_queue(struct request_queue *q)
404 spinlock_t *lock = q->queue_lock;
406 /* mark @q DEAD, no new request or merges will be allowed afterwards */
407 mutex_lock(&q->sysfs_lock);
408 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
411 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
412 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
413 queue_flag_set(QUEUE_FLAG_DEAD, q);
415 if (q->queue_lock != &q->__queue_lock)
416 q->queue_lock = &q->__queue_lock;
418 spin_unlock_irq(lock);
419 mutex_unlock(&q->sysfs_lock);
421 /* drain all requests queued before DEAD marking */
424 /* @q won't process any more request, flush async actions */
425 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
428 /* @q is and will stay empty, shutdown and put */
431 EXPORT_SYMBOL(blk_cleanup_queue);
433 static int blk_init_free_list(struct request_queue *q)
435 struct request_list *rl = &q->rq;
437 if (unlikely(rl->rq_pool))
440 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
441 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
443 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
444 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
446 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
447 mempool_free_slab, request_cachep, q->node);
455 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
457 return blk_alloc_queue_node(gfp_mask, -1);
459 EXPORT_SYMBOL(blk_alloc_queue);
461 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
463 struct request_queue *q;
466 q = kmem_cache_alloc_node(blk_requestq_cachep,
467 gfp_mask | __GFP_ZERO, node_id);
471 q->backing_dev_info.ra_pages =
472 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
473 q->backing_dev_info.state = 0;
474 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
475 q->backing_dev_info.name = "block";
478 err = bdi_init(&q->backing_dev_info);
480 kmem_cache_free(blk_requestq_cachep, q);
484 if (blk_throtl_init(q)) {
485 kmem_cache_free(blk_requestq_cachep, q);
489 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
490 laptop_mode_timer_fn, (unsigned long) q);
491 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
492 INIT_LIST_HEAD(&q->timeout_list);
493 INIT_LIST_HEAD(&q->flush_queue[0]);
494 INIT_LIST_HEAD(&q->flush_queue[1]);
495 INIT_LIST_HEAD(&q->flush_data_in_flight);
496 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
498 kobject_init(&q->kobj, &blk_queue_ktype);
500 mutex_init(&q->sysfs_lock);
501 spin_lock_init(&q->__queue_lock);
504 * By default initialize queue_lock to internal lock and driver can
505 * override it later if need be.
507 q->queue_lock = &q->__queue_lock;
511 EXPORT_SYMBOL(blk_alloc_queue_node);
514 * blk_init_queue - prepare a request queue for use with a block device
515 * @rfn: The function to be called to process requests that have been
516 * placed on the queue.
517 * @lock: Request queue spin lock
520 * If a block device wishes to use the standard request handling procedures,
521 * which sorts requests and coalesces adjacent requests, then it must
522 * call blk_init_queue(). The function @rfn will be called when there
523 * are requests on the queue that need to be processed. If the device
524 * supports plugging, then @rfn may not be called immediately when requests
525 * are available on the queue, but may be called at some time later instead.
526 * Plugged queues are generally unplugged when a buffer belonging to one
527 * of the requests on the queue is needed, or due to memory pressure.
529 * @rfn is not required, or even expected, to remove all requests off the
530 * queue, but only as many as it can handle at a time. If it does leave
531 * requests on the queue, it is responsible for arranging that the requests
532 * get dealt with eventually.
534 * The queue spin lock must be held while manipulating the requests on the
535 * request queue; this lock will be taken also from interrupt context, so irq
536 * disabling is needed for it.
538 * Function returns a pointer to the initialized request queue, or %NULL if
542 * blk_init_queue() must be paired with a blk_cleanup_queue() call
543 * when the block device is deactivated (such as at module unload).
546 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
548 return blk_init_queue_node(rfn, lock, -1);
550 EXPORT_SYMBOL(blk_init_queue);
552 struct request_queue *
553 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
555 struct request_queue *uninit_q, *q;
557 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
561 q = blk_init_allocated_queue(uninit_q, rfn, lock);
563 blk_cleanup_queue(uninit_q);
567 EXPORT_SYMBOL(blk_init_queue_node);
569 struct request_queue *
570 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
576 if (blk_init_free_list(q))
580 q->prep_rq_fn = NULL;
581 q->unprep_rq_fn = NULL;
582 q->queue_flags = QUEUE_FLAG_DEFAULT;
584 /* Override internal queue lock with supplied lock pointer */
586 q->queue_lock = lock;
589 * This also sets hw/phys segments, boundary and size
591 blk_queue_make_request(q, __make_request);
593 q->sg_reserved_size = INT_MAX;
598 if (!elevator_init(q, NULL)) {
599 blk_queue_congestion_threshold(q);
605 EXPORT_SYMBOL(blk_init_allocated_queue);
607 int blk_get_queue(struct request_queue *q)
609 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
610 kobject_get(&q->kobj);
616 EXPORT_SYMBOL(blk_get_queue);
618 static inline void blk_free_request(struct request_queue *q, struct request *rq)
620 if (rq->cmd_flags & REQ_ELVPRIV)
621 elv_put_request(q, rq);
622 mempool_free(rq, q->rq.rq_pool);
625 static struct request *
626 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
628 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
635 rq->cmd_flags = flags | REQ_ALLOCED;
638 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
639 mempool_free(rq, q->rq.rq_pool);
642 rq->cmd_flags |= REQ_ELVPRIV;
649 * ioc_batching returns true if the ioc is a valid batching request and
650 * should be given priority access to a request.
652 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
658 * Make sure the process is able to allocate at least 1 request
659 * even if the batch times out, otherwise we could theoretically
662 return ioc->nr_batch_requests == q->nr_batching ||
663 (ioc->nr_batch_requests > 0
664 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
668 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
669 * will cause the process to be a "batcher" on all queues in the system. This
670 * is the behaviour we want though - once it gets a wakeup it should be given
673 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
675 if (!ioc || ioc_batching(q, ioc))
678 ioc->nr_batch_requests = q->nr_batching;
679 ioc->last_waited = jiffies;
682 static void __freed_request(struct request_queue *q, int sync)
684 struct request_list *rl = &q->rq;
686 if (rl->count[sync] < queue_congestion_off_threshold(q))
687 blk_clear_queue_congested(q, sync);
689 if (rl->count[sync] + 1 <= q->nr_requests) {
690 if (waitqueue_active(&rl->wait[sync]))
691 wake_up(&rl->wait[sync]);
693 blk_clear_queue_full(q, sync);
698 * A request has just been released. Account for it, update the full and
699 * congestion status, wake up any waiters. Called under q->queue_lock.
701 static void freed_request(struct request_queue *q, int sync, int priv)
703 struct request_list *rl = &q->rq;
709 __freed_request(q, sync);
711 if (unlikely(rl->starved[sync ^ 1]))
712 __freed_request(q, sync ^ 1);
716 * Determine if elevator data should be initialized when allocating the
717 * request associated with @bio.
719 static bool blk_rq_should_init_elevator(struct bio *bio)
725 * Flush requests do not use the elevator so skip initialization.
726 * This allows a request to share the flush and elevator data.
728 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
735 * get_request - get a free request
736 * @q: request_queue to allocate request from
737 * @rw_flags: RW and SYNC flags
738 * @bio: bio to allocate request for (can be %NULL)
739 * @gfp_mask: allocation mask
741 * Get a free request from @q. This function may fail under memory
742 * pressure or if @q is dead.
744 * Must be callled with @q->queue_lock held and,
745 * Returns %NULL on failure, with @q->queue_lock held.
746 * Returns !%NULL on success, with @q->queue_lock *not held*.
748 static struct request *get_request(struct request_queue *q, int rw_flags,
749 struct bio *bio, gfp_t gfp_mask)
751 struct request *rq = NULL;
752 struct request_list *rl = &q->rq;
753 struct io_context *ioc = NULL;
754 const bool is_sync = rw_is_sync(rw_flags) != 0;
755 int may_queue, priv = 0;
757 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
760 may_queue = elv_may_queue(q, rw_flags);
761 if (may_queue == ELV_MQUEUE_NO)
764 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
765 if (rl->count[is_sync]+1 >= q->nr_requests) {
766 ioc = current_io_context(GFP_ATOMIC, q->node);
768 * The queue will fill after this allocation, so set
769 * it as full, and mark this process as "batching".
770 * This process will be allowed to complete a batch of
771 * requests, others will be blocked.
773 if (!blk_queue_full(q, is_sync)) {
774 ioc_set_batching(q, ioc);
775 blk_set_queue_full(q, is_sync);
777 if (may_queue != ELV_MQUEUE_MUST
778 && !ioc_batching(q, ioc)) {
780 * The queue is full and the allocating
781 * process is not a "batcher", and not
782 * exempted by the IO scheduler
788 blk_set_queue_congested(q, is_sync);
792 * Only allow batching queuers to allocate up to 50% over the defined
793 * limit of requests, otherwise we could have thousands of requests
794 * allocated with any setting of ->nr_requests
796 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
799 rl->count[is_sync]++;
800 rl->starved[is_sync] = 0;
802 if (blk_rq_should_init_elevator(bio)) {
803 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
808 if (blk_queue_io_stat(q))
809 rw_flags |= REQ_IO_STAT;
810 spin_unlock_irq(q->queue_lock);
812 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
815 * Allocation failed presumably due to memory. Undo anything
816 * we might have messed up.
818 * Allocating task should really be put onto the front of the
819 * wait queue, but this is pretty rare.
821 spin_lock_irq(q->queue_lock);
822 freed_request(q, is_sync, priv);
825 * in the very unlikely event that allocation failed and no
826 * requests for this direction was pending, mark us starved
827 * so that freeing of a request in the other direction will
828 * notice us. another possible fix would be to split the
829 * rq mempool into READ and WRITE
832 if (unlikely(rl->count[is_sync] == 0))
833 rl->starved[is_sync] = 1;
839 * ioc may be NULL here, and ioc_batching will be false. That's
840 * OK, if the queue is under the request limit then requests need
841 * not count toward the nr_batch_requests limit. There will always
842 * be some limit enforced by BLK_BATCH_TIME.
844 if (ioc_batching(q, ioc))
845 ioc->nr_batch_requests--;
847 trace_block_getrq(q, bio, rw_flags & 1);
853 * get_request_wait - get a free request with retry
854 * @q: request_queue to allocate request from
855 * @rw_flags: RW and SYNC flags
856 * @bio: bio to allocate request for (can be %NULL)
858 * Get a free request from @q. This function keeps retrying under memory
859 * pressure and fails iff @q is dead.
861 * Must be callled with @q->queue_lock held and,
862 * Returns %NULL on failure, with @q->queue_lock held.
863 * Returns !%NULL on success, with @q->queue_lock *not held*.
865 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
868 const bool is_sync = rw_is_sync(rw_flags) != 0;
871 rq = get_request(q, rw_flags, bio, GFP_NOIO);
874 struct io_context *ioc;
875 struct request_list *rl = &q->rq;
877 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
880 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
881 TASK_UNINTERRUPTIBLE);
883 trace_block_sleeprq(q, bio, rw_flags & 1);
885 spin_unlock_irq(q->queue_lock);
889 * After sleeping, we become a "batching" process and
890 * will be able to allocate at least one request, and
891 * up to a big batch of them for a small period time.
892 * See ioc_batching, ioc_set_batching
894 ioc = current_io_context(GFP_NOIO, q->node);
895 ioc_set_batching(q, ioc);
897 spin_lock_irq(q->queue_lock);
898 finish_wait(&rl->wait[is_sync], &wait);
900 rq = get_request(q, rw_flags, bio, GFP_NOIO);
906 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
910 BUG_ON(rw != READ && rw != WRITE);
912 spin_lock_irq(q->queue_lock);
913 if (gfp_mask & __GFP_WAIT)
914 rq = get_request_wait(q, rw, NULL);
916 rq = get_request(q, rw, NULL, gfp_mask);
918 spin_unlock_irq(q->queue_lock);
919 /* q->queue_lock is unlocked at this point */
923 EXPORT_SYMBOL(blk_get_request);
926 * blk_make_request - given a bio, allocate a corresponding struct request.
927 * @q: target request queue
928 * @bio: The bio describing the memory mappings that will be submitted for IO.
929 * It may be a chained-bio properly constructed by block/bio layer.
930 * @gfp_mask: gfp flags to be used for memory allocation
932 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
933 * type commands. Where the struct request needs to be farther initialized by
934 * the caller. It is passed a &struct bio, which describes the memory info of
937 * The caller of blk_make_request must make sure that bi_io_vec
938 * are set to describe the memory buffers. That bio_data_dir() will return
939 * the needed direction of the request. (And all bio's in the passed bio-chain
940 * are properly set accordingly)
942 * If called under none-sleepable conditions, mapped bio buffers must not
943 * need bouncing, by calling the appropriate masked or flagged allocator,
944 * suitable for the target device. Otherwise the call to blk_queue_bounce will
947 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
948 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
949 * anything but the first bio in the chain. Otherwise you risk waiting for IO
950 * completion of a bio that hasn't been submitted yet, thus resulting in a
951 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
952 * of bio_alloc(), as that avoids the mempool deadlock.
953 * If possible a big IO should be split into smaller parts when allocation
954 * fails. Partial allocation should not be an error, or you risk a live-lock.
956 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
959 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
962 return ERR_PTR(-ENOMEM);
965 struct bio *bounce_bio = bio;
968 blk_queue_bounce(q, &bounce_bio);
969 ret = blk_rq_append_bio(q, rq, bounce_bio);
978 EXPORT_SYMBOL(blk_make_request);
981 * blk_requeue_request - put a request back on queue
982 * @q: request queue where request should be inserted
983 * @rq: request to be inserted
986 * Drivers often keep queueing requests until the hardware cannot accept
987 * more, when that condition happens we need to put the request back
988 * on the queue. Must be called with queue lock held.
990 void blk_requeue_request(struct request_queue *q, struct request *rq)
992 blk_delete_timer(rq);
993 blk_clear_rq_complete(rq);
994 trace_block_rq_requeue(q, rq);
996 if (blk_rq_tagged(rq))
997 blk_queue_end_tag(q, rq);
999 BUG_ON(blk_queued_rq(rq));
1001 elv_requeue_request(q, rq);
1003 EXPORT_SYMBOL(blk_requeue_request);
1005 static void add_acct_request(struct request_queue *q, struct request *rq,
1008 drive_stat_acct(rq, 1);
1009 __elv_add_request(q, rq, where);
1013 * blk_insert_request - insert a special request into a request queue
1014 * @q: request queue where request should be inserted
1015 * @rq: request to be inserted
1016 * @at_head: insert request at head or tail of queue
1017 * @data: private data
1020 * Many block devices need to execute commands asynchronously, so they don't
1021 * block the whole kernel from preemption during request execution. This is
1022 * accomplished normally by inserting aritficial requests tagged as
1023 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1024 * be scheduled for actual execution by the request queue.
1026 * We have the option of inserting the head or the tail of the queue.
1027 * Typically we use the tail for new ioctls and so forth. We use the head
1028 * of the queue for things like a QUEUE_FULL message from a device, or a
1029 * host that is unable to accept a particular command.
1031 void blk_insert_request(struct request_queue *q, struct request *rq,
1032 int at_head, void *data)
1034 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1035 unsigned long flags;
1038 * tell I/O scheduler that this isn't a regular read/write (ie it
1039 * must not attempt merges on this) and that it acts as a soft
1042 rq->cmd_type = REQ_TYPE_SPECIAL;
1046 spin_lock_irqsave(q->queue_lock, flags);
1049 * If command is tagged, release the tag
1051 if (blk_rq_tagged(rq))
1052 blk_queue_end_tag(q, rq);
1054 add_acct_request(q, rq, where);
1056 spin_unlock_irqrestore(q->queue_lock, flags);
1058 EXPORT_SYMBOL(blk_insert_request);
1060 static void part_round_stats_single(int cpu, struct hd_struct *part,
1063 if (now == part->stamp)
1066 if (part_in_flight(part)) {
1067 __part_stat_add(cpu, part, time_in_queue,
1068 part_in_flight(part) * (now - part->stamp));
1069 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1075 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1076 * @cpu: cpu number for stats access
1077 * @part: target partition
1079 * The average IO queue length and utilisation statistics are maintained
1080 * by observing the current state of the queue length and the amount of
1081 * time it has been in this state for.
1083 * Normally, that accounting is done on IO completion, but that can result
1084 * in more than a second's worth of IO being accounted for within any one
1085 * second, leading to >100% utilisation. To deal with that, we call this
1086 * function to do a round-off before returning the results when reading
1087 * /proc/diskstats. This accounts immediately for all queue usage up to
1088 * the current jiffies and restarts the counters again.
1090 void part_round_stats(int cpu, struct hd_struct *part)
1092 unsigned long now = jiffies;
1095 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1096 part_round_stats_single(cpu, part, now);
1098 EXPORT_SYMBOL_GPL(part_round_stats);
1101 * queue lock must be held
1103 void __blk_put_request(struct request_queue *q, struct request *req)
1107 if (unlikely(--req->ref_count))
1110 elv_completed_request(q, req);
1112 /* this is a bio leak */
1113 WARN_ON(req->bio != NULL);
1116 * Request may not have originated from ll_rw_blk. if not,
1117 * it didn't come out of our reserved rq pools
1119 if (req->cmd_flags & REQ_ALLOCED) {
1120 int is_sync = rq_is_sync(req) != 0;
1121 int priv = req->cmd_flags & REQ_ELVPRIV;
1123 BUG_ON(!list_empty(&req->queuelist));
1124 BUG_ON(!hlist_unhashed(&req->hash));
1126 blk_free_request(q, req);
1127 freed_request(q, is_sync, priv);
1130 EXPORT_SYMBOL_GPL(__blk_put_request);
1132 void blk_put_request(struct request *req)
1134 unsigned long flags;
1135 struct request_queue *q = req->q;
1137 spin_lock_irqsave(q->queue_lock, flags);
1138 __blk_put_request(q, req);
1139 spin_unlock_irqrestore(q->queue_lock, flags);
1141 EXPORT_SYMBOL(blk_put_request);
1144 * blk_add_request_payload - add a payload to a request
1145 * @rq: request to update
1146 * @page: page backing the payload
1147 * @len: length of the payload.
1149 * This allows to later add a payload to an already submitted request by
1150 * a block driver. The driver needs to take care of freeing the payload
1153 * Note that this is a quite horrible hack and nothing but handling of
1154 * discard requests should ever use it.
1156 void blk_add_request_payload(struct request *rq, struct page *page,
1159 struct bio *bio = rq->bio;
1161 bio->bi_io_vec->bv_page = page;
1162 bio->bi_io_vec->bv_offset = 0;
1163 bio->bi_io_vec->bv_len = len;
1167 bio->bi_phys_segments = 1;
1169 rq->__data_len = rq->resid_len = len;
1170 rq->nr_phys_segments = 1;
1171 rq->buffer = bio_data(bio);
1173 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1175 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1178 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1180 if (!ll_back_merge_fn(q, req, bio))
1183 trace_block_bio_backmerge(q, bio);
1185 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1186 blk_rq_set_mixed_merge(req);
1188 req->biotail->bi_next = bio;
1190 req->__data_len += bio->bi_size;
1191 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1193 drive_stat_acct(req, 0);
1194 elv_bio_merged(q, req, bio);
1198 static bool bio_attempt_front_merge(struct request_queue *q,
1199 struct request *req, struct bio *bio)
1201 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1203 if (!ll_front_merge_fn(q, req, bio))
1206 trace_block_bio_frontmerge(q, bio);
1208 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1209 blk_rq_set_mixed_merge(req);
1211 bio->bi_next = req->bio;
1215 * may not be valid. if the low level driver said
1216 * it didn't need a bounce buffer then it better
1217 * not touch req->buffer either...
1219 req->buffer = bio_data(bio);
1220 req->__sector = bio->bi_sector;
1221 req->__data_len += bio->bi_size;
1222 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1224 drive_stat_acct(req, 0);
1225 elv_bio_merged(q, req, bio);
1230 * Attempts to merge with the plugged list in the current process. Returns
1231 * true if merge was successful, otherwise false.
1233 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1236 struct blk_plug *plug;
1244 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1250 el_ret = elv_try_merge(rq, bio);
1251 if (el_ret == ELEVATOR_BACK_MERGE) {
1252 ret = bio_attempt_back_merge(q, rq, bio);
1255 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1256 ret = bio_attempt_front_merge(q, rq, bio);
1265 void init_request_from_bio(struct request *req, struct bio *bio)
1267 req->cpu = bio->bi_comp_cpu;
1268 req->cmd_type = REQ_TYPE_FS;
1270 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1271 if (bio->bi_rw & REQ_RAHEAD)
1272 req->cmd_flags |= REQ_FAILFAST_MASK;
1275 req->__sector = bio->bi_sector;
1276 req->ioprio = bio_prio(bio);
1277 blk_rq_bio_prep(req->q, req, bio);
1280 static int __make_request(struct request_queue *q, struct bio *bio)
1282 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1283 struct blk_plug *plug;
1284 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1285 struct request *req;
1288 * low level driver can indicate that it wants pages above a
1289 * certain limit bounced to low memory (ie for highmem, or even
1290 * ISA dma in theory)
1292 blk_queue_bounce(q, &bio);
1294 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1295 spin_lock_irq(q->queue_lock);
1296 where = ELEVATOR_INSERT_FLUSH;
1301 * Check if we can merge with the plugged list before grabbing
1304 if (attempt_plug_merge(current, q, bio))
1307 spin_lock_irq(q->queue_lock);
1309 el_ret = elv_merge(q, &req, bio);
1310 if (el_ret == ELEVATOR_BACK_MERGE) {
1311 if (bio_attempt_back_merge(q, req, bio)) {
1312 if (!attempt_back_merge(q, req))
1313 elv_merged_request(q, req, el_ret);
1316 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1317 if (bio_attempt_front_merge(q, req, bio)) {
1318 if (!attempt_front_merge(q, req))
1319 elv_merged_request(q, req, el_ret);
1326 * This sync check and mask will be re-done in init_request_from_bio(),
1327 * but we need to set it earlier to expose the sync flag to the
1328 * rq allocator and io schedulers.
1330 rw_flags = bio_data_dir(bio);
1332 rw_flags |= REQ_SYNC;
1335 * Grab a free request. This is might sleep but can not fail.
1336 * Returns with the queue unlocked.
1338 req = get_request_wait(q, rw_flags, bio);
1339 if (unlikely(!req)) {
1340 bio_endio(bio, -ENODEV); /* @q is dead */
1345 * After dropping the lock and possibly sleeping here, our request
1346 * may now be mergeable after it had proven unmergeable (above).
1347 * We don't worry about that case for efficiency. It won't happen
1348 * often, and the elevators are able to handle it.
1350 init_request_from_bio(req, bio);
1352 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1353 bio_flagged(bio, BIO_CPU_AFFINE)) {
1354 req->cpu = blk_cpu_to_group(get_cpu());
1358 plug = current->plug;
1361 * If this is the first request added after a plug, fire
1362 * of a plug trace. If others have been added before, check
1363 * if we have multiple devices in this plug. If so, make a
1364 * note to sort the list before dispatch.
1366 if (list_empty(&plug->list))
1367 trace_block_plug(q);
1368 else if (!plug->should_sort) {
1369 struct request *__rq;
1371 __rq = list_entry_rq(plug->list.prev);
1373 plug->should_sort = 1;
1375 list_add_tail(&req->queuelist, &plug->list);
1376 drive_stat_acct(req, 1);
1378 spin_lock_irq(q->queue_lock);
1379 add_acct_request(q, req, where);
1382 spin_unlock_irq(q->queue_lock);
1389 * If bio->bi_dev is a partition, remap the location
1391 static inline void blk_partition_remap(struct bio *bio)
1393 struct block_device *bdev = bio->bi_bdev;
1395 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1396 struct hd_struct *p = bdev->bd_part;
1398 bio->bi_sector += p->start_sect;
1399 bio->bi_bdev = bdev->bd_contains;
1401 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1403 bio->bi_sector - p->start_sect);
1407 static void handle_bad_sector(struct bio *bio)
1409 char b[BDEVNAME_SIZE];
1411 printk(KERN_INFO "attempt to access beyond end of device\n");
1412 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1413 bdevname(bio->bi_bdev, b),
1415 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1416 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1418 set_bit(BIO_EOF, &bio->bi_flags);
1421 #ifdef CONFIG_FAIL_MAKE_REQUEST
1423 static DECLARE_FAULT_ATTR(fail_make_request);
1425 static int __init setup_fail_make_request(char *str)
1427 return setup_fault_attr(&fail_make_request, str);
1429 __setup("fail_make_request=", setup_fail_make_request);
1431 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1433 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1436 static int __init fail_make_request_debugfs(void)
1438 return init_fault_attr_dentries(&fail_make_request,
1439 "fail_make_request");
1442 late_initcall(fail_make_request_debugfs);
1444 #else /* CONFIG_FAIL_MAKE_REQUEST */
1446 static inline bool should_fail_request(struct hd_struct *part,
1452 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1455 * Check whether this bio extends beyond the end of the device.
1457 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1464 /* Test device or partition size, when known. */
1465 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1467 sector_t sector = bio->bi_sector;
1469 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1471 * This may well happen - the kernel calls bread()
1472 * without checking the size of the device, e.g., when
1473 * mounting a device.
1475 handle_bad_sector(bio);
1484 * generic_make_request - hand a buffer to its device driver for I/O
1485 * @bio: The bio describing the location in memory and on the device.
1487 * generic_make_request() is used to make I/O requests of block
1488 * devices. It is passed a &struct bio, which describes the I/O that needs
1491 * generic_make_request() does not return any status. The
1492 * success/failure status of the request, along with notification of
1493 * completion, is delivered asynchronously through the bio->bi_end_io
1494 * function described (one day) else where.
1496 * The caller of generic_make_request must make sure that bi_io_vec
1497 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1498 * set to describe the device address, and the
1499 * bi_end_io and optionally bi_private are set to describe how
1500 * completion notification should be signaled.
1502 * generic_make_request and the drivers it calls may use bi_next if this
1503 * bio happens to be merged with someone else, and may change bi_dev and
1504 * bi_sector for remaps as it sees fit. So the values of these fields
1505 * should NOT be depended on after the call to generic_make_request.
1507 static inline void __generic_make_request(struct bio *bio)
1509 struct request_queue *q;
1510 sector_t old_sector;
1511 int ret, nr_sectors = bio_sectors(bio);
1517 if (bio_check_eod(bio, nr_sectors))
1521 * Resolve the mapping until finished. (drivers are
1522 * still free to implement/resolve their own stacking
1523 * by explicitly returning 0)
1525 * NOTE: we don't repeat the blk_size check for each new device.
1526 * Stacking drivers are expected to know what they are doing.
1531 char b[BDEVNAME_SIZE];
1532 struct hd_struct *part;
1534 q = bdev_get_queue(bio->bi_bdev);
1537 "generic_make_request: Trying to access "
1538 "nonexistent block-device %s (%Lu)\n",
1539 bdevname(bio->bi_bdev, b),
1540 (long long) bio->bi_sector);
1544 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1545 nr_sectors > queue_max_hw_sectors(q))) {
1546 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1547 bdevname(bio->bi_bdev, b),
1549 queue_max_hw_sectors(q));
1553 part = bio->bi_bdev->bd_part;
1554 if (should_fail_request(part, bio->bi_size) ||
1555 should_fail_request(&part_to_disk(part)->part0,
1560 * If this device has partitions, remap block n
1561 * of partition p to block n+start(p) of the disk.
1563 blk_partition_remap(bio);
1565 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1568 if (old_sector != -1)
1569 trace_block_bio_remap(q, bio, old_dev, old_sector);
1571 old_sector = bio->bi_sector;
1572 old_dev = bio->bi_bdev->bd_dev;
1574 if (bio_check_eod(bio, nr_sectors))
1578 * Filter flush bio's early so that make_request based
1579 * drivers without flush support don't have to worry
1582 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1583 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1590 if ((bio->bi_rw & REQ_DISCARD) &&
1591 (!blk_queue_discard(q) ||
1592 ((bio->bi_rw & REQ_SECURE) &&
1593 !blk_queue_secdiscard(q)))) {
1598 if (blk_throtl_bio(q, &bio))
1602 * If bio = NULL, bio has been throttled and will be submitted
1608 trace_block_bio_queue(q, bio);
1610 ret = q->make_request_fn(q, bio);
1616 bio_endio(bio, err);
1620 * We only want one ->make_request_fn to be active at a time,
1621 * else stack usage with stacked devices could be a problem.
1622 * So use current->bio_list to keep a list of requests
1623 * submited by a make_request_fn function.
1624 * current->bio_list is also used as a flag to say if
1625 * generic_make_request is currently active in this task or not.
1626 * If it is NULL, then no make_request is active. If it is non-NULL,
1627 * then a make_request is active, and new requests should be added
1630 void generic_make_request(struct bio *bio)
1632 struct bio_list bio_list_on_stack;
1634 if (current->bio_list) {
1635 /* make_request is active */
1636 bio_list_add(current->bio_list, bio);
1639 /* following loop may be a bit non-obvious, and so deserves some
1641 * Before entering the loop, bio->bi_next is NULL (as all callers
1642 * ensure that) so we have a list with a single bio.
1643 * We pretend that we have just taken it off a longer list, so
1644 * we assign bio_list to a pointer to the bio_list_on_stack,
1645 * thus initialising the bio_list of new bios to be
1646 * added. __generic_make_request may indeed add some more bios
1647 * through a recursive call to generic_make_request. If it
1648 * did, we find a non-NULL value in bio_list and re-enter the loop
1649 * from the top. In this case we really did just take the bio
1650 * of the top of the list (no pretending) and so remove it from
1651 * bio_list, and call into __generic_make_request again.
1653 * The loop was structured like this to make only one call to
1654 * __generic_make_request (which is important as it is large and
1655 * inlined) and to keep the structure simple.
1657 BUG_ON(bio->bi_next);
1658 bio_list_init(&bio_list_on_stack);
1659 current->bio_list = &bio_list_on_stack;
1661 __generic_make_request(bio);
1662 bio = bio_list_pop(current->bio_list);
1664 current->bio_list = NULL; /* deactivate */
1666 EXPORT_SYMBOL(generic_make_request);
1669 * submit_bio - submit a bio to the block device layer for I/O
1670 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1671 * @bio: The &struct bio which describes the I/O
1673 * submit_bio() is very similar in purpose to generic_make_request(), and
1674 * uses that function to do most of the work. Both are fairly rough
1675 * interfaces; @bio must be presetup and ready for I/O.
1678 void submit_bio(int rw, struct bio *bio)
1680 int count = bio_sectors(bio);
1685 * If it's a regular read/write or a barrier with data attached,
1686 * go through the normal accounting stuff before submission.
1688 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1690 count_vm_events(PGPGOUT, count);
1692 task_io_account_read(bio->bi_size);
1693 count_vm_events(PGPGIN, count);
1696 if (unlikely(block_dump)) {
1697 char b[BDEVNAME_SIZE];
1698 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1699 current->comm, task_pid_nr(current),
1700 (rw & WRITE) ? "WRITE" : "READ",
1701 (unsigned long long)bio->bi_sector,
1702 bdevname(bio->bi_bdev, b),
1707 generic_make_request(bio);
1709 EXPORT_SYMBOL(submit_bio);
1712 * blk_rq_check_limits - Helper function to check a request for the queue limit
1714 * @rq: the request being checked
1717 * @rq may have been made based on weaker limitations of upper-level queues
1718 * in request stacking drivers, and it may violate the limitation of @q.
1719 * Since the block layer and the underlying device driver trust @rq
1720 * after it is inserted to @q, it should be checked against @q before
1721 * the insertion using this generic function.
1723 * This function should also be useful for request stacking drivers
1724 * in some cases below, so export this function.
1725 * Request stacking drivers like request-based dm may change the queue
1726 * limits while requests are in the queue (e.g. dm's table swapping).
1727 * Such request stacking drivers should check those requests agaist
1728 * the new queue limits again when they dispatch those requests,
1729 * although such checkings are also done against the old queue limits
1730 * when submitting requests.
1732 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1734 if (rq->cmd_flags & REQ_DISCARD)
1737 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1738 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1739 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1744 * queue's settings related to segment counting like q->bounce_pfn
1745 * may differ from that of other stacking queues.
1746 * Recalculate it to check the request correctly on this queue's
1749 blk_recalc_rq_segments(rq);
1750 if (rq->nr_phys_segments > queue_max_segments(q)) {
1751 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1757 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1760 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1761 * @q: the queue to submit the request
1762 * @rq: the request being queued
1764 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1766 unsigned long flags;
1768 if (blk_rq_check_limits(q, rq))
1772 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1775 spin_lock_irqsave(q->queue_lock, flags);
1778 * Submitting request must be dequeued before calling this function
1779 * because it will be linked to another request_queue
1781 BUG_ON(blk_queued_rq(rq));
1783 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1784 spin_unlock_irqrestore(q->queue_lock, flags);
1788 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1791 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1792 * @rq: request to examine
1795 * A request could be merge of IOs which require different failure
1796 * handling. This function determines the number of bytes which
1797 * can be failed from the beginning of the request without
1798 * crossing into area which need to be retried further.
1801 * The number of bytes to fail.
1804 * queue_lock must be held.
1806 unsigned int blk_rq_err_bytes(const struct request *rq)
1808 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1809 unsigned int bytes = 0;
1812 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1813 return blk_rq_bytes(rq);
1816 * Currently the only 'mixing' which can happen is between
1817 * different fastfail types. We can safely fail portions
1818 * which have all the failfast bits that the first one has -
1819 * the ones which are at least as eager to fail as the first
1822 for (bio = rq->bio; bio; bio = bio->bi_next) {
1823 if ((bio->bi_rw & ff) != ff)
1825 bytes += bio->bi_size;
1828 /* this could lead to infinite loop */
1829 BUG_ON(blk_rq_bytes(rq) && !bytes);
1832 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1834 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1836 if (blk_do_io_stat(req)) {
1837 const int rw = rq_data_dir(req);
1838 struct hd_struct *part;
1841 cpu = part_stat_lock();
1843 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1848 static void blk_account_io_done(struct request *req)
1851 * Account IO completion. flush_rq isn't accounted as a
1852 * normal IO on queueing nor completion. Accounting the
1853 * containing request is enough.
1855 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1856 unsigned long duration = jiffies - req->start_time;
1857 const int rw = rq_data_dir(req);
1858 struct hd_struct *part;
1861 cpu = part_stat_lock();
1864 part_stat_inc(cpu, part, ios[rw]);
1865 part_stat_add(cpu, part, ticks[rw], duration);
1866 part_round_stats(cpu, part);
1867 part_dec_in_flight(part, rw);
1869 hd_struct_put(part);
1875 * blk_peek_request - peek at the top of a request queue
1876 * @q: request queue to peek at
1879 * Return the request at the top of @q. The returned request
1880 * should be started using blk_start_request() before LLD starts
1884 * Pointer to the request at the top of @q if available. Null
1888 * queue_lock must be held.
1890 struct request *blk_peek_request(struct request_queue *q)
1895 while ((rq = __elv_next_request(q)) != NULL) {
1896 if (!(rq->cmd_flags & REQ_STARTED)) {
1898 * This is the first time the device driver
1899 * sees this request (possibly after
1900 * requeueing). Notify IO scheduler.
1902 if (rq->cmd_flags & REQ_SORTED)
1903 elv_activate_rq(q, rq);
1906 * just mark as started even if we don't start
1907 * it, a request that has been delayed should
1908 * not be passed by new incoming requests
1910 rq->cmd_flags |= REQ_STARTED;
1911 trace_block_rq_issue(q, rq);
1914 if (!q->boundary_rq || q->boundary_rq == rq) {
1915 q->end_sector = rq_end_sector(rq);
1916 q->boundary_rq = NULL;
1919 if (rq->cmd_flags & REQ_DONTPREP)
1922 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1924 * make sure space for the drain appears we
1925 * know we can do this because max_hw_segments
1926 * has been adjusted to be one fewer than the
1929 rq->nr_phys_segments++;
1935 ret = q->prep_rq_fn(q, rq);
1936 if (ret == BLKPREP_OK) {
1938 } else if (ret == BLKPREP_DEFER) {
1940 * the request may have been (partially) prepped.
1941 * we need to keep this request in the front to
1942 * avoid resource deadlock. REQ_STARTED will
1943 * prevent other fs requests from passing this one.
1945 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1946 !(rq->cmd_flags & REQ_DONTPREP)) {
1948 * remove the space for the drain we added
1949 * so that we don't add it again
1951 --rq->nr_phys_segments;
1956 } else if (ret == BLKPREP_KILL) {
1957 rq->cmd_flags |= REQ_QUIET;
1959 * Mark this request as started so we don't trigger
1960 * any debug logic in the end I/O path.
1962 blk_start_request(rq);
1963 __blk_end_request_all(rq, -EIO);
1965 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1972 EXPORT_SYMBOL(blk_peek_request);
1974 void blk_dequeue_request(struct request *rq)
1976 struct request_queue *q = rq->q;
1978 BUG_ON(list_empty(&rq->queuelist));
1979 BUG_ON(ELV_ON_HASH(rq));
1981 list_del_init(&rq->queuelist);
1984 * the time frame between a request being removed from the lists
1985 * and to it is freed is accounted as io that is in progress at
1988 if (blk_account_rq(rq)) {
1989 q->in_flight[rq_is_sync(rq)]++;
1990 set_io_start_time_ns(rq);
1995 * blk_start_request - start request processing on the driver
1996 * @req: request to dequeue
1999 * Dequeue @req and start timeout timer on it. This hands off the
2000 * request to the driver.
2002 * Block internal functions which don't want to start timer should
2003 * call blk_dequeue_request().
2006 * queue_lock must be held.
2008 void blk_start_request(struct request *req)
2010 blk_dequeue_request(req);
2013 * We are now handing the request to the hardware, initialize
2014 * resid_len to full count and add the timeout handler.
2016 req->resid_len = blk_rq_bytes(req);
2017 if (unlikely(blk_bidi_rq(req)))
2018 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2022 EXPORT_SYMBOL(blk_start_request);
2025 * blk_fetch_request - fetch a request from a request queue
2026 * @q: request queue to fetch a request from
2029 * Return the request at the top of @q. The request is started on
2030 * return and LLD can start processing it immediately.
2033 * Pointer to the request at the top of @q if available. Null
2037 * queue_lock must be held.
2039 struct request *blk_fetch_request(struct request_queue *q)
2043 rq = blk_peek_request(q);
2045 blk_start_request(rq);
2048 EXPORT_SYMBOL(blk_fetch_request);
2051 * blk_update_request - Special helper function for request stacking drivers
2052 * @req: the request being processed
2053 * @error: %0 for success, < %0 for error
2054 * @nr_bytes: number of bytes to complete @req
2057 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2058 * the request structure even if @req doesn't have leftover.
2059 * If @req has leftover, sets it up for the next range of segments.
2061 * This special helper function is only for request stacking drivers
2062 * (e.g. request-based dm) so that they can handle partial completion.
2063 * Actual device drivers should use blk_end_request instead.
2065 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2066 * %false return from this function.
2069 * %false - this request doesn't have any more data
2070 * %true - this request has more data
2072 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2074 int total_bytes, bio_nbytes, next_idx = 0;
2080 trace_block_rq_complete(req->q, req);
2083 * For fs requests, rq is just carrier of independent bio's
2084 * and each partial completion should be handled separately.
2085 * Reset per-request error on each partial completion.
2087 * TODO: tj: This is too subtle. It would be better to let
2088 * low level drivers do what they see fit.
2090 if (req->cmd_type == REQ_TYPE_FS)
2093 if (error && req->cmd_type == REQ_TYPE_FS &&
2094 !(req->cmd_flags & REQ_QUIET)) {
2099 error_type = "recoverable transport";
2102 error_type = "critical target";
2105 error_type = "critical nexus";
2112 printk(KERN_DEBUG "end_request: %s error, dev %s, sector %llu\n",
2113 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2114 (unsigned long long)blk_rq_pos(req));
2117 blk_account_io_completion(req, nr_bytes);
2119 total_bytes = bio_nbytes = 0;
2120 while ((bio = req->bio) != NULL) {
2123 if (nr_bytes >= bio->bi_size) {
2124 req->bio = bio->bi_next;
2125 nbytes = bio->bi_size;
2126 req_bio_endio(req, bio, nbytes, error);
2130 int idx = bio->bi_idx + next_idx;
2132 if (unlikely(idx >= bio->bi_vcnt)) {
2133 blk_dump_rq_flags(req, "__end_that");
2134 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2135 __func__, idx, bio->bi_vcnt);
2139 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2140 BIO_BUG_ON(nbytes > bio->bi_size);
2143 * not a complete bvec done
2145 if (unlikely(nbytes > nr_bytes)) {
2146 bio_nbytes += nr_bytes;
2147 total_bytes += nr_bytes;
2152 * advance to the next vector
2155 bio_nbytes += nbytes;
2158 total_bytes += nbytes;
2164 * end more in this run, or just return 'not-done'
2166 if (unlikely(nr_bytes <= 0))
2176 * Reset counters so that the request stacking driver
2177 * can find how many bytes remain in the request
2180 req->__data_len = 0;
2185 * if the request wasn't completed, update state
2188 req_bio_endio(req, bio, bio_nbytes, error);
2189 bio->bi_idx += next_idx;
2190 bio_iovec(bio)->bv_offset += nr_bytes;
2191 bio_iovec(bio)->bv_len -= nr_bytes;
2194 req->__data_len -= total_bytes;
2195 req->buffer = bio_data(req->bio);
2197 /* update sector only for requests with clear definition of sector */
2198 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2199 req->__sector += total_bytes >> 9;
2201 /* mixed attributes always follow the first bio */
2202 if (req->cmd_flags & REQ_MIXED_MERGE) {
2203 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2204 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2208 * If total number of sectors is less than the first segment
2209 * size, something has gone terribly wrong.
2211 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2212 blk_dump_rq_flags(req, "request botched");
2213 req->__data_len = blk_rq_cur_bytes(req);
2216 /* recalculate the number of segments */
2217 blk_recalc_rq_segments(req);
2221 EXPORT_SYMBOL_GPL(blk_update_request);
2223 static bool blk_update_bidi_request(struct request *rq, int error,
2224 unsigned int nr_bytes,
2225 unsigned int bidi_bytes)
2227 if (blk_update_request(rq, error, nr_bytes))
2230 /* Bidi request must be completed as a whole */
2231 if (unlikely(blk_bidi_rq(rq)) &&
2232 blk_update_request(rq->next_rq, error, bidi_bytes))
2235 if (blk_queue_add_random(rq->q))
2236 add_disk_randomness(rq->rq_disk);
2242 * blk_unprep_request - unprepare a request
2245 * This function makes a request ready for complete resubmission (or
2246 * completion). It happens only after all error handling is complete,
2247 * so represents the appropriate moment to deallocate any resources
2248 * that were allocated to the request in the prep_rq_fn. The queue
2249 * lock is held when calling this.
2251 void blk_unprep_request(struct request *req)
2253 struct request_queue *q = req->q;
2255 req->cmd_flags &= ~REQ_DONTPREP;
2256 if (q->unprep_rq_fn)
2257 q->unprep_rq_fn(q, req);
2259 EXPORT_SYMBOL_GPL(blk_unprep_request);
2262 * queue lock must be held
2264 static void blk_finish_request(struct request *req, int error)
2266 if (blk_rq_tagged(req))
2267 blk_queue_end_tag(req->q, req);
2269 BUG_ON(blk_queued_rq(req));
2271 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2272 laptop_io_completion(&req->q->backing_dev_info);
2274 blk_delete_timer(req);
2276 if (req->cmd_flags & REQ_DONTPREP)
2277 blk_unprep_request(req);
2280 blk_account_io_done(req);
2283 req->end_io(req, error);
2285 if (blk_bidi_rq(req))
2286 __blk_put_request(req->next_rq->q, req->next_rq);
2288 __blk_put_request(req->q, req);
2293 * blk_end_bidi_request - Complete a bidi request
2294 * @rq: the request to complete
2295 * @error: %0 for success, < %0 for error
2296 * @nr_bytes: number of bytes to complete @rq
2297 * @bidi_bytes: number of bytes to complete @rq->next_rq
2300 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2301 * Drivers that supports bidi can safely call this member for any
2302 * type of request, bidi or uni. In the later case @bidi_bytes is
2306 * %false - we are done with this request
2307 * %true - still buffers pending for this request
2309 static bool blk_end_bidi_request(struct request *rq, int error,
2310 unsigned int nr_bytes, unsigned int bidi_bytes)
2312 struct request_queue *q = rq->q;
2313 unsigned long flags;
2315 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2318 spin_lock_irqsave(q->queue_lock, flags);
2319 blk_finish_request(rq, error);
2320 spin_unlock_irqrestore(q->queue_lock, flags);
2326 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2327 * @rq: the request to complete
2328 * @error: %0 for success, < %0 for error
2329 * @nr_bytes: number of bytes to complete @rq
2330 * @bidi_bytes: number of bytes to complete @rq->next_rq
2333 * Identical to blk_end_bidi_request() except that queue lock is
2334 * assumed to be locked on entry and remains so on return.
2337 * %false - we are done with this request
2338 * %true - still buffers pending for this request
2340 static bool __blk_end_bidi_request(struct request *rq, int error,
2341 unsigned int nr_bytes, unsigned int bidi_bytes)
2343 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2346 blk_finish_request(rq, error);
2352 * blk_end_request - Helper function for drivers to complete the request.
2353 * @rq: the request being processed
2354 * @error: %0 for success, < %0 for error
2355 * @nr_bytes: number of bytes to complete
2358 * Ends I/O on a number of bytes attached to @rq.
2359 * If @rq has leftover, sets it up for the next range of segments.
2362 * %false - we are done with this request
2363 * %true - still buffers pending for this request
2365 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2367 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2369 EXPORT_SYMBOL(blk_end_request);
2372 * blk_end_request_all - Helper function for drives to finish the request.
2373 * @rq: the request to finish
2374 * @error: %0 for success, < %0 for error
2377 * Completely finish @rq.
2379 void blk_end_request_all(struct request *rq, int error)
2382 unsigned int bidi_bytes = 0;
2384 if (unlikely(blk_bidi_rq(rq)))
2385 bidi_bytes = blk_rq_bytes(rq->next_rq);
2387 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2390 EXPORT_SYMBOL(blk_end_request_all);
2393 * blk_end_request_cur - Helper function to finish the current request chunk.
2394 * @rq: the request to finish the current chunk for
2395 * @error: %0 for success, < %0 for error
2398 * Complete the current consecutively mapped chunk from @rq.
2401 * %false - we are done with this request
2402 * %true - still buffers pending for this request
2404 bool blk_end_request_cur(struct request *rq, int error)
2406 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2408 EXPORT_SYMBOL(blk_end_request_cur);
2411 * blk_end_request_err - Finish a request till the next failure boundary.
2412 * @rq: the request to finish till the next failure boundary for
2413 * @error: must be negative errno
2416 * Complete @rq till the next failure boundary.
2419 * %false - we are done with this request
2420 * %true - still buffers pending for this request
2422 bool blk_end_request_err(struct request *rq, int error)
2424 WARN_ON(error >= 0);
2425 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2427 EXPORT_SYMBOL_GPL(blk_end_request_err);
2430 * __blk_end_request - Helper function for drivers to complete the request.
2431 * @rq: the request being processed
2432 * @error: %0 for success, < %0 for error
2433 * @nr_bytes: number of bytes to complete
2436 * Must be called with queue lock held unlike blk_end_request().
2439 * %false - we are done with this request
2440 * %true - still buffers pending for this request
2442 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2444 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2446 EXPORT_SYMBOL(__blk_end_request);
2449 * __blk_end_request_all - Helper function for drives to finish the request.
2450 * @rq: the request to finish
2451 * @error: %0 for success, < %0 for error
2454 * Completely finish @rq. Must be called with queue lock held.
2456 void __blk_end_request_all(struct request *rq, int error)
2459 unsigned int bidi_bytes = 0;
2461 if (unlikely(blk_bidi_rq(rq)))
2462 bidi_bytes = blk_rq_bytes(rq->next_rq);
2464 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2467 EXPORT_SYMBOL(__blk_end_request_all);
2470 * __blk_end_request_cur - Helper function to finish the current request chunk.
2471 * @rq: the request to finish the current chunk for
2472 * @error: %0 for success, < %0 for error
2475 * Complete the current consecutively mapped chunk from @rq. Must
2476 * be called with queue lock held.
2479 * %false - we are done with this request
2480 * %true - still buffers pending for this request
2482 bool __blk_end_request_cur(struct request *rq, int error)
2484 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2486 EXPORT_SYMBOL(__blk_end_request_cur);
2489 * __blk_end_request_err - Finish a request till the next failure boundary.
2490 * @rq: the request to finish till the next failure boundary for
2491 * @error: must be negative errno
2494 * Complete @rq till the next failure boundary. Must be called
2495 * with queue lock held.
2498 * %false - we are done with this request
2499 * %true - still buffers pending for this request
2501 bool __blk_end_request_err(struct request *rq, int error)
2503 WARN_ON(error >= 0);
2504 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2506 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2508 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2511 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2512 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2514 if (bio_has_data(bio)) {
2515 rq->nr_phys_segments = bio_phys_segments(q, bio);
2516 rq->buffer = bio_data(bio);
2518 rq->__data_len = bio->bi_size;
2519 rq->bio = rq->biotail = bio;
2522 rq->rq_disk = bio->bi_bdev->bd_disk;
2525 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2527 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2528 * @rq: the request to be flushed
2531 * Flush all pages in @rq.
2533 void rq_flush_dcache_pages(struct request *rq)
2535 struct req_iterator iter;
2536 struct bio_vec *bvec;
2538 rq_for_each_segment(bvec, rq, iter)
2539 flush_dcache_page(bvec->bv_page);
2541 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2545 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2546 * @q : the queue of the device being checked
2549 * Check if underlying low-level drivers of a device are busy.
2550 * If the drivers want to export their busy state, they must set own
2551 * exporting function using blk_queue_lld_busy() first.
2553 * Basically, this function is used only by request stacking drivers
2554 * to stop dispatching requests to underlying devices when underlying
2555 * devices are busy. This behavior helps more I/O merging on the queue
2556 * of the request stacking driver and prevents I/O throughput regression
2557 * on burst I/O load.
2560 * 0 - Not busy (The request stacking driver should dispatch request)
2561 * 1 - Busy (The request stacking driver should stop dispatching request)
2563 int blk_lld_busy(struct request_queue *q)
2566 return q->lld_busy_fn(q);
2570 EXPORT_SYMBOL_GPL(blk_lld_busy);
2573 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2574 * @rq: the clone request to be cleaned up
2577 * Free all bios in @rq for a cloned request.
2579 void blk_rq_unprep_clone(struct request *rq)
2583 while ((bio = rq->bio) != NULL) {
2584 rq->bio = bio->bi_next;
2589 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2592 * Copy attributes of the original request to the clone request.
2593 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2595 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2597 dst->cpu = src->cpu;
2598 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2599 dst->cmd_type = src->cmd_type;
2600 dst->__sector = blk_rq_pos(src);
2601 dst->__data_len = blk_rq_bytes(src);
2602 dst->nr_phys_segments = src->nr_phys_segments;
2603 dst->ioprio = src->ioprio;
2604 dst->extra_len = src->extra_len;
2608 * blk_rq_prep_clone - Helper function to setup clone request
2609 * @rq: the request to be setup
2610 * @rq_src: original request to be cloned
2611 * @bs: bio_set that bios for clone are allocated from
2612 * @gfp_mask: memory allocation mask for bio
2613 * @bio_ctr: setup function to be called for each clone bio.
2614 * Returns %0 for success, non %0 for failure.
2615 * @data: private data to be passed to @bio_ctr
2618 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2619 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2620 * are not copied, and copying such parts is the caller's responsibility.
2621 * Also, pages which the original bios are pointing to are not copied
2622 * and the cloned bios just point same pages.
2623 * So cloned bios must be completed before original bios, which means
2624 * the caller must complete @rq before @rq_src.
2626 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2627 struct bio_set *bs, gfp_t gfp_mask,
2628 int (*bio_ctr)(struct bio *, struct bio *, void *),
2631 struct bio *bio, *bio_src;
2636 blk_rq_init(NULL, rq);
2638 __rq_for_each_bio(bio_src, rq_src) {
2639 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2643 __bio_clone(bio, bio_src);
2645 if (bio_integrity(bio_src) &&
2646 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2649 if (bio_ctr && bio_ctr(bio, bio_src, data))
2653 rq->biotail->bi_next = bio;
2656 rq->bio = rq->biotail = bio;
2659 __blk_rq_prep_clone(rq, rq_src);
2666 blk_rq_unprep_clone(rq);
2670 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2672 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2674 return queue_work(kblockd_workqueue, work);
2676 EXPORT_SYMBOL(kblockd_schedule_work);
2678 int kblockd_schedule_delayed_work(struct request_queue *q,
2679 struct delayed_work *dwork, unsigned long delay)
2681 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2683 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2685 #define PLUG_MAGIC 0x91827364
2687 void blk_start_plug(struct blk_plug *plug)
2689 struct task_struct *tsk = current;
2691 plug->magic = PLUG_MAGIC;
2692 INIT_LIST_HEAD(&plug->list);
2693 INIT_LIST_HEAD(&plug->cb_list);
2694 plug->should_sort = 0;
2697 * If this is a nested plug, don't actually assign it. It will be
2698 * flushed on its own.
2702 * Store ordering should not be needed here, since a potential
2703 * preempt will imply a full memory barrier
2708 EXPORT_SYMBOL(blk_start_plug);
2710 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2712 struct request *rqa = container_of(a, struct request, queuelist);
2713 struct request *rqb = container_of(b, struct request, queuelist);
2715 return !(rqa->q <= rqb->q);
2719 * If 'from_schedule' is true, then postpone the dispatch of requests
2720 * until a safe kblockd context. We due this to avoid accidental big
2721 * additional stack usage in driver dispatch, in places where the originally
2722 * plugger did not intend it.
2724 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2726 __releases(q->queue_lock)
2728 trace_block_unplug(q, depth, !from_schedule);
2731 * If we are punting this to kblockd, then we can safely drop
2732 * the queue_lock before waking kblockd (which needs to take
2735 if (from_schedule) {
2736 spin_unlock(q->queue_lock);
2737 blk_run_queue_async(q);
2740 spin_unlock(q->queue_lock);
2745 static void flush_plug_callbacks(struct blk_plug *plug)
2747 LIST_HEAD(callbacks);
2749 if (list_empty(&plug->cb_list))
2752 list_splice_init(&plug->cb_list, &callbacks);
2754 while (!list_empty(&callbacks)) {
2755 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2758 list_del(&cb->list);
2763 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2765 struct request_queue *q;
2766 unsigned long flags;
2771 BUG_ON(plug->magic != PLUG_MAGIC);
2773 flush_plug_callbacks(plug);
2774 if (list_empty(&plug->list))
2777 list_splice_init(&plug->list, &list);
2779 if (plug->should_sort) {
2780 list_sort(NULL, &list, plug_rq_cmp);
2781 plug->should_sort = 0;
2788 * Save and disable interrupts here, to avoid doing it for every
2789 * queue lock we have to take.
2791 local_irq_save(flags);
2792 while (!list_empty(&list)) {
2793 rq = list_entry_rq(list.next);
2794 list_del_init(&rq->queuelist);
2798 * This drops the queue lock
2801 queue_unplugged(q, depth, from_schedule);
2804 spin_lock(q->queue_lock);
2807 * rq is already accounted, so use raw insert
2809 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2810 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2812 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2818 * This drops the queue lock
2821 queue_unplugged(q, depth, from_schedule);
2823 local_irq_restore(flags);
2826 void blk_finish_plug(struct blk_plug *plug)
2828 blk_flush_plug_list(plug, false);
2830 if (plug == current->plug)
2831 current->plug = NULL;
2833 EXPORT_SYMBOL(blk_finish_plug);
2835 int __init blk_dev_init(void)
2837 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2838 sizeof(((struct request *)0)->cmd_flags));
2840 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2841 kblockd_workqueue = alloc_workqueue("kblockd",
2842 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2843 if (!kblockd_workqueue)
2844 panic("Failed to create kblockd\n");
2846 request_cachep = kmem_cache_create("blkdev_requests",
2847 sizeof(struct request), 0, SLAB_PANIC, NULL);
2849 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2850 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);