2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/block.h>
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
43 * For the allocated request tables
45 static struct kmem_cache *request_cachep;
48 * For queue allocation
50 struct kmem_cache *blk_requestq_cachep;
53 * Controlling structure to kblockd
55 static struct workqueue_struct *kblockd_workqueue;
57 static void drive_stat_acct(struct request *rq, int new_io)
59 struct hd_struct *part;
60 int rw = rq_data_dir(rq);
63 if (!blk_do_io_stat(rq))
66 cpu = part_stat_lock();
70 part_stat_inc(cpu, part, merges[rw]);
72 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
73 if (!hd_struct_try_get(part)) {
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
82 part = &rq->rq_disk->part0;
85 part_round_stats(cpu, part);
86 part_inc_in_flight(part, rw);
93 void blk_queue_congestion_threshold(struct request_queue *q)
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
100 q->nr_congestion_on = nr;
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 q->nr_congestion_off = nr;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
115 * Will return NULL if the request queue cannot be located.
117 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119 struct backing_dev_info *ret = NULL;
120 struct request_queue *q = bdev_get_queue(bdev);
123 ret = &q->backing_dev_info;
126 EXPORT_SYMBOL(blk_get_backing_dev_info);
128 void blk_rq_init(struct request_queue *q, struct request *rq)
130 memset(rq, 0, sizeof(*rq));
132 INIT_LIST_HEAD(&rq->queuelist);
133 INIT_LIST_HEAD(&rq->timeout_list);
136 rq->__sector = (sector_t) -1;
137 INIT_HLIST_NODE(&rq->hash);
138 RB_CLEAR_NODE(&rq->rb_node);
140 rq->cmd_len = BLK_MAX_CDB;
143 rq->start_time = jiffies;
144 set_start_time_ns(rq);
147 EXPORT_SYMBOL(blk_rq_init);
149 static void req_bio_endio(struct request *rq, struct bio *bio,
150 unsigned int nbytes, int error)
153 clear_bit(BIO_UPTODATE, &bio->bi_flags);
154 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
157 if (unlikely(nbytes > bio->bi_size)) {
158 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
159 __func__, nbytes, bio->bi_size);
160 nbytes = bio->bi_size;
163 if (unlikely(rq->cmd_flags & REQ_QUIET))
164 set_bit(BIO_QUIET, &bio->bi_flags);
166 bio->bi_size -= nbytes;
167 bio->bi_sector += (nbytes >> 9);
169 if (bio_integrity(bio))
170 bio_integrity_advance(bio, nbytes);
172 /* don't actually finish bio if it's part of flush sequence */
173 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
174 bio_endio(bio, error);
177 void blk_dump_rq_flags(struct request *rq, char *msg)
181 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
182 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
185 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
186 (unsigned long long)blk_rq_pos(rq),
187 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
188 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
189 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
191 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
192 printk(KERN_INFO " cdb: ");
193 for (bit = 0; bit < BLK_MAX_CDB; bit++)
194 printk("%02x ", rq->cmd[bit]);
198 EXPORT_SYMBOL(blk_dump_rq_flags);
200 static void blk_delay_work(struct work_struct *work)
202 struct request_queue *q;
204 q = container_of(work, struct request_queue, delay_work.work);
205 spin_lock_irq(q->queue_lock);
207 spin_unlock_irq(q->queue_lock);
211 * blk_delay_queue - restart queueing after defined interval
212 * @q: The &struct request_queue in question
213 * @msecs: Delay in msecs
216 * Sometimes queueing needs to be postponed for a little while, to allow
217 * resources to come back. This function will make sure that queueing is
218 * restarted around the specified time.
220 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
222 queue_delayed_work(kblockd_workqueue, &q->delay_work,
223 msecs_to_jiffies(msecs));
225 EXPORT_SYMBOL(blk_delay_queue);
228 * blk_start_queue - restart a previously stopped queue
229 * @q: The &struct request_queue in question
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 void blk_start_queue(struct request_queue *q)
238 WARN_ON(!irqs_disabled());
240 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
243 EXPORT_SYMBOL(blk_start_queue);
246 * blk_stop_queue - stop a queue
247 * @q: The &struct request_queue in question
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 void blk_stop_queue(struct request_queue *q)
261 __cancel_delayed_work(&q->delay_work);
262 queue_flag_set(QUEUE_FLAG_STOPPED, q);
264 EXPORT_SYMBOL(blk_stop_queue);
267 * blk_sync_queue - cancel any pending callbacks on a queue
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
275 * that the callbacks might use. The caller must already have made sure
276 * that its ->make_request_fn will not re-add plugging prior to calling
279 * This function does not cancel any asynchronous activity arising
280 * out of elevator or throttling code. That would require elevaotor_exit()
281 * and blk_throtl_exit() to be called with queue lock initialized.
284 void blk_sync_queue(struct request_queue *q)
286 del_timer_sync(&q->timeout);
287 cancel_delayed_work_sync(&q->delay_work);
289 EXPORT_SYMBOL(blk_sync_queue);
292 * __blk_run_queue - run a single device queue
293 * @q: The queue to run
296 * See @blk_run_queue. This variant must be called with the queue lock
297 * held and interrupts disabled.
299 void __blk_run_queue(struct request_queue *q)
301 if (unlikely(blk_queue_stopped(q)))
306 EXPORT_SYMBOL(__blk_run_queue);
309 * blk_run_queue_async - run a single device queue in workqueue context
310 * @q: The queue to run
313 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
316 void blk_run_queue_async(struct request_queue *q)
318 if (likely(!blk_queue_stopped(q))) {
319 __cancel_delayed_work(&q->delay_work);
320 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
323 EXPORT_SYMBOL(blk_run_queue_async);
326 * blk_run_queue - run a single device queue
327 * @q: The queue to run
330 * Invoke request handling on this queue, if it has pending work to do.
331 * May be used to restart queueing when a request has completed.
333 void blk_run_queue(struct request_queue *q)
337 spin_lock_irqsave(q->queue_lock, flags);
339 spin_unlock_irqrestore(q->queue_lock, flags);
341 EXPORT_SYMBOL(blk_run_queue);
343 void blk_put_queue(struct request_queue *q)
345 kobject_put(&q->kobj);
347 EXPORT_SYMBOL(blk_put_queue);
350 * blk_drain_queue - drain requests from request_queue
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);
372 nr_rqs = q->rq.count[0] + q->rq.count[1];
374 nr_rqs = q->rq.elvpriv;
376 spin_unlock_irq(q->queue_lock);
385 * blk_cleanup_queue - shutdown a request queue
386 * @q: request queue to shutdown
388 * Mark @q DEAD, drain all pending requests, destroy and put it. All
389 * future requests will be failed immediately with -ENODEV.
391 void blk_cleanup_queue(struct request_queue *q)
393 spinlock_t *lock = q->queue_lock;
395 /* mark @q DEAD, no new request or merges will be allowed afterwards */
396 mutex_lock(&q->sysfs_lock);
397 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
400 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
401 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
402 queue_flag_set(QUEUE_FLAG_DEAD, q);
404 if (q->queue_lock != &q->__queue_lock)
405 q->queue_lock = &q->__queue_lock;
407 spin_unlock_irq(lock);
408 mutex_unlock(&q->sysfs_lock);
411 * Drain all requests queued before DEAD marking. The caller might
412 * be trying to tear down @q before its elevator is initialized, in
413 * which case we don't want to call into draining.
416 blk_drain_queue(q, true);
418 /* @q won't process any more request, flush async actions */
419 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
422 /* @q is and will stay empty, shutdown and put */
425 EXPORT_SYMBOL(blk_cleanup_queue);
427 static int blk_init_free_list(struct request_queue *q)
429 struct request_list *rl = &q->rq;
431 if (unlikely(rl->rq_pool))
434 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
435 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
437 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
438 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
440 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
441 mempool_free_slab, request_cachep, q->node);
449 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
451 return blk_alloc_queue_node(gfp_mask, -1);
453 EXPORT_SYMBOL(blk_alloc_queue);
455 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
457 struct request_queue *q;
460 q = kmem_cache_alloc_node(blk_requestq_cachep,
461 gfp_mask | __GFP_ZERO, node_id);
465 q->backing_dev_info.ra_pages =
466 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
467 q->backing_dev_info.state = 0;
468 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
469 q->backing_dev_info.name = "block";
472 err = bdi_init(&q->backing_dev_info);
474 kmem_cache_free(blk_requestq_cachep, q);
478 if (blk_throtl_init(q)) {
479 kmem_cache_free(blk_requestq_cachep, q);
483 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
484 laptop_mode_timer_fn, (unsigned long) q);
485 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
486 INIT_LIST_HEAD(&q->timeout_list);
487 INIT_LIST_HEAD(&q->flush_queue[0]);
488 INIT_LIST_HEAD(&q->flush_queue[1]);
489 INIT_LIST_HEAD(&q->flush_data_in_flight);
490 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
492 kobject_init(&q->kobj, &blk_queue_ktype);
494 mutex_init(&q->sysfs_lock);
495 spin_lock_init(&q->__queue_lock);
498 * By default initialize queue_lock to internal lock and driver can
499 * override it later if need be.
501 q->queue_lock = &q->__queue_lock;
505 EXPORT_SYMBOL(blk_alloc_queue_node);
508 * blk_init_queue - prepare a request queue for use with a block device
509 * @rfn: The function to be called to process requests that have been
510 * placed on the queue.
511 * @lock: Request queue spin lock
514 * If a block device wishes to use the standard request handling procedures,
515 * which sorts requests and coalesces adjacent requests, then it must
516 * call blk_init_queue(). The function @rfn will be called when there
517 * are requests on the queue that need to be processed. If the device
518 * supports plugging, then @rfn may not be called immediately when requests
519 * are available on the queue, but may be called at some time later instead.
520 * Plugged queues are generally unplugged when a buffer belonging to one
521 * of the requests on the queue is needed, or due to memory pressure.
523 * @rfn is not required, or even expected, to remove all requests off the
524 * queue, but only as many as it can handle at a time. If it does leave
525 * requests on the queue, it is responsible for arranging that the requests
526 * get dealt with eventually.
528 * The queue spin lock must be held while manipulating the requests on the
529 * request queue; this lock will be taken also from interrupt context, so irq
530 * disabling is needed for it.
532 * Function returns a pointer to the initialized request queue, or %NULL if
536 * blk_init_queue() must be paired with a blk_cleanup_queue() call
537 * when the block device is deactivated (such as at module unload).
540 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
542 return blk_init_queue_node(rfn, lock, -1);
544 EXPORT_SYMBOL(blk_init_queue);
546 struct request_queue *
547 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
549 struct request_queue *uninit_q, *q;
551 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
555 q = blk_init_allocated_queue(uninit_q, rfn, lock);
557 blk_cleanup_queue(uninit_q);
561 EXPORT_SYMBOL(blk_init_queue_node);
563 struct request_queue *
564 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
570 if (blk_init_free_list(q))
574 q->prep_rq_fn = NULL;
575 q->unprep_rq_fn = NULL;
576 q->queue_flags = QUEUE_FLAG_DEFAULT;
578 /* Override internal queue lock with supplied lock pointer */
580 q->queue_lock = lock;
583 * This also sets hw/phys segments, boundary and size
585 blk_queue_make_request(q, blk_queue_bio);
587 q->sg_reserved_size = INT_MAX;
592 if (!elevator_init(q, NULL)) {
593 blk_queue_congestion_threshold(q);
599 EXPORT_SYMBOL(blk_init_allocated_queue);
601 int blk_get_queue(struct request_queue *q)
603 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
604 kobject_get(&q->kobj);
610 EXPORT_SYMBOL(blk_get_queue);
612 static inline void blk_free_request(struct request_queue *q, struct request *rq)
614 if (rq->cmd_flags & REQ_ELVPRIV)
615 elv_put_request(q, rq);
616 mempool_free(rq, q->rq.rq_pool);
619 static struct request *
620 blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
622 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
629 rq->cmd_flags = flags | REQ_ALLOCED;
631 if ((flags & REQ_ELVPRIV) &&
632 unlikely(elv_set_request(q, rq, gfp_mask))) {
633 mempool_free(rq, q->rq.rq_pool);
641 * ioc_batching returns true if the ioc is a valid batching request and
642 * should be given priority access to a request.
644 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
650 * Make sure the process is able to allocate at least 1 request
651 * even if the batch times out, otherwise we could theoretically
654 return ioc->nr_batch_requests == q->nr_batching ||
655 (ioc->nr_batch_requests > 0
656 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
660 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
661 * will cause the process to be a "batcher" on all queues in the system. This
662 * is the behaviour we want though - once it gets a wakeup it should be given
665 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
667 if (!ioc || ioc_batching(q, ioc))
670 ioc->nr_batch_requests = q->nr_batching;
671 ioc->last_waited = jiffies;
674 static void __freed_request(struct request_queue *q, int sync)
676 struct request_list *rl = &q->rq;
678 if (rl->count[sync] < queue_congestion_off_threshold(q))
679 blk_clear_queue_congested(q, sync);
681 if (rl->count[sync] + 1 <= q->nr_requests) {
682 if (waitqueue_active(&rl->wait[sync]))
683 wake_up(&rl->wait[sync]);
685 blk_clear_queue_full(q, sync);
690 * A request has just been released. Account for it, update the full and
691 * congestion status, wake up any waiters. Called under q->queue_lock.
693 static void freed_request(struct request_queue *q, unsigned int flags)
695 struct request_list *rl = &q->rq;
696 int sync = rw_is_sync(flags);
699 if (flags & REQ_ELVPRIV)
702 __freed_request(q, sync);
704 if (unlikely(rl->starved[sync ^ 1]))
705 __freed_request(q, sync ^ 1);
709 * Determine if elevator data should be initialized when allocating the
710 * request associated with @bio.
712 static bool blk_rq_should_init_elevator(struct bio *bio)
718 * Flush requests do not use the elevator so skip initialization.
719 * This allows a request to share the flush and elevator data.
721 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
728 * get_request - get a free request
729 * @q: request_queue to allocate request from
730 * @rw_flags: RW and SYNC flags
731 * @bio: bio to allocate request for (can be %NULL)
732 * @gfp_mask: allocation mask
734 * Get a free request from @q. This function may fail under memory
735 * pressure or if @q is dead.
737 * Must be callled with @q->queue_lock held and,
738 * Returns %NULL on failure, with @q->queue_lock held.
739 * Returns !%NULL on success, with @q->queue_lock *not held*.
741 static struct request *get_request(struct request_queue *q, int rw_flags,
742 struct bio *bio, gfp_t gfp_mask)
744 struct request *rq = NULL;
745 struct request_list *rl = &q->rq;
746 struct io_context *ioc = NULL;
747 const bool is_sync = rw_is_sync(rw_flags) != 0;
750 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
753 may_queue = elv_may_queue(q, rw_flags);
754 if (may_queue == ELV_MQUEUE_NO)
757 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
758 if (rl->count[is_sync]+1 >= q->nr_requests) {
759 ioc = current_io_context(GFP_ATOMIC, q->node);
761 * The queue will fill after this allocation, so set
762 * it as full, and mark this process as "batching".
763 * This process will be allowed to complete a batch of
764 * requests, others will be blocked.
766 if (!blk_queue_full(q, is_sync)) {
767 ioc_set_batching(q, ioc);
768 blk_set_queue_full(q, is_sync);
770 if (may_queue != ELV_MQUEUE_MUST
771 && !ioc_batching(q, ioc)) {
773 * The queue is full and the allocating
774 * process is not a "batcher", and not
775 * exempted by the IO scheduler
781 blk_set_queue_congested(q, is_sync);
785 * Only allow batching queuers to allocate up to 50% over the defined
786 * limit of requests, otherwise we could have thousands of requests
787 * allocated with any setting of ->nr_requests
789 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
792 rl->count[is_sync]++;
793 rl->starved[is_sync] = 0;
795 if (blk_rq_should_init_elevator(bio) &&
796 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
797 rw_flags |= REQ_ELVPRIV;
801 if (blk_queue_io_stat(q))
802 rw_flags |= REQ_IO_STAT;
803 spin_unlock_irq(q->queue_lock);
805 rq = blk_alloc_request(q, rw_flags, gfp_mask);
808 * Allocation failed presumably due to memory. Undo anything
809 * we might have messed up.
811 * Allocating task should really be put onto the front of the
812 * wait queue, but this is pretty rare.
814 spin_lock_irq(q->queue_lock);
815 freed_request(q, rw_flags);
818 * in the very unlikely event that allocation failed and no
819 * requests for this direction was pending, mark us starved
820 * so that freeing of a request in the other direction will
821 * notice us. another possible fix would be to split the
822 * rq mempool into READ and WRITE
825 if (unlikely(rl->count[is_sync] == 0))
826 rl->starved[is_sync] = 1;
832 * ioc may be NULL here, and ioc_batching will be false. That's
833 * OK, if the queue is under the request limit then requests need
834 * not count toward the nr_batch_requests limit. There will always
835 * be some limit enforced by BLK_BATCH_TIME.
837 if (ioc_batching(q, ioc))
838 ioc->nr_batch_requests--;
840 trace_block_getrq(q, bio, rw_flags & 1);
846 * get_request_wait - get a free request with retry
847 * @q: request_queue to allocate request from
848 * @rw_flags: RW and SYNC flags
849 * @bio: bio to allocate request for (can be %NULL)
851 * Get a free request from @q. This function keeps retrying under memory
852 * pressure and fails iff @q is dead.
854 * Must be callled with @q->queue_lock held and,
855 * Returns %NULL on failure, with @q->queue_lock held.
856 * Returns !%NULL on success, with @q->queue_lock *not held*.
858 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
861 const bool is_sync = rw_is_sync(rw_flags) != 0;
864 rq = get_request(q, rw_flags, bio, GFP_NOIO);
867 struct io_context *ioc;
868 struct request_list *rl = &q->rq;
870 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
873 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
874 TASK_UNINTERRUPTIBLE);
876 trace_block_sleeprq(q, bio, rw_flags & 1);
878 spin_unlock_irq(q->queue_lock);
882 * After sleeping, we become a "batching" process and
883 * will be able to allocate at least one request, and
884 * up to a big batch of them for a small period time.
885 * See ioc_batching, ioc_set_batching
887 ioc = current_io_context(GFP_NOIO, q->node);
888 ioc_set_batching(q, ioc);
890 spin_lock_irq(q->queue_lock);
891 finish_wait(&rl->wait[is_sync], &wait);
893 rq = get_request(q, rw_flags, bio, GFP_NOIO);
899 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
903 BUG_ON(rw != READ && rw != WRITE);
905 spin_lock_irq(q->queue_lock);
906 if (gfp_mask & __GFP_WAIT)
907 rq = get_request_wait(q, rw, NULL);
909 rq = get_request(q, rw, NULL, gfp_mask);
911 spin_unlock_irq(q->queue_lock);
912 /* q->queue_lock is unlocked at this point */
916 EXPORT_SYMBOL(blk_get_request);
919 * blk_make_request - given a bio, allocate a corresponding struct request.
920 * @q: target request queue
921 * @bio: The bio describing the memory mappings that will be submitted for IO.
922 * It may be a chained-bio properly constructed by block/bio layer.
923 * @gfp_mask: gfp flags to be used for memory allocation
925 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
926 * type commands. Where the struct request needs to be farther initialized by
927 * the caller. It is passed a &struct bio, which describes the memory info of
930 * The caller of blk_make_request must make sure that bi_io_vec
931 * are set to describe the memory buffers. That bio_data_dir() will return
932 * the needed direction of the request. (And all bio's in the passed bio-chain
933 * are properly set accordingly)
935 * If called under none-sleepable conditions, mapped bio buffers must not
936 * need bouncing, by calling the appropriate masked or flagged allocator,
937 * suitable for the target device. Otherwise the call to blk_queue_bounce will
940 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
941 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
942 * anything but the first bio in the chain. Otherwise you risk waiting for IO
943 * completion of a bio that hasn't been submitted yet, thus resulting in a
944 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
945 * of bio_alloc(), as that avoids the mempool deadlock.
946 * If possible a big IO should be split into smaller parts when allocation
947 * fails. Partial allocation should not be an error, or you risk a live-lock.
949 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
952 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
955 return ERR_PTR(-ENOMEM);
958 struct bio *bounce_bio = bio;
961 blk_queue_bounce(q, &bounce_bio);
962 ret = blk_rq_append_bio(q, rq, bounce_bio);
971 EXPORT_SYMBOL(blk_make_request);
974 * blk_requeue_request - put a request back on queue
975 * @q: request queue where request should be inserted
976 * @rq: request to be inserted
979 * Drivers often keep queueing requests until the hardware cannot accept
980 * more, when that condition happens we need to put the request back
981 * on the queue. Must be called with queue lock held.
983 void blk_requeue_request(struct request_queue *q, struct request *rq)
985 blk_delete_timer(rq);
986 blk_clear_rq_complete(rq);
987 trace_block_rq_requeue(q, rq);
989 if (blk_rq_tagged(rq))
990 blk_queue_end_tag(q, rq);
992 BUG_ON(blk_queued_rq(rq));
994 elv_requeue_request(q, rq);
996 EXPORT_SYMBOL(blk_requeue_request);
998 static void add_acct_request(struct request_queue *q, struct request *rq,
1001 drive_stat_acct(rq, 1);
1002 __elv_add_request(q, rq, where);
1006 * blk_insert_request - insert a special request into a request queue
1007 * @q: request queue where request should be inserted
1008 * @rq: request to be inserted
1009 * @at_head: insert request at head or tail of queue
1010 * @data: private data
1013 * Many block devices need to execute commands asynchronously, so they don't
1014 * block the whole kernel from preemption during request execution. This is
1015 * accomplished normally by inserting aritficial requests tagged as
1016 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1017 * be scheduled for actual execution by the request queue.
1019 * We have the option of inserting the head or the tail of the queue.
1020 * Typically we use the tail for new ioctls and so forth. We use the head
1021 * of the queue for things like a QUEUE_FULL message from a device, or a
1022 * host that is unable to accept a particular command.
1024 void blk_insert_request(struct request_queue *q, struct request *rq,
1025 int at_head, void *data)
1027 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1028 unsigned long flags;
1031 * tell I/O scheduler that this isn't a regular read/write (ie it
1032 * must not attempt merges on this) and that it acts as a soft
1035 rq->cmd_type = REQ_TYPE_SPECIAL;
1039 spin_lock_irqsave(q->queue_lock, flags);
1042 * If command is tagged, release the tag
1044 if (blk_rq_tagged(rq))
1045 blk_queue_end_tag(q, rq);
1047 add_acct_request(q, rq, where);
1049 spin_unlock_irqrestore(q->queue_lock, flags);
1051 EXPORT_SYMBOL(blk_insert_request);
1053 static void part_round_stats_single(int cpu, struct hd_struct *part,
1056 if (now == part->stamp)
1059 if (part_in_flight(part)) {
1060 __part_stat_add(cpu, part, time_in_queue,
1061 part_in_flight(part) * (now - part->stamp));
1062 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1068 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1069 * @cpu: cpu number for stats access
1070 * @part: target partition
1072 * The average IO queue length and utilisation statistics are maintained
1073 * by observing the current state of the queue length and the amount of
1074 * time it has been in this state for.
1076 * Normally, that accounting is done on IO completion, but that can result
1077 * in more than a second's worth of IO being accounted for within any one
1078 * second, leading to >100% utilisation. To deal with that, we call this
1079 * function to do a round-off before returning the results when reading
1080 * /proc/diskstats. This accounts immediately for all queue usage up to
1081 * the current jiffies and restarts the counters again.
1083 void part_round_stats(int cpu, struct hd_struct *part)
1085 unsigned long now = jiffies;
1088 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1089 part_round_stats_single(cpu, part, now);
1091 EXPORT_SYMBOL_GPL(part_round_stats);
1094 * queue lock must be held
1096 void __blk_put_request(struct request_queue *q, struct request *req)
1100 if (unlikely(--req->ref_count))
1103 elv_completed_request(q, req);
1105 /* this is a bio leak */
1106 WARN_ON(req->bio != NULL);
1109 * Request may not have originated from ll_rw_blk. if not,
1110 * it didn't come out of our reserved rq pools
1112 if (req->cmd_flags & REQ_ALLOCED) {
1113 unsigned int flags = req->cmd_flags;
1115 BUG_ON(!list_empty(&req->queuelist));
1116 BUG_ON(!hlist_unhashed(&req->hash));
1118 blk_free_request(q, req);
1119 freed_request(q, flags);
1122 EXPORT_SYMBOL_GPL(__blk_put_request);
1124 void blk_put_request(struct request *req)
1126 unsigned long flags;
1127 struct request_queue *q = req->q;
1129 spin_lock_irqsave(q->queue_lock, flags);
1130 __blk_put_request(q, req);
1131 spin_unlock_irqrestore(q->queue_lock, flags);
1133 EXPORT_SYMBOL(blk_put_request);
1136 * blk_add_request_payload - add a payload to a request
1137 * @rq: request to update
1138 * @page: page backing the payload
1139 * @len: length of the payload.
1141 * This allows to later add a payload to an already submitted request by
1142 * a block driver. The driver needs to take care of freeing the payload
1145 * Note that this is a quite horrible hack and nothing but handling of
1146 * discard requests should ever use it.
1148 void blk_add_request_payload(struct request *rq, struct page *page,
1151 struct bio *bio = rq->bio;
1153 bio->bi_io_vec->bv_page = page;
1154 bio->bi_io_vec->bv_offset = 0;
1155 bio->bi_io_vec->bv_len = len;
1159 bio->bi_phys_segments = 1;
1161 rq->__data_len = rq->resid_len = len;
1162 rq->nr_phys_segments = 1;
1163 rq->buffer = bio_data(bio);
1165 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1167 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1170 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1172 if (!ll_back_merge_fn(q, req, bio))
1175 trace_block_bio_backmerge(q, bio);
1177 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1178 blk_rq_set_mixed_merge(req);
1180 req->biotail->bi_next = bio;
1182 req->__data_len += bio->bi_size;
1183 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1185 drive_stat_acct(req, 0);
1186 elv_bio_merged(q, req, bio);
1190 static bool bio_attempt_front_merge(struct request_queue *q,
1191 struct request *req, struct bio *bio)
1193 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1195 if (!ll_front_merge_fn(q, req, bio))
1198 trace_block_bio_frontmerge(q, bio);
1200 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1201 blk_rq_set_mixed_merge(req);
1203 bio->bi_next = req->bio;
1207 * may not be valid. if the low level driver said
1208 * it didn't need a bounce buffer then it better
1209 * not touch req->buffer either...
1211 req->buffer = bio_data(bio);
1212 req->__sector = bio->bi_sector;
1213 req->__data_len += bio->bi_size;
1214 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1216 drive_stat_acct(req, 0);
1217 elv_bio_merged(q, req, bio);
1222 * attempt_plug_merge - try to merge with %current's plugged list
1223 * @q: request_queue new bio is being queued at
1224 * @bio: new bio being queued
1225 * @request_count: out parameter for number of traversed plugged requests
1227 * Determine whether @bio being queued on @q can be merged with a request
1228 * on %current's plugged list. Returns %true if merge was successful,
1231 * This function is called without @q->queue_lock; however, elevator is
1232 * accessed iff there already are requests on the plugged list which in
1233 * turn guarantees validity of the elevator.
1235 * Note that, on successful merge, elevator operation
1236 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1237 * must be ready for this.
1239 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1240 unsigned int *request_count)
1242 struct blk_plug *plug;
1246 plug = current->plug;
1251 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1259 el_ret = elv_try_merge(rq, bio);
1260 if (el_ret == ELEVATOR_BACK_MERGE) {
1261 ret = bio_attempt_back_merge(q, rq, bio);
1264 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1265 ret = bio_attempt_front_merge(q, rq, bio);
1274 void init_request_from_bio(struct request *req, struct bio *bio)
1276 req->cmd_type = REQ_TYPE_FS;
1278 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1279 if (bio->bi_rw & REQ_RAHEAD)
1280 req->cmd_flags |= REQ_FAILFAST_MASK;
1283 req->__sector = bio->bi_sector;
1284 req->ioprio = bio_prio(bio);
1285 blk_rq_bio_prep(req->q, req, bio);
1288 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1290 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1291 struct blk_plug *plug;
1292 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1293 struct request *req;
1294 unsigned int request_count = 0;
1297 * low level driver can indicate that it wants pages above a
1298 * certain limit bounced to low memory (ie for highmem, or even
1299 * ISA dma in theory)
1301 blk_queue_bounce(q, &bio);
1303 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1304 spin_lock_irq(q->queue_lock);
1305 where = ELEVATOR_INSERT_FLUSH;
1310 * Check if we can merge with the plugged list before grabbing
1313 if (attempt_plug_merge(q, bio, &request_count))
1316 spin_lock_irq(q->queue_lock);
1318 el_ret = elv_merge(q, &req, bio);
1319 if (el_ret == ELEVATOR_BACK_MERGE) {
1320 if (bio_attempt_back_merge(q, req, bio)) {
1321 if (!attempt_back_merge(q, req))
1322 elv_merged_request(q, req, el_ret);
1325 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1326 if (bio_attempt_front_merge(q, req, bio)) {
1327 if (!attempt_front_merge(q, req))
1328 elv_merged_request(q, req, el_ret);
1335 * This sync check and mask will be re-done in init_request_from_bio(),
1336 * but we need to set it earlier to expose the sync flag to the
1337 * rq allocator and io schedulers.
1339 rw_flags = bio_data_dir(bio);
1341 rw_flags |= REQ_SYNC;
1344 * Grab a free request. This is might sleep but can not fail.
1345 * Returns with the queue unlocked.
1347 req = get_request_wait(q, rw_flags, bio);
1348 if (unlikely(!req)) {
1349 bio_endio(bio, -ENODEV); /* @q is dead */
1354 * After dropping the lock and possibly sleeping here, our request
1355 * may now be mergeable after it had proven unmergeable (above).
1356 * We don't worry about that case for efficiency. It won't happen
1357 * often, and the elevators are able to handle it.
1359 init_request_from_bio(req, bio);
1361 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1362 req->cpu = raw_smp_processor_id();
1364 plug = current->plug;
1367 * If this is the first request added after a plug, fire
1368 * of a plug trace. If others have been added before, check
1369 * if we have multiple devices in this plug. If so, make a
1370 * note to sort the list before dispatch.
1372 if (list_empty(&plug->list))
1373 trace_block_plug(q);
1375 if (!plug->should_sort) {
1376 struct request *__rq;
1378 __rq = list_entry_rq(plug->list.prev);
1380 plug->should_sort = 1;
1382 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1383 blk_flush_plug_list(plug, false);
1384 trace_block_plug(q);
1387 list_add_tail(&req->queuelist, &plug->list);
1388 drive_stat_acct(req, 1);
1390 spin_lock_irq(q->queue_lock);
1391 add_acct_request(q, req, where);
1394 spin_unlock_irq(q->queue_lock);
1397 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1400 * If bio->bi_dev is a partition, remap the location
1402 static inline void blk_partition_remap(struct bio *bio)
1404 struct block_device *bdev = bio->bi_bdev;
1406 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1407 struct hd_struct *p = bdev->bd_part;
1409 bio->bi_sector += p->start_sect;
1410 bio->bi_bdev = bdev->bd_contains;
1412 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1414 bio->bi_sector - p->start_sect);
1418 static void handle_bad_sector(struct bio *bio)
1420 char b[BDEVNAME_SIZE];
1422 printk(KERN_INFO "attempt to access beyond end of device\n");
1423 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1424 bdevname(bio->bi_bdev, b),
1426 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1427 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1429 set_bit(BIO_EOF, &bio->bi_flags);
1432 #ifdef CONFIG_FAIL_MAKE_REQUEST
1434 static DECLARE_FAULT_ATTR(fail_make_request);
1436 static int __init setup_fail_make_request(char *str)
1438 return setup_fault_attr(&fail_make_request, str);
1440 __setup("fail_make_request=", setup_fail_make_request);
1442 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1444 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1447 static int __init fail_make_request_debugfs(void)
1449 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1450 NULL, &fail_make_request);
1452 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1455 late_initcall(fail_make_request_debugfs);
1457 #else /* CONFIG_FAIL_MAKE_REQUEST */
1459 static inline bool should_fail_request(struct hd_struct *part,
1465 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1468 * Check whether this bio extends beyond the end of the device.
1470 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1477 /* Test device or partition size, when known. */
1478 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1480 sector_t sector = bio->bi_sector;
1482 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1484 * This may well happen - the kernel calls bread()
1485 * without checking the size of the device, e.g., when
1486 * mounting a device.
1488 handle_bad_sector(bio);
1496 static noinline_for_stack bool
1497 generic_make_request_checks(struct bio *bio)
1499 struct request_queue *q;
1500 int nr_sectors = bio_sectors(bio);
1502 char b[BDEVNAME_SIZE];
1503 struct hd_struct *part;
1507 if (bio_check_eod(bio, nr_sectors))
1510 q = bdev_get_queue(bio->bi_bdev);
1513 "generic_make_request: Trying to access "
1514 "nonexistent block-device %s (%Lu)\n",
1515 bdevname(bio->bi_bdev, b),
1516 (long long) bio->bi_sector);
1520 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1521 nr_sectors > queue_max_hw_sectors(q))) {
1522 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1523 bdevname(bio->bi_bdev, b),
1525 queue_max_hw_sectors(q));
1529 part = bio->bi_bdev->bd_part;
1530 if (should_fail_request(part, bio->bi_size) ||
1531 should_fail_request(&part_to_disk(part)->part0,
1536 * If this device has partitions, remap block n
1537 * of partition p to block n+start(p) of the disk.
1539 blk_partition_remap(bio);
1541 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1544 if (bio_check_eod(bio, nr_sectors))
1548 * Filter flush bio's early so that make_request based
1549 * drivers without flush support don't have to worry
1552 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1553 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1560 if ((bio->bi_rw & REQ_DISCARD) &&
1561 (!blk_queue_discard(q) ||
1562 ((bio->bi_rw & REQ_SECURE) &&
1563 !blk_queue_secdiscard(q)))) {
1568 if (blk_throtl_bio(q, bio))
1569 return false; /* throttled, will be resubmitted later */
1571 trace_block_bio_queue(q, bio);
1575 bio_endio(bio, err);
1580 * generic_make_request - hand a buffer to its device driver for I/O
1581 * @bio: The bio describing the location in memory and on the device.
1583 * generic_make_request() is used to make I/O requests of block
1584 * devices. It is passed a &struct bio, which describes the I/O that needs
1587 * generic_make_request() does not return any status. The
1588 * success/failure status of the request, along with notification of
1589 * completion, is delivered asynchronously through the bio->bi_end_io
1590 * function described (one day) else where.
1592 * The caller of generic_make_request must make sure that bi_io_vec
1593 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1594 * set to describe the device address, and the
1595 * bi_end_io and optionally bi_private are set to describe how
1596 * completion notification should be signaled.
1598 * generic_make_request and the drivers it calls may use bi_next if this
1599 * bio happens to be merged with someone else, and may resubmit the bio to
1600 * a lower device by calling into generic_make_request recursively, which
1601 * means the bio should NOT be touched after the call to ->make_request_fn.
1603 void generic_make_request(struct bio *bio)
1605 struct bio_list bio_list_on_stack;
1607 if (!generic_make_request_checks(bio))
1611 * We only want one ->make_request_fn to be active at a time, else
1612 * stack usage with stacked devices could be a problem. So use
1613 * current->bio_list to keep a list of requests submited by a
1614 * make_request_fn function. current->bio_list is also used as a
1615 * flag to say if generic_make_request is currently active in this
1616 * task or not. If it is NULL, then no make_request is active. If
1617 * it is non-NULL, then a make_request is active, and new requests
1618 * should be added at the tail
1620 if (current->bio_list) {
1621 bio_list_add(current->bio_list, bio);
1625 /* following loop may be a bit non-obvious, and so deserves some
1627 * Before entering the loop, bio->bi_next is NULL (as all callers
1628 * ensure that) so we have a list with a single bio.
1629 * We pretend that we have just taken it off a longer list, so
1630 * we assign bio_list to a pointer to the bio_list_on_stack,
1631 * thus initialising the bio_list of new bios to be
1632 * added. ->make_request() may indeed add some more bios
1633 * through a recursive call to generic_make_request. If it
1634 * did, we find a non-NULL value in bio_list and re-enter the loop
1635 * from the top. In this case we really did just take the bio
1636 * of the top of the list (no pretending) and so remove it from
1637 * bio_list, and call into ->make_request() again.
1639 BUG_ON(bio->bi_next);
1640 bio_list_init(&bio_list_on_stack);
1641 current->bio_list = &bio_list_on_stack;
1643 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1645 q->make_request_fn(q, bio);
1647 bio = bio_list_pop(current->bio_list);
1649 current->bio_list = NULL; /* deactivate */
1651 EXPORT_SYMBOL(generic_make_request);
1654 * submit_bio - submit a bio to the block device layer for I/O
1655 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1656 * @bio: The &struct bio which describes the I/O
1658 * submit_bio() is very similar in purpose to generic_make_request(), and
1659 * uses that function to do most of the work. Both are fairly rough
1660 * interfaces; @bio must be presetup and ready for I/O.
1663 void submit_bio(int rw, struct bio *bio)
1665 int count = bio_sectors(bio);
1670 * If it's a regular read/write or a barrier with data attached,
1671 * go through the normal accounting stuff before submission.
1673 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1675 count_vm_events(PGPGOUT, count);
1677 task_io_account_read(bio->bi_size);
1678 count_vm_events(PGPGIN, count);
1681 if (unlikely(block_dump)) {
1682 char b[BDEVNAME_SIZE];
1683 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1684 current->comm, task_pid_nr(current),
1685 (rw & WRITE) ? "WRITE" : "READ",
1686 (unsigned long long)bio->bi_sector,
1687 bdevname(bio->bi_bdev, b),
1692 generic_make_request(bio);
1694 EXPORT_SYMBOL(submit_bio);
1697 * blk_rq_check_limits - Helper function to check a request for the queue limit
1699 * @rq: the request being checked
1702 * @rq may have been made based on weaker limitations of upper-level queues
1703 * in request stacking drivers, and it may violate the limitation of @q.
1704 * Since the block layer and the underlying device driver trust @rq
1705 * after it is inserted to @q, it should be checked against @q before
1706 * the insertion using this generic function.
1708 * This function should also be useful for request stacking drivers
1709 * in some cases below, so export this function.
1710 * Request stacking drivers like request-based dm may change the queue
1711 * limits while requests are in the queue (e.g. dm's table swapping).
1712 * Such request stacking drivers should check those requests agaist
1713 * the new queue limits again when they dispatch those requests,
1714 * although such checkings are also done against the old queue limits
1715 * when submitting requests.
1717 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1719 if (rq->cmd_flags & REQ_DISCARD)
1722 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1723 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1724 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1729 * queue's settings related to segment counting like q->bounce_pfn
1730 * may differ from that of other stacking queues.
1731 * Recalculate it to check the request correctly on this queue's
1734 blk_recalc_rq_segments(rq);
1735 if (rq->nr_phys_segments > queue_max_segments(q)) {
1736 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1742 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1745 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1746 * @q: the queue to submit the request
1747 * @rq: the request being queued
1749 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1751 unsigned long flags;
1752 int where = ELEVATOR_INSERT_BACK;
1754 if (blk_rq_check_limits(q, rq))
1758 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1761 spin_lock_irqsave(q->queue_lock, flags);
1764 * Submitting request must be dequeued before calling this function
1765 * because it will be linked to another request_queue
1767 BUG_ON(blk_queued_rq(rq));
1769 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1770 where = ELEVATOR_INSERT_FLUSH;
1772 add_acct_request(q, rq, where);
1773 if (where == ELEVATOR_INSERT_FLUSH)
1775 spin_unlock_irqrestore(q->queue_lock, flags);
1779 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1782 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1783 * @rq: request to examine
1786 * A request could be merge of IOs which require different failure
1787 * handling. This function determines the number of bytes which
1788 * can be failed from the beginning of the request without
1789 * crossing into area which need to be retried further.
1792 * The number of bytes to fail.
1795 * queue_lock must be held.
1797 unsigned int blk_rq_err_bytes(const struct request *rq)
1799 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1800 unsigned int bytes = 0;
1803 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1804 return blk_rq_bytes(rq);
1807 * Currently the only 'mixing' which can happen is between
1808 * different fastfail types. We can safely fail portions
1809 * which have all the failfast bits that the first one has -
1810 * the ones which are at least as eager to fail as the first
1813 for (bio = rq->bio; bio; bio = bio->bi_next) {
1814 if ((bio->bi_rw & ff) != ff)
1816 bytes += bio->bi_size;
1819 /* this could lead to infinite loop */
1820 BUG_ON(blk_rq_bytes(rq) && !bytes);
1823 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1825 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1827 if (blk_do_io_stat(req)) {
1828 const int rw = rq_data_dir(req);
1829 struct hd_struct *part;
1832 cpu = part_stat_lock();
1834 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1839 static void blk_account_io_done(struct request *req)
1842 * Account IO completion. flush_rq isn't accounted as a
1843 * normal IO on queueing nor completion. Accounting the
1844 * containing request is enough.
1846 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1847 unsigned long duration = jiffies - req->start_time;
1848 const int rw = rq_data_dir(req);
1849 struct hd_struct *part;
1852 cpu = part_stat_lock();
1855 part_stat_inc(cpu, part, ios[rw]);
1856 part_stat_add(cpu, part, ticks[rw], duration);
1857 part_round_stats(cpu, part);
1858 part_dec_in_flight(part, rw);
1860 hd_struct_put(part);
1866 * blk_peek_request - peek at the top of a request queue
1867 * @q: request queue to peek at
1870 * Return the request at the top of @q. The returned request
1871 * should be started using blk_start_request() before LLD starts
1875 * Pointer to the request at the top of @q if available. Null
1879 * queue_lock must be held.
1881 struct request *blk_peek_request(struct request_queue *q)
1886 while ((rq = __elv_next_request(q)) != NULL) {
1887 if (!(rq->cmd_flags & REQ_STARTED)) {
1889 * This is the first time the device driver
1890 * sees this request (possibly after
1891 * requeueing). Notify IO scheduler.
1893 if (rq->cmd_flags & REQ_SORTED)
1894 elv_activate_rq(q, rq);
1897 * just mark as started even if we don't start
1898 * it, a request that has been delayed should
1899 * not be passed by new incoming requests
1901 rq->cmd_flags |= REQ_STARTED;
1902 trace_block_rq_issue(q, rq);
1905 if (!q->boundary_rq || q->boundary_rq == rq) {
1906 q->end_sector = rq_end_sector(rq);
1907 q->boundary_rq = NULL;
1910 if (rq->cmd_flags & REQ_DONTPREP)
1913 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1915 * make sure space for the drain appears we
1916 * know we can do this because max_hw_segments
1917 * has been adjusted to be one fewer than the
1920 rq->nr_phys_segments++;
1926 ret = q->prep_rq_fn(q, rq);
1927 if (ret == BLKPREP_OK) {
1929 } else if (ret == BLKPREP_DEFER) {
1931 * the request may have been (partially) prepped.
1932 * we need to keep this request in the front to
1933 * avoid resource deadlock. REQ_STARTED will
1934 * prevent other fs requests from passing this one.
1936 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1937 !(rq->cmd_flags & REQ_DONTPREP)) {
1939 * remove the space for the drain we added
1940 * so that we don't add it again
1942 --rq->nr_phys_segments;
1947 } else if (ret == BLKPREP_KILL) {
1948 rq->cmd_flags |= REQ_QUIET;
1950 * Mark this request as started so we don't trigger
1951 * any debug logic in the end I/O path.
1953 blk_start_request(rq);
1954 __blk_end_request_all(rq, -EIO);
1956 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1963 EXPORT_SYMBOL(blk_peek_request);
1965 void blk_dequeue_request(struct request *rq)
1967 struct request_queue *q = rq->q;
1969 BUG_ON(list_empty(&rq->queuelist));
1970 BUG_ON(ELV_ON_HASH(rq));
1972 list_del_init(&rq->queuelist);
1975 * the time frame between a request being removed from the lists
1976 * and to it is freed is accounted as io that is in progress at
1979 if (blk_account_rq(rq)) {
1980 q->in_flight[rq_is_sync(rq)]++;
1981 set_io_start_time_ns(rq);
1986 * blk_start_request - start request processing on the driver
1987 * @req: request to dequeue
1990 * Dequeue @req and start timeout timer on it. This hands off the
1991 * request to the driver.
1993 * Block internal functions which don't want to start timer should
1994 * call blk_dequeue_request().
1997 * queue_lock must be held.
1999 void blk_start_request(struct request *req)
2001 blk_dequeue_request(req);
2004 * We are now handing the request to the hardware, initialize
2005 * resid_len to full count and add the timeout handler.
2007 req->resid_len = blk_rq_bytes(req);
2008 if (unlikely(blk_bidi_rq(req)))
2009 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2013 EXPORT_SYMBOL(blk_start_request);
2016 * blk_fetch_request - fetch a request from a request queue
2017 * @q: request queue to fetch a request from
2020 * Return the request at the top of @q. The request is started on
2021 * return and LLD can start processing it immediately.
2024 * Pointer to the request at the top of @q if available. Null
2028 * queue_lock must be held.
2030 struct request *blk_fetch_request(struct request_queue *q)
2034 rq = blk_peek_request(q);
2036 blk_start_request(rq);
2039 EXPORT_SYMBOL(blk_fetch_request);
2042 * blk_update_request - Special helper function for request stacking drivers
2043 * @req: the request being processed
2044 * @error: %0 for success, < %0 for error
2045 * @nr_bytes: number of bytes to complete @req
2048 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2049 * the request structure even if @req doesn't have leftover.
2050 * If @req has leftover, sets it up for the next range of segments.
2052 * This special helper function is only for request stacking drivers
2053 * (e.g. request-based dm) so that they can handle partial completion.
2054 * Actual device drivers should use blk_end_request instead.
2056 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2057 * %false return from this function.
2060 * %false - this request doesn't have any more data
2061 * %true - this request has more data
2063 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2065 int total_bytes, bio_nbytes, next_idx = 0;
2071 trace_block_rq_complete(req->q, req);
2074 * For fs requests, rq is just carrier of independent bio's
2075 * and each partial completion should be handled separately.
2076 * Reset per-request error on each partial completion.
2078 * TODO: tj: This is too subtle. It would be better to let
2079 * low level drivers do what they see fit.
2081 if (req->cmd_type == REQ_TYPE_FS)
2084 if (error && req->cmd_type == REQ_TYPE_FS &&
2085 !(req->cmd_flags & REQ_QUIET)) {
2090 error_type = "recoverable transport";
2093 error_type = "critical target";
2096 error_type = "critical nexus";
2103 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2104 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2105 (unsigned long long)blk_rq_pos(req));
2108 blk_account_io_completion(req, nr_bytes);
2110 total_bytes = bio_nbytes = 0;
2111 while ((bio = req->bio) != NULL) {
2114 if (nr_bytes >= bio->bi_size) {
2115 req->bio = bio->bi_next;
2116 nbytes = bio->bi_size;
2117 req_bio_endio(req, bio, nbytes, error);
2121 int idx = bio->bi_idx + next_idx;
2123 if (unlikely(idx >= bio->bi_vcnt)) {
2124 blk_dump_rq_flags(req, "__end_that");
2125 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2126 __func__, idx, bio->bi_vcnt);
2130 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2131 BIO_BUG_ON(nbytes > bio->bi_size);
2134 * not a complete bvec done
2136 if (unlikely(nbytes > nr_bytes)) {
2137 bio_nbytes += nr_bytes;
2138 total_bytes += nr_bytes;
2143 * advance to the next vector
2146 bio_nbytes += nbytes;
2149 total_bytes += nbytes;
2155 * end more in this run, or just return 'not-done'
2157 if (unlikely(nr_bytes <= 0))
2167 * Reset counters so that the request stacking driver
2168 * can find how many bytes remain in the request
2171 req->__data_len = 0;
2176 * if the request wasn't completed, update state
2179 req_bio_endio(req, bio, bio_nbytes, error);
2180 bio->bi_idx += next_idx;
2181 bio_iovec(bio)->bv_offset += nr_bytes;
2182 bio_iovec(bio)->bv_len -= nr_bytes;
2185 req->__data_len -= total_bytes;
2186 req->buffer = bio_data(req->bio);
2188 /* update sector only for requests with clear definition of sector */
2189 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2190 req->__sector += total_bytes >> 9;
2192 /* mixed attributes always follow the first bio */
2193 if (req->cmd_flags & REQ_MIXED_MERGE) {
2194 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2195 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2199 * If total number of sectors is less than the first segment
2200 * size, something has gone terribly wrong.
2202 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2203 blk_dump_rq_flags(req, "request botched");
2204 req->__data_len = blk_rq_cur_bytes(req);
2207 /* recalculate the number of segments */
2208 blk_recalc_rq_segments(req);
2212 EXPORT_SYMBOL_GPL(blk_update_request);
2214 static bool blk_update_bidi_request(struct request *rq, int error,
2215 unsigned int nr_bytes,
2216 unsigned int bidi_bytes)
2218 if (blk_update_request(rq, error, nr_bytes))
2221 /* Bidi request must be completed as a whole */
2222 if (unlikely(blk_bidi_rq(rq)) &&
2223 blk_update_request(rq->next_rq, error, bidi_bytes))
2226 if (blk_queue_add_random(rq->q))
2227 add_disk_randomness(rq->rq_disk);
2233 * blk_unprep_request - unprepare a request
2236 * This function makes a request ready for complete resubmission (or
2237 * completion). It happens only after all error handling is complete,
2238 * so represents the appropriate moment to deallocate any resources
2239 * that were allocated to the request in the prep_rq_fn. The queue
2240 * lock is held when calling this.
2242 void blk_unprep_request(struct request *req)
2244 struct request_queue *q = req->q;
2246 req->cmd_flags &= ~REQ_DONTPREP;
2247 if (q->unprep_rq_fn)
2248 q->unprep_rq_fn(q, req);
2250 EXPORT_SYMBOL_GPL(blk_unprep_request);
2253 * queue lock must be held
2255 static void blk_finish_request(struct request *req, int error)
2257 if (blk_rq_tagged(req))
2258 blk_queue_end_tag(req->q, req);
2260 BUG_ON(blk_queued_rq(req));
2262 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2263 laptop_io_completion(&req->q->backing_dev_info);
2265 blk_delete_timer(req);
2267 if (req->cmd_flags & REQ_DONTPREP)
2268 blk_unprep_request(req);
2271 blk_account_io_done(req);
2274 req->end_io(req, error);
2276 if (blk_bidi_rq(req))
2277 __blk_put_request(req->next_rq->q, req->next_rq);
2279 __blk_put_request(req->q, req);
2284 * blk_end_bidi_request - Complete a bidi request
2285 * @rq: the request to complete
2286 * @error: %0 for success, < %0 for error
2287 * @nr_bytes: number of bytes to complete @rq
2288 * @bidi_bytes: number of bytes to complete @rq->next_rq
2291 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2292 * Drivers that supports bidi can safely call this member for any
2293 * type of request, bidi or uni. In the later case @bidi_bytes is
2297 * %false - we are done with this request
2298 * %true - still buffers pending for this request
2300 static bool blk_end_bidi_request(struct request *rq, int error,
2301 unsigned int nr_bytes, unsigned int bidi_bytes)
2303 struct request_queue *q = rq->q;
2304 unsigned long flags;
2306 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2309 spin_lock_irqsave(q->queue_lock, flags);
2310 blk_finish_request(rq, error);
2311 spin_unlock_irqrestore(q->queue_lock, flags);
2317 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2318 * @rq: the request to complete
2319 * @error: %0 for success, < %0 for error
2320 * @nr_bytes: number of bytes to complete @rq
2321 * @bidi_bytes: number of bytes to complete @rq->next_rq
2324 * Identical to blk_end_bidi_request() except that queue lock is
2325 * assumed to be locked on entry and remains so on return.
2328 * %false - we are done with this request
2329 * %true - still buffers pending for this request
2331 bool __blk_end_bidi_request(struct request *rq, int error,
2332 unsigned int nr_bytes, unsigned int bidi_bytes)
2334 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2337 blk_finish_request(rq, error);
2343 * blk_end_request - Helper function for drivers to complete the request.
2344 * @rq: the request being processed
2345 * @error: %0 for success, < %0 for error
2346 * @nr_bytes: number of bytes to complete
2349 * Ends I/O on a number of bytes attached to @rq.
2350 * If @rq has leftover, sets it up for the next range of segments.
2353 * %false - we are done with this request
2354 * %true - still buffers pending for this request
2356 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2358 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2360 EXPORT_SYMBOL(blk_end_request);
2363 * blk_end_request_all - Helper function for drives to finish the request.
2364 * @rq: the request to finish
2365 * @error: %0 for success, < %0 for error
2368 * Completely finish @rq.
2370 void blk_end_request_all(struct request *rq, int error)
2373 unsigned int bidi_bytes = 0;
2375 if (unlikely(blk_bidi_rq(rq)))
2376 bidi_bytes = blk_rq_bytes(rq->next_rq);
2378 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2381 EXPORT_SYMBOL(blk_end_request_all);
2384 * blk_end_request_cur - Helper function to finish the current request chunk.
2385 * @rq: the request to finish the current chunk for
2386 * @error: %0 for success, < %0 for error
2389 * Complete the current consecutively mapped chunk from @rq.
2392 * %false - we are done with this request
2393 * %true - still buffers pending for this request
2395 bool blk_end_request_cur(struct request *rq, int error)
2397 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2399 EXPORT_SYMBOL(blk_end_request_cur);
2402 * blk_end_request_err - Finish a request till the next failure boundary.
2403 * @rq: the request to finish till the next failure boundary for
2404 * @error: must be negative errno
2407 * Complete @rq till the next failure boundary.
2410 * %false - we are done with this request
2411 * %true - still buffers pending for this request
2413 bool blk_end_request_err(struct request *rq, int error)
2415 WARN_ON(error >= 0);
2416 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2418 EXPORT_SYMBOL_GPL(blk_end_request_err);
2421 * __blk_end_request - Helper function for drivers to complete the request.
2422 * @rq: the request being processed
2423 * @error: %0 for success, < %0 for error
2424 * @nr_bytes: number of bytes to complete
2427 * Must be called with queue lock held unlike blk_end_request().
2430 * %false - we are done with this request
2431 * %true - still buffers pending for this request
2433 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2435 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2437 EXPORT_SYMBOL(__blk_end_request);
2440 * __blk_end_request_all - Helper function for drives to finish the request.
2441 * @rq: the request to finish
2442 * @error: %0 for success, < %0 for error
2445 * Completely finish @rq. Must be called with queue lock held.
2447 void __blk_end_request_all(struct request *rq, int error)
2450 unsigned int bidi_bytes = 0;
2452 if (unlikely(blk_bidi_rq(rq)))
2453 bidi_bytes = blk_rq_bytes(rq->next_rq);
2455 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2458 EXPORT_SYMBOL(__blk_end_request_all);
2461 * __blk_end_request_cur - Helper function to finish the current request chunk.
2462 * @rq: the request to finish the current chunk for
2463 * @error: %0 for success, < %0 for error
2466 * Complete the current consecutively mapped chunk from @rq. Must
2467 * be called with queue lock held.
2470 * %false - we are done with this request
2471 * %true - still buffers pending for this request
2473 bool __blk_end_request_cur(struct request *rq, int error)
2475 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2477 EXPORT_SYMBOL(__blk_end_request_cur);
2480 * __blk_end_request_err - Finish a request till the next failure boundary.
2481 * @rq: the request to finish till the next failure boundary for
2482 * @error: must be negative errno
2485 * Complete @rq till the next failure boundary. Must be called
2486 * with queue lock held.
2489 * %false - we are done with this request
2490 * %true - still buffers pending for this request
2492 bool __blk_end_request_err(struct request *rq, int error)
2494 WARN_ON(error >= 0);
2495 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2497 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2499 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2502 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2503 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2505 if (bio_has_data(bio)) {
2506 rq->nr_phys_segments = bio_phys_segments(q, bio);
2507 rq->buffer = bio_data(bio);
2509 rq->__data_len = bio->bi_size;
2510 rq->bio = rq->biotail = bio;
2513 rq->rq_disk = bio->bi_bdev->bd_disk;
2516 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2518 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2519 * @rq: the request to be flushed
2522 * Flush all pages in @rq.
2524 void rq_flush_dcache_pages(struct request *rq)
2526 struct req_iterator iter;
2527 struct bio_vec *bvec;
2529 rq_for_each_segment(bvec, rq, iter)
2530 flush_dcache_page(bvec->bv_page);
2532 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2536 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2537 * @q : the queue of the device being checked
2540 * Check if underlying low-level drivers of a device are busy.
2541 * If the drivers want to export their busy state, they must set own
2542 * exporting function using blk_queue_lld_busy() first.
2544 * Basically, this function is used only by request stacking drivers
2545 * to stop dispatching requests to underlying devices when underlying
2546 * devices are busy. This behavior helps more I/O merging on the queue
2547 * of the request stacking driver and prevents I/O throughput regression
2548 * on burst I/O load.
2551 * 0 - Not busy (The request stacking driver should dispatch request)
2552 * 1 - Busy (The request stacking driver should stop dispatching request)
2554 int blk_lld_busy(struct request_queue *q)
2557 return q->lld_busy_fn(q);
2561 EXPORT_SYMBOL_GPL(blk_lld_busy);
2564 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2565 * @rq: the clone request to be cleaned up
2568 * Free all bios in @rq for a cloned request.
2570 void blk_rq_unprep_clone(struct request *rq)
2574 while ((bio = rq->bio) != NULL) {
2575 rq->bio = bio->bi_next;
2580 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2583 * Copy attributes of the original request to the clone request.
2584 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2586 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2588 dst->cpu = src->cpu;
2589 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2590 dst->cmd_type = src->cmd_type;
2591 dst->__sector = blk_rq_pos(src);
2592 dst->__data_len = blk_rq_bytes(src);
2593 dst->nr_phys_segments = src->nr_phys_segments;
2594 dst->ioprio = src->ioprio;
2595 dst->extra_len = src->extra_len;
2599 * blk_rq_prep_clone - Helper function to setup clone request
2600 * @rq: the request to be setup
2601 * @rq_src: original request to be cloned
2602 * @bs: bio_set that bios for clone are allocated from
2603 * @gfp_mask: memory allocation mask for bio
2604 * @bio_ctr: setup function to be called for each clone bio.
2605 * Returns %0 for success, non %0 for failure.
2606 * @data: private data to be passed to @bio_ctr
2609 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2610 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2611 * are not copied, and copying such parts is the caller's responsibility.
2612 * Also, pages which the original bios are pointing to are not copied
2613 * and the cloned bios just point same pages.
2614 * So cloned bios must be completed before original bios, which means
2615 * the caller must complete @rq before @rq_src.
2617 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2618 struct bio_set *bs, gfp_t gfp_mask,
2619 int (*bio_ctr)(struct bio *, struct bio *, void *),
2622 struct bio *bio, *bio_src;
2627 blk_rq_init(NULL, rq);
2629 __rq_for_each_bio(bio_src, rq_src) {
2630 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2634 __bio_clone(bio, bio_src);
2636 if (bio_integrity(bio_src) &&
2637 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2640 if (bio_ctr && bio_ctr(bio, bio_src, data))
2644 rq->biotail->bi_next = bio;
2647 rq->bio = rq->biotail = bio;
2650 __blk_rq_prep_clone(rq, rq_src);
2657 blk_rq_unprep_clone(rq);
2661 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2663 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2665 return queue_work(kblockd_workqueue, work);
2667 EXPORT_SYMBOL(kblockd_schedule_work);
2669 int kblockd_schedule_delayed_work(struct request_queue *q,
2670 struct delayed_work *dwork, unsigned long delay)
2672 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2674 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2676 #define PLUG_MAGIC 0x91827364
2679 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2680 * @plug: The &struct blk_plug that needs to be initialized
2683 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2684 * pending I/O should the task end up blocking between blk_start_plug() and
2685 * blk_finish_plug(). This is important from a performance perspective, but
2686 * also ensures that we don't deadlock. For instance, if the task is blocking
2687 * for a memory allocation, memory reclaim could end up wanting to free a
2688 * page belonging to that request that is currently residing in our private
2689 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2690 * this kind of deadlock.
2692 void blk_start_plug(struct blk_plug *plug)
2694 struct task_struct *tsk = current;
2696 plug->magic = PLUG_MAGIC;
2697 INIT_LIST_HEAD(&plug->list);
2698 INIT_LIST_HEAD(&plug->cb_list);
2699 plug->should_sort = 0;
2702 * If this is a nested plug, don't actually assign it. It will be
2703 * flushed on its own.
2707 * Store ordering should not be needed here, since a potential
2708 * preempt will imply a full memory barrier
2713 EXPORT_SYMBOL(blk_start_plug);
2715 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2717 struct request *rqa = container_of(a, struct request, queuelist);
2718 struct request *rqb = container_of(b, struct request, queuelist);
2720 return !(rqa->q <= rqb->q);
2724 * If 'from_schedule' is true, then postpone the dispatch of requests
2725 * until a safe kblockd context. We due this to avoid accidental big
2726 * additional stack usage in driver dispatch, in places where the originally
2727 * plugger did not intend it.
2729 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2731 __releases(q->queue_lock)
2733 trace_block_unplug(q, depth, !from_schedule);
2736 * If we are punting this to kblockd, then we can safely drop
2737 * the queue_lock before waking kblockd (which needs to take
2740 if (from_schedule) {
2741 spin_unlock(q->queue_lock);
2742 blk_run_queue_async(q);
2745 spin_unlock(q->queue_lock);
2750 static void flush_plug_callbacks(struct blk_plug *plug)
2752 LIST_HEAD(callbacks);
2754 if (list_empty(&plug->cb_list))
2757 list_splice_init(&plug->cb_list, &callbacks);
2759 while (!list_empty(&callbacks)) {
2760 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2763 list_del(&cb->list);
2768 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2770 struct request_queue *q;
2771 unsigned long flags;
2776 BUG_ON(plug->magic != PLUG_MAGIC);
2778 flush_plug_callbacks(plug);
2779 if (list_empty(&plug->list))
2782 list_splice_init(&plug->list, &list);
2784 if (plug->should_sort) {
2785 list_sort(NULL, &list, plug_rq_cmp);
2786 plug->should_sort = 0;
2793 * Save and disable interrupts here, to avoid doing it for every
2794 * queue lock we have to take.
2796 local_irq_save(flags);
2797 while (!list_empty(&list)) {
2798 rq = list_entry_rq(list.next);
2799 list_del_init(&rq->queuelist);
2803 * This drops the queue lock
2806 queue_unplugged(q, depth, from_schedule);
2809 spin_lock(q->queue_lock);
2812 * rq is already accounted, so use raw insert
2814 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2815 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2817 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2823 * This drops the queue lock
2826 queue_unplugged(q, depth, from_schedule);
2828 local_irq_restore(flags);
2831 void blk_finish_plug(struct blk_plug *plug)
2833 blk_flush_plug_list(plug, false);
2835 if (plug == current->plug)
2836 current->plug = NULL;
2838 EXPORT_SYMBOL(blk_finish_plug);
2840 int __init blk_dev_init(void)
2842 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2843 sizeof(((struct request *)0)->cmd_flags));
2845 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2846 kblockd_workqueue = alloc_workqueue("kblockd",
2847 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2848 if (!kblockd_workqueue)
2849 panic("Failed to create kblockd\n");
2851 request_cachep = kmem_cache_create("blkdev_requests",
2852 sizeof(struct request), 0, SLAB_PANIC, NULL);
2854 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2855 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);