2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
41 static int __make_request(struct request_queue *q, struct bio *bio);
44 * For the allocated request tables
46 static struct kmem_cache *request_cachep;
49 * For queue allocation
51 struct kmem_cache *blk_requestq_cachep;
54 * Controlling structure to kblockd
56 static struct workqueue_struct *kblockd_workqueue;
58 static void drive_stat_acct(struct request *rq, int new_io)
60 struct hd_struct *part;
61 int rw = rq_data_dir(rq);
64 if (!blk_do_io_stat(rq))
67 cpu = part_stat_lock();
71 part_stat_inc(cpu, part, merges[rw]);
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
74 if (!hd_struct_try_get(part)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part = &rq->rq_disk->part0;
86 part_round_stats(cpu, part);
87 part_inc_in_flight(part, rw);
94 void blk_queue_congestion_threshold(struct request_queue *q)
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
101 q->nr_congestion_on = nr;
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
106 q->nr_congestion_off = nr;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120 struct backing_dev_info *ret = NULL;
121 struct request_queue *q = bdev_get_queue(bdev);
124 ret = &q->backing_dev_info;
127 EXPORT_SYMBOL(blk_get_backing_dev_info);
129 void blk_rq_init(struct request_queue *q, struct request *rq)
131 memset(rq, 0, sizeof(*rq));
133 INIT_LIST_HEAD(&rq->queuelist);
134 INIT_LIST_HEAD(&rq->timeout_list);
137 rq->__sector = (sector_t) -1;
138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
141 rq->cmd_len = BLK_MAX_CDB;
144 rq->start_time = jiffies;
145 set_start_time_ns(rq);
148 EXPORT_SYMBOL(blk_rq_init);
150 static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
178 void blk_dump_rq_flags(struct request *rq, char *msg)
182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
193 printk(KERN_INFO " cdb: ");
194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
195 printk("%02x ", rq->cmd[bit]);
199 EXPORT_SYMBOL(blk_dump_rq_flags);
201 static void blk_delay_work(struct work_struct *work)
203 struct request_queue *q;
205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
208 spin_unlock_irq(q->queue_lock);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
226 EXPORT_SYMBOL(blk_delay_queue);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue *q)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
244 EXPORT_SYMBOL(blk_start_queue);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue *q)
262 __cancel_delayed_work(&q->delay_work);
263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
265 EXPORT_SYMBOL(blk_stop_queue);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue *q)
287 del_timer_sync(&q->timeout);
288 cancel_delayed_work_sync(&q->delay_work);
290 EXPORT_SYMBOL(blk_sync_queue);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue *q)
302 if (unlikely(blk_queue_stopped(q)))
307 EXPORT_SYMBOL(__blk_run_queue);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
317 void blk_run_queue_async(struct request_queue *q)
319 if (likely(!blk_queue_stopped(q))) {
320 __cancel_delayed_work(&q->delay_work);
321 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
324 EXPORT_SYMBOL(blk_run_queue_async);
327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
331 * Invoke request handling on this queue, if it has pending work to do.
332 * May be used to restart queueing when a request has completed.
334 void blk_run_queue(struct request_queue *q)
338 spin_lock_irqsave(q->queue_lock, flags);
340 spin_unlock_irqrestore(q->queue_lock, flags);
342 EXPORT_SYMBOL(blk_run_queue);
344 void blk_put_queue(struct request_queue *q)
346 kobject_put(&q->kobj);
348 EXPORT_SYMBOL(blk_put_queue);
351 * Note: If a driver supplied the queue lock, it is disconnected
352 * by this function. The actual state of the lock doesn't matter
353 * here as the request_queue isn't accessible after this point
354 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
356 void blk_cleanup_queue(struct request_queue *q)
359 * We know we have process context here, so we can be a little
360 * cautious and ensure that pending block actions on this device
361 * are done before moving on. Going into this function, we should
362 * not have processes doing IO to this device.
366 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
367 mutex_lock(&q->sysfs_lock);
368 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
369 mutex_unlock(&q->sysfs_lock);
371 if (q->queue_lock != &q->__queue_lock)
372 q->queue_lock = &q->__queue_lock;
376 EXPORT_SYMBOL(blk_cleanup_queue);
378 static int blk_init_free_list(struct request_queue *q)
380 struct request_list *rl = &q->rq;
382 if (unlikely(rl->rq_pool))
385 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
386 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
388 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
389 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
391 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
392 mempool_free_slab, request_cachep, q->node);
400 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
402 return blk_alloc_queue_node(gfp_mask, -1);
404 EXPORT_SYMBOL(blk_alloc_queue);
406 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
408 struct request_queue *q;
411 q = kmem_cache_alloc_node(blk_requestq_cachep,
412 gfp_mask | __GFP_ZERO, node_id);
416 q->backing_dev_info.ra_pages =
417 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 q->backing_dev_info.state = 0;
419 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
420 q->backing_dev_info.name = "block";
422 err = bdi_init(&q->backing_dev_info);
424 kmem_cache_free(blk_requestq_cachep, q);
428 if (blk_throtl_init(q)) {
429 kmem_cache_free(blk_requestq_cachep, q);
433 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
434 laptop_mode_timer_fn, (unsigned long) q);
435 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
436 INIT_LIST_HEAD(&q->timeout_list);
437 INIT_LIST_HEAD(&q->flush_queue[0]);
438 INIT_LIST_HEAD(&q->flush_queue[1]);
439 INIT_LIST_HEAD(&q->flush_data_in_flight);
440 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
442 kobject_init(&q->kobj, &blk_queue_ktype);
444 mutex_init(&q->sysfs_lock);
445 spin_lock_init(&q->__queue_lock);
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
451 q->queue_lock = &q->__queue_lock;
455 EXPORT_SYMBOL(blk_alloc_queue_node);
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
478 * The queue spin lock must be held while manipulating the requests on the
479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
482 * Function returns a pointer to the initialized request queue, or %NULL if
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
490 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
492 return blk_init_queue_node(rfn, lock, -1);
494 EXPORT_SYMBOL(blk_init_queue);
496 struct request_queue *
497 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
499 struct request_queue *uninit_q, *q;
501 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
505 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
507 blk_cleanup_queue(uninit_q);
511 EXPORT_SYMBOL(blk_init_queue_node);
513 struct request_queue *
514 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
517 return blk_init_allocated_queue_node(q, rfn, lock, -1);
519 EXPORT_SYMBOL(blk_init_allocated_queue);
521 struct request_queue *
522 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
523 spinlock_t *lock, int node_id)
529 if (blk_init_free_list(q))
533 q->prep_rq_fn = NULL;
534 q->unprep_rq_fn = NULL;
535 q->queue_flags = QUEUE_FLAG_DEFAULT;
537 /* Override internal queue lock with supplied lock pointer */
539 q->queue_lock = lock;
542 * This also sets hw/phys segments, boundary and size
544 blk_queue_make_request(q, __make_request);
546 q->sg_reserved_size = INT_MAX;
551 if (!elevator_init(q, NULL)) {
552 blk_queue_congestion_threshold(q);
558 EXPORT_SYMBOL(blk_init_allocated_queue_node);
560 int blk_get_queue(struct request_queue *q)
562 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
563 kobject_get(&q->kobj);
569 EXPORT_SYMBOL(blk_get_queue);
571 static inline void blk_free_request(struct request_queue *q, struct request *rq)
573 if (rq->cmd_flags & REQ_ELVPRIV)
574 elv_put_request(q, rq);
575 mempool_free(rq, q->rq.rq_pool);
578 static struct request *
579 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
581 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
588 rq->cmd_flags = flags | REQ_ALLOCED;
591 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
592 mempool_free(rq, q->rq.rq_pool);
595 rq->cmd_flags |= REQ_ELVPRIV;
602 * ioc_batching returns true if the ioc is a valid batching request and
603 * should be given priority access to a request.
605 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
611 * Make sure the process is able to allocate at least 1 request
612 * even if the batch times out, otherwise we could theoretically
615 return ioc->nr_batch_requests == q->nr_batching ||
616 (ioc->nr_batch_requests > 0
617 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
621 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
622 * will cause the process to be a "batcher" on all queues in the system. This
623 * is the behaviour we want though - once it gets a wakeup it should be given
626 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
628 if (!ioc || ioc_batching(q, ioc))
631 ioc->nr_batch_requests = q->nr_batching;
632 ioc->last_waited = jiffies;
635 static void __freed_request(struct request_queue *q, int sync)
637 struct request_list *rl = &q->rq;
639 if (rl->count[sync] < queue_congestion_off_threshold(q))
640 blk_clear_queue_congested(q, sync);
642 if (rl->count[sync] + 1 <= q->nr_requests) {
643 if (waitqueue_active(&rl->wait[sync]))
644 wake_up(&rl->wait[sync]);
646 blk_clear_queue_full(q, sync);
651 * A request has just been released. Account for it, update the full and
652 * congestion status, wake up any waiters. Called under q->queue_lock.
654 static void freed_request(struct request_queue *q, int sync, int priv)
656 struct request_list *rl = &q->rq;
662 __freed_request(q, sync);
664 if (unlikely(rl->starved[sync ^ 1]))
665 __freed_request(q, sync ^ 1);
669 * Determine if elevator data should be initialized when allocating the
670 * request associated with @bio.
672 static bool blk_rq_should_init_elevator(struct bio *bio)
678 * Flush requests do not use the elevator so skip initialization.
679 * This allows a request to share the flush and elevator data.
681 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
688 * Get a free request, queue_lock must be held.
689 * Returns NULL on failure, with queue_lock held.
690 * Returns !NULL on success, with queue_lock *not held*.
692 static struct request *get_request(struct request_queue *q, int rw_flags,
693 struct bio *bio, gfp_t gfp_mask)
695 struct request *rq = NULL;
696 struct request_list *rl = &q->rq;
697 struct io_context *ioc = NULL;
698 const bool is_sync = rw_is_sync(rw_flags) != 0;
699 int may_queue, priv = 0;
701 may_queue = elv_may_queue(q, rw_flags);
702 if (may_queue == ELV_MQUEUE_NO)
705 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
706 if (rl->count[is_sync]+1 >= q->nr_requests) {
707 ioc = current_io_context(GFP_ATOMIC, q->node);
709 * The queue will fill after this allocation, so set
710 * it as full, and mark this process as "batching".
711 * This process will be allowed to complete a batch of
712 * requests, others will be blocked.
714 if (!blk_queue_full(q, is_sync)) {
715 ioc_set_batching(q, ioc);
716 blk_set_queue_full(q, is_sync);
718 if (may_queue != ELV_MQUEUE_MUST
719 && !ioc_batching(q, ioc)) {
721 * The queue is full and the allocating
722 * process is not a "batcher", and not
723 * exempted by the IO scheduler
729 blk_set_queue_congested(q, is_sync);
733 * Only allow batching queuers to allocate up to 50% over the defined
734 * limit of requests, otherwise we could have thousands of requests
735 * allocated with any setting of ->nr_requests
737 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
740 rl->count[is_sync]++;
741 rl->starved[is_sync] = 0;
743 if (blk_rq_should_init_elevator(bio)) {
744 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
749 if (blk_queue_io_stat(q))
750 rw_flags |= REQ_IO_STAT;
751 spin_unlock_irq(q->queue_lock);
753 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
756 * Allocation failed presumably due to memory. Undo anything
757 * we might have messed up.
759 * Allocating task should really be put onto the front of the
760 * wait queue, but this is pretty rare.
762 spin_lock_irq(q->queue_lock);
763 freed_request(q, is_sync, priv);
766 * in the very unlikely event that allocation failed and no
767 * requests for this direction was pending, mark us starved
768 * so that freeing of a request in the other direction will
769 * notice us. another possible fix would be to split the
770 * rq mempool into READ and WRITE
773 if (unlikely(rl->count[is_sync] == 0))
774 rl->starved[is_sync] = 1;
780 * ioc may be NULL here, and ioc_batching will be false. That's
781 * OK, if the queue is under the request limit then requests need
782 * not count toward the nr_batch_requests limit. There will always
783 * be some limit enforced by BLK_BATCH_TIME.
785 if (ioc_batching(q, ioc))
786 ioc->nr_batch_requests--;
788 trace_block_getrq(q, bio, rw_flags & 1);
794 * No available requests for this queue, wait for some requests to become
797 * Called with q->queue_lock held, and returns with it unlocked.
799 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
802 const bool is_sync = rw_is_sync(rw_flags) != 0;
805 rq = get_request(q, rw_flags, bio, GFP_NOIO);
808 struct io_context *ioc;
809 struct request_list *rl = &q->rq;
811 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
812 TASK_UNINTERRUPTIBLE);
814 trace_block_sleeprq(q, bio, rw_flags & 1);
816 spin_unlock_irq(q->queue_lock);
820 * After sleeping, we become a "batching" process and
821 * will be able to allocate at least one request, and
822 * up to a big batch of them for a small period time.
823 * See ioc_batching, ioc_set_batching
825 ioc = current_io_context(GFP_NOIO, q->node);
826 ioc_set_batching(q, ioc);
828 spin_lock_irq(q->queue_lock);
829 finish_wait(&rl->wait[is_sync], &wait);
831 rq = get_request(q, rw_flags, bio, GFP_NOIO);
837 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
841 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
844 BUG_ON(rw != READ && rw != WRITE);
846 spin_lock_irq(q->queue_lock);
847 if (gfp_mask & __GFP_WAIT) {
848 rq = get_request_wait(q, rw, NULL);
850 rq = get_request(q, rw, NULL, gfp_mask);
852 spin_unlock_irq(q->queue_lock);
854 /* q->queue_lock is unlocked at this point */
858 EXPORT_SYMBOL(blk_get_request);
861 * blk_make_request - given a bio, allocate a corresponding struct request.
862 * @q: target request queue
863 * @bio: The bio describing the memory mappings that will be submitted for IO.
864 * It may be a chained-bio properly constructed by block/bio layer.
865 * @gfp_mask: gfp flags to be used for memory allocation
867 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
868 * type commands. Where the struct request needs to be farther initialized by
869 * the caller. It is passed a &struct bio, which describes the memory info of
872 * The caller of blk_make_request must make sure that bi_io_vec
873 * are set to describe the memory buffers. That bio_data_dir() will return
874 * the needed direction of the request. (And all bio's in the passed bio-chain
875 * are properly set accordingly)
877 * If called under none-sleepable conditions, mapped bio buffers must not
878 * need bouncing, by calling the appropriate masked or flagged allocator,
879 * suitable for the target device. Otherwise the call to blk_queue_bounce will
882 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
883 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
884 * anything but the first bio in the chain. Otherwise you risk waiting for IO
885 * completion of a bio that hasn't been submitted yet, thus resulting in a
886 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
887 * of bio_alloc(), as that avoids the mempool deadlock.
888 * If possible a big IO should be split into smaller parts when allocation
889 * fails. Partial allocation should not be an error, or you risk a live-lock.
891 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
894 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
897 return ERR_PTR(-ENOMEM);
900 struct bio *bounce_bio = bio;
903 blk_queue_bounce(q, &bounce_bio);
904 ret = blk_rq_append_bio(q, rq, bounce_bio);
913 EXPORT_SYMBOL(blk_make_request);
916 * blk_requeue_request - put a request back on queue
917 * @q: request queue where request should be inserted
918 * @rq: request to be inserted
921 * Drivers often keep queueing requests until the hardware cannot accept
922 * more, when that condition happens we need to put the request back
923 * on the queue. Must be called with queue lock held.
925 void blk_requeue_request(struct request_queue *q, struct request *rq)
927 blk_delete_timer(rq);
928 blk_clear_rq_complete(rq);
929 trace_block_rq_requeue(q, rq);
931 if (blk_rq_tagged(rq))
932 blk_queue_end_tag(q, rq);
934 BUG_ON(blk_queued_rq(rq));
936 elv_requeue_request(q, rq);
938 EXPORT_SYMBOL(blk_requeue_request);
940 static void add_acct_request(struct request_queue *q, struct request *rq,
943 drive_stat_acct(rq, 1);
944 __elv_add_request(q, rq, where);
948 * blk_insert_request - insert a special request into a request queue
949 * @q: request queue where request should be inserted
950 * @rq: request to be inserted
951 * @at_head: insert request at head or tail of queue
952 * @data: private data
955 * Many block devices need to execute commands asynchronously, so they don't
956 * block the whole kernel from preemption during request execution. This is
957 * accomplished normally by inserting aritficial requests tagged as
958 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
959 * be scheduled for actual execution by the request queue.
961 * We have the option of inserting the head or the tail of the queue.
962 * Typically we use the tail for new ioctls and so forth. We use the head
963 * of the queue for things like a QUEUE_FULL message from a device, or a
964 * host that is unable to accept a particular command.
966 void blk_insert_request(struct request_queue *q, struct request *rq,
967 int at_head, void *data)
969 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
973 * tell I/O scheduler that this isn't a regular read/write (ie it
974 * must not attempt merges on this) and that it acts as a soft
977 rq->cmd_type = REQ_TYPE_SPECIAL;
981 spin_lock_irqsave(q->queue_lock, flags);
984 * If command is tagged, release the tag
986 if (blk_rq_tagged(rq))
987 blk_queue_end_tag(q, rq);
989 add_acct_request(q, rq, where);
991 spin_unlock_irqrestore(q->queue_lock, flags);
993 EXPORT_SYMBOL(blk_insert_request);
995 static void part_round_stats_single(int cpu, struct hd_struct *part,
998 if (now == part->stamp)
1001 if (part_in_flight(part)) {
1002 __part_stat_add(cpu, part, time_in_queue,
1003 part_in_flight(part) * (now - part->stamp));
1004 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1010 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1011 * @cpu: cpu number for stats access
1012 * @part: target partition
1014 * The average IO queue length and utilisation statistics are maintained
1015 * by observing the current state of the queue length and the amount of
1016 * time it has been in this state for.
1018 * Normally, that accounting is done on IO completion, but that can result
1019 * in more than a second's worth of IO being accounted for within any one
1020 * second, leading to >100% utilisation. To deal with that, we call this
1021 * function to do a round-off before returning the results when reading
1022 * /proc/diskstats. This accounts immediately for all queue usage up to
1023 * the current jiffies and restarts the counters again.
1025 void part_round_stats(int cpu, struct hd_struct *part)
1027 unsigned long now = jiffies;
1030 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1031 part_round_stats_single(cpu, part, now);
1033 EXPORT_SYMBOL_GPL(part_round_stats);
1036 * queue lock must be held
1038 void __blk_put_request(struct request_queue *q, struct request *req)
1042 if (unlikely(--req->ref_count))
1045 elv_completed_request(q, req);
1047 /* this is a bio leak */
1048 WARN_ON(req->bio != NULL);
1051 * Request may not have originated from ll_rw_blk. if not,
1052 * it didn't come out of our reserved rq pools
1054 if (req->cmd_flags & REQ_ALLOCED) {
1055 int is_sync = rq_is_sync(req) != 0;
1056 int priv = req->cmd_flags & REQ_ELVPRIV;
1058 BUG_ON(!list_empty(&req->queuelist));
1059 BUG_ON(!hlist_unhashed(&req->hash));
1061 blk_free_request(q, req);
1062 freed_request(q, is_sync, priv);
1065 EXPORT_SYMBOL_GPL(__blk_put_request);
1067 void blk_put_request(struct request *req)
1069 unsigned long flags;
1070 struct request_queue *q = req->q;
1072 spin_lock_irqsave(q->queue_lock, flags);
1073 __blk_put_request(q, req);
1074 spin_unlock_irqrestore(q->queue_lock, flags);
1076 EXPORT_SYMBOL(blk_put_request);
1079 * blk_add_request_payload - add a payload to a request
1080 * @rq: request to update
1081 * @page: page backing the payload
1082 * @len: length of the payload.
1084 * This allows to later add a payload to an already submitted request by
1085 * a block driver. The driver needs to take care of freeing the payload
1088 * Note that this is a quite horrible hack and nothing but handling of
1089 * discard requests should ever use it.
1091 void blk_add_request_payload(struct request *rq, struct page *page,
1094 struct bio *bio = rq->bio;
1096 bio->bi_io_vec->bv_page = page;
1097 bio->bi_io_vec->bv_offset = 0;
1098 bio->bi_io_vec->bv_len = len;
1102 bio->bi_phys_segments = 1;
1104 rq->__data_len = rq->resid_len = len;
1105 rq->nr_phys_segments = 1;
1106 rq->buffer = bio_data(bio);
1108 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1110 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1113 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1115 if (!ll_back_merge_fn(q, req, bio))
1118 trace_block_bio_backmerge(q, bio);
1120 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1121 blk_rq_set_mixed_merge(req);
1123 req->biotail->bi_next = bio;
1125 req->__data_len += bio->bi_size;
1126 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1128 drive_stat_acct(req, 0);
1129 elv_bio_merged(q, req, bio);
1133 static bool bio_attempt_front_merge(struct request_queue *q,
1134 struct request *req, struct bio *bio)
1136 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1138 if (!ll_front_merge_fn(q, req, bio))
1141 trace_block_bio_frontmerge(q, bio);
1143 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1144 blk_rq_set_mixed_merge(req);
1146 bio->bi_next = req->bio;
1150 * may not be valid. if the low level driver said
1151 * it didn't need a bounce buffer then it better
1152 * not touch req->buffer either...
1154 req->buffer = bio_data(bio);
1155 req->__sector = bio->bi_sector;
1156 req->__data_len += bio->bi_size;
1157 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1159 drive_stat_acct(req, 0);
1160 elv_bio_merged(q, req, bio);
1165 * Attempts to merge with the plugged list in the current process. Returns
1166 * true if merge was successful, otherwise false.
1168 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1169 struct bio *bio, unsigned int *request_count)
1171 struct blk_plug *plug;
1180 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1188 el_ret = elv_try_merge(rq, bio);
1189 if (el_ret == ELEVATOR_BACK_MERGE) {
1190 ret = bio_attempt_back_merge(q, rq, bio);
1193 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1194 ret = bio_attempt_front_merge(q, rq, bio);
1203 void init_request_from_bio(struct request *req, struct bio *bio)
1205 req->cpu = bio->bi_comp_cpu;
1206 req->cmd_type = REQ_TYPE_FS;
1208 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1209 if (bio->bi_rw & REQ_RAHEAD)
1210 req->cmd_flags |= REQ_FAILFAST_MASK;
1213 req->__sector = bio->bi_sector;
1214 req->ioprio = bio_prio(bio);
1215 blk_rq_bio_prep(req->q, req, bio);
1218 static int __make_request(struct request_queue *q, struct bio *bio)
1220 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1221 struct blk_plug *plug;
1222 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1223 struct request *req;
1224 unsigned int request_count = 0;
1227 * low level driver can indicate that it wants pages above a
1228 * certain limit bounced to low memory (ie for highmem, or even
1229 * ISA dma in theory)
1231 blk_queue_bounce(q, &bio);
1233 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1234 spin_lock_irq(q->queue_lock);
1235 where = ELEVATOR_INSERT_FLUSH;
1240 * Check if we can merge with the plugged list before grabbing
1243 if (attempt_plug_merge(current, q, bio, &request_count))
1246 spin_lock_irq(q->queue_lock);
1248 el_ret = elv_merge(q, &req, bio);
1249 if (el_ret == ELEVATOR_BACK_MERGE) {
1250 if (bio_attempt_back_merge(q, req, bio)) {
1251 if (!attempt_back_merge(q, req))
1252 elv_merged_request(q, req, el_ret);
1255 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1256 if (bio_attempt_front_merge(q, req, bio)) {
1257 if (!attempt_front_merge(q, req))
1258 elv_merged_request(q, req, el_ret);
1265 * This sync check and mask will be re-done in init_request_from_bio(),
1266 * but we need to set it earlier to expose the sync flag to the
1267 * rq allocator and io schedulers.
1269 rw_flags = bio_data_dir(bio);
1271 rw_flags |= REQ_SYNC;
1274 * Grab a free request. This is might sleep but can not fail.
1275 * Returns with the queue unlocked.
1277 req = get_request_wait(q, rw_flags, bio);
1280 * After dropping the lock and possibly sleeping here, our request
1281 * may now be mergeable after it had proven unmergeable (above).
1282 * We don't worry about that case for efficiency. It won't happen
1283 * often, and the elevators are able to handle it.
1285 init_request_from_bio(req, bio);
1287 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1288 bio_flagged(bio, BIO_CPU_AFFINE))
1289 req->cpu = raw_smp_processor_id();
1291 plug = current->plug;
1294 * If this is the first request added after a plug, fire
1295 * of a plug trace. If others have been added before, check
1296 * if we have multiple devices in this plug. If so, make a
1297 * note to sort the list before dispatch.
1299 if (list_empty(&plug->list))
1300 trace_block_plug(q);
1301 else if (!plug->should_sort) {
1302 struct request *__rq;
1304 __rq = list_entry_rq(plug->list.prev);
1306 plug->should_sort = 1;
1308 if (request_count >= BLK_MAX_REQUEST_COUNT)
1309 blk_flush_plug_list(plug, false);
1310 list_add_tail(&req->queuelist, &plug->list);
1311 drive_stat_acct(req, 1);
1313 spin_lock_irq(q->queue_lock);
1314 add_acct_request(q, req, where);
1317 spin_unlock_irq(q->queue_lock);
1324 * If bio->bi_dev is a partition, remap the location
1326 static inline void blk_partition_remap(struct bio *bio)
1328 struct block_device *bdev = bio->bi_bdev;
1330 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1331 struct hd_struct *p = bdev->bd_part;
1333 bio->bi_sector += p->start_sect;
1334 bio->bi_bdev = bdev->bd_contains;
1336 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1338 bio->bi_sector - p->start_sect);
1342 static void handle_bad_sector(struct bio *bio)
1344 char b[BDEVNAME_SIZE];
1346 printk(KERN_INFO "attempt to access beyond end of device\n");
1347 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1348 bdevname(bio->bi_bdev, b),
1350 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1351 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1353 set_bit(BIO_EOF, &bio->bi_flags);
1356 #ifdef CONFIG_FAIL_MAKE_REQUEST
1358 static DECLARE_FAULT_ATTR(fail_make_request);
1360 static int __init setup_fail_make_request(char *str)
1362 return setup_fault_attr(&fail_make_request, str);
1364 __setup("fail_make_request=", setup_fail_make_request);
1366 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1368 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1371 static int __init fail_make_request_debugfs(void)
1373 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1374 NULL, &fail_make_request);
1376 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1379 late_initcall(fail_make_request_debugfs);
1381 #else /* CONFIG_FAIL_MAKE_REQUEST */
1383 static inline bool should_fail_request(struct hd_struct *part,
1389 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1392 * Check whether this bio extends beyond the end of the device.
1394 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1401 /* Test device or partition size, when known. */
1402 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1404 sector_t sector = bio->bi_sector;
1406 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1408 * This may well happen - the kernel calls bread()
1409 * without checking the size of the device, e.g., when
1410 * mounting a device.
1412 handle_bad_sector(bio);
1421 * generic_make_request - hand a buffer to its device driver for I/O
1422 * @bio: The bio describing the location in memory and on the device.
1424 * generic_make_request() is used to make I/O requests of block
1425 * devices. It is passed a &struct bio, which describes the I/O that needs
1428 * generic_make_request() does not return any status. The
1429 * success/failure status of the request, along with notification of
1430 * completion, is delivered asynchronously through the bio->bi_end_io
1431 * function described (one day) else where.
1433 * The caller of generic_make_request must make sure that bi_io_vec
1434 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1435 * set to describe the device address, and the
1436 * bi_end_io and optionally bi_private are set to describe how
1437 * completion notification should be signaled.
1439 * generic_make_request and the drivers it calls may use bi_next if this
1440 * bio happens to be merged with someone else, and may change bi_dev and
1441 * bi_sector for remaps as it sees fit. So the values of these fields
1442 * should NOT be depended on after the call to generic_make_request.
1444 static inline void __generic_make_request(struct bio *bio)
1446 struct request_queue *q;
1447 sector_t old_sector;
1448 int ret, nr_sectors = bio_sectors(bio);
1454 if (bio_check_eod(bio, nr_sectors))
1458 * Resolve the mapping until finished. (drivers are
1459 * still free to implement/resolve their own stacking
1460 * by explicitly returning 0)
1462 * NOTE: we don't repeat the blk_size check for each new device.
1463 * Stacking drivers are expected to know what they are doing.
1468 char b[BDEVNAME_SIZE];
1469 struct hd_struct *part;
1471 q = bdev_get_queue(bio->bi_bdev);
1474 "generic_make_request: Trying to access "
1475 "nonexistent block-device %s (%Lu)\n",
1476 bdevname(bio->bi_bdev, b),
1477 (long long) bio->bi_sector);
1481 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1482 nr_sectors > queue_max_hw_sectors(q))) {
1483 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1484 bdevname(bio->bi_bdev, b),
1486 queue_max_hw_sectors(q));
1490 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1493 part = bio->bi_bdev->bd_part;
1494 if (should_fail_request(part, bio->bi_size) ||
1495 should_fail_request(&part_to_disk(part)->part0,
1500 * If this device has partitions, remap block n
1501 * of partition p to block n+start(p) of the disk.
1503 blk_partition_remap(bio);
1505 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1508 if (old_sector != -1)
1509 trace_block_bio_remap(q, bio, old_dev, old_sector);
1511 old_sector = bio->bi_sector;
1512 old_dev = bio->bi_bdev->bd_dev;
1514 if (bio_check_eod(bio, nr_sectors))
1518 * Filter flush bio's early so that make_request based
1519 * drivers without flush support don't have to worry
1522 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1523 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1530 if ((bio->bi_rw & REQ_DISCARD) &&
1531 (!blk_queue_discard(q) ||
1532 ((bio->bi_rw & REQ_SECURE) &&
1533 !blk_queue_secdiscard(q)))) {
1538 if (blk_throtl_bio(q, &bio))
1542 * If bio = NULL, bio has been throttled and will be submitted
1548 trace_block_bio_queue(q, bio);
1550 ret = q->make_request_fn(q, bio);
1556 bio_endio(bio, err);
1560 * We only want one ->make_request_fn to be active at a time,
1561 * else stack usage with stacked devices could be a problem.
1562 * So use current->bio_list to keep a list of requests
1563 * submited by a make_request_fn function.
1564 * current->bio_list is also used as a flag to say if
1565 * generic_make_request is currently active in this task or not.
1566 * If it is NULL, then no make_request is active. If it is non-NULL,
1567 * then a make_request is active, and new requests should be added
1570 void generic_make_request(struct bio *bio)
1572 struct bio_list bio_list_on_stack;
1574 if (current->bio_list) {
1575 /* make_request is active */
1576 bio_list_add(current->bio_list, bio);
1579 /* following loop may be a bit non-obvious, and so deserves some
1581 * Before entering the loop, bio->bi_next is NULL (as all callers
1582 * ensure that) so we have a list with a single bio.
1583 * We pretend that we have just taken it off a longer list, so
1584 * we assign bio_list to a pointer to the bio_list_on_stack,
1585 * thus initialising the bio_list of new bios to be
1586 * added. __generic_make_request may indeed add some more bios
1587 * through a recursive call to generic_make_request. If it
1588 * did, we find a non-NULL value in bio_list and re-enter the loop
1589 * from the top. In this case we really did just take the bio
1590 * of the top of the list (no pretending) and so remove it from
1591 * bio_list, and call into __generic_make_request again.
1593 * The loop was structured like this to make only one call to
1594 * __generic_make_request (which is important as it is large and
1595 * inlined) and to keep the structure simple.
1597 BUG_ON(bio->bi_next);
1598 bio_list_init(&bio_list_on_stack);
1599 current->bio_list = &bio_list_on_stack;
1601 __generic_make_request(bio);
1602 bio = bio_list_pop(current->bio_list);
1604 current->bio_list = NULL; /* deactivate */
1606 EXPORT_SYMBOL(generic_make_request);
1609 * submit_bio - submit a bio to the block device layer for I/O
1610 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1611 * @bio: The &struct bio which describes the I/O
1613 * submit_bio() is very similar in purpose to generic_make_request(), and
1614 * uses that function to do most of the work. Both are fairly rough
1615 * interfaces; @bio must be presetup and ready for I/O.
1618 void submit_bio(int rw, struct bio *bio)
1620 int count = bio_sectors(bio);
1625 * If it's a regular read/write or a barrier with data attached,
1626 * go through the normal accounting stuff before submission.
1628 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1630 count_vm_events(PGPGOUT, count);
1632 task_io_account_read(bio->bi_size);
1633 count_vm_events(PGPGIN, count);
1636 if (unlikely(block_dump)) {
1637 char b[BDEVNAME_SIZE];
1638 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1639 current->comm, task_pid_nr(current),
1640 (rw & WRITE) ? "WRITE" : "READ",
1641 (unsigned long long)bio->bi_sector,
1642 bdevname(bio->bi_bdev, b),
1647 generic_make_request(bio);
1649 EXPORT_SYMBOL(submit_bio);
1652 * blk_rq_check_limits - Helper function to check a request for the queue limit
1654 * @rq: the request being checked
1657 * @rq may have been made based on weaker limitations of upper-level queues
1658 * in request stacking drivers, and it may violate the limitation of @q.
1659 * Since the block layer and the underlying device driver trust @rq
1660 * after it is inserted to @q, it should be checked against @q before
1661 * the insertion using this generic function.
1663 * This function should also be useful for request stacking drivers
1664 * in some cases below, so export this function.
1665 * Request stacking drivers like request-based dm may change the queue
1666 * limits while requests are in the queue (e.g. dm's table swapping).
1667 * Such request stacking drivers should check those requests agaist
1668 * the new queue limits again when they dispatch those requests,
1669 * although such checkings are also done against the old queue limits
1670 * when submitting requests.
1672 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1674 if (rq->cmd_flags & REQ_DISCARD)
1677 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1678 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1679 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1684 * queue's settings related to segment counting like q->bounce_pfn
1685 * may differ from that of other stacking queues.
1686 * Recalculate it to check the request correctly on this queue's
1689 blk_recalc_rq_segments(rq);
1690 if (rq->nr_phys_segments > queue_max_segments(q)) {
1691 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1697 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1700 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1701 * @q: the queue to submit the request
1702 * @rq: the request being queued
1704 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1706 unsigned long flags;
1707 int where = ELEVATOR_INSERT_BACK;
1709 if (blk_rq_check_limits(q, rq))
1713 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1716 spin_lock_irqsave(q->queue_lock, flags);
1719 * Submitting request must be dequeued before calling this function
1720 * because it will be linked to another request_queue
1722 BUG_ON(blk_queued_rq(rq));
1724 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1725 where = ELEVATOR_INSERT_FLUSH;
1727 add_acct_request(q, rq, where);
1728 spin_unlock_irqrestore(q->queue_lock, flags);
1732 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1735 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1736 * @rq: request to examine
1739 * A request could be merge of IOs which require different failure
1740 * handling. This function determines the number of bytes which
1741 * can be failed from the beginning of the request without
1742 * crossing into area which need to be retried further.
1745 * The number of bytes to fail.
1748 * queue_lock must be held.
1750 unsigned int blk_rq_err_bytes(const struct request *rq)
1752 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1753 unsigned int bytes = 0;
1756 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1757 return blk_rq_bytes(rq);
1760 * Currently the only 'mixing' which can happen is between
1761 * different fastfail types. We can safely fail portions
1762 * which have all the failfast bits that the first one has -
1763 * the ones which are at least as eager to fail as the first
1766 for (bio = rq->bio; bio; bio = bio->bi_next) {
1767 if ((bio->bi_rw & ff) != ff)
1769 bytes += bio->bi_size;
1772 /* this could lead to infinite loop */
1773 BUG_ON(blk_rq_bytes(rq) && !bytes);
1776 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1778 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1780 if (blk_do_io_stat(req)) {
1781 const int rw = rq_data_dir(req);
1782 struct hd_struct *part;
1785 cpu = part_stat_lock();
1787 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1792 static void blk_account_io_done(struct request *req)
1795 * Account IO completion. flush_rq isn't accounted as a
1796 * normal IO on queueing nor completion. Accounting the
1797 * containing request is enough.
1799 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1800 unsigned long duration = jiffies - req->start_time;
1801 const int rw = rq_data_dir(req);
1802 struct hd_struct *part;
1805 cpu = part_stat_lock();
1808 part_stat_inc(cpu, part, ios[rw]);
1809 part_stat_add(cpu, part, ticks[rw], duration);
1810 part_round_stats(cpu, part);
1811 part_dec_in_flight(part, rw);
1813 hd_struct_put(part);
1819 * blk_peek_request - peek at the top of a request queue
1820 * @q: request queue to peek at
1823 * Return the request at the top of @q. The returned request
1824 * should be started using blk_start_request() before LLD starts
1828 * Pointer to the request at the top of @q if available. Null
1832 * queue_lock must be held.
1834 struct request *blk_peek_request(struct request_queue *q)
1839 while ((rq = __elv_next_request(q)) != NULL) {
1840 if (!(rq->cmd_flags & REQ_STARTED)) {
1842 * This is the first time the device driver
1843 * sees this request (possibly after
1844 * requeueing). Notify IO scheduler.
1846 if (rq->cmd_flags & REQ_SORTED)
1847 elv_activate_rq(q, rq);
1850 * just mark as started even if we don't start
1851 * it, a request that has been delayed should
1852 * not be passed by new incoming requests
1854 rq->cmd_flags |= REQ_STARTED;
1855 trace_block_rq_issue(q, rq);
1858 if (!q->boundary_rq || q->boundary_rq == rq) {
1859 q->end_sector = rq_end_sector(rq);
1860 q->boundary_rq = NULL;
1863 if (rq->cmd_flags & REQ_DONTPREP)
1866 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1868 * make sure space for the drain appears we
1869 * know we can do this because max_hw_segments
1870 * has been adjusted to be one fewer than the
1873 rq->nr_phys_segments++;
1879 ret = q->prep_rq_fn(q, rq);
1880 if (ret == BLKPREP_OK) {
1882 } else if (ret == BLKPREP_DEFER) {
1884 * the request may have been (partially) prepped.
1885 * we need to keep this request in the front to
1886 * avoid resource deadlock. REQ_STARTED will
1887 * prevent other fs requests from passing this one.
1889 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1890 !(rq->cmd_flags & REQ_DONTPREP)) {
1892 * remove the space for the drain we added
1893 * so that we don't add it again
1895 --rq->nr_phys_segments;
1900 } else if (ret == BLKPREP_KILL) {
1901 rq->cmd_flags |= REQ_QUIET;
1903 * Mark this request as started so we don't trigger
1904 * any debug logic in the end I/O path.
1906 blk_start_request(rq);
1907 __blk_end_request_all(rq, -EIO);
1909 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1916 EXPORT_SYMBOL(blk_peek_request);
1918 void blk_dequeue_request(struct request *rq)
1920 struct request_queue *q = rq->q;
1922 BUG_ON(list_empty(&rq->queuelist));
1923 BUG_ON(ELV_ON_HASH(rq));
1925 list_del_init(&rq->queuelist);
1928 * the time frame between a request being removed from the lists
1929 * and to it is freed is accounted as io that is in progress at
1932 if (blk_account_rq(rq)) {
1933 q->in_flight[rq_is_sync(rq)]++;
1934 set_io_start_time_ns(rq);
1939 * blk_start_request - start request processing on the driver
1940 * @req: request to dequeue
1943 * Dequeue @req and start timeout timer on it. This hands off the
1944 * request to the driver.
1946 * Block internal functions which don't want to start timer should
1947 * call blk_dequeue_request().
1950 * queue_lock must be held.
1952 void blk_start_request(struct request *req)
1954 blk_dequeue_request(req);
1957 * We are now handing the request to the hardware, initialize
1958 * resid_len to full count and add the timeout handler.
1960 req->resid_len = blk_rq_bytes(req);
1961 if (unlikely(blk_bidi_rq(req)))
1962 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1966 EXPORT_SYMBOL(blk_start_request);
1969 * blk_fetch_request - fetch a request from a request queue
1970 * @q: request queue to fetch a request from
1973 * Return the request at the top of @q. The request is started on
1974 * return and LLD can start processing it immediately.
1977 * Pointer to the request at the top of @q if available. Null
1981 * queue_lock must be held.
1983 struct request *blk_fetch_request(struct request_queue *q)
1987 rq = blk_peek_request(q);
1989 blk_start_request(rq);
1992 EXPORT_SYMBOL(blk_fetch_request);
1995 * blk_update_request - Special helper function for request stacking drivers
1996 * @req: the request being processed
1997 * @error: %0 for success, < %0 for error
1998 * @nr_bytes: number of bytes to complete @req
2001 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2002 * the request structure even if @req doesn't have leftover.
2003 * If @req has leftover, sets it up for the next range of segments.
2005 * This special helper function is only for request stacking drivers
2006 * (e.g. request-based dm) so that they can handle partial completion.
2007 * Actual device drivers should use blk_end_request instead.
2009 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2010 * %false return from this function.
2013 * %false - this request doesn't have any more data
2014 * %true - this request has more data
2016 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2018 int total_bytes, bio_nbytes, next_idx = 0;
2024 trace_block_rq_complete(req->q, req);
2027 * For fs requests, rq is just carrier of independent bio's
2028 * and each partial completion should be handled separately.
2029 * Reset per-request error on each partial completion.
2031 * TODO: tj: This is too subtle. It would be better to let
2032 * low level drivers do what they see fit.
2034 if (req->cmd_type == REQ_TYPE_FS)
2037 if (error && req->cmd_type == REQ_TYPE_FS &&
2038 !(req->cmd_flags & REQ_QUIET)) {
2043 error_type = "recoverable transport";
2046 error_type = "critical target";
2049 error_type = "critical nexus";
2056 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2057 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2058 (unsigned long long)blk_rq_pos(req));
2061 blk_account_io_completion(req, nr_bytes);
2063 total_bytes = bio_nbytes = 0;
2064 while ((bio = req->bio) != NULL) {
2067 if (nr_bytes >= bio->bi_size) {
2068 req->bio = bio->bi_next;
2069 nbytes = bio->bi_size;
2070 req_bio_endio(req, bio, nbytes, error);
2074 int idx = bio->bi_idx + next_idx;
2076 if (unlikely(idx >= bio->bi_vcnt)) {
2077 blk_dump_rq_flags(req, "__end_that");
2078 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2079 __func__, idx, bio->bi_vcnt);
2083 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2084 BIO_BUG_ON(nbytes > bio->bi_size);
2087 * not a complete bvec done
2089 if (unlikely(nbytes > nr_bytes)) {
2090 bio_nbytes += nr_bytes;
2091 total_bytes += nr_bytes;
2096 * advance to the next vector
2099 bio_nbytes += nbytes;
2102 total_bytes += nbytes;
2108 * end more in this run, or just return 'not-done'
2110 if (unlikely(nr_bytes <= 0))
2120 * Reset counters so that the request stacking driver
2121 * can find how many bytes remain in the request
2124 req->__data_len = 0;
2129 * if the request wasn't completed, update state
2132 req_bio_endio(req, bio, bio_nbytes, error);
2133 bio->bi_idx += next_idx;
2134 bio_iovec(bio)->bv_offset += nr_bytes;
2135 bio_iovec(bio)->bv_len -= nr_bytes;
2138 req->__data_len -= total_bytes;
2139 req->buffer = bio_data(req->bio);
2141 /* update sector only for requests with clear definition of sector */
2142 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2143 req->__sector += total_bytes >> 9;
2145 /* mixed attributes always follow the first bio */
2146 if (req->cmd_flags & REQ_MIXED_MERGE) {
2147 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2148 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2152 * If total number of sectors is less than the first segment
2153 * size, something has gone terribly wrong.
2155 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2156 blk_dump_rq_flags(req, "request botched");
2157 req->__data_len = blk_rq_cur_bytes(req);
2160 /* recalculate the number of segments */
2161 blk_recalc_rq_segments(req);
2165 EXPORT_SYMBOL_GPL(blk_update_request);
2167 static bool blk_update_bidi_request(struct request *rq, int error,
2168 unsigned int nr_bytes,
2169 unsigned int bidi_bytes)
2171 if (blk_update_request(rq, error, nr_bytes))
2174 /* Bidi request must be completed as a whole */
2175 if (unlikely(blk_bidi_rq(rq)) &&
2176 blk_update_request(rq->next_rq, error, bidi_bytes))
2179 if (blk_queue_add_random(rq->q))
2180 add_disk_randomness(rq->rq_disk);
2186 * blk_unprep_request - unprepare a request
2189 * This function makes a request ready for complete resubmission (or
2190 * completion). It happens only after all error handling is complete,
2191 * so represents the appropriate moment to deallocate any resources
2192 * that were allocated to the request in the prep_rq_fn. The queue
2193 * lock is held when calling this.
2195 void blk_unprep_request(struct request *req)
2197 struct request_queue *q = req->q;
2199 req->cmd_flags &= ~REQ_DONTPREP;
2200 if (q->unprep_rq_fn)
2201 q->unprep_rq_fn(q, req);
2203 EXPORT_SYMBOL_GPL(blk_unprep_request);
2206 * queue lock must be held
2208 static void blk_finish_request(struct request *req, int error)
2210 if (blk_rq_tagged(req))
2211 blk_queue_end_tag(req->q, req);
2213 BUG_ON(blk_queued_rq(req));
2215 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2216 laptop_io_completion(&req->q->backing_dev_info);
2218 blk_delete_timer(req);
2220 if (req->cmd_flags & REQ_DONTPREP)
2221 blk_unprep_request(req);
2224 blk_account_io_done(req);
2227 req->end_io(req, error);
2229 if (blk_bidi_rq(req))
2230 __blk_put_request(req->next_rq->q, req->next_rq);
2232 __blk_put_request(req->q, req);
2237 * blk_end_bidi_request - Complete a bidi request
2238 * @rq: the request to complete
2239 * @error: %0 for success, < %0 for error
2240 * @nr_bytes: number of bytes to complete @rq
2241 * @bidi_bytes: number of bytes to complete @rq->next_rq
2244 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2245 * Drivers that supports bidi can safely call this member for any
2246 * type of request, bidi or uni. In the later case @bidi_bytes is
2250 * %false - we are done with this request
2251 * %true - still buffers pending for this request
2253 static bool blk_end_bidi_request(struct request *rq, int error,
2254 unsigned int nr_bytes, unsigned int bidi_bytes)
2256 struct request_queue *q = rq->q;
2257 unsigned long flags;
2259 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2262 spin_lock_irqsave(q->queue_lock, flags);
2263 blk_finish_request(rq, error);
2264 spin_unlock_irqrestore(q->queue_lock, flags);
2270 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2271 * @rq: the request to complete
2272 * @error: %0 for success, < %0 for error
2273 * @nr_bytes: number of bytes to complete @rq
2274 * @bidi_bytes: number of bytes to complete @rq->next_rq
2277 * Identical to blk_end_bidi_request() except that queue lock is
2278 * assumed to be locked on entry and remains so on return.
2281 * %false - we are done with this request
2282 * %true - still buffers pending for this request
2284 bool __blk_end_bidi_request(struct request *rq, int error,
2285 unsigned int nr_bytes, unsigned int bidi_bytes)
2287 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2290 blk_finish_request(rq, error);
2296 * blk_end_request - Helper function for drivers to complete the request.
2297 * @rq: the request being processed
2298 * @error: %0 for success, < %0 for error
2299 * @nr_bytes: number of bytes to complete
2302 * Ends I/O on a number of bytes attached to @rq.
2303 * If @rq has leftover, sets it up for the next range of segments.
2306 * %false - we are done with this request
2307 * %true - still buffers pending for this request
2309 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2311 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2313 EXPORT_SYMBOL(blk_end_request);
2316 * blk_end_request_all - Helper function for drives to finish the request.
2317 * @rq: the request to finish
2318 * @error: %0 for success, < %0 for error
2321 * Completely finish @rq.
2323 void blk_end_request_all(struct request *rq, int error)
2326 unsigned int bidi_bytes = 0;
2328 if (unlikely(blk_bidi_rq(rq)))
2329 bidi_bytes = blk_rq_bytes(rq->next_rq);
2331 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2334 EXPORT_SYMBOL(blk_end_request_all);
2337 * blk_end_request_cur - Helper function to finish the current request chunk.
2338 * @rq: the request to finish the current chunk for
2339 * @error: %0 for success, < %0 for error
2342 * Complete the current consecutively mapped chunk from @rq.
2345 * %false - we are done with this request
2346 * %true - still buffers pending for this request
2348 bool blk_end_request_cur(struct request *rq, int error)
2350 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2352 EXPORT_SYMBOL(blk_end_request_cur);
2355 * blk_end_request_err - Finish a request till the next failure boundary.
2356 * @rq: the request to finish till the next failure boundary for
2357 * @error: must be negative errno
2360 * Complete @rq till the next failure boundary.
2363 * %false - we are done with this request
2364 * %true - still buffers pending for this request
2366 bool blk_end_request_err(struct request *rq, int error)
2368 WARN_ON(error >= 0);
2369 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2371 EXPORT_SYMBOL_GPL(blk_end_request_err);
2374 * __blk_end_request - Helper function for drivers to complete the request.
2375 * @rq: the request being processed
2376 * @error: %0 for success, < %0 for error
2377 * @nr_bytes: number of bytes to complete
2380 * Must be called with queue lock held unlike blk_end_request().
2383 * %false - we are done with this request
2384 * %true - still buffers pending for this request
2386 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2388 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2390 EXPORT_SYMBOL(__blk_end_request);
2393 * __blk_end_request_all - Helper function for drives to finish the request.
2394 * @rq: the request to finish
2395 * @error: %0 for success, < %0 for error
2398 * Completely finish @rq. Must be called with queue lock held.
2400 void __blk_end_request_all(struct request *rq, int error)
2403 unsigned int bidi_bytes = 0;
2405 if (unlikely(blk_bidi_rq(rq)))
2406 bidi_bytes = blk_rq_bytes(rq->next_rq);
2408 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2411 EXPORT_SYMBOL(__blk_end_request_all);
2414 * __blk_end_request_cur - Helper function to finish the current request chunk.
2415 * @rq: the request to finish the current chunk for
2416 * @error: %0 for success, < %0 for error
2419 * Complete the current consecutively mapped chunk from @rq. Must
2420 * be called with queue lock held.
2423 * %false - we are done with this request
2424 * %true - still buffers pending for this request
2426 bool __blk_end_request_cur(struct request *rq, int error)
2428 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2430 EXPORT_SYMBOL(__blk_end_request_cur);
2433 * __blk_end_request_err - Finish a request till the next failure boundary.
2434 * @rq: the request to finish till the next failure boundary for
2435 * @error: must be negative errno
2438 * Complete @rq till the next failure boundary. Must be called
2439 * with queue lock held.
2442 * %false - we are done with this request
2443 * %true - still buffers pending for this request
2445 bool __blk_end_request_err(struct request *rq, int error)
2447 WARN_ON(error >= 0);
2448 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2450 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2452 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2455 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2456 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2458 if (bio_has_data(bio)) {
2459 rq->nr_phys_segments = bio_phys_segments(q, bio);
2460 rq->buffer = bio_data(bio);
2462 rq->__data_len = bio->bi_size;
2463 rq->bio = rq->biotail = bio;
2466 rq->rq_disk = bio->bi_bdev->bd_disk;
2469 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2471 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2472 * @rq: the request to be flushed
2475 * Flush all pages in @rq.
2477 void rq_flush_dcache_pages(struct request *rq)
2479 struct req_iterator iter;
2480 struct bio_vec *bvec;
2482 rq_for_each_segment(bvec, rq, iter)
2483 flush_dcache_page(bvec->bv_page);
2485 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2489 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2490 * @q : the queue of the device being checked
2493 * Check if underlying low-level drivers of a device are busy.
2494 * If the drivers want to export their busy state, they must set own
2495 * exporting function using blk_queue_lld_busy() first.
2497 * Basically, this function is used only by request stacking drivers
2498 * to stop dispatching requests to underlying devices when underlying
2499 * devices are busy. This behavior helps more I/O merging on the queue
2500 * of the request stacking driver and prevents I/O throughput regression
2501 * on burst I/O load.
2504 * 0 - Not busy (The request stacking driver should dispatch request)
2505 * 1 - Busy (The request stacking driver should stop dispatching request)
2507 int blk_lld_busy(struct request_queue *q)
2510 return q->lld_busy_fn(q);
2514 EXPORT_SYMBOL_GPL(blk_lld_busy);
2517 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2518 * @rq: the clone request to be cleaned up
2521 * Free all bios in @rq for a cloned request.
2523 void blk_rq_unprep_clone(struct request *rq)
2527 while ((bio = rq->bio) != NULL) {
2528 rq->bio = bio->bi_next;
2533 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2536 * Copy attributes of the original request to the clone request.
2537 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2539 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2541 dst->cpu = src->cpu;
2542 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2543 dst->cmd_type = src->cmd_type;
2544 dst->__sector = blk_rq_pos(src);
2545 dst->__data_len = blk_rq_bytes(src);
2546 dst->nr_phys_segments = src->nr_phys_segments;
2547 dst->ioprio = src->ioprio;
2548 dst->extra_len = src->extra_len;
2552 * blk_rq_prep_clone - Helper function to setup clone request
2553 * @rq: the request to be setup
2554 * @rq_src: original request to be cloned
2555 * @bs: bio_set that bios for clone are allocated from
2556 * @gfp_mask: memory allocation mask for bio
2557 * @bio_ctr: setup function to be called for each clone bio.
2558 * Returns %0 for success, non %0 for failure.
2559 * @data: private data to be passed to @bio_ctr
2562 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2563 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2564 * are not copied, and copying such parts is the caller's responsibility.
2565 * Also, pages which the original bios are pointing to are not copied
2566 * and the cloned bios just point same pages.
2567 * So cloned bios must be completed before original bios, which means
2568 * the caller must complete @rq before @rq_src.
2570 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2571 struct bio_set *bs, gfp_t gfp_mask,
2572 int (*bio_ctr)(struct bio *, struct bio *, void *),
2575 struct bio *bio, *bio_src;
2580 blk_rq_init(NULL, rq);
2582 __rq_for_each_bio(bio_src, rq_src) {
2583 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2587 __bio_clone(bio, bio_src);
2589 if (bio_integrity(bio_src) &&
2590 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2593 if (bio_ctr && bio_ctr(bio, bio_src, data))
2597 rq->biotail->bi_next = bio;
2600 rq->bio = rq->biotail = bio;
2603 __blk_rq_prep_clone(rq, rq_src);
2610 blk_rq_unprep_clone(rq);
2614 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2616 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2618 return queue_work(kblockd_workqueue, work);
2620 EXPORT_SYMBOL(kblockd_schedule_work);
2622 int kblockd_schedule_delayed_work(struct request_queue *q,
2623 struct delayed_work *dwork, unsigned long delay)
2625 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2627 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2629 #define PLUG_MAGIC 0x91827364
2631 void blk_start_plug(struct blk_plug *plug)
2633 struct task_struct *tsk = current;
2635 plug->magic = PLUG_MAGIC;
2636 INIT_LIST_HEAD(&plug->list);
2637 INIT_LIST_HEAD(&plug->cb_list);
2638 plug->should_sort = 0;
2641 * If this is a nested plug, don't actually assign it. It will be
2642 * flushed on its own.
2646 * Store ordering should not be needed here, since a potential
2647 * preempt will imply a full memory barrier
2652 EXPORT_SYMBOL(blk_start_plug);
2654 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2656 struct request *rqa = container_of(a, struct request, queuelist);
2657 struct request *rqb = container_of(b, struct request, queuelist);
2659 return !(rqa->q <= rqb->q);
2663 * If 'from_schedule' is true, then postpone the dispatch of requests
2664 * until a safe kblockd context. We due this to avoid accidental big
2665 * additional stack usage in driver dispatch, in places where the originally
2666 * plugger did not intend it.
2668 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2670 __releases(q->queue_lock)
2672 trace_block_unplug(q, depth, !from_schedule);
2675 * If we are punting this to kblockd, then we can safely drop
2676 * the queue_lock before waking kblockd (which needs to take
2679 if (from_schedule) {
2680 spin_unlock(q->queue_lock);
2681 blk_run_queue_async(q);
2684 spin_unlock(q->queue_lock);
2689 static void flush_plug_callbacks(struct blk_plug *plug)
2691 LIST_HEAD(callbacks);
2693 if (list_empty(&plug->cb_list))
2696 list_splice_init(&plug->cb_list, &callbacks);
2698 while (!list_empty(&callbacks)) {
2699 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2702 list_del(&cb->list);
2707 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2709 struct request_queue *q;
2710 unsigned long flags;
2715 BUG_ON(plug->magic != PLUG_MAGIC);
2717 flush_plug_callbacks(plug);
2718 if (list_empty(&plug->list))
2721 list_splice_init(&plug->list, &list);
2723 if (plug->should_sort) {
2724 list_sort(NULL, &list, plug_rq_cmp);
2725 plug->should_sort = 0;
2732 * Save and disable interrupts here, to avoid doing it for every
2733 * queue lock we have to take.
2735 local_irq_save(flags);
2736 while (!list_empty(&list)) {
2737 rq = list_entry_rq(list.next);
2738 list_del_init(&rq->queuelist);
2742 * This drops the queue lock
2745 queue_unplugged(q, depth, from_schedule);
2748 spin_lock(q->queue_lock);
2751 * rq is already accounted, so use raw insert
2753 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2754 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2756 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2762 * This drops the queue lock
2765 queue_unplugged(q, depth, from_schedule);
2767 local_irq_restore(flags);
2770 void blk_finish_plug(struct blk_plug *plug)
2772 blk_flush_plug_list(plug, false);
2774 if (plug == current->plug)
2775 current->plug = NULL;
2777 EXPORT_SYMBOL(blk_finish_plug);
2779 int __init blk_dev_init(void)
2781 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2782 sizeof(((struct request *)0)->cmd_flags));
2784 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2785 kblockd_workqueue = alloc_workqueue("kblockd",
2786 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2787 if (!kblockd_workqueue)
2788 panic("Failed to create kblockd\n");
2790 request_cachep = kmem_cache_create("blkdev_requests",
2791 sizeof(struct request), 0, SLAB_PANIC, NULL);
2793 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2794 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);