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);
207 __blk_run_queue(q, false);
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 schedule_delayed_work(&q->delay_work, 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);
241 __blk_run_queue(q, false);
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
294 * @force_kblockd: Don't run @q->request_fn directly. Use kblockd.
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled. If force_kblockd is true, then it is
299 * safe to call this without holding the queue lock.
302 void __blk_run_queue(struct request_queue *q, bool force_kblockd)
304 if (unlikely(blk_queue_stopped(q)))
308 * Only recurse once to avoid overrunning the stack, let the unplug
309 * handling reinvoke the handler shortly if we already got there.
311 if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
313 queue_flag_clear(QUEUE_FLAG_REENTER, q);
315 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
317 EXPORT_SYMBOL(__blk_run_queue);
320 * blk_run_queue - run a single device queue
321 * @q: The queue to run
324 * Invoke request handling on this queue, if it has pending work to do.
325 * May be used to restart queueing when a request has completed.
327 void blk_run_queue(struct request_queue *q)
331 spin_lock_irqsave(q->queue_lock, flags);
332 __blk_run_queue(q, false);
333 spin_unlock_irqrestore(q->queue_lock, flags);
335 EXPORT_SYMBOL(blk_run_queue);
337 void blk_put_queue(struct request_queue *q)
339 kobject_put(&q->kobj);
343 * Note: If a driver supplied the queue lock, it should not zap that lock
344 * unexpectedly as some queue cleanup components like elevator_exit() and
345 * blk_throtl_exit() need queue lock.
347 void blk_cleanup_queue(struct request_queue *q)
350 * We know we have process context here, so we can be a little
351 * cautious and ensure that pending block actions on this device
352 * are done before moving on. Going into this function, we should
353 * not have processes doing IO to this device.
357 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
358 mutex_lock(&q->sysfs_lock);
359 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
360 mutex_unlock(&q->sysfs_lock);
363 elevator_exit(q->elevator);
369 EXPORT_SYMBOL(blk_cleanup_queue);
371 static int blk_init_free_list(struct request_queue *q)
373 struct request_list *rl = &q->rq;
375 if (unlikely(rl->rq_pool))
378 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
379 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
381 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
382 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
384 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
385 mempool_free_slab, request_cachep, q->node);
393 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
395 return blk_alloc_queue_node(gfp_mask, -1);
397 EXPORT_SYMBOL(blk_alloc_queue);
399 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
401 struct request_queue *q;
404 q = kmem_cache_alloc_node(blk_requestq_cachep,
405 gfp_mask | __GFP_ZERO, node_id);
409 q->backing_dev_info.ra_pages =
410 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
411 q->backing_dev_info.state = 0;
412 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
413 q->backing_dev_info.name = "block";
415 err = bdi_init(&q->backing_dev_info);
417 kmem_cache_free(blk_requestq_cachep, q);
421 if (blk_throtl_init(q)) {
422 kmem_cache_free(blk_requestq_cachep, q);
426 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
427 laptop_mode_timer_fn, (unsigned long) q);
428 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
429 INIT_LIST_HEAD(&q->timeout_list);
430 INIT_LIST_HEAD(&q->flush_queue[0]);
431 INIT_LIST_HEAD(&q->flush_queue[1]);
432 INIT_LIST_HEAD(&q->flush_data_in_flight);
433 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
435 kobject_init(&q->kobj, &blk_queue_ktype);
437 mutex_init(&q->sysfs_lock);
438 spin_lock_init(&q->__queue_lock);
441 * By default initialize queue_lock to internal lock and driver can
442 * override it later if need be.
444 q->queue_lock = &q->__queue_lock;
448 EXPORT_SYMBOL(blk_alloc_queue_node);
451 * blk_init_queue - prepare a request queue for use with a block device
452 * @rfn: The function to be called to process requests that have been
453 * placed on the queue.
454 * @lock: Request queue spin lock
457 * If a block device wishes to use the standard request handling procedures,
458 * which sorts requests and coalesces adjacent requests, then it must
459 * call blk_init_queue(). The function @rfn will be called when there
460 * are requests on the queue that need to be processed. If the device
461 * supports plugging, then @rfn may not be called immediately when requests
462 * are available on the queue, but may be called at some time later instead.
463 * Plugged queues are generally unplugged when a buffer belonging to one
464 * of the requests on the queue is needed, or due to memory pressure.
466 * @rfn is not required, or even expected, to remove all requests off the
467 * queue, but only as many as it can handle at a time. If it does leave
468 * requests on the queue, it is responsible for arranging that the requests
469 * get dealt with eventually.
471 * The queue spin lock must be held while manipulating the requests on the
472 * request queue; this lock will be taken also from interrupt context, so irq
473 * disabling is needed for it.
475 * Function returns a pointer to the initialized request queue, or %NULL if
479 * blk_init_queue() must be paired with a blk_cleanup_queue() call
480 * when the block device is deactivated (such as at module unload).
483 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
485 return blk_init_queue_node(rfn, lock, -1);
487 EXPORT_SYMBOL(blk_init_queue);
489 struct request_queue *
490 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
492 struct request_queue *uninit_q, *q;
494 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
498 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
500 blk_cleanup_queue(uninit_q);
504 EXPORT_SYMBOL(blk_init_queue_node);
506 struct request_queue *
507 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
510 return blk_init_allocated_queue_node(q, rfn, lock, -1);
512 EXPORT_SYMBOL(blk_init_allocated_queue);
514 struct request_queue *
515 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
516 spinlock_t *lock, int node_id)
522 if (blk_init_free_list(q))
526 q->prep_rq_fn = NULL;
527 q->unprep_rq_fn = NULL;
528 q->queue_flags = QUEUE_FLAG_DEFAULT;
530 /* Override internal queue lock with supplied lock pointer */
532 q->queue_lock = lock;
535 * This also sets hw/phys segments, boundary and size
537 blk_queue_make_request(q, __make_request);
539 q->sg_reserved_size = INT_MAX;
544 if (!elevator_init(q, NULL)) {
545 blk_queue_congestion_threshold(q);
551 EXPORT_SYMBOL(blk_init_allocated_queue_node);
553 int blk_get_queue(struct request_queue *q)
555 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
556 kobject_get(&q->kobj);
563 static inline void blk_free_request(struct request_queue *q, struct request *rq)
565 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
567 if (rq->cmd_flags & REQ_ELVPRIV)
568 elv_put_request(q, rq);
569 mempool_free(rq, q->rq.rq_pool);
572 static struct request *
573 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
575 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
582 rq->cmd_flags = flags | REQ_ALLOCED;
585 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
586 mempool_free(rq, q->rq.rq_pool);
589 rq->cmd_flags |= REQ_ELVPRIV;
596 * ioc_batching returns true if the ioc is a valid batching request and
597 * should be given priority access to a request.
599 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
605 * Make sure the process is able to allocate at least 1 request
606 * even if the batch times out, otherwise we could theoretically
609 return ioc->nr_batch_requests == q->nr_batching ||
610 (ioc->nr_batch_requests > 0
611 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
615 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
616 * will cause the process to be a "batcher" on all queues in the system. This
617 * is the behaviour we want though - once it gets a wakeup it should be given
620 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
622 if (!ioc || ioc_batching(q, ioc))
625 ioc->nr_batch_requests = q->nr_batching;
626 ioc->last_waited = jiffies;
629 static void __freed_request(struct request_queue *q, int sync)
631 struct request_list *rl = &q->rq;
633 if (rl->count[sync] < queue_congestion_off_threshold(q))
634 blk_clear_queue_congested(q, sync);
636 if (rl->count[sync] + 1 <= q->nr_requests) {
637 if (waitqueue_active(&rl->wait[sync]))
638 wake_up(&rl->wait[sync]);
640 blk_clear_queue_full(q, sync);
645 * A request has just been released. Account for it, update the full and
646 * congestion status, wake up any waiters. Called under q->queue_lock.
648 static void freed_request(struct request_queue *q, int sync, int priv)
650 struct request_list *rl = &q->rq;
656 __freed_request(q, sync);
658 if (unlikely(rl->starved[sync ^ 1]))
659 __freed_request(q, sync ^ 1);
663 * Determine if elevator data should be initialized when allocating the
664 * request associated with @bio.
666 static bool blk_rq_should_init_elevator(struct bio *bio)
672 * Flush requests do not use the elevator so skip initialization.
673 * This allows a request to share the flush and elevator data.
675 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
682 * Get a free request, queue_lock must be held.
683 * Returns NULL on failure, with queue_lock held.
684 * Returns !NULL on success, with queue_lock *not held*.
686 static struct request *get_request(struct request_queue *q, int rw_flags,
687 struct bio *bio, gfp_t gfp_mask)
689 struct request *rq = NULL;
690 struct request_list *rl = &q->rq;
691 struct io_context *ioc = NULL;
692 const bool is_sync = rw_is_sync(rw_flags) != 0;
693 int may_queue, priv = 0;
695 may_queue = elv_may_queue(q, rw_flags);
696 if (may_queue == ELV_MQUEUE_NO)
699 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
700 if (rl->count[is_sync]+1 >= q->nr_requests) {
701 ioc = current_io_context(GFP_ATOMIC, q->node);
703 * The queue will fill after this allocation, so set
704 * it as full, and mark this process as "batching".
705 * This process will be allowed to complete a batch of
706 * requests, others will be blocked.
708 if (!blk_queue_full(q, is_sync)) {
709 ioc_set_batching(q, ioc);
710 blk_set_queue_full(q, is_sync);
712 if (may_queue != ELV_MQUEUE_MUST
713 && !ioc_batching(q, ioc)) {
715 * The queue is full and the allocating
716 * process is not a "batcher", and not
717 * exempted by the IO scheduler
723 blk_set_queue_congested(q, is_sync);
727 * Only allow batching queuers to allocate up to 50% over the defined
728 * limit of requests, otherwise we could have thousands of requests
729 * allocated with any setting of ->nr_requests
731 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
734 rl->count[is_sync]++;
735 rl->starved[is_sync] = 0;
737 if (blk_rq_should_init_elevator(bio)) {
738 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
743 if (blk_queue_io_stat(q))
744 rw_flags |= REQ_IO_STAT;
745 spin_unlock_irq(q->queue_lock);
747 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
750 * Allocation failed presumably due to memory. Undo anything
751 * we might have messed up.
753 * Allocating task should really be put onto the front of the
754 * wait queue, but this is pretty rare.
756 spin_lock_irq(q->queue_lock);
757 freed_request(q, is_sync, priv);
760 * in the very unlikely event that allocation failed and no
761 * requests for this direction was pending, mark us starved
762 * so that freeing of a request in the other direction will
763 * notice us. another possible fix would be to split the
764 * rq mempool into READ and WRITE
767 if (unlikely(rl->count[is_sync] == 0))
768 rl->starved[is_sync] = 1;
774 * ioc may be NULL here, and ioc_batching will be false. That's
775 * OK, if the queue is under the request limit then requests need
776 * not count toward the nr_batch_requests limit. There will always
777 * be some limit enforced by BLK_BATCH_TIME.
779 if (ioc_batching(q, ioc))
780 ioc->nr_batch_requests--;
782 trace_block_getrq(q, bio, rw_flags & 1);
788 * No available requests for this queue, wait for some requests to become
791 * Called with q->queue_lock held, and returns with it unlocked.
793 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
796 const bool is_sync = rw_is_sync(rw_flags) != 0;
799 rq = get_request(q, rw_flags, bio, GFP_NOIO);
802 struct io_context *ioc;
803 struct request_list *rl = &q->rq;
805 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
806 TASK_UNINTERRUPTIBLE);
808 trace_block_sleeprq(q, bio, rw_flags & 1);
810 spin_unlock_irq(q->queue_lock);
814 * After sleeping, we become a "batching" process and
815 * will be able to allocate at least one request, and
816 * up to a big batch of them for a small period time.
817 * See ioc_batching, ioc_set_batching
819 ioc = current_io_context(GFP_NOIO, q->node);
820 ioc_set_batching(q, ioc);
822 spin_lock_irq(q->queue_lock);
823 finish_wait(&rl->wait[is_sync], &wait);
825 rq = get_request(q, rw_flags, bio, GFP_NOIO);
831 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
835 BUG_ON(rw != READ && rw != WRITE);
837 spin_lock_irq(q->queue_lock);
838 if (gfp_mask & __GFP_WAIT) {
839 rq = get_request_wait(q, rw, NULL);
841 rq = get_request(q, rw, NULL, gfp_mask);
843 spin_unlock_irq(q->queue_lock);
845 /* q->queue_lock is unlocked at this point */
849 EXPORT_SYMBOL(blk_get_request);
852 * blk_make_request - given a bio, allocate a corresponding struct request.
853 * @q: target request queue
854 * @bio: The bio describing the memory mappings that will be submitted for IO.
855 * It may be a chained-bio properly constructed by block/bio layer.
856 * @gfp_mask: gfp flags to be used for memory allocation
858 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
859 * type commands. Where the struct request needs to be farther initialized by
860 * the caller. It is passed a &struct bio, which describes the memory info of
863 * The caller of blk_make_request must make sure that bi_io_vec
864 * are set to describe the memory buffers. That bio_data_dir() will return
865 * the needed direction of the request. (And all bio's in the passed bio-chain
866 * are properly set accordingly)
868 * If called under none-sleepable conditions, mapped bio buffers must not
869 * need bouncing, by calling the appropriate masked or flagged allocator,
870 * suitable for the target device. Otherwise the call to blk_queue_bounce will
873 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
874 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
875 * anything but the first bio in the chain. Otherwise you risk waiting for IO
876 * completion of a bio that hasn't been submitted yet, thus resulting in a
877 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
878 * of bio_alloc(), as that avoids the mempool deadlock.
879 * If possible a big IO should be split into smaller parts when allocation
880 * fails. Partial allocation should not be an error, or you risk a live-lock.
882 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
885 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
888 return ERR_PTR(-ENOMEM);
891 struct bio *bounce_bio = bio;
894 blk_queue_bounce(q, &bounce_bio);
895 ret = blk_rq_append_bio(q, rq, bounce_bio);
904 EXPORT_SYMBOL(blk_make_request);
907 * blk_requeue_request - put a request back on queue
908 * @q: request queue where request should be inserted
909 * @rq: request to be inserted
912 * Drivers often keep queueing requests until the hardware cannot accept
913 * more, when that condition happens we need to put the request back
914 * on the queue. Must be called with queue lock held.
916 void blk_requeue_request(struct request_queue *q, struct request *rq)
918 blk_delete_timer(rq);
919 blk_clear_rq_complete(rq);
920 trace_block_rq_requeue(q, rq);
922 if (blk_rq_tagged(rq))
923 blk_queue_end_tag(q, rq);
925 BUG_ON(blk_queued_rq(rq));
927 elv_requeue_request(q, rq);
929 EXPORT_SYMBOL(blk_requeue_request);
931 static void add_acct_request(struct request_queue *q, struct request *rq,
934 drive_stat_acct(rq, 1);
935 __elv_add_request(q, rq, where);
939 * blk_insert_request - insert a special request into a request queue
940 * @q: request queue where request should be inserted
941 * @rq: request to be inserted
942 * @at_head: insert request at head or tail of queue
943 * @data: private data
946 * Many block devices need to execute commands asynchronously, so they don't
947 * block the whole kernel from preemption during request execution. This is
948 * accomplished normally by inserting aritficial requests tagged as
949 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
950 * be scheduled for actual execution by the request queue.
952 * We have the option of inserting the head or the tail of the queue.
953 * Typically we use the tail for new ioctls and so forth. We use the head
954 * of the queue for things like a QUEUE_FULL message from a device, or a
955 * host that is unable to accept a particular command.
957 void blk_insert_request(struct request_queue *q, struct request *rq,
958 int at_head, void *data)
960 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
964 * tell I/O scheduler that this isn't a regular read/write (ie it
965 * must not attempt merges on this) and that it acts as a soft
968 rq->cmd_type = REQ_TYPE_SPECIAL;
972 spin_lock_irqsave(q->queue_lock, flags);
975 * If command is tagged, release the tag
977 if (blk_rq_tagged(rq))
978 blk_queue_end_tag(q, rq);
980 add_acct_request(q, rq, where);
981 __blk_run_queue(q, false);
982 spin_unlock_irqrestore(q->queue_lock, flags);
984 EXPORT_SYMBOL(blk_insert_request);
986 static void part_round_stats_single(int cpu, struct hd_struct *part,
989 if (now == part->stamp)
992 if (part_in_flight(part)) {
993 __part_stat_add(cpu, part, time_in_queue,
994 part_in_flight(part) * (now - part->stamp));
995 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1001 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1002 * @cpu: cpu number for stats access
1003 * @part: target partition
1005 * The average IO queue length and utilisation statistics are maintained
1006 * by observing the current state of the queue length and the amount of
1007 * time it has been in this state for.
1009 * Normally, that accounting is done on IO completion, but that can result
1010 * in more than a second's worth of IO being accounted for within any one
1011 * second, leading to >100% utilisation. To deal with that, we call this
1012 * function to do a round-off before returning the results when reading
1013 * /proc/diskstats. This accounts immediately for all queue usage up to
1014 * the current jiffies and restarts the counters again.
1016 void part_round_stats(int cpu, struct hd_struct *part)
1018 unsigned long now = jiffies;
1021 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1022 part_round_stats_single(cpu, part, now);
1024 EXPORT_SYMBOL_GPL(part_round_stats);
1027 * queue lock must be held
1029 void __blk_put_request(struct request_queue *q, struct request *req)
1033 if (unlikely(--req->ref_count))
1036 elv_completed_request(q, req);
1038 /* this is a bio leak */
1039 WARN_ON(req->bio != NULL);
1042 * Request may not have originated from ll_rw_blk. if not,
1043 * it didn't come out of our reserved rq pools
1045 if (req->cmd_flags & REQ_ALLOCED) {
1046 int is_sync = rq_is_sync(req) != 0;
1047 int priv = req->cmd_flags & REQ_ELVPRIV;
1049 BUG_ON(!list_empty(&req->queuelist));
1050 BUG_ON(!hlist_unhashed(&req->hash));
1052 blk_free_request(q, req);
1053 freed_request(q, is_sync, priv);
1056 EXPORT_SYMBOL_GPL(__blk_put_request);
1058 void blk_put_request(struct request *req)
1060 unsigned long flags;
1061 struct request_queue *q = req->q;
1063 spin_lock_irqsave(q->queue_lock, flags);
1064 __blk_put_request(q, req);
1065 spin_unlock_irqrestore(q->queue_lock, flags);
1067 EXPORT_SYMBOL(blk_put_request);
1070 * blk_add_request_payload - add a payload to a request
1071 * @rq: request to update
1072 * @page: page backing the payload
1073 * @len: length of the payload.
1075 * This allows to later add a payload to an already submitted request by
1076 * a block driver. The driver needs to take care of freeing the payload
1079 * Note that this is a quite horrible hack and nothing but handling of
1080 * discard requests should ever use it.
1082 void blk_add_request_payload(struct request *rq, struct page *page,
1085 struct bio *bio = rq->bio;
1087 bio->bi_io_vec->bv_page = page;
1088 bio->bi_io_vec->bv_offset = 0;
1089 bio->bi_io_vec->bv_len = len;
1093 bio->bi_phys_segments = 1;
1095 rq->__data_len = rq->resid_len = len;
1096 rq->nr_phys_segments = 1;
1097 rq->buffer = bio_data(bio);
1099 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1101 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1104 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1107 * Debug stuff, kill later
1109 if (!rq_mergeable(req)) {
1110 blk_dump_rq_flags(req, "back");
1114 if (!ll_back_merge_fn(q, req, bio))
1117 trace_block_bio_backmerge(q, bio);
1119 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1120 blk_rq_set_mixed_merge(req);
1122 req->biotail->bi_next = bio;
1124 req->__data_len += bio->bi_size;
1125 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1127 drive_stat_acct(req, 0);
1131 static bool bio_attempt_front_merge(struct request_queue *q,
1132 struct request *req, struct bio *bio)
1134 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1138 * Debug stuff, kill later
1140 if (!rq_mergeable(req)) {
1141 blk_dump_rq_flags(req, "front");
1145 if (!ll_front_merge_fn(q, req, bio))
1148 trace_block_bio_frontmerge(q, bio);
1150 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1151 blk_rq_set_mixed_merge(req);
1153 sector = bio->bi_sector;
1155 bio->bi_next = req->bio;
1159 * may not be valid. if the low level driver said
1160 * it didn't need a bounce buffer then it better
1161 * not touch req->buffer either...
1163 req->buffer = bio_data(bio);
1164 req->__sector = bio->bi_sector;
1165 req->__data_len += bio->bi_size;
1166 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1168 drive_stat_acct(req, 0);
1173 * Attempts to merge with the plugged list in the current process. Returns
1174 * true if merge was successful, otherwise false.
1176 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1179 struct blk_plug *plug;
1187 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1193 el_ret = elv_try_merge(rq, bio);
1194 if (el_ret == ELEVATOR_BACK_MERGE) {
1195 ret = bio_attempt_back_merge(q, rq, bio);
1198 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1199 ret = bio_attempt_front_merge(q, rq, bio);
1208 void init_request_from_bio(struct request *req, struct bio *bio)
1210 req->cpu = bio->bi_comp_cpu;
1211 req->cmd_type = REQ_TYPE_FS;
1213 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1214 if (bio->bi_rw & REQ_RAHEAD)
1215 req->cmd_flags |= REQ_FAILFAST_MASK;
1218 req->__sector = bio->bi_sector;
1219 req->ioprio = bio_prio(bio);
1220 blk_rq_bio_prep(req->q, req, bio);
1223 static int __make_request(struct request_queue *q, struct bio *bio)
1225 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1226 struct blk_plug *plug;
1227 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1228 struct request *req;
1231 * low level driver can indicate that it wants pages above a
1232 * certain limit bounced to low memory (ie for highmem, or even
1233 * ISA dma in theory)
1235 blk_queue_bounce(q, &bio);
1237 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1238 spin_lock_irq(q->queue_lock);
1239 where = ELEVATOR_INSERT_FLUSH;
1244 * Check if we can merge with the plugged list before grabbing
1247 if (attempt_plug_merge(current, q, bio))
1250 spin_lock_irq(q->queue_lock);
1252 el_ret = elv_merge(q, &req, bio);
1253 if (el_ret == ELEVATOR_BACK_MERGE) {
1254 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1255 if (bio_attempt_back_merge(q, req, bio)) {
1256 if (!attempt_back_merge(q, req))
1257 elv_merged_request(q, req, el_ret);
1260 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1261 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1262 if (bio_attempt_front_merge(q, req, bio)) {
1263 if (!attempt_front_merge(q, req))
1264 elv_merged_request(q, req, el_ret);
1271 * This sync check and mask will be re-done in init_request_from_bio(),
1272 * but we need to set it earlier to expose the sync flag to the
1273 * rq allocator and io schedulers.
1275 rw_flags = bio_data_dir(bio);
1277 rw_flags |= REQ_SYNC;
1280 * Grab a free request. This is might sleep but can not fail.
1281 * Returns with the queue unlocked.
1283 req = get_request_wait(q, rw_flags, bio);
1286 * After dropping the lock and possibly sleeping here, our request
1287 * may now be mergeable after it had proven unmergeable (above).
1288 * We don't worry about that case for efficiency. It won't happen
1289 * often, and the elevators are able to handle it.
1291 init_request_from_bio(req, bio);
1293 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1294 bio_flagged(bio, BIO_CPU_AFFINE)) {
1295 req->cpu = blk_cpu_to_group(get_cpu());
1299 plug = current->plug;
1302 * If this is the first request added after a plug, fire
1303 * of a plug trace. If others have been added before, check
1304 * if we have multiple devices in this plug. If so, make a
1305 * note to sort the list before dispatch.
1307 if (list_empty(&plug->list))
1308 trace_block_plug(q);
1309 else if (!plug->should_sort) {
1310 struct request *__rq;
1312 __rq = list_entry_rq(plug->list.prev);
1314 plug->should_sort = 1;
1317 * Debug flag, kill later
1319 req->cmd_flags |= REQ_ON_PLUG;
1320 list_add_tail(&req->queuelist, &plug->list);
1321 drive_stat_acct(req, 1);
1323 spin_lock_irq(q->queue_lock);
1324 add_acct_request(q, req, where);
1325 __blk_run_queue(q, false);
1327 spin_unlock_irq(q->queue_lock);
1334 * If bio->bi_dev is a partition, remap the location
1336 static inline void blk_partition_remap(struct bio *bio)
1338 struct block_device *bdev = bio->bi_bdev;
1340 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1341 struct hd_struct *p = bdev->bd_part;
1343 bio->bi_sector += p->start_sect;
1344 bio->bi_bdev = bdev->bd_contains;
1346 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1348 bio->bi_sector - p->start_sect);
1352 static void handle_bad_sector(struct bio *bio)
1354 char b[BDEVNAME_SIZE];
1356 printk(KERN_INFO "attempt to access beyond end of device\n");
1357 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1358 bdevname(bio->bi_bdev, b),
1360 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1361 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1363 set_bit(BIO_EOF, &bio->bi_flags);
1366 #ifdef CONFIG_FAIL_MAKE_REQUEST
1368 static DECLARE_FAULT_ATTR(fail_make_request);
1370 static int __init setup_fail_make_request(char *str)
1372 return setup_fault_attr(&fail_make_request, str);
1374 __setup("fail_make_request=", setup_fail_make_request);
1376 static int should_fail_request(struct bio *bio)
1378 struct hd_struct *part = bio->bi_bdev->bd_part;
1380 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1381 return should_fail(&fail_make_request, bio->bi_size);
1386 static int __init fail_make_request_debugfs(void)
1388 return init_fault_attr_dentries(&fail_make_request,
1389 "fail_make_request");
1392 late_initcall(fail_make_request_debugfs);
1394 #else /* CONFIG_FAIL_MAKE_REQUEST */
1396 static inline int should_fail_request(struct bio *bio)
1401 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1404 * Check whether this bio extends beyond the end of the device.
1406 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1413 /* Test device or partition size, when known. */
1414 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1416 sector_t sector = bio->bi_sector;
1418 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1420 * This may well happen - the kernel calls bread()
1421 * without checking the size of the device, e.g., when
1422 * mounting a device.
1424 handle_bad_sector(bio);
1433 * generic_make_request - hand a buffer to its device driver for I/O
1434 * @bio: The bio describing the location in memory and on the device.
1436 * generic_make_request() is used to make I/O requests of block
1437 * devices. It is passed a &struct bio, which describes the I/O that needs
1440 * generic_make_request() does not return any status. The
1441 * success/failure status of the request, along with notification of
1442 * completion, is delivered asynchronously through the bio->bi_end_io
1443 * function described (one day) else where.
1445 * The caller of generic_make_request must make sure that bi_io_vec
1446 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1447 * set to describe the device address, and the
1448 * bi_end_io and optionally bi_private are set to describe how
1449 * completion notification should be signaled.
1451 * generic_make_request and the drivers it calls may use bi_next if this
1452 * bio happens to be merged with someone else, and may change bi_dev and
1453 * bi_sector for remaps as it sees fit. So the values of these fields
1454 * should NOT be depended on after the call to generic_make_request.
1456 static inline void __generic_make_request(struct bio *bio)
1458 struct request_queue *q;
1459 sector_t old_sector;
1460 int ret, nr_sectors = bio_sectors(bio);
1466 if (bio_check_eod(bio, nr_sectors))
1470 * Resolve the mapping until finished. (drivers are
1471 * still free to implement/resolve their own stacking
1472 * by explicitly returning 0)
1474 * NOTE: we don't repeat the blk_size check for each new device.
1475 * Stacking drivers are expected to know what they are doing.
1480 char b[BDEVNAME_SIZE];
1482 q = bdev_get_queue(bio->bi_bdev);
1485 "generic_make_request: Trying to access "
1486 "nonexistent block-device %s (%Lu)\n",
1487 bdevname(bio->bi_bdev, b),
1488 (long long) bio->bi_sector);
1492 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1493 nr_sectors > queue_max_hw_sectors(q))) {
1494 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1495 bdevname(bio->bi_bdev, b),
1497 queue_max_hw_sectors(q));
1501 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1504 if (should_fail_request(bio))
1508 * If this device has partitions, remap block n
1509 * of partition p to block n+start(p) of the disk.
1511 blk_partition_remap(bio);
1513 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1516 if (old_sector != -1)
1517 trace_block_bio_remap(q, bio, old_dev, old_sector);
1519 old_sector = bio->bi_sector;
1520 old_dev = bio->bi_bdev->bd_dev;
1522 if (bio_check_eod(bio, nr_sectors))
1526 * Filter flush bio's early so that make_request based
1527 * drivers without flush support don't have to worry
1530 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1531 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1538 if ((bio->bi_rw & REQ_DISCARD) &&
1539 (!blk_queue_discard(q) ||
1540 ((bio->bi_rw & REQ_SECURE) &&
1541 !blk_queue_secdiscard(q)))) {
1546 blk_throtl_bio(q, &bio);
1549 * If bio = NULL, bio has been throttled and will be submitted
1555 trace_block_bio_queue(q, bio);
1557 ret = q->make_request_fn(q, bio);
1563 bio_endio(bio, err);
1567 * We only want one ->make_request_fn to be active at a time,
1568 * else stack usage with stacked devices could be a problem.
1569 * So use current->bio_list to keep a list of requests
1570 * submited by a make_request_fn function.
1571 * current->bio_list is also used as a flag to say if
1572 * generic_make_request is currently active in this task or not.
1573 * If it is NULL, then no make_request is active. If it is non-NULL,
1574 * then a make_request is active, and new requests should be added
1577 void generic_make_request(struct bio *bio)
1579 struct bio_list bio_list_on_stack;
1581 if (current->bio_list) {
1582 /* make_request is active */
1583 bio_list_add(current->bio_list, bio);
1586 /* following loop may be a bit non-obvious, and so deserves some
1588 * Before entering the loop, bio->bi_next is NULL (as all callers
1589 * ensure that) so we have a list with a single bio.
1590 * We pretend that we have just taken it off a longer list, so
1591 * we assign bio_list to a pointer to the bio_list_on_stack,
1592 * thus initialising the bio_list of new bios to be
1593 * added. __generic_make_request may indeed add some more bios
1594 * through a recursive call to generic_make_request. If it
1595 * did, we find a non-NULL value in bio_list and re-enter the loop
1596 * from the top. In this case we really did just take the bio
1597 * of the top of the list (no pretending) and so remove it from
1598 * bio_list, and call into __generic_make_request again.
1600 * The loop was structured like this to make only one call to
1601 * __generic_make_request (which is important as it is large and
1602 * inlined) and to keep the structure simple.
1604 BUG_ON(bio->bi_next);
1605 bio_list_init(&bio_list_on_stack);
1606 current->bio_list = &bio_list_on_stack;
1608 __generic_make_request(bio);
1609 bio = bio_list_pop(current->bio_list);
1611 current->bio_list = NULL; /* deactivate */
1613 EXPORT_SYMBOL(generic_make_request);
1616 * submit_bio - submit a bio to the block device layer for I/O
1617 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1618 * @bio: The &struct bio which describes the I/O
1620 * submit_bio() is very similar in purpose to generic_make_request(), and
1621 * uses that function to do most of the work. Both are fairly rough
1622 * interfaces; @bio must be presetup and ready for I/O.
1625 void submit_bio(int rw, struct bio *bio)
1627 int count = bio_sectors(bio);
1632 * If it's a regular read/write or a barrier with data attached,
1633 * go through the normal accounting stuff before submission.
1635 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1637 count_vm_events(PGPGOUT, count);
1639 task_io_account_read(bio->bi_size);
1640 count_vm_events(PGPGIN, count);
1643 if (unlikely(block_dump)) {
1644 char b[BDEVNAME_SIZE];
1645 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1646 current->comm, task_pid_nr(current),
1647 (rw & WRITE) ? "WRITE" : "READ",
1648 (unsigned long long)bio->bi_sector,
1649 bdevname(bio->bi_bdev, b),
1654 generic_make_request(bio);
1656 EXPORT_SYMBOL(submit_bio);
1659 * blk_rq_check_limits - Helper function to check a request for the queue limit
1661 * @rq: the request being checked
1664 * @rq may have been made based on weaker limitations of upper-level queues
1665 * in request stacking drivers, and it may violate the limitation of @q.
1666 * Since the block layer and the underlying device driver trust @rq
1667 * after it is inserted to @q, it should be checked against @q before
1668 * the insertion using this generic function.
1670 * This function should also be useful for request stacking drivers
1671 * in some cases below, so export this function.
1672 * Request stacking drivers like request-based dm may change the queue
1673 * limits while requests are in the queue (e.g. dm's table swapping).
1674 * Such request stacking drivers should check those requests agaist
1675 * the new queue limits again when they dispatch those requests,
1676 * although such checkings are also done against the old queue limits
1677 * when submitting requests.
1679 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1681 if (rq->cmd_flags & REQ_DISCARD)
1684 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1685 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1686 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1691 * queue's settings related to segment counting like q->bounce_pfn
1692 * may differ from that of other stacking queues.
1693 * Recalculate it to check the request correctly on this queue's
1696 blk_recalc_rq_segments(rq);
1697 if (rq->nr_phys_segments > queue_max_segments(q)) {
1698 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1704 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1707 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1708 * @q: the queue to submit the request
1709 * @rq: the request being queued
1711 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1713 unsigned long flags;
1715 if (blk_rq_check_limits(q, rq))
1718 #ifdef CONFIG_FAIL_MAKE_REQUEST
1719 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1720 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1724 spin_lock_irqsave(q->queue_lock, flags);
1727 * Submitting request must be dequeued before calling this function
1728 * because it will be linked to another request_queue
1730 BUG_ON(blk_queued_rq(rq));
1732 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1733 spin_unlock_irqrestore(q->queue_lock, flags);
1737 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1740 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1741 * @rq: request to examine
1744 * A request could be merge of IOs which require different failure
1745 * handling. This function determines the number of bytes which
1746 * can be failed from the beginning of the request without
1747 * crossing into area which need to be retried further.
1750 * The number of bytes to fail.
1753 * queue_lock must be held.
1755 unsigned int blk_rq_err_bytes(const struct request *rq)
1757 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1758 unsigned int bytes = 0;
1761 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1762 return blk_rq_bytes(rq);
1765 * Currently the only 'mixing' which can happen is between
1766 * different fastfail types. We can safely fail portions
1767 * which have all the failfast bits that the first one has -
1768 * the ones which are at least as eager to fail as the first
1771 for (bio = rq->bio; bio; bio = bio->bi_next) {
1772 if ((bio->bi_rw & ff) != ff)
1774 bytes += bio->bi_size;
1777 /* this could lead to infinite loop */
1778 BUG_ON(blk_rq_bytes(rq) && !bytes);
1781 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1783 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1785 if (blk_do_io_stat(req)) {
1786 const int rw = rq_data_dir(req);
1787 struct hd_struct *part;
1790 cpu = part_stat_lock();
1792 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1797 static void blk_account_io_done(struct request *req)
1800 * Account IO completion. flush_rq isn't accounted as a
1801 * normal IO on queueing nor completion. Accounting the
1802 * containing request is enough.
1804 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1805 unsigned long duration = jiffies - req->start_time;
1806 const int rw = rq_data_dir(req);
1807 struct hd_struct *part;
1810 cpu = part_stat_lock();
1813 part_stat_inc(cpu, part, ios[rw]);
1814 part_stat_add(cpu, part, ticks[rw], duration);
1815 part_round_stats(cpu, part);
1816 part_dec_in_flight(part, rw);
1818 hd_struct_put(part);
1824 * blk_peek_request - peek at the top of a request queue
1825 * @q: request queue to peek at
1828 * Return the request at the top of @q. The returned request
1829 * should be started using blk_start_request() before LLD starts
1833 * Pointer to the request at the top of @q if available. Null
1837 * queue_lock must be held.
1839 struct request *blk_peek_request(struct request_queue *q)
1844 while ((rq = __elv_next_request(q)) != NULL) {
1845 if (!(rq->cmd_flags & REQ_STARTED)) {
1847 * This is the first time the device driver
1848 * sees this request (possibly after
1849 * requeueing). Notify IO scheduler.
1851 if (rq->cmd_flags & REQ_SORTED)
1852 elv_activate_rq(q, rq);
1855 * just mark as started even if we don't start
1856 * it, a request that has been delayed should
1857 * not be passed by new incoming requests
1859 rq->cmd_flags |= REQ_STARTED;
1860 trace_block_rq_issue(q, rq);
1863 if (!q->boundary_rq || q->boundary_rq == rq) {
1864 q->end_sector = rq_end_sector(rq);
1865 q->boundary_rq = NULL;
1868 if (rq->cmd_flags & REQ_DONTPREP)
1871 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1873 * make sure space for the drain appears we
1874 * know we can do this because max_hw_segments
1875 * has been adjusted to be one fewer than the
1878 rq->nr_phys_segments++;
1884 ret = q->prep_rq_fn(q, rq);
1885 if (ret == BLKPREP_OK) {
1887 } else if (ret == BLKPREP_DEFER) {
1889 * the request may have been (partially) prepped.
1890 * we need to keep this request in the front to
1891 * avoid resource deadlock. REQ_STARTED will
1892 * prevent other fs requests from passing this one.
1894 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1895 !(rq->cmd_flags & REQ_DONTPREP)) {
1897 * remove the space for the drain we added
1898 * so that we don't add it again
1900 --rq->nr_phys_segments;
1905 } else if (ret == BLKPREP_KILL) {
1906 rq->cmd_flags |= REQ_QUIET;
1908 * Mark this request as started so we don't trigger
1909 * any debug logic in the end I/O path.
1911 blk_start_request(rq);
1912 __blk_end_request_all(rq, -EIO);
1914 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1921 EXPORT_SYMBOL(blk_peek_request);
1923 void blk_dequeue_request(struct request *rq)
1925 struct request_queue *q = rq->q;
1927 BUG_ON(list_empty(&rq->queuelist));
1928 BUG_ON(ELV_ON_HASH(rq));
1930 list_del_init(&rq->queuelist);
1933 * the time frame between a request being removed from the lists
1934 * and to it is freed is accounted as io that is in progress at
1937 if (blk_account_rq(rq)) {
1938 q->in_flight[rq_is_sync(rq)]++;
1939 set_io_start_time_ns(rq);
1944 * blk_start_request - start request processing on the driver
1945 * @req: request to dequeue
1948 * Dequeue @req and start timeout timer on it. This hands off the
1949 * request to the driver.
1951 * Block internal functions which don't want to start timer should
1952 * call blk_dequeue_request().
1955 * queue_lock must be held.
1957 void blk_start_request(struct request *req)
1959 blk_dequeue_request(req);
1962 * We are now handing the request to the hardware, initialize
1963 * resid_len to full count and add the timeout handler.
1965 req->resid_len = blk_rq_bytes(req);
1966 if (unlikely(blk_bidi_rq(req)))
1967 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1971 EXPORT_SYMBOL(blk_start_request);
1974 * blk_fetch_request - fetch a request from a request queue
1975 * @q: request queue to fetch a request from
1978 * Return the request at the top of @q. The request is started on
1979 * return and LLD can start processing it immediately.
1982 * Pointer to the request at the top of @q if available. Null
1986 * queue_lock must be held.
1988 struct request *blk_fetch_request(struct request_queue *q)
1992 rq = blk_peek_request(q);
1994 blk_start_request(rq);
1997 EXPORT_SYMBOL(blk_fetch_request);
2000 * blk_update_request - Special helper function for request stacking drivers
2001 * @req: the request being processed
2002 * @error: %0 for success, < %0 for error
2003 * @nr_bytes: number of bytes to complete @req
2006 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2007 * the request structure even if @req doesn't have leftover.
2008 * If @req has leftover, sets it up for the next range of segments.
2010 * This special helper function is only for request stacking drivers
2011 * (e.g. request-based dm) so that they can handle partial completion.
2012 * Actual device drivers should use blk_end_request instead.
2014 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2015 * %false return from this function.
2018 * %false - this request doesn't have any more data
2019 * %true - this request has more data
2021 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2023 int total_bytes, bio_nbytes, next_idx = 0;
2029 trace_block_rq_complete(req->q, req);
2032 * For fs requests, rq is just carrier of independent bio's
2033 * and each partial completion should be handled separately.
2034 * Reset per-request error on each partial completion.
2036 * TODO: tj: This is too subtle. It would be better to let
2037 * low level drivers do what they see fit.
2039 if (req->cmd_type == REQ_TYPE_FS)
2042 if (error && req->cmd_type == REQ_TYPE_FS &&
2043 !(req->cmd_flags & REQ_QUIET)) {
2048 error_type = "recoverable transport";
2051 error_type = "critical target";
2054 error_type = "critical nexus";
2061 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2062 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2063 (unsigned long long)blk_rq_pos(req));
2066 blk_account_io_completion(req, nr_bytes);
2068 total_bytes = bio_nbytes = 0;
2069 while ((bio = req->bio) != NULL) {
2072 if (nr_bytes >= bio->bi_size) {
2073 req->bio = bio->bi_next;
2074 nbytes = bio->bi_size;
2075 req_bio_endio(req, bio, nbytes, error);
2079 int idx = bio->bi_idx + next_idx;
2081 if (unlikely(idx >= bio->bi_vcnt)) {
2082 blk_dump_rq_flags(req, "__end_that");
2083 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2084 __func__, idx, bio->bi_vcnt);
2088 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2089 BIO_BUG_ON(nbytes > bio->bi_size);
2092 * not a complete bvec done
2094 if (unlikely(nbytes > nr_bytes)) {
2095 bio_nbytes += nr_bytes;
2096 total_bytes += nr_bytes;
2101 * advance to the next vector
2104 bio_nbytes += nbytes;
2107 total_bytes += nbytes;
2113 * end more in this run, or just return 'not-done'
2115 if (unlikely(nr_bytes <= 0))
2125 * Reset counters so that the request stacking driver
2126 * can find how many bytes remain in the request
2129 req->__data_len = 0;
2134 * if the request wasn't completed, update state
2137 req_bio_endio(req, bio, bio_nbytes, error);
2138 bio->bi_idx += next_idx;
2139 bio_iovec(bio)->bv_offset += nr_bytes;
2140 bio_iovec(bio)->bv_len -= nr_bytes;
2143 req->__data_len -= total_bytes;
2144 req->buffer = bio_data(req->bio);
2146 /* update sector only for requests with clear definition of sector */
2147 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2148 req->__sector += total_bytes >> 9;
2150 /* mixed attributes always follow the first bio */
2151 if (req->cmd_flags & REQ_MIXED_MERGE) {
2152 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2153 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2157 * If total number of sectors is less than the first segment
2158 * size, something has gone terribly wrong.
2160 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2161 blk_dump_rq_flags(req, "request botched");
2162 req->__data_len = blk_rq_cur_bytes(req);
2165 /* recalculate the number of segments */
2166 blk_recalc_rq_segments(req);
2170 EXPORT_SYMBOL_GPL(blk_update_request);
2172 static bool blk_update_bidi_request(struct request *rq, int error,
2173 unsigned int nr_bytes,
2174 unsigned int bidi_bytes)
2176 if (blk_update_request(rq, error, nr_bytes))
2179 /* Bidi request must be completed as a whole */
2180 if (unlikely(blk_bidi_rq(rq)) &&
2181 blk_update_request(rq->next_rq, error, bidi_bytes))
2184 if (blk_queue_add_random(rq->q))
2185 add_disk_randomness(rq->rq_disk);
2191 * blk_unprep_request - unprepare a request
2194 * This function makes a request ready for complete resubmission (or
2195 * completion). It happens only after all error handling is complete,
2196 * so represents the appropriate moment to deallocate any resources
2197 * that were allocated to the request in the prep_rq_fn. The queue
2198 * lock is held when calling this.
2200 void blk_unprep_request(struct request *req)
2202 struct request_queue *q = req->q;
2204 req->cmd_flags &= ~REQ_DONTPREP;
2205 if (q->unprep_rq_fn)
2206 q->unprep_rq_fn(q, req);
2208 EXPORT_SYMBOL_GPL(blk_unprep_request);
2211 * queue lock must be held
2213 static void blk_finish_request(struct request *req, int error)
2215 if (blk_rq_tagged(req))
2216 blk_queue_end_tag(req->q, req);
2218 BUG_ON(blk_queued_rq(req));
2220 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2221 laptop_io_completion(&req->q->backing_dev_info);
2223 blk_delete_timer(req);
2225 if (req->cmd_flags & REQ_DONTPREP)
2226 blk_unprep_request(req);
2229 blk_account_io_done(req);
2232 req->end_io(req, error);
2234 if (blk_bidi_rq(req))
2235 __blk_put_request(req->next_rq->q, req->next_rq);
2237 __blk_put_request(req->q, req);
2242 * blk_end_bidi_request - Complete a bidi request
2243 * @rq: the request to complete
2244 * @error: %0 for success, < %0 for error
2245 * @nr_bytes: number of bytes to complete @rq
2246 * @bidi_bytes: number of bytes to complete @rq->next_rq
2249 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2250 * Drivers that supports bidi can safely call this member for any
2251 * type of request, bidi or uni. In the later case @bidi_bytes is
2255 * %false - we are done with this request
2256 * %true - still buffers pending for this request
2258 static bool blk_end_bidi_request(struct request *rq, int error,
2259 unsigned int nr_bytes, unsigned int bidi_bytes)
2261 struct request_queue *q = rq->q;
2262 unsigned long flags;
2264 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2267 spin_lock_irqsave(q->queue_lock, flags);
2268 blk_finish_request(rq, error);
2269 spin_unlock_irqrestore(q->queue_lock, flags);
2275 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2276 * @rq: the request to complete
2277 * @error: %0 for success, < %0 for error
2278 * @nr_bytes: number of bytes to complete @rq
2279 * @bidi_bytes: number of bytes to complete @rq->next_rq
2282 * Identical to blk_end_bidi_request() except that queue lock is
2283 * assumed to be locked on entry and remains so on return.
2286 * %false - we are done with this request
2287 * %true - still buffers pending for this request
2289 static bool __blk_end_bidi_request(struct request *rq, int error,
2290 unsigned int nr_bytes, unsigned int bidi_bytes)
2292 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2295 blk_finish_request(rq, error);
2301 * blk_end_request - Helper function for drivers to complete the request.
2302 * @rq: the request being processed
2303 * @error: %0 for success, < %0 for error
2304 * @nr_bytes: number of bytes to complete
2307 * Ends I/O on a number of bytes attached to @rq.
2308 * If @rq has leftover, sets it up for the next range of segments.
2311 * %false - we are done with this request
2312 * %true - still buffers pending for this request
2314 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2316 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2318 EXPORT_SYMBOL(blk_end_request);
2321 * blk_end_request_all - Helper function for drives to finish the request.
2322 * @rq: the request to finish
2323 * @error: %0 for success, < %0 for error
2326 * Completely finish @rq.
2328 void blk_end_request_all(struct request *rq, int error)
2331 unsigned int bidi_bytes = 0;
2333 if (unlikely(blk_bidi_rq(rq)))
2334 bidi_bytes = blk_rq_bytes(rq->next_rq);
2336 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2339 EXPORT_SYMBOL(blk_end_request_all);
2342 * blk_end_request_cur - Helper function to finish the current request chunk.
2343 * @rq: the request to finish the current chunk for
2344 * @error: %0 for success, < %0 for error
2347 * Complete the current consecutively mapped chunk from @rq.
2350 * %false - we are done with this request
2351 * %true - still buffers pending for this request
2353 bool blk_end_request_cur(struct request *rq, int error)
2355 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2357 EXPORT_SYMBOL(blk_end_request_cur);
2360 * blk_end_request_err - Finish a request till the next failure boundary.
2361 * @rq: the request to finish till the next failure boundary for
2362 * @error: must be negative errno
2365 * Complete @rq till the next failure boundary.
2368 * %false - we are done with this request
2369 * %true - still buffers pending for this request
2371 bool blk_end_request_err(struct request *rq, int error)
2373 WARN_ON(error >= 0);
2374 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2376 EXPORT_SYMBOL_GPL(blk_end_request_err);
2379 * __blk_end_request - Helper function for drivers to complete the request.
2380 * @rq: the request being processed
2381 * @error: %0 for success, < %0 for error
2382 * @nr_bytes: number of bytes to complete
2385 * Must be called with queue lock held unlike blk_end_request().
2388 * %false - we are done with this request
2389 * %true - still buffers pending for this request
2391 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2393 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2395 EXPORT_SYMBOL(__blk_end_request);
2398 * __blk_end_request_all - Helper function for drives to finish the request.
2399 * @rq: the request to finish
2400 * @error: %0 for success, < %0 for error
2403 * Completely finish @rq. Must be called with queue lock held.
2405 void __blk_end_request_all(struct request *rq, int error)
2408 unsigned int bidi_bytes = 0;
2410 if (unlikely(blk_bidi_rq(rq)))
2411 bidi_bytes = blk_rq_bytes(rq->next_rq);
2413 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2416 EXPORT_SYMBOL(__blk_end_request_all);
2419 * __blk_end_request_cur - Helper function to finish the current request chunk.
2420 * @rq: the request to finish the current chunk for
2421 * @error: %0 for success, < %0 for error
2424 * Complete the current consecutively mapped chunk from @rq. Must
2425 * be called with queue lock held.
2428 * %false - we are done with this request
2429 * %true - still buffers pending for this request
2431 bool __blk_end_request_cur(struct request *rq, int error)
2433 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2435 EXPORT_SYMBOL(__blk_end_request_cur);
2438 * __blk_end_request_err - Finish a request till the next failure boundary.
2439 * @rq: the request to finish till the next failure boundary for
2440 * @error: must be negative errno
2443 * Complete @rq till the next failure boundary. Must be called
2444 * with queue lock held.
2447 * %false - we are done with this request
2448 * %true - still buffers pending for this request
2450 bool __blk_end_request_err(struct request *rq, int error)
2452 WARN_ON(error >= 0);
2453 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2455 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2457 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2460 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2461 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2463 if (bio_has_data(bio)) {
2464 rq->nr_phys_segments = bio_phys_segments(q, bio);
2465 rq->buffer = bio_data(bio);
2467 rq->__data_len = bio->bi_size;
2468 rq->bio = rq->biotail = bio;
2471 rq->rq_disk = bio->bi_bdev->bd_disk;
2474 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2476 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2477 * @rq: the request to be flushed
2480 * Flush all pages in @rq.
2482 void rq_flush_dcache_pages(struct request *rq)
2484 struct req_iterator iter;
2485 struct bio_vec *bvec;
2487 rq_for_each_segment(bvec, rq, iter)
2488 flush_dcache_page(bvec->bv_page);
2490 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2494 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2495 * @q : the queue of the device being checked
2498 * Check if underlying low-level drivers of a device are busy.
2499 * If the drivers want to export their busy state, they must set own
2500 * exporting function using blk_queue_lld_busy() first.
2502 * Basically, this function is used only by request stacking drivers
2503 * to stop dispatching requests to underlying devices when underlying
2504 * devices are busy. This behavior helps more I/O merging on the queue
2505 * of the request stacking driver and prevents I/O throughput regression
2506 * on burst I/O load.
2509 * 0 - Not busy (The request stacking driver should dispatch request)
2510 * 1 - Busy (The request stacking driver should stop dispatching request)
2512 int blk_lld_busy(struct request_queue *q)
2515 return q->lld_busy_fn(q);
2519 EXPORT_SYMBOL_GPL(blk_lld_busy);
2522 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2523 * @rq: the clone request to be cleaned up
2526 * Free all bios in @rq for a cloned request.
2528 void blk_rq_unprep_clone(struct request *rq)
2532 while ((bio = rq->bio) != NULL) {
2533 rq->bio = bio->bi_next;
2538 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2541 * Copy attributes of the original request to the clone request.
2542 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2544 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2546 dst->cpu = src->cpu;
2547 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2548 dst->cmd_type = src->cmd_type;
2549 dst->__sector = blk_rq_pos(src);
2550 dst->__data_len = blk_rq_bytes(src);
2551 dst->nr_phys_segments = src->nr_phys_segments;
2552 dst->ioprio = src->ioprio;
2553 dst->extra_len = src->extra_len;
2557 * blk_rq_prep_clone - Helper function to setup clone request
2558 * @rq: the request to be setup
2559 * @rq_src: original request to be cloned
2560 * @bs: bio_set that bios for clone are allocated from
2561 * @gfp_mask: memory allocation mask for bio
2562 * @bio_ctr: setup function to be called for each clone bio.
2563 * Returns %0 for success, non %0 for failure.
2564 * @data: private data to be passed to @bio_ctr
2567 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2568 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2569 * are not copied, and copying such parts is the caller's responsibility.
2570 * Also, pages which the original bios are pointing to are not copied
2571 * and the cloned bios just point same pages.
2572 * So cloned bios must be completed before original bios, which means
2573 * the caller must complete @rq before @rq_src.
2575 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2576 struct bio_set *bs, gfp_t gfp_mask,
2577 int (*bio_ctr)(struct bio *, struct bio *, void *),
2580 struct bio *bio, *bio_src;
2585 blk_rq_init(NULL, rq);
2587 __rq_for_each_bio(bio_src, rq_src) {
2588 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2592 __bio_clone(bio, bio_src);
2594 if (bio_integrity(bio_src) &&
2595 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2598 if (bio_ctr && bio_ctr(bio, bio_src, data))
2602 rq->biotail->bi_next = bio;
2605 rq->bio = rq->biotail = bio;
2608 __blk_rq_prep_clone(rq, rq_src);
2615 blk_rq_unprep_clone(rq);
2619 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2621 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2623 return queue_work(kblockd_workqueue, work);
2625 EXPORT_SYMBOL(kblockd_schedule_work);
2627 int kblockd_schedule_delayed_work(struct request_queue *q,
2628 struct delayed_work *dwork, unsigned long delay)
2630 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2632 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2634 #define PLUG_MAGIC 0x91827364
2636 void blk_start_plug(struct blk_plug *plug)
2638 struct task_struct *tsk = current;
2640 plug->magic = PLUG_MAGIC;
2641 INIT_LIST_HEAD(&plug->list);
2642 INIT_LIST_HEAD(&plug->cb_list);
2643 plug->should_sort = 0;
2646 * If this is a nested plug, don't actually assign it. It will be
2647 * flushed on its own.
2651 * Store ordering should not be needed here, since a potential
2652 * preempt will imply a full memory barrier
2657 EXPORT_SYMBOL(blk_start_plug);
2659 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2661 struct request *rqa = container_of(a, struct request, queuelist);
2662 struct request *rqb = container_of(b, struct request, queuelist);
2664 return !(rqa->q <= rqb->q);
2668 * If 'from_schedule' is true, then postpone the dispatch of requests
2669 * until a safe kblockd context. We due this to avoid accidental big
2670 * additional stack usage in driver dispatch, in places where the originally
2671 * plugger did not intend it.
2673 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2675 __releases(q->queue_lock)
2677 trace_block_unplug(q, depth, !from_schedule);
2680 * If we are punting this to kblockd, then we can safely drop
2681 * the queue_lock before waking kblockd (which needs to take
2684 if (from_schedule) {
2685 spin_unlock(q->queue_lock);
2686 __blk_run_queue(q, true);
2688 __blk_run_queue(q, false);
2689 spin_unlock(q->queue_lock);
2694 static void flush_plug_callbacks(struct blk_plug *plug)
2696 LIST_HEAD(callbacks);
2698 if (list_empty(&plug->cb_list))
2701 list_splice_init(&plug->cb_list, &callbacks);
2703 while (!list_empty(&callbacks)) {
2704 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2707 list_del(&cb->list);
2712 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2714 struct request_queue *q;
2715 unsigned long flags;
2720 BUG_ON(plug->magic != PLUG_MAGIC);
2722 flush_plug_callbacks(plug);
2723 if (list_empty(&plug->list))
2726 list_splice_init(&plug->list, &list);
2728 if (plug->should_sort) {
2729 list_sort(NULL, &list, plug_rq_cmp);
2730 plug->should_sort = 0;
2737 * Save and disable interrupts here, to avoid doing it for every
2738 * queue lock we have to take.
2740 local_irq_save(flags);
2741 while (!list_empty(&list)) {
2742 rq = list_entry_rq(list.next);
2743 list_del_init(&rq->queuelist);
2744 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2748 * This drops the queue lock
2751 queue_unplugged(q, depth, from_schedule);
2754 spin_lock(q->queue_lock);
2756 rq->cmd_flags &= ~REQ_ON_PLUG;
2759 * rq is already accounted, so use raw insert
2761 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2762 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2764 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2770 * This drops the queue lock
2773 queue_unplugged(q, depth, from_schedule);
2775 local_irq_restore(flags);
2777 EXPORT_SYMBOL(blk_flush_plug_list);
2779 void blk_finish_plug(struct blk_plug *plug)
2781 blk_flush_plug_list(plug, false);
2783 if (plug == current->plug)
2784 current->plug = NULL;
2786 EXPORT_SYMBOL(blk_finish_plug);
2788 int __init blk_dev_init(void)
2790 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2791 sizeof(((struct request *)0)->cmd_flags));
2793 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2794 kblockd_workqueue = alloc_workqueue("kblockd",
2795 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2796 if (!kblockd_workqueue)
2797 panic("Failed to create kblockd\n");
2799 request_cachep = kmem_cache_create("blkdev_requests",
2800 sizeof(struct request), 0, SLAB_PANIC, NULL);
2802 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2803 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);