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>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/block.h>
36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
40 static int __make_request(struct request_queue *q, struct bio *bio);
43 * For the allocated request tables
45 static struct kmem_cache *request_cachep;
48 * For queue allocation
50 struct kmem_cache *blk_requestq_cachep;
53 * Controlling structure to kblockd
55 static struct workqueue_struct *kblockd_workqueue;
57 static void drive_stat_acct(struct request *rq, int new_io)
59 struct hd_struct *part;
60 int rw = rq_data_dir(rq);
63 if (!blk_do_io_stat(rq))
66 cpu = part_stat_lock();
70 part_stat_inc(cpu, part, merges[rw]);
72 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
73 part_round_stats(cpu, part);
74 part_inc_in_flight(part, rw);
81 void blk_queue_congestion_threshold(struct request_queue *q)
85 nr = q->nr_requests - (q->nr_requests / 8) + 1;
86 if (nr > q->nr_requests)
88 q->nr_congestion_on = nr;
90 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
93 q->nr_congestion_off = nr;
97 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
100 * Locates the passed device's request queue and returns the address of its
103 * Will return NULL if the request queue cannot be located.
105 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
107 struct backing_dev_info *ret = NULL;
108 struct request_queue *q = bdev_get_queue(bdev);
111 ret = &q->backing_dev_info;
114 EXPORT_SYMBOL(blk_get_backing_dev_info);
116 void blk_rq_init(struct request_queue *q, struct request *rq)
118 memset(rq, 0, sizeof(*rq));
120 INIT_LIST_HEAD(&rq->queuelist);
121 INIT_LIST_HEAD(&rq->timeout_list);
124 rq->__sector = (sector_t) -1;
125 INIT_HLIST_NODE(&rq->hash);
126 RB_CLEAR_NODE(&rq->rb_node);
128 rq->cmd_len = BLK_MAX_CDB;
131 rq->start_time = jiffies;
132 set_start_time_ns(rq);
135 EXPORT_SYMBOL(blk_rq_init);
137 static void req_bio_endio(struct request *rq, struct bio *bio,
138 unsigned int nbytes, int error)
140 struct request_queue *q = rq->q;
142 if (&q->bar_rq != rq) {
144 clear_bit(BIO_UPTODATE, &bio->bi_flags);
145 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
148 if (unlikely(nbytes > bio->bi_size)) {
149 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
150 __func__, nbytes, bio->bi_size);
151 nbytes = bio->bi_size;
154 if (unlikely(rq->cmd_flags & REQ_QUIET))
155 set_bit(BIO_QUIET, &bio->bi_flags);
157 bio->bi_size -= nbytes;
158 bio->bi_sector += (nbytes >> 9);
160 if (bio_integrity(bio))
161 bio_integrity_advance(bio, nbytes);
163 if (bio->bi_size == 0)
164 bio_endio(bio, error);
168 * Okay, this is the barrier request in progress, just
171 if (error && !q->orderr)
176 void blk_dump_rq_flags(struct request *rq, char *msg)
180 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
181 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
185 (unsigned long long)blk_rq_pos(rq),
186 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
187 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
188 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
190 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
191 printk(KERN_INFO " cdb: ");
192 for (bit = 0; bit < BLK_MAX_CDB; bit++)
193 printk("%02x ", rq->cmd[bit]);
197 EXPORT_SYMBOL(blk_dump_rq_flags);
200 * "plug" the device if there are no outstanding requests: this will
201 * force the transfer to start only after we have put all the requests
204 * This is called with interrupts off and no requests on the queue and
205 * with the queue lock held.
207 void blk_plug_device(struct request_queue *q)
209 WARN_ON(!irqs_disabled());
212 * don't plug a stopped queue, it must be paired with blk_start_queue()
213 * which will restart the queueing
215 if (blk_queue_stopped(q))
218 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
219 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
223 EXPORT_SYMBOL(blk_plug_device);
226 * blk_plug_device_unlocked - plug a device without queue lock held
227 * @q: The &struct request_queue to plug
230 * Like @blk_plug_device(), but grabs the queue lock and disables
233 void blk_plug_device_unlocked(struct request_queue *q)
237 spin_lock_irqsave(q->queue_lock, flags);
239 spin_unlock_irqrestore(q->queue_lock, flags);
241 EXPORT_SYMBOL(blk_plug_device_unlocked);
244 * remove the queue from the plugged list, if present. called with
245 * queue lock held and interrupts disabled.
247 int blk_remove_plug(struct request_queue *q)
249 WARN_ON(!irqs_disabled());
251 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
254 del_timer(&q->unplug_timer);
257 EXPORT_SYMBOL(blk_remove_plug);
260 * remove the plug and let it rip..
262 void __generic_unplug_device(struct request_queue *q)
264 if (unlikely(blk_queue_stopped(q)))
266 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
273 * generic_unplug_device - fire a request queue
274 * @q: The &struct request_queue in question
277 * Linux uses plugging to build bigger requests queues before letting
278 * the device have at them. If a queue is plugged, the I/O scheduler
279 * is still adding and merging requests on the queue. Once the queue
280 * gets unplugged, the request_fn defined for the queue is invoked and
283 void generic_unplug_device(struct request_queue *q)
285 if (blk_queue_plugged(q)) {
286 spin_lock_irq(q->queue_lock);
287 __generic_unplug_device(q);
288 spin_unlock_irq(q->queue_lock);
291 EXPORT_SYMBOL(generic_unplug_device);
293 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
296 struct request_queue *q = bdi->unplug_io_data;
301 void blk_unplug_work(struct work_struct *work)
303 struct request_queue *q =
304 container_of(work, struct request_queue, unplug_work);
306 trace_block_unplug_io(q);
310 void blk_unplug_timeout(unsigned long data)
312 struct request_queue *q = (struct request_queue *)data;
314 trace_block_unplug_timer(q);
315 kblockd_schedule_work(q, &q->unplug_work);
318 void blk_unplug(struct request_queue *q)
321 * devices don't necessarily have an ->unplug_fn defined
324 trace_block_unplug_io(q);
328 EXPORT_SYMBOL(blk_unplug);
331 * blk_start_queue - restart a previously stopped queue
332 * @q: The &struct request_queue in question
335 * blk_start_queue() will clear the stop flag on the queue, and call
336 * the request_fn for the queue if it was in a stopped state when
337 * entered. Also see blk_stop_queue(). Queue lock must be held.
339 void blk_start_queue(struct request_queue *q)
341 WARN_ON(!irqs_disabled());
343 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
346 EXPORT_SYMBOL(blk_start_queue);
349 * blk_stop_queue - stop a queue
350 * @q: The &struct request_queue in question
353 * The Linux block layer assumes that a block driver will consume all
354 * entries on the request queue when the request_fn strategy is called.
355 * Often this will not happen, because of hardware limitations (queue
356 * depth settings). If a device driver gets a 'queue full' response,
357 * or if it simply chooses not to queue more I/O at one point, it can
358 * call this function to prevent the request_fn from being called until
359 * the driver has signalled it's ready to go again. This happens by calling
360 * blk_start_queue() to restart queue operations. Queue lock must be held.
362 void blk_stop_queue(struct request_queue *q)
365 queue_flag_set(QUEUE_FLAG_STOPPED, q);
367 EXPORT_SYMBOL(blk_stop_queue);
370 * blk_sync_queue - cancel any pending callbacks on a queue
374 * The block layer may perform asynchronous callback activity
375 * on a queue, such as calling the unplug function after a timeout.
376 * A block device may call blk_sync_queue to ensure that any
377 * such activity is cancelled, thus allowing it to release resources
378 * that the callbacks might use. The caller must already have made sure
379 * that its ->make_request_fn will not re-add plugging prior to calling
383 void blk_sync_queue(struct request_queue *q)
385 del_timer_sync(&q->unplug_timer);
386 del_timer_sync(&q->timeout);
387 cancel_work_sync(&q->unplug_work);
388 throtl_shutdown_timer_wq(q);
390 EXPORT_SYMBOL(blk_sync_queue);
393 * __blk_run_queue - run a single device queue
394 * @q: The queue to run
397 * See @blk_run_queue. This variant must be called with the queue lock
398 * held and interrupts disabled.
401 void __blk_run_queue(struct request_queue *q)
405 if (unlikely(blk_queue_stopped(q)))
408 if (elv_queue_empty(q))
412 * Only recurse once to avoid overrunning the stack, let the unplug
413 * handling reinvoke the handler shortly if we already got there.
415 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
417 queue_flag_clear(QUEUE_FLAG_REENTER, q);
419 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
420 kblockd_schedule_work(q, &q->unplug_work);
423 EXPORT_SYMBOL(__blk_run_queue);
426 * blk_run_queue - run a single device queue
427 * @q: The queue to run
430 * Invoke request handling on this queue, if it has pending work to do.
431 * May be used to restart queueing when a request has completed.
433 void blk_run_queue(struct request_queue *q)
437 spin_lock_irqsave(q->queue_lock, flags);
439 spin_unlock_irqrestore(q->queue_lock, flags);
441 EXPORT_SYMBOL(blk_run_queue);
443 void blk_put_queue(struct request_queue *q)
445 kobject_put(&q->kobj);
448 void blk_cleanup_queue(struct request_queue *q)
451 * We know we have process context here, so we can be a little
452 * cautious and ensure that pending block actions on this device
453 * are done before moving on. Going into this function, we should
454 * not have processes doing IO to this device.
458 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
459 mutex_lock(&q->sysfs_lock);
460 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
461 mutex_unlock(&q->sysfs_lock);
464 elevator_exit(q->elevator);
470 EXPORT_SYMBOL(blk_cleanup_queue);
472 static int blk_init_free_list(struct request_queue *q)
474 struct request_list *rl = &q->rq;
476 if (unlikely(rl->rq_pool))
479 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
480 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
482 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
483 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
485 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
486 mempool_free_slab, request_cachep, q->node);
494 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
496 return blk_alloc_queue_node(gfp_mask, -1);
498 EXPORT_SYMBOL(blk_alloc_queue);
500 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
502 struct request_queue *q;
505 q = kmem_cache_alloc_node(blk_requestq_cachep,
506 gfp_mask | __GFP_ZERO, node_id);
510 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
511 q->backing_dev_info.unplug_io_data = q;
512 q->backing_dev_info.ra_pages =
513 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
514 q->backing_dev_info.state = 0;
515 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
516 q->backing_dev_info.name = "block";
518 err = bdi_init(&q->backing_dev_info);
520 kmem_cache_free(blk_requestq_cachep, q);
524 if (blk_throtl_init(q)) {
525 kmem_cache_free(blk_requestq_cachep, q);
529 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
530 laptop_mode_timer_fn, (unsigned long) q);
531 init_timer(&q->unplug_timer);
532 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
533 INIT_LIST_HEAD(&q->timeout_list);
534 INIT_WORK(&q->unplug_work, blk_unplug_work);
536 kobject_init(&q->kobj, &blk_queue_ktype);
538 mutex_init(&q->sysfs_lock);
539 spin_lock_init(&q->__queue_lock);
543 EXPORT_SYMBOL(blk_alloc_queue_node);
546 * blk_init_queue - prepare a request queue for use with a block device
547 * @rfn: The function to be called to process requests that have been
548 * placed on the queue.
549 * @lock: Request queue spin lock
552 * If a block device wishes to use the standard request handling procedures,
553 * which sorts requests and coalesces adjacent requests, then it must
554 * call blk_init_queue(). The function @rfn will be called when there
555 * are requests on the queue that need to be processed. If the device
556 * supports plugging, then @rfn may not be called immediately when requests
557 * are available on the queue, but may be called at some time later instead.
558 * Plugged queues are generally unplugged when a buffer belonging to one
559 * of the requests on the queue is needed, or due to memory pressure.
561 * @rfn is not required, or even expected, to remove all requests off the
562 * queue, but only as many as it can handle at a time. If it does leave
563 * requests on the queue, it is responsible for arranging that the requests
564 * get dealt with eventually.
566 * The queue spin lock must be held while manipulating the requests on the
567 * request queue; this lock will be taken also from interrupt context, so irq
568 * disabling is needed for it.
570 * Function returns a pointer to the initialized request queue, or %NULL if
574 * blk_init_queue() must be paired with a blk_cleanup_queue() call
575 * when the block device is deactivated (such as at module unload).
578 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
580 return blk_init_queue_node(rfn, lock, -1);
582 EXPORT_SYMBOL(blk_init_queue);
584 struct request_queue *
585 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
587 struct request_queue *uninit_q, *q;
589 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
593 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
595 blk_cleanup_queue(uninit_q);
599 EXPORT_SYMBOL(blk_init_queue_node);
601 struct request_queue *
602 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
605 return blk_init_allocated_queue_node(q, rfn, lock, -1);
607 EXPORT_SYMBOL(blk_init_allocated_queue);
609 struct request_queue *
610 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
611 spinlock_t *lock, int node_id)
617 if (blk_init_free_list(q))
621 q->prep_rq_fn = NULL;
622 q->unprep_rq_fn = NULL;
623 q->unplug_fn = generic_unplug_device;
624 q->queue_flags = QUEUE_FLAG_DEFAULT;
625 q->queue_lock = lock;
628 * This also sets hw/phys segments, boundary and size
630 blk_queue_make_request(q, __make_request);
632 q->sg_reserved_size = INT_MAX;
637 if (!elevator_init(q, NULL)) {
638 blk_queue_congestion_threshold(q);
644 EXPORT_SYMBOL(blk_init_allocated_queue_node);
646 int blk_get_queue(struct request_queue *q)
648 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
649 kobject_get(&q->kobj);
656 static inline void blk_free_request(struct request_queue *q, struct request *rq)
658 if (rq->cmd_flags & REQ_ELVPRIV)
659 elv_put_request(q, rq);
660 mempool_free(rq, q->rq.rq_pool);
663 static struct request *
664 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
666 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
673 rq->cmd_flags = flags | REQ_ALLOCED;
676 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
677 mempool_free(rq, q->rq.rq_pool);
680 rq->cmd_flags |= REQ_ELVPRIV;
687 * ioc_batching returns true if the ioc is a valid batching request and
688 * should be given priority access to a request.
690 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
696 * Make sure the process is able to allocate at least 1 request
697 * even if the batch times out, otherwise we could theoretically
700 return ioc->nr_batch_requests == q->nr_batching ||
701 (ioc->nr_batch_requests > 0
702 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
706 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
707 * will cause the process to be a "batcher" on all queues in the system. This
708 * is the behaviour we want though - once it gets a wakeup it should be given
711 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
713 if (!ioc || ioc_batching(q, ioc))
716 ioc->nr_batch_requests = q->nr_batching;
717 ioc->last_waited = jiffies;
720 static void __freed_request(struct request_queue *q, int sync)
722 struct request_list *rl = &q->rq;
724 if (rl->count[sync] < queue_congestion_off_threshold(q))
725 blk_clear_queue_congested(q, sync);
727 if (rl->count[sync] + 1 <= q->nr_requests) {
728 if (waitqueue_active(&rl->wait[sync]))
729 wake_up(&rl->wait[sync]);
731 blk_clear_queue_full(q, sync);
736 * A request has just been released. Account for it, update the full and
737 * congestion status, wake up any waiters. Called under q->queue_lock.
739 static void freed_request(struct request_queue *q, int sync, int priv)
741 struct request_list *rl = &q->rq;
747 __freed_request(q, sync);
749 if (unlikely(rl->starved[sync ^ 1]))
750 __freed_request(q, sync ^ 1);
754 * Get a free request, queue_lock must be held.
755 * Returns NULL on failure, with queue_lock held.
756 * Returns !NULL on success, with queue_lock *not held*.
758 static struct request *get_request(struct request_queue *q, int rw_flags,
759 struct bio *bio, gfp_t gfp_mask)
761 struct request *rq = NULL;
762 struct request_list *rl = &q->rq;
763 struct io_context *ioc = NULL;
764 const bool is_sync = rw_is_sync(rw_flags) != 0;
767 may_queue = elv_may_queue(q, rw_flags);
768 if (may_queue == ELV_MQUEUE_NO)
771 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
772 if (rl->count[is_sync]+1 >= q->nr_requests) {
773 ioc = current_io_context(GFP_ATOMIC, q->node);
775 * The queue will fill after this allocation, so set
776 * it as full, and mark this process as "batching".
777 * This process will be allowed to complete a batch of
778 * requests, others will be blocked.
780 if (!blk_queue_full(q, is_sync)) {
781 ioc_set_batching(q, ioc);
782 blk_set_queue_full(q, is_sync);
784 if (may_queue != ELV_MQUEUE_MUST
785 && !ioc_batching(q, ioc)) {
787 * The queue is full and the allocating
788 * process is not a "batcher", and not
789 * exempted by the IO scheduler
795 blk_set_queue_congested(q, is_sync);
799 * Only allow batching queuers to allocate up to 50% over the defined
800 * limit of requests, otherwise we could have thousands of requests
801 * allocated with any setting of ->nr_requests
803 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
806 rl->count[is_sync]++;
807 rl->starved[is_sync] = 0;
809 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
814 * Don't do stats for non-priv requests
816 if (blk_queue_io_stat(q))
817 rw_flags |= REQ_IO_STAT;
820 spin_unlock_irq(q->queue_lock);
822 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
825 * Allocation failed presumably due to memory. Undo anything
826 * we might have messed up.
828 * Allocating task should really be put onto the front of the
829 * wait queue, but this is pretty rare.
831 spin_lock_irq(q->queue_lock);
832 freed_request(q, is_sync, priv);
835 * in the very unlikely event that allocation failed and no
836 * requests for this direction was pending, mark us starved
837 * so that freeing of a request in the other direction will
838 * notice us. another possible fix would be to split the
839 * rq mempool into READ and WRITE
842 if (unlikely(rl->count[is_sync] == 0))
843 rl->starved[is_sync] = 1;
849 * ioc may be NULL here, and ioc_batching will be false. That's
850 * OK, if the queue is under the request limit then requests need
851 * not count toward the nr_batch_requests limit. There will always
852 * be some limit enforced by BLK_BATCH_TIME.
854 if (ioc_batching(q, ioc))
855 ioc->nr_batch_requests--;
857 trace_block_getrq(q, bio, rw_flags & 1);
863 * No available requests for this queue, unplug the device and wait for some
864 * requests to become available.
866 * Called with q->queue_lock held, and returns with it unlocked.
868 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
871 const bool is_sync = rw_is_sync(rw_flags) != 0;
874 rq = get_request(q, rw_flags, bio, GFP_NOIO);
877 struct io_context *ioc;
878 struct request_list *rl = &q->rq;
880 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
881 TASK_UNINTERRUPTIBLE);
883 trace_block_sleeprq(q, bio, rw_flags & 1);
885 __generic_unplug_device(q);
886 spin_unlock_irq(q->queue_lock);
890 * After sleeping, we become a "batching" process and
891 * will be able to allocate at least one request, and
892 * up to a big batch of them for a small period time.
893 * See ioc_batching, ioc_set_batching
895 ioc = current_io_context(GFP_NOIO, q->node);
896 ioc_set_batching(q, ioc);
898 spin_lock_irq(q->queue_lock);
899 finish_wait(&rl->wait[is_sync], &wait);
901 rq = get_request(q, rw_flags, bio, GFP_NOIO);
907 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
911 BUG_ON(rw != READ && rw != WRITE);
913 spin_lock_irq(q->queue_lock);
914 if (gfp_mask & __GFP_WAIT) {
915 rq = get_request_wait(q, rw, NULL);
917 rq = get_request(q, rw, NULL, gfp_mask);
919 spin_unlock_irq(q->queue_lock);
921 /* q->queue_lock is unlocked at this point */
925 EXPORT_SYMBOL(blk_get_request);
928 * blk_make_request - given a bio, allocate a corresponding struct request.
929 * @q: target request queue
930 * @bio: The bio describing the memory mappings that will be submitted for IO.
931 * It may be a chained-bio properly constructed by block/bio layer.
932 * @gfp_mask: gfp flags to be used for memory allocation
934 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
935 * type commands. Where the struct request needs to be farther initialized by
936 * the caller. It is passed a &struct bio, which describes the memory info of
939 * The caller of blk_make_request must make sure that bi_io_vec
940 * are set to describe the memory buffers. That bio_data_dir() will return
941 * the needed direction of the request. (And all bio's in the passed bio-chain
942 * are properly set accordingly)
944 * If called under none-sleepable conditions, mapped bio buffers must not
945 * need bouncing, by calling the appropriate masked or flagged allocator,
946 * suitable for the target device. Otherwise the call to blk_queue_bounce will
949 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
950 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
951 * anything but the first bio in the chain. Otherwise you risk waiting for IO
952 * completion of a bio that hasn't been submitted yet, thus resulting in a
953 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
954 * of bio_alloc(), as that avoids the mempool deadlock.
955 * If possible a big IO should be split into smaller parts when allocation
956 * fails. Partial allocation should not be an error, or you risk a live-lock.
958 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
961 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
964 return ERR_PTR(-ENOMEM);
967 struct bio *bounce_bio = bio;
970 blk_queue_bounce(q, &bounce_bio);
971 ret = blk_rq_append_bio(q, rq, bounce_bio);
980 EXPORT_SYMBOL(blk_make_request);
983 * blk_requeue_request - put a request back on queue
984 * @q: request queue where request should be inserted
985 * @rq: request to be inserted
988 * Drivers often keep queueing requests until the hardware cannot accept
989 * more, when that condition happens we need to put the request back
990 * on the queue. Must be called with queue lock held.
992 void blk_requeue_request(struct request_queue *q, struct request *rq)
994 blk_delete_timer(rq);
995 blk_clear_rq_complete(rq);
996 trace_block_rq_requeue(q, rq);
998 if (blk_rq_tagged(rq))
999 blk_queue_end_tag(q, rq);
1001 BUG_ON(blk_queued_rq(rq));
1003 elv_requeue_request(q, rq);
1005 EXPORT_SYMBOL(blk_requeue_request);
1008 * blk_insert_request - insert a special request into a request queue
1009 * @q: request queue where request should be inserted
1010 * @rq: request to be inserted
1011 * @at_head: insert request at head or tail of queue
1012 * @data: private data
1015 * Many block devices need to execute commands asynchronously, so they don't
1016 * block the whole kernel from preemption during request execution. This is
1017 * accomplished normally by inserting aritficial requests tagged as
1018 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1019 * be scheduled for actual execution by the request queue.
1021 * We have the option of inserting the head or the tail of the queue.
1022 * Typically we use the tail for new ioctls and so forth. We use the head
1023 * of the queue for things like a QUEUE_FULL message from a device, or a
1024 * host that is unable to accept a particular command.
1026 void blk_insert_request(struct request_queue *q, struct request *rq,
1027 int at_head, void *data)
1029 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1030 unsigned long flags;
1033 * tell I/O scheduler that this isn't a regular read/write (ie it
1034 * must not attempt merges on this) and that it acts as a soft
1037 rq->cmd_type = REQ_TYPE_SPECIAL;
1041 spin_lock_irqsave(q->queue_lock, flags);
1044 * If command is tagged, release the tag
1046 if (blk_rq_tagged(rq))
1047 blk_queue_end_tag(q, rq);
1049 drive_stat_acct(rq, 1);
1050 __elv_add_request(q, rq, where, 0);
1052 spin_unlock_irqrestore(q->queue_lock, flags);
1054 EXPORT_SYMBOL(blk_insert_request);
1057 * add-request adds a request to the linked list.
1058 * queue lock is held and interrupts disabled, as we muck with the
1059 * request queue list.
1061 static inline void add_request(struct request_queue *q, struct request *req)
1063 drive_stat_acct(req, 1);
1066 * elevator indicated where it wants this request to be
1067 * inserted at elevator_merge time
1069 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1072 static void part_round_stats_single(int cpu, struct hd_struct *part,
1075 if (now == part->stamp)
1078 if (part_in_flight(part)) {
1079 __part_stat_add(cpu, part, time_in_queue,
1080 part_in_flight(part) * (now - part->stamp));
1081 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1087 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1088 * @cpu: cpu number for stats access
1089 * @part: target partition
1091 * The average IO queue length and utilisation statistics are maintained
1092 * by observing the current state of the queue length and the amount of
1093 * time it has been in this state for.
1095 * Normally, that accounting is done on IO completion, but that can result
1096 * in more than a second's worth of IO being accounted for within any one
1097 * second, leading to >100% utilisation. To deal with that, we call this
1098 * function to do a round-off before returning the results when reading
1099 * /proc/diskstats. This accounts immediately for all queue usage up to
1100 * the current jiffies and restarts the counters again.
1102 void part_round_stats(int cpu, struct hd_struct *part)
1104 unsigned long now = jiffies;
1107 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1108 part_round_stats_single(cpu, part, now);
1110 EXPORT_SYMBOL_GPL(part_round_stats);
1113 * queue lock must be held
1115 void __blk_put_request(struct request_queue *q, struct request *req)
1119 if (unlikely(--req->ref_count))
1122 elv_completed_request(q, req);
1124 /* this is a bio leak */
1125 WARN_ON(req->bio != NULL);
1128 * Request may not have originated from ll_rw_blk. if not,
1129 * it didn't come out of our reserved rq pools
1131 if (req->cmd_flags & REQ_ALLOCED) {
1132 int is_sync = rq_is_sync(req) != 0;
1133 int priv = req->cmd_flags & REQ_ELVPRIV;
1135 BUG_ON(!list_empty(&req->queuelist));
1136 BUG_ON(!hlist_unhashed(&req->hash));
1138 blk_free_request(q, req);
1139 freed_request(q, is_sync, priv);
1142 EXPORT_SYMBOL_GPL(__blk_put_request);
1144 void blk_put_request(struct request *req)
1146 unsigned long flags;
1147 struct request_queue *q = req->q;
1149 spin_lock_irqsave(q->queue_lock, flags);
1150 __blk_put_request(q, req);
1151 spin_unlock_irqrestore(q->queue_lock, flags);
1153 EXPORT_SYMBOL(blk_put_request);
1156 * blk_add_request_payload - add a payload to a request
1157 * @rq: request to update
1158 * @page: page backing the payload
1159 * @len: length of the payload.
1161 * This allows to later add a payload to an already submitted request by
1162 * a block driver. The driver needs to take care of freeing the payload
1165 * Note that this is a quite horrible hack and nothing but handling of
1166 * discard requests should ever use it.
1168 void blk_add_request_payload(struct request *rq, struct page *page,
1171 struct bio *bio = rq->bio;
1173 bio->bi_io_vec->bv_page = page;
1174 bio->bi_io_vec->bv_offset = 0;
1175 bio->bi_io_vec->bv_len = len;
1179 bio->bi_phys_segments = 1;
1181 rq->__data_len = rq->resid_len = len;
1182 rq->nr_phys_segments = 1;
1183 rq->buffer = bio_data(bio);
1185 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1187 void init_request_from_bio(struct request *req, struct bio *bio)
1189 req->cpu = bio->bi_comp_cpu;
1190 req->cmd_type = REQ_TYPE_FS;
1192 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1193 if (bio->bi_rw & REQ_RAHEAD)
1194 req->cmd_flags |= REQ_FAILFAST_MASK;
1197 req->__sector = bio->bi_sector;
1198 req->ioprio = bio_prio(bio);
1199 blk_rq_bio_prep(req->q, req, bio);
1203 * Only disabling plugging for non-rotational devices if it does tagging
1204 * as well, otherwise we do need the proper merging
1206 static inline bool queue_should_plug(struct request_queue *q)
1208 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1211 static int __make_request(struct request_queue *q, struct bio *bio)
1213 struct request *req;
1215 unsigned int bytes = bio->bi_size;
1216 const unsigned short prio = bio_prio(bio);
1217 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1218 const bool unplug = !!(bio->bi_rw & REQ_UNPLUG);
1219 const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK;
1222 if ((bio->bi_rw & REQ_HARDBARRIER) &&
1223 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1224 bio_endio(bio, -EOPNOTSUPP);
1228 * low level driver can indicate that it wants pages above a
1229 * certain limit bounced to low memory (ie for highmem, or even
1230 * ISA dma in theory)
1232 blk_queue_bounce(q, &bio);
1234 spin_lock_irq(q->queue_lock);
1236 if (unlikely((bio->bi_rw & REQ_HARDBARRIER)) || elv_queue_empty(q))
1239 el_ret = elv_merge(q, &req, bio);
1241 case ELEVATOR_BACK_MERGE:
1242 BUG_ON(!rq_mergeable(req));
1244 if (!ll_back_merge_fn(q, req, bio))
1247 trace_block_bio_backmerge(q, bio);
1249 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1250 blk_rq_set_mixed_merge(req);
1252 req->biotail->bi_next = bio;
1254 req->__data_len += bytes;
1255 req->ioprio = ioprio_best(req->ioprio, prio);
1256 if (!blk_rq_cpu_valid(req))
1257 req->cpu = bio->bi_comp_cpu;
1258 drive_stat_acct(req, 0);
1259 elv_bio_merged(q, req, bio);
1260 if (!attempt_back_merge(q, req))
1261 elv_merged_request(q, req, el_ret);
1264 case ELEVATOR_FRONT_MERGE:
1265 BUG_ON(!rq_mergeable(req));
1267 if (!ll_front_merge_fn(q, req, bio))
1270 trace_block_bio_frontmerge(q, bio);
1272 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1273 blk_rq_set_mixed_merge(req);
1274 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1275 req->cmd_flags |= ff;
1278 bio->bi_next = req->bio;
1282 * may not be valid. if the low level driver said
1283 * it didn't need a bounce buffer then it better
1284 * not touch req->buffer either...
1286 req->buffer = bio_data(bio);
1287 req->__sector = bio->bi_sector;
1288 req->__data_len += bytes;
1289 req->ioprio = ioprio_best(req->ioprio, prio);
1290 if (!blk_rq_cpu_valid(req))
1291 req->cpu = bio->bi_comp_cpu;
1292 drive_stat_acct(req, 0);
1293 elv_bio_merged(q, req, bio);
1294 if (!attempt_front_merge(q, req))
1295 elv_merged_request(q, req, el_ret);
1298 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1305 * This sync check and mask will be re-done in init_request_from_bio(),
1306 * but we need to set it earlier to expose the sync flag to the
1307 * rq allocator and io schedulers.
1309 rw_flags = bio_data_dir(bio);
1311 rw_flags |= REQ_SYNC;
1314 * Grab a free request. This is might sleep but can not fail.
1315 * Returns with the queue unlocked.
1317 req = get_request_wait(q, rw_flags, bio);
1320 * After dropping the lock and possibly sleeping here, our request
1321 * may now be mergeable after it had proven unmergeable (above).
1322 * We don't worry about that case for efficiency. It won't happen
1323 * often, and the elevators are able to handle it.
1325 init_request_from_bio(req, bio);
1327 spin_lock_irq(q->queue_lock);
1328 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1329 bio_flagged(bio, BIO_CPU_AFFINE))
1330 req->cpu = blk_cpu_to_group(smp_processor_id());
1331 if (queue_should_plug(q) && elv_queue_empty(q))
1333 add_request(q, req);
1335 if (unplug || !queue_should_plug(q))
1336 __generic_unplug_device(q);
1337 spin_unlock_irq(q->queue_lock);
1342 * If bio->bi_dev is a partition, remap the location
1344 static inline void blk_partition_remap(struct bio *bio)
1346 struct block_device *bdev = bio->bi_bdev;
1348 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1349 struct hd_struct *p = bdev->bd_part;
1351 bio->bi_sector += p->start_sect;
1352 bio->bi_bdev = bdev->bd_contains;
1354 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1356 bio->bi_sector - p->start_sect);
1360 static void handle_bad_sector(struct bio *bio)
1362 char b[BDEVNAME_SIZE];
1364 printk(KERN_INFO "attempt to access beyond end of device\n");
1365 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1366 bdevname(bio->bi_bdev, b),
1368 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1369 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1371 set_bit(BIO_EOF, &bio->bi_flags);
1374 #ifdef CONFIG_FAIL_MAKE_REQUEST
1376 static DECLARE_FAULT_ATTR(fail_make_request);
1378 static int __init setup_fail_make_request(char *str)
1380 return setup_fault_attr(&fail_make_request, str);
1382 __setup("fail_make_request=", setup_fail_make_request);
1384 static int should_fail_request(struct bio *bio)
1386 struct hd_struct *part = bio->bi_bdev->bd_part;
1388 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1389 return should_fail(&fail_make_request, bio->bi_size);
1394 static int __init fail_make_request_debugfs(void)
1396 return init_fault_attr_dentries(&fail_make_request,
1397 "fail_make_request");
1400 late_initcall(fail_make_request_debugfs);
1402 #else /* CONFIG_FAIL_MAKE_REQUEST */
1404 static inline int should_fail_request(struct bio *bio)
1409 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1412 * Check whether this bio extends beyond the end of the device.
1414 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1421 /* Test device or partition size, when known. */
1422 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1424 sector_t sector = bio->bi_sector;
1426 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1428 * This may well happen - the kernel calls bread()
1429 * without checking the size of the device, e.g., when
1430 * mounting a device.
1432 handle_bad_sector(bio);
1441 * generic_make_request - hand a buffer to its device driver for I/O
1442 * @bio: The bio describing the location in memory and on the device.
1444 * generic_make_request() is used to make I/O requests of block
1445 * devices. It is passed a &struct bio, which describes the I/O that needs
1448 * generic_make_request() does not return any status. The
1449 * success/failure status of the request, along with notification of
1450 * completion, is delivered asynchronously through the bio->bi_end_io
1451 * function described (one day) else where.
1453 * The caller of generic_make_request must make sure that bi_io_vec
1454 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1455 * set to describe the device address, and the
1456 * bi_end_io and optionally bi_private are set to describe how
1457 * completion notification should be signaled.
1459 * generic_make_request and the drivers it calls may use bi_next if this
1460 * bio happens to be merged with someone else, and may change bi_dev and
1461 * bi_sector for remaps as it sees fit. So the values of these fields
1462 * should NOT be depended on after the call to generic_make_request.
1464 static inline void __generic_make_request(struct bio *bio)
1466 struct request_queue *q;
1467 sector_t old_sector;
1468 int ret, nr_sectors = bio_sectors(bio);
1474 if (bio_check_eod(bio, nr_sectors))
1478 * Resolve the mapping until finished. (drivers are
1479 * still free to implement/resolve their own stacking
1480 * by explicitly returning 0)
1482 * NOTE: we don't repeat the blk_size check for each new device.
1483 * Stacking drivers are expected to know what they are doing.
1488 char b[BDEVNAME_SIZE];
1490 q = bdev_get_queue(bio->bi_bdev);
1493 "generic_make_request: Trying to access "
1494 "nonexistent block-device %s (%Lu)\n",
1495 bdevname(bio->bi_bdev, b),
1496 (long long) bio->bi_sector);
1500 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1501 nr_sectors > queue_max_hw_sectors(q))) {
1502 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1503 bdevname(bio->bi_bdev, b),
1505 queue_max_hw_sectors(q));
1509 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1512 if (should_fail_request(bio))
1516 * If this device has partitions, remap block n
1517 * of partition p to block n+start(p) of the disk.
1519 blk_partition_remap(bio);
1521 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1524 if (old_sector != -1)
1525 trace_block_remap(q, bio, old_dev, old_sector);
1527 old_sector = bio->bi_sector;
1528 old_dev = bio->bi_bdev->bd_dev;
1530 if (bio_check_eod(bio, nr_sectors))
1533 if ((bio->bi_rw & REQ_DISCARD) &&
1534 (!blk_queue_discard(q) ||
1535 ((bio->bi_rw & REQ_SECURE) &&
1536 !blk_queue_secdiscard(q)))) {
1541 blk_throtl_bio(q, &bio);
1544 * If bio = NULL, bio has been throttled and will be submitted
1550 trace_block_bio_queue(q, bio);
1552 ret = q->make_request_fn(q, bio);
1558 bio_endio(bio, err);
1562 * We only want one ->make_request_fn to be active at a time,
1563 * else stack usage with stacked devices could be a problem.
1564 * So use current->bio_list to keep a list of requests
1565 * submited by a make_request_fn function.
1566 * current->bio_list is also used as a flag to say if
1567 * generic_make_request is currently active in this task or not.
1568 * If it is NULL, then no make_request is active. If it is non-NULL,
1569 * then a make_request is active, and new requests should be added
1572 void generic_make_request(struct bio *bio)
1574 struct bio_list bio_list_on_stack;
1576 if (current->bio_list) {
1577 /* make_request is active */
1578 bio_list_add(current->bio_list, bio);
1581 /* following loop may be a bit non-obvious, and so deserves some
1583 * Before entering the loop, bio->bi_next is NULL (as all callers
1584 * ensure that) so we have a list with a single bio.
1585 * We pretend that we have just taken it off a longer list, so
1586 * we assign bio_list to a pointer to the bio_list_on_stack,
1587 * thus initialising the bio_list of new bios to be
1588 * added. __generic_make_request may indeed add some more bios
1589 * through a recursive call to generic_make_request. If it
1590 * did, we find a non-NULL value in bio_list and re-enter the loop
1591 * from the top. In this case we really did just take the bio
1592 * of the top of the list (no pretending) and so remove it from
1593 * bio_list, and call into __generic_make_request again.
1595 * The loop was structured like this to make only one call to
1596 * __generic_make_request (which is important as it is large and
1597 * inlined) and to keep the structure simple.
1599 BUG_ON(bio->bi_next);
1600 bio_list_init(&bio_list_on_stack);
1601 current->bio_list = &bio_list_on_stack;
1603 __generic_make_request(bio);
1604 bio = bio_list_pop(current->bio_list);
1606 current->bio_list = NULL; /* deactivate */
1608 EXPORT_SYMBOL(generic_make_request);
1611 * submit_bio - submit a bio to the block device layer for I/O
1612 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1613 * @bio: The &struct bio which describes the I/O
1615 * submit_bio() is very similar in purpose to generic_make_request(), and
1616 * uses that function to do most of the work. Both are fairly rough
1617 * interfaces; @bio must be presetup and ready for I/O.
1620 void submit_bio(int rw, struct bio *bio)
1622 int count = bio_sectors(bio);
1627 * If it's a regular read/write or a barrier with data attached,
1628 * go through the normal accounting stuff before submission.
1630 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1632 count_vm_events(PGPGOUT, count);
1634 task_io_account_read(bio->bi_size);
1635 count_vm_events(PGPGIN, count);
1638 if (unlikely(block_dump)) {
1639 char b[BDEVNAME_SIZE];
1640 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1641 current->comm, task_pid_nr(current),
1642 (rw & WRITE) ? "WRITE" : "READ",
1643 (unsigned long long)bio->bi_sector,
1644 bdevname(bio->bi_bdev, b),
1649 generic_make_request(bio);
1651 EXPORT_SYMBOL(submit_bio);
1654 * blk_rq_check_limits - Helper function to check a request for the queue limit
1656 * @rq: the request being checked
1659 * @rq may have been made based on weaker limitations of upper-level queues
1660 * in request stacking drivers, and it may violate the limitation of @q.
1661 * Since the block layer and the underlying device driver trust @rq
1662 * after it is inserted to @q, it should be checked against @q before
1663 * the insertion using this generic function.
1665 * This function should also be useful for request stacking drivers
1666 * in some cases below, so export this fuction.
1667 * Request stacking drivers like request-based dm may change the queue
1668 * limits while requests are in the queue (e.g. dm's table swapping).
1669 * Such request stacking drivers should check those requests agaist
1670 * the new queue limits again when they dispatch those requests,
1671 * although such checkings are also done against the old queue limits
1672 * when submitting requests.
1674 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1676 if (rq->cmd_flags & REQ_DISCARD)
1679 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1680 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1681 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1686 * queue's settings related to segment counting like q->bounce_pfn
1687 * may differ from that of other stacking queues.
1688 * Recalculate it to check the request correctly on this queue's
1691 blk_recalc_rq_segments(rq);
1692 if (rq->nr_phys_segments > queue_max_segments(q)) {
1693 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1699 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1702 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1703 * @q: the queue to submit the request
1704 * @rq: the request being queued
1706 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1708 unsigned long flags;
1710 if (blk_rq_check_limits(q, rq))
1713 #ifdef CONFIG_FAIL_MAKE_REQUEST
1714 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1715 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1719 spin_lock_irqsave(q->queue_lock, flags);
1722 * Submitting request must be dequeued before calling this function
1723 * because it will be linked to another request_queue
1725 BUG_ON(blk_queued_rq(rq));
1727 drive_stat_acct(rq, 1);
1728 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1730 spin_unlock_irqrestore(q->queue_lock, flags);
1734 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1737 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1738 * @rq: request to examine
1741 * A request could be merge of IOs which require different failure
1742 * handling. This function determines the number of bytes which
1743 * can be failed from the beginning of the request without
1744 * crossing into area which need to be retried further.
1747 * The number of bytes to fail.
1750 * queue_lock must be held.
1752 unsigned int blk_rq_err_bytes(const struct request *rq)
1754 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1755 unsigned int bytes = 0;
1758 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1759 return blk_rq_bytes(rq);
1762 * Currently the only 'mixing' which can happen is between
1763 * different fastfail types. We can safely fail portions
1764 * which have all the failfast bits that the first one has -
1765 * the ones which are at least as eager to fail as the first
1768 for (bio = rq->bio; bio; bio = bio->bi_next) {
1769 if ((bio->bi_rw & ff) != ff)
1771 bytes += bio->bi_size;
1774 /* this could lead to infinite loop */
1775 BUG_ON(blk_rq_bytes(rq) && !bytes);
1778 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1780 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1782 if (blk_do_io_stat(req)) {
1783 const int rw = rq_data_dir(req);
1784 struct hd_struct *part;
1787 cpu = part_stat_lock();
1789 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1794 static void blk_account_io_done(struct request *req)
1797 * Account IO completion. bar_rq isn't accounted as a normal
1798 * IO on queueing nor completion. Accounting the containing
1799 * request is enough.
1801 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
1802 unsigned long duration = jiffies - req->start_time;
1803 const int rw = rq_data_dir(req);
1804 struct hd_struct *part;
1807 cpu = part_stat_lock();
1810 part_stat_inc(cpu, part, ios[rw]);
1811 part_stat_add(cpu, part, ticks[rw], duration);
1812 part_round_stats(cpu, part);
1813 part_dec_in_flight(part, rw);
1820 * blk_peek_request - peek at the top of a request queue
1821 * @q: request queue to peek at
1824 * Return the request at the top of @q. The returned request
1825 * should be started using blk_start_request() before LLD starts
1829 * Pointer to the request at the top of @q if available. Null
1833 * queue_lock must be held.
1835 struct request *blk_peek_request(struct request_queue *q)
1840 while ((rq = __elv_next_request(q)) != NULL) {
1841 if (!(rq->cmd_flags & REQ_STARTED)) {
1843 * This is the first time the device driver
1844 * sees this request (possibly after
1845 * requeueing). Notify IO scheduler.
1847 if (rq->cmd_flags & REQ_SORTED)
1848 elv_activate_rq(q, rq);
1851 * just mark as started even if we don't start
1852 * it, a request that has been delayed should
1853 * not be passed by new incoming requests
1855 rq->cmd_flags |= REQ_STARTED;
1856 trace_block_rq_issue(q, rq);
1859 if (!q->boundary_rq || q->boundary_rq == rq) {
1860 q->end_sector = rq_end_sector(rq);
1861 q->boundary_rq = NULL;
1864 if (rq->cmd_flags & REQ_DONTPREP)
1867 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1869 * make sure space for the drain appears we
1870 * know we can do this because max_hw_segments
1871 * has been adjusted to be one fewer than the
1874 rq->nr_phys_segments++;
1880 ret = q->prep_rq_fn(q, rq);
1881 if (ret == BLKPREP_OK) {
1883 } else if (ret == BLKPREP_DEFER) {
1885 * the request may have been (partially) prepped.
1886 * we need to keep this request in the front to
1887 * avoid resource deadlock. REQ_STARTED will
1888 * prevent other fs requests from passing this one.
1890 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1891 !(rq->cmd_flags & REQ_DONTPREP)) {
1893 * remove the space for the drain we added
1894 * so that we don't add it again
1896 --rq->nr_phys_segments;
1901 } else if (ret == BLKPREP_KILL) {
1902 rq->cmd_flags |= REQ_QUIET;
1904 * Mark this request as started so we don't trigger
1905 * any debug logic in the end I/O path.
1907 blk_start_request(rq);
1908 __blk_end_request_all(rq, -EIO);
1910 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1917 EXPORT_SYMBOL(blk_peek_request);
1919 void blk_dequeue_request(struct request *rq)
1921 struct request_queue *q = rq->q;
1923 BUG_ON(list_empty(&rq->queuelist));
1924 BUG_ON(ELV_ON_HASH(rq));
1926 list_del_init(&rq->queuelist);
1929 * the time frame between a request being removed from the lists
1930 * and to it is freed is accounted as io that is in progress at
1933 if (blk_account_rq(rq)) {
1934 q->in_flight[rq_is_sync(rq)]++;
1935 set_io_start_time_ns(rq);
1940 * blk_start_request - start request processing on the driver
1941 * @req: request to dequeue
1944 * Dequeue @req and start timeout timer on it. This hands off the
1945 * request to the driver.
1947 * Block internal functions which don't want to start timer should
1948 * call blk_dequeue_request().
1951 * queue_lock must be held.
1953 void blk_start_request(struct request *req)
1955 blk_dequeue_request(req);
1958 * We are now handing the request to the hardware, initialize
1959 * resid_len to full count and add the timeout handler.
1961 req->resid_len = blk_rq_bytes(req);
1962 if (unlikely(blk_bidi_rq(req)))
1963 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1967 EXPORT_SYMBOL(blk_start_request);
1970 * blk_fetch_request - fetch a request from a request queue
1971 * @q: request queue to fetch a request from
1974 * Return the request at the top of @q. The request is started on
1975 * return and LLD can start processing it immediately.
1978 * Pointer to the request at the top of @q if available. Null
1982 * queue_lock must be held.
1984 struct request *blk_fetch_request(struct request_queue *q)
1988 rq = blk_peek_request(q);
1990 blk_start_request(rq);
1993 EXPORT_SYMBOL(blk_fetch_request);
1996 * blk_update_request - Special helper function for request stacking drivers
1997 * @req: the request being processed
1998 * @error: %0 for success, < %0 for error
1999 * @nr_bytes: number of bytes to complete @req
2002 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2003 * the request structure even if @req doesn't have leftover.
2004 * If @req has leftover, sets it up for the next range of segments.
2006 * This special helper function is only for request stacking drivers
2007 * (e.g. request-based dm) so that they can handle partial completion.
2008 * Actual device drivers should use blk_end_request instead.
2010 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2011 * %false return from this function.
2014 * %false - this request doesn't have any more data
2015 * %true - this request has more data
2017 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2019 int total_bytes, bio_nbytes, next_idx = 0;
2025 trace_block_rq_complete(req->q, req);
2028 * For fs requests, rq is just carrier of independent bio's
2029 * and each partial completion should be handled separately.
2030 * Reset per-request error on each partial completion.
2032 * TODO: tj: This is too subtle. It would be better to let
2033 * low level drivers do what they see fit.
2035 if (req->cmd_type == REQ_TYPE_FS)
2038 if (error && req->cmd_type == REQ_TYPE_FS &&
2039 !(req->cmd_flags & REQ_QUIET)) {
2040 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
2041 req->rq_disk ? req->rq_disk->disk_name : "?",
2042 (unsigned long long)blk_rq_pos(req));
2045 blk_account_io_completion(req, nr_bytes);
2047 total_bytes = bio_nbytes = 0;
2048 while ((bio = req->bio) != NULL) {
2051 if (nr_bytes >= bio->bi_size) {
2052 req->bio = bio->bi_next;
2053 nbytes = bio->bi_size;
2054 req_bio_endio(req, bio, nbytes, error);
2058 int idx = bio->bi_idx + next_idx;
2060 if (unlikely(idx >= bio->bi_vcnt)) {
2061 blk_dump_rq_flags(req, "__end_that");
2062 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2063 __func__, idx, bio->bi_vcnt);
2067 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2068 BIO_BUG_ON(nbytes > bio->bi_size);
2071 * not a complete bvec done
2073 if (unlikely(nbytes > nr_bytes)) {
2074 bio_nbytes += nr_bytes;
2075 total_bytes += nr_bytes;
2080 * advance to the next vector
2083 bio_nbytes += nbytes;
2086 total_bytes += nbytes;
2092 * end more in this run, or just return 'not-done'
2094 if (unlikely(nr_bytes <= 0))
2104 * Reset counters so that the request stacking driver
2105 * can find how many bytes remain in the request
2108 req->__data_len = 0;
2113 * if the request wasn't completed, update state
2116 req_bio_endio(req, bio, bio_nbytes, error);
2117 bio->bi_idx += next_idx;
2118 bio_iovec(bio)->bv_offset += nr_bytes;
2119 bio_iovec(bio)->bv_len -= nr_bytes;
2122 req->__data_len -= total_bytes;
2123 req->buffer = bio_data(req->bio);
2125 /* update sector only for requests with clear definition of sector */
2126 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2127 req->__sector += total_bytes >> 9;
2129 /* mixed attributes always follow the first bio */
2130 if (req->cmd_flags & REQ_MIXED_MERGE) {
2131 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2132 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2136 * If total number of sectors is less than the first segment
2137 * size, something has gone terribly wrong.
2139 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2140 printk(KERN_ERR "blk: request botched\n");
2141 req->__data_len = blk_rq_cur_bytes(req);
2144 /* recalculate the number of segments */
2145 blk_recalc_rq_segments(req);
2149 EXPORT_SYMBOL_GPL(blk_update_request);
2151 static bool blk_update_bidi_request(struct request *rq, int error,
2152 unsigned int nr_bytes,
2153 unsigned int bidi_bytes)
2155 if (blk_update_request(rq, error, nr_bytes))
2158 /* Bidi request must be completed as a whole */
2159 if (unlikely(blk_bidi_rq(rq)) &&
2160 blk_update_request(rq->next_rq, error, bidi_bytes))
2163 if (blk_queue_add_random(rq->q))
2164 add_disk_randomness(rq->rq_disk);
2170 * blk_unprep_request - unprepare a request
2173 * This function makes a request ready for complete resubmission (or
2174 * completion). It happens only after all error handling is complete,
2175 * so represents the appropriate moment to deallocate any resources
2176 * that were allocated to the request in the prep_rq_fn. The queue
2177 * lock is held when calling this.
2179 void blk_unprep_request(struct request *req)
2181 struct request_queue *q = req->q;
2183 req->cmd_flags &= ~REQ_DONTPREP;
2184 if (q->unprep_rq_fn)
2185 q->unprep_rq_fn(q, req);
2187 EXPORT_SYMBOL_GPL(blk_unprep_request);
2190 * queue lock must be held
2192 static void blk_finish_request(struct request *req, int error)
2194 if (blk_rq_tagged(req))
2195 blk_queue_end_tag(req->q, req);
2197 BUG_ON(blk_queued_rq(req));
2199 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2200 laptop_io_completion(&req->q->backing_dev_info);
2202 blk_delete_timer(req);
2204 if (req->cmd_flags & REQ_DONTPREP)
2205 blk_unprep_request(req);
2208 blk_account_io_done(req);
2211 req->end_io(req, error);
2213 if (blk_bidi_rq(req))
2214 __blk_put_request(req->next_rq->q, req->next_rq);
2216 __blk_put_request(req->q, req);
2221 * blk_end_bidi_request - Complete a bidi request
2222 * @rq: the request to complete
2223 * @error: %0 for success, < %0 for error
2224 * @nr_bytes: number of bytes to complete @rq
2225 * @bidi_bytes: number of bytes to complete @rq->next_rq
2228 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2229 * Drivers that supports bidi can safely call this member for any
2230 * type of request, bidi or uni. In the later case @bidi_bytes is
2234 * %false - we are done with this request
2235 * %true - still buffers pending for this request
2237 static bool blk_end_bidi_request(struct request *rq, int error,
2238 unsigned int nr_bytes, unsigned int bidi_bytes)
2240 struct request_queue *q = rq->q;
2241 unsigned long flags;
2243 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2246 spin_lock_irqsave(q->queue_lock, flags);
2247 blk_finish_request(rq, error);
2248 spin_unlock_irqrestore(q->queue_lock, flags);
2254 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2255 * @rq: the request to complete
2256 * @error: %0 for success, < %0 for error
2257 * @nr_bytes: number of bytes to complete @rq
2258 * @bidi_bytes: number of bytes to complete @rq->next_rq
2261 * Identical to blk_end_bidi_request() except that queue lock is
2262 * assumed to be locked on entry and remains so on return.
2265 * %false - we are done with this request
2266 * %true - still buffers pending for this request
2268 static bool __blk_end_bidi_request(struct request *rq, int error,
2269 unsigned int nr_bytes, unsigned int bidi_bytes)
2271 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2274 blk_finish_request(rq, error);
2280 * blk_end_request - Helper function for drivers to complete the request.
2281 * @rq: the request being processed
2282 * @error: %0 for success, < %0 for error
2283 * @nr_bytes: number of bytes to complete
2286 * Ends I/O on a number of bytes attached to @rq.
2287 * If @rq has leftover, sets it up for the next range of segments.
2290 * %false - we are done with this request
2291 * %true - still buffers pending for this request
2293 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2295 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2297 EXPORT_SYMBOL(blk_end_request);
2300 * blk_end_request_all - Helper function for drives to finish the request.
2301 * @rq: the request to finish
2302 * @error: %0 for success, < %0 for error
2305 * Completely finish @rq.
2307 void blk_end_request_all(struct request *rq, int error)
2310 unsigned int bidi_bytes = 0;
2312 if (unlikely(blk_bidi_rq(rq)))
2313 bidi_bytes = blk_rq_bytes(rq->next_rq);
2315 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2318 EXPORT_SYMBOL(blk_end_request_all);
2321 * blk_end_request_cur - Helper function to finish the current request chunk.
2322 * @rq: the request to finish the current chunk for
2323 * @error: %0 for success, < %0 for error
2326 * Complete the current consecutively mapped chunk from @rq.
2329 * %false - we are done with this request
2330 * %true - still buffers pending for this request
2332 bool blk_end_request_cur(struct request *rq, int error)
2334 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2336 EXPORT_SYMBOL(blk_end_request_cur);
2339 * blk_end_request_err - Finish a request till the next failure boundary.
2340 * @rq: the request to finish till the next failure boundary for
2341 * @error: must be negative errno
2344 * Complete @rq till the next failure boundary.
2347 * %false - we are done with this request
2348 * %true - still buffers pending for this request
2350 bool blk_end_request_err(struct request *rq, int error)
2352 WARN_ON(error >= 0);
2353 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2355 EXPORT_SYMBOL_GPL(blk_end_request_err);
2358 * __blk_end_request - Helper function for drivers to complete the request.
2359 * @rq: the request being processed
2360 * @error: %0 for success, < %0 for error
2361 * @nr_bytes: number of bytes to complete
2364 * Must be called with queue lock held unlike blk_end_request().
2367 * %false - we are done with this request
2368 * %true - still buffers pending for this request
2370 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2372 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2374 EXPORT_SYMBOL(__blk_end_request);
2377 * __blk_end_request_all - Helper function for drives to finish the request.
2378 * @rq: the request to finish
2379 * @error: %0 for success, < %0 for error
2382 * Completely finish @rq. Must be called with queue lock held.
2384 void __blk_end_request_all(struct request *rq, int error)
2387 unsigned int bidi_bytes = 0;
2389 if (unlikely(blk_bidi_rq(rq)))
2390 bidi_bytes = blk_rq_bytes(rq->next_rq);
2392 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2395 EXPORT_SYMBOL(__blk_end_request_all);
2398 * __blk_end_request_cur - Helper function to finish the current request chunk.
2399 * @rq: the request to finish the current chunk for
2400 * @error: %0 for success, < %0 for error
2403 * Complete the current consecutively mapped chunk from @rq. Must
2404 * be called with queue lock held.
2407 * %false - we are done with this request
2408 * %true - still buffers pending for this request
2410 bool __blk_end_request_cur(struct request *rq, int error)
2412 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2414 EXPORT_SYMBOL(__blk_end_request_cur);
2417 * __blk_end_request_err - Finish a request till the next failure boundary.
2418 * @rq: the request to finish till the next failure boundary for
2419 * @error: must be negative errno
2422 * Complete @rq till the next failure boundary. Must be called
2423 * with queue lock held.
2426 * %false - we are done with this request
2427 * %true - still buffers pending for this request
2429 bool __blk_end_request_err(struct request *rq, int error)
2431 WARN_ON(error >= 0);
2432 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2434 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2436 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2439 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2440 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2442 if (bio_has_data(bio)) {
2443 rq->nr_phys_segments = bio_phys_segments(q, bio);
2444 rq->buffer = bio_data(bio);
2446 rq->__data_len = bio->bi_size;
2447 rq->bio = rq->biotail = bio;
2450 rq->rq_disk = bio->bi_bdev->bd_disk;
2453 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2455 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2456 * @rq: the request to be flushed
2459 * Flush all pages in @rq.
2461 void rq_flush_dcache_pages(struct request *rq)
2463 struct req_iterator iter;
2464 struct bio_vec *bvec;
2466 rq_for_each_segment(bvec, rq, iter)
2467 flush_dcache_page(bvec->bv_page);
2469 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2473 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2474 * @q : the queue of the device being checked
2477 * Check if underlying low-level drivers of a device are busy.
2478 * If the drivers want to export their busy state, they must set own
2479 * exporting function using blk_queue_lld_busy() first.
2481 * Basically, this function is used only by request stacking drivers
2482 * to stop dispatching requests to underlying devices when underlying
2483 * devices are busy. This behavior helps more I/O merging on the queue
2484 * of the request stacking driver and prevents I/O throughput regression
2485 * on burst I/O load.
2488 * 0 - Not busy (The request stacking driver should dispatch request)
2489 * 1 - Busy (The request stacking driver should stop dispatching request)
2491 int blk_lld_busy(struct request_queue *q)
2494 return q->lld_busy_fn(q);
2498 EXPORT_SYMBOL_GPL(blk_lld_busy);
2501 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2502 * @rq: the clone request to be cleaned up
2505 * Free all bios in @rq for a cloned request.
2507 void blk_rq_unprep_clone(struct request *rq)
2511 while ((bio = rq->bio) != NULL) {
2512 rq->bio = bio->bi_next;
2517 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2520 * Copy attributes of the original request to the clone request.
2521 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2523 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2525 dst->cpu = src->cpu;
2526 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2527 if (src->cmd_flags & REQ_DISCARD)
2528 dst->cmd_flags |= REQ_DISCARD;
2529 dst->cmd_type = src->cmd_type;
2530 dst->__sector = blk_rq_pos(src);
2531 dst->__data_len = blk_rq_bytes(src);
2532 dst->nr_phys_segments = src->nr_phys_segments;
2533 dst->ioprio = src->ioprio;
2534 dst->extra_len = src->extra_len;
2538 * blk_rq_prep_clone - Helper function to setup clone request
2539 * @rq: the request to be setup
2540 * @rq_src: original request to be cloned
2541 * @bs: bio_set that bios for clone are allocated from
2542 * @gfp_mask: memory allocation mask for bio
2543 * @bio_ctr: setup function to be called for each clone bio.
2544 * Returns %0 for success, non %0 for failure.
2545 * @data: private data to be passed to @bio_ctr
2548 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2549 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2550 * are not copied, and copying such parts is the caller's responsibility.
2551 * Also, pages which the original bios are pointing to are not copied
2552 * and the cloned bios just point same pages.
2553 * So cloned bios must be completed before original bios, which means
2554 * the caller must complete @rq before @rq_src.
2556 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2557 struct bio_set *bs, gfp_t gfp_mask,
2558 int (*bio_ctr)(struct bio *, struct bio *, void *),
2561 struct bio *bio, *bio_src;
2566 blk_rq_init(NULL, rq);
2568 __rq_for_each_bio(bio_src, rq_src) {
2569 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2573 __bio_clone(bio, bio_src);
2575 if (bio_integrity(bio_src) &&
2576 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2579 if (bio_ctr && bio_ctr(bio, bio_src, data))
2583 rq->biotail->bi_next = bio;
2586 rq->bio = rq->biotail = bio;
2589 __blk_rq_prep_clone(rq, rq_src);
2596 blk_rq_unprep_clone(rq);
2600 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2602 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2604 return queue_work(kblockd_workqueue, work);
2606 EXPORT_SYMBOL(kblockd_schedule_work);
2608 int kblockd_schedule_delayed_work(struct request_queue *q,
2609 struct delayed_work *dwork, unsigned long delay)
2611 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2613 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2615 int __init blk_dev_init(void)
2617 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2618 sizeof(((struct request *)0)->cmd_flags));
2620 kblockd_workqueue = create_workqueue("kblockd");
2621 if (!kblockd_workqueue)
2622 panic("Failed to create kblockd\n");
2624 request_cachep = kmem_cache_create("blkdev_requests",
2625 sizeof(struct request), 0, SLAB_PANIC, NULL);
2627 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2628 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);