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.
301 void __blk_run_queue(struct request_queue *q, bool force_kblockd)
303 if (unlikely(blk_queue_stopped(q)))
307 * Only recurse once to avoid overrunning the stack, let the unplug
308 * handling reinvoke the handler shortly if we already got there.
310 if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
312 queue_flag_clear(QUEUE_FLAG_REENTER, q);
314 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
316 EXPORT_SYMBOL(__blk_run_queue);
319 * blk_run_queue - run a single device queue
320 * @q: The queue to run
323 * Invoke request handling on this queue, if it has pending work to do.
324 * May be used to restart queueing when a request has completed.
326 void blk_run_queue(struct request_queue *q)
330 spin_lock_irqsave(q->queue_lock, flags);
331 __blk_run_queue(q, false);
332 spin_unlock_irqrestore(q->queue_lock, flags);
334 EXPORT_SYMBOL(blk_run_queue);
336 void blk_put_queue(struct request_queue *q)
338 kobject_put(&q->kobj);
342 * Note: If a driver supplied the queue lock, it should not zap that lock
343 * unexpectedly as some queue cleanup components like elevator_exit() and
344 * blk_throtl_exit() need queue lock.
346 void blk_cleanup_queue(struct request_queue *q)
349 * We know we have process context here, so we can be a little
350 * cautious and ensure that pending block actions on this device
351 * are done before moving on. Going into this function, we should
352 * not have processes doing IO to this device.
356 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
357 mutex_lock(&q->sysfs_lock);
358 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
359 mutex_unlock(&q->sysfs_lock);
362 elevator_exit(q->elevator);
368 EXPORT_SYMBOL(blk_cleanup_queue);
370 static int blk_init_free_list(struct request_queue *q)
372 struct request_list *rl = &q->rq;
374 if (unlikely(rl->rq_pool))
377 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
378 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
380 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
381 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
383 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
384 mempool_free_slab, request_cachep, q->node);
392 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
394 return blk_alloc_queue_node(gfp_mask, -1);
396 EXPORT_SYMBOL(blk_alloc_queue);
398 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
400 struct request_queue *q;
403 q = kmem_cache_alloc_node(blk_requestq_cachep,
404 gfp_mask | __GFP_ZERO, node_id);
408 q->backing_dev_info.ra_pages =
409 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
410 q->backing_dev_info.state = 0;
411 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
412 q->backing_dev_info.name = "block";
414 err = bdi_init(&q->backing_dev_info);
416 kmem_cache_free(blk_requestq_cachep, q);
420 if (blk_throtl_init(q)) {
421 kmem_cache_free(blk_requestq_cachep, q);
425 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
426 laptop_mode_timer_fn, (unsigned long) q);
427 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
428 INIT_LIST_HEAD(&q->timeout_list);
429 INIT_LIST_HEAD(&q->flush_queue[0]);
430 INIT_LIST_HEAD(&q->flush_queue[1]);
431 INIT_LIST_HEAD(&q->flush_data_in_flight);
432 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
434 kobject_init(&q->kobj, &blk_queue_ktype);
436 mutex_init(&q->sysfs_lock);
437 spin_lock_init(&q->__queue_lock);
440 * By default initialize queue_lock to internal lock and driver can
441 * override it later if need be.
443 q->queue_lock = &q->__queue_lock;
447 EXPORT_SYMBOL(blk_alloc_queue_node);
450 * blk_init_queue - prepare a request queue for use with a block device
451 * @rfn: The function to be called to process requests that have been
452 * placed on the queue.
453 * @lock: Request queue spin lock
456 * If a block device wishes to use the standard request handling procedures,
457 * which sorts requests and coalesces adjacent requests, then it must
458 * call blk_init_queue(). The function @rfn will be called when there
459 * are requests on the queue that need to be processed. If the device
460 * supports plugging, then @rfn may not be called immediately when requests
461 * are available on the queue, but may be called at some time later instead.
462 * Plugged queues are generally unplugged when a buffer belonging to one
463 * of the requests on the queue is needed, or due to memory pressure.
465 * @rfn is not required, or even expected, to remove all requests off the
466 * queue, but only as many as it can handle at a time. If it does leave
467 * requests on the queue, it is responsible for arranging that the requests
468 * get dealt with eventually.
470 * The queue spin lock must be held while manipulating the requests on the
471 * request queue; this lock will be taken also from interrupt context, so irq
472 * disabling is needed for it.
474 * Function returns a pointer to the initialized request queue, or %NULL if
478 * blk_init_queue() must be paired with a blk_cleanup_queue() call
479 * when the block device is deactivated (such as at module unload).
482 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
484 return blk_init_queue_node(rfn, lock, -1);
486 EXPORT_SYMBOL(blk_init_queue);
488 struct request_queue *
489 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
491 struct request_queue *uninit_q, *q;
493 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
497 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
499 blk_cleanup_queue(uninit_q);
503 EXPORT_SYMBOL(blk_init_queue_node);
505 struct request_queue *
506 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
509 return blk_init_allocated_queue_node(q, rfn, lock, -1);
511 EXPORT_SYMBOL(blk_init_allocated_queue);
513 struct request_queue *
514 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
515 spinlock_t *lock, int node_id)
521 if (blk_init_free_list(q))
525 q->prep_rq_fn = NULL;
526 q->unprep_rq_fn = NULL;
527 q->queue_flags = QUEUE_FLAG_DEFAULT;
529 /* Override internal queue lock with supplied lock pointer */
531 q->queue_lock = lock;
534 * This also sets hw/phys segments, boundary and size
536 blk_queue_make_request(q, __make_request);
538 q->sg_reserved_size = INT_MAX;
543 if (!elevator_init(q, NULL)) {
544 blk_queue_congestion_threshold(q);
550 EXPORT_SYMBOL(blk_init_allocated_queue_node);
552 int blk_get_queue(struct request_queue *q)
554 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
555 kobject_get(&q->kobj);
562 static inline void blk_free_request(struct request_queue *q, struct request *rq)
564 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
566 if (rq->cmd_flags & REQ_ELVPRIV)
567 elv_put_request(q, rq);
568 mempool_free(rq, q->rq.rq_pool);
571 static struct request *
572 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
574 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
581 rq->cmd_flags = flags | REQ_ALLOCED;
584 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
585 mempool_free(rq, q->rq.rq_pool);
588 rq->cmd_flags |= REQ_ELVPRIV;
595 * ioc_batching returns true if the ioc is a valid batching request and
596 * should be given priority access to a request.
598 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
604 * Make sure the process is able to allocate at least 1 request
605 * even if the batch times out, otherwise we could theoretically
608 return ioc->nr_batch_requests == q->nr_batching ||
609 (ioc->nr_batch_requests > 0
610 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
614 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
615 * will cause the process to be a "batcher" on all queues in the system. This
616 * is the behaviour we want though - once it gets a wakeup it should be given
619 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
621 if (!ioc || ioc_batching(q, ioc))
624 ioc->nr_batch_requests = q->nr_batching;
625 ioc->last_waited = jiffies;
628 static void __freed_request(struct request_queue *q, int sync)
630 struct request_list *rl = &q->rq;
632 if (rl->count[sync] < queue_congestion_off_threshold(q))
633 blk_clear_queue_congested(q, sync);
635 if (rl->count[sync] + 1 <= q->nr_requests) {
636 if (waitqueue_active(&rl->wait[sync]))
637 wake_up(&rl->wait[sync]);
639 blk_clear_queue_full(q, sync);
644 * A request has just been released. Account for it, update the full and
645 * congestion status, wake up any waiters. Called under q->queue_lock.
647 static void freed_request(struct request_queue *q, int sync, int priv)
649 struct request_list *rl = &q->rq;
655 __freed_request(q, sync);
657 if (unlikely(rl->starved[sync ^ 1]))
658 __freed_request(q, sync ^ 1);
662 * Determine if elevator data should be initialized when allocating the
663 * request associated with @bio.
665 static bool blk_rq_should_init_elevator(struct bio *bio)
671 * Flush requests do not use the elevator so skip initialization.
672 * This allows a request to share the flush and elevator data.
674 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
681 * Get a free request, queue_lock must be held.
682 * Returns NULL on failure, with queue_lock held.
683 * Returns !NULL on success, with queue_lock *not held*.
685 static struct request *get_request(struct request_queue *q, int rw_flags,
686 struct bio *bio, gfp_t gfp_mask)
688 struct request *rq = NULL;
689 struct request_list *rl = &q->rq;
690 struct io_context *ioc = NULL;
691 const bool is_sync = rw_is_sync(rw_flags) != 0;
692 int may_queue, priv = 0;
694 may_queue = elv_may_queue(q, rw_flags);
695 if (may_queue == ELV_MQUEUE_NO)
698 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
699 if (rl->count[is_sync]+1 >= q->nr_requests) {
700 ioc = current_io_context(GFP_ATOMIC, q->node);
702 * The queue will fill after this allocation, so set
703 * it as full, and mark this process as "batching".
704 * This process will be allowed to complete a batch of
705 * requests, others will be blocked.
707 if (!blk_queue_full(q, is_sync)) {
708 ioc_set_batching(q, ioc);
709 blk_set_queue_full(q, is_sync);
711 if (may_queue != ELV_MQUEUE_MUST
712 && !ioc_batching(q, ioc)) {
714 * The queue is full and the allocating
715 * process is not a "batcher", and not
716 * exempted by the IO scheduler
722 blk_set_queue_congested(q, is_sync);
726 * Only allow batching queuers to allocate up to 50% over the defined
727 * limit of requests, otherwise we could have thousands of requests
728 * allocated with any setting of ->nr_requests
730 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
733 rl->count[is_sync]++;
734 rl->starved[is_sync] = 0;
736 if (blk_rq_should_init_elevator(bio)) {
737 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
742 if (blk_queue_io_stat(q))
743 rw_flags |= REQ_IO_STAT;
744 spin_unlock_irq(q->queue_lock);
746 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
749 * Allocation failed presumably due to memory. Undo anything
750 * we might have messed up.
752 * Allocating task should really be put onto the front of the
753 * wait queue, but this is pretty rare.
755 spin_lock_irq(q->queue_lock);
756 freed_request(q, is_sync, priv);
759 * in the very unlikely event that allocation failed and no
760 * requests for this direction was pending, mark us starved
761 * so that freeing of a request in the other direction will
762 * notice us. another possible fix would be to split the
763 * rq mempool into READ and WRITE
766 if (unlikely(rl->count[is_sync] == 0))
767 rl->starved[is_sync] = 1;
773 * ioc may be NULL here, and ioc_batching will be false. That's
774 * OK, if the queue is under the request limit then requests need
775 * not count toward the nr_batch_requests limit. There will always
776 * be some limit enforced by BLK_BATCH_TIME.
778 if (ioc_batching(q, ioc))
779 ioc->nr_batch_requests--;
781 trace_block_getrq(q, bio, rw_flags & 1);
787 * No available requests for this queue, wait for some requests to become
790 * Called with q->queue_lock held, and returns with it unlocked.
792 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
795 const bool is_sync = rw_is_sync(rw_flags) != 0;
798 rq = get_request(q, rw_flags, bio, GFP_NOIO);
801 struct io_context *ioc;
802 struct request_list *rl = &q->rq;
804 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
805 TASK_UNINTERRUPTIBLE);
807 trace_block_sleeprq(q, bio, rw_flags & 1);
809 spin_unlock_irq(q->queue_lock);
813 * After sleeping, we become a "batching" process and
814 * will be able to allocate at least one request, and
815 * up to a big batch of them for a small period time.
816 * See ioc_batching, ioc_set_batching
818 ioc = current_io_context(GFP_NOIO, q->node);
819 ioc_set_batching(q, ioc);
821 spin_lock_irq(q->queue_lock);
822 finish_wait(&rl->wait[is_sync], &wait);
824 rq = get_request(q, rw_flags, bio, GFP_NOIO);
830 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
834 BUG_ON(rw != READ && rw != WRITE);
836 spin_lock_irq(q->queue_lock);
837 if (gfp_mask & __GFP_WAIT) {
838 rq = get_request_wait(q, rw, NULL);
840 rq = get_request(q, rw, NULL, gfp_mask);
842 spin_unlock_irq(q->queue_lock);
844 /* q->queue_lock is unlocked at this point */
848 EXPORT_SYMBOL(blk_get_request);
851 * blk_make_request - given a bio, allocate a corresponding struct request.
852 * @q: target request queue
853 * @bio: The bio describing the memory mappings that will be submitted for IO.
854 * It may be a chained-bio properly constructed by block/bio layer.
855 * @gfp_mask: gfp flags to be used for memory allocation
857 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
858 * type commands. Where the struct request needs to be farther initialized by
859 * the caller. It is passed a &struct bio, which describes the memory info of
862 * The caller of blk_make_request must make sure that bi_io_vec
863 * are set to describe the memory buffers. That bio_data_dir() will return
864 * the needed direction of the request. (And all bio's in the passed bio-chain
865 * are properly set accordingly)
867 * If called under none-sleepable conditions, mapped bio buffers must not
868 * need bouncing, by calling the appropriate masked or flagged allocator,
869 * suitable for the target device. Otherwise the call to blk_queue_bounce will
872 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
873 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
874 * anything but the first bio in the chain. Otherwise you risk waiting for IO
875 * completion of a bio that hasn't been submitted yet, thus resulting in a
876 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
877 * of bio_alloc(), as that avoids the mempool deadlock.
878 * If possible a big IO should be split into smaller parts when allocation
879 * fails. Partial allocation should not be an error, or you risk a live-lock.
881 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
884 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
887 return ERR_PTR(-ENOMEM);
890 struct bio *bounce_bio = bio;
893 blk_queue_bounce(q, &bounce_bio);
894 ret = blk_rq_append_bio(q, rq, bounce_bio);
903 EXPORT_SYMBOL(blk_make_request);
906 * blk_requeue_request - put a request back on queue
907 * @q: request queue where request should be inserted
908 * @rq: request to be inserted
911 * Drivers often keep queueing requests until the hardware cannot accept
912 * more, when that condition happens we need to put the request back
913 * on the queue. Must be called with queue lock held.
915 void blk_requeue_request(struct request_queue *q, struct request *rq)
917 blk_delete_timer(rq);
918 blk_clear_rq_complete(rq);
919 trace_block_rq_requeue(q, rq);
921 if (blk_rq_tagged(rq))
922 blk_queue_end_tag(q, rq);
924 BUG_ON(blk_queued_rq(rq));
926 elv_requeue_request(q, rq);
928 EXPORT_SYMBOL(blk_requeue_request);
930 static void add_acct_request(struct request_queue *q, struct request *rq,
933 drive_stat_acct(rq, 1);
934 __elv_add_request(q, rq, where);
938 * blk_insert_request - insert a special request into a request queue
939 * @q: request queue where request should be inserted
940 * @rq: request to be inserted
941 * @at_head: insert request at head or tail of queue
942 * @data: private data
945 * Many block devices need to execute commands asynchronously, so they don't
946 * block the whole kernel from preemption during request execution. This is
947 * accomplished normally by inserting aritficial requests tagged as
948 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
949 * be scheduled for actual execution by the request queue.
951 * We have the option of inserting the head or the tail of the queue.
952 * Typically we use the tail for new ioctls and so forth. We use the head
953 * of the queue for things like a QUEUE_FULL message from a device, or a
954 * host that is unable to accept a particular command.
956 void blk_insert_request(struct request_queue *q, struct request *rq,
957 int at_head, void *data)
959 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
963 * tell I/O scheduler that this isn't a regular read/write (ie it
964 * must not attempt merges on this) and that it acts as a soft
967 rq->cmd_type = REQ_TYPE_SPECIAL;
971 spin_lock_irqsave(q->queue_lock, flags);
974 * If command is tagged, release the tag
976 if (blk_rq_tagged(rq))
977 blk_queue_end_tag(q, rq);
979 add_acct_request(q, rq, where);
980 __blk_run_queue(q, false);
981 spin_unlock_irqrestore(q->queue_lock, flags);
983 EXPORT_SYMBOL(blk_insert_request);
985 static void part_round_stats_single(int cpu, struct hd_struct *part,
988 if (now == part->stamp)
991 if (part_in_flight(part)) {
992 __part_stat_add(cpu, part, time_in_queue,
993 part_in_flight(part) * (now - part->stamp));
994 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1000 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1001 * @cpu: cpu number for stats access
1002 * @part: target partition
1004 * The average IO queue length and utilisation statistics are maintained
1005 * by observing the current state of the queue length and the amount of
1006 * time it has been in this state for.
1008 * Normally, that accounting is done on IO completion, but that can result
1009 * in more than a second's worth of IO being accounted for within any one
1010 * second, leading to >100% utilisation. To deal with that, we call this
1011 * function to do a round-off before returning the results when reading
1012 * /proc/diskstats. This accounts immediately for all queue usage up to
1013 * the current jiffies and restarts the counters again.
1015 void part_round_stats(int cpu, struct hd_struct *part)
1017 unsigned long now = jiffies;
1020 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1021 part_round_stats_single(cpu, part, now);
1023 EXPORT_SYMBOL_GPL(part_round_stats);
1026 * queue lock must be held
1028 void __blk_put_request(struct request_queue *q, struct request *req)
1032 if (unlikely(--req->ref_count))
1035 elv_completed_request(q, req);
1037 /* this is a bio leak */
1038 WARN_ON(req->bio != NULL);
1041 * Request may not have originated from ll_rw_blk. if not,
1042 * it didn't come out of our reserved rq pools
1044 if (req->cmd_flags & REQ_ALLOCED) {
1045 int is_sync = rq_is_sync(req) != 0;
1046 int priv = req->cmd_flags & REQ_ELVPRIV;
1048 BUG_ON(!list_empty(&req->queuelist));
1049 BUG_ON(!hlist_unhashed(&req->hash));
1051 blk_free_request(q, req);
1052 freed_request(q, is_sync, priv);
1055 EXPORT_SYMBOL_GPL(__blk_put_request);
1057 void blk_put_request(struct request *req)
1059 unsigned long flags;
1060 struct request_queue *q = req->q;
1062 spin_lock_irqsave(q->queue_lock, flags);
1063 __blk_put_request(q, req);
1064 spin_unlock_irqrestore(q->queue_lock, flags);
1066 EXPORT_SYMBOL(blk_put_request);
1069 * blk_add_request_payload - add a payload to a request
1070 * @rq: request to update
1071 * @page: page backing the payload
1072 * @len: length of the payload.
1074 * This allows to later add a payload to an already submitted request by
1075 * a block driver. The driver needs to take care of freeing the payload
1078 * Note that this is a quite horrible hack and nothing but handling of
1079 * discard requests should ever use it.
1081 void blk_add_request_payload(struct request *rq, struct page *page,
1084 struct bio *bio = rq->bio;
1086 bio->bi_io_vec->bv_page = page;
1087 bio->bi_io_vec->bv_offset = 0;
1088 bio->bi_io_vec->bv_len = len;
1092 bio->bi_phys_segments = 1;
1094 rq->__data_len = rq->resid_len = len;
1095 rq->nr_phys_segments = 1;
1096 rq->buffer = bio_data(bio);
1098 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1100 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1103 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1106 * Debug stuff, kill later
1108 if (!rq_mergeable(req)) {
1109 blk_dump_rq_flags(req, "back");
1113 if (!ll_back_merge_fn(q, req, bio))
1116 trace_block_bio_backmerge(q, bio);
1118 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1119 blk_rq_set_mixed_merge(req);
1121 req->biotail->bi_next = bio;
1123 req->__data_len += bio->bi_size;
1124 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1126 drive_stat_acct(req, 0);
1130 static bool bio_attempt_front_merge(struct request_queue *q,
1131 struct request *req, struct bio *bio)
1133 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1137 * Debug stuff, kill later
1139 if (!rq_mergeable(req)) {
1140 blk_dump_rq_flags(req, "front");
1144 if (!ll_front_merge_fn(q, req, bio))
1147 trace_block_bio_frontmerge(q, bio);
1149 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1150 blk_rq_set_mixed_merge(req);
1152 sector = bio->bi_sector;
1154 bio->bi_next = req->bio;
1158 * may not be valid. if the low level driver said
1159 * it didn't need a bounce buffer then it better
1160 * not touch req->buffer either...
1162 req->buffer = bio_data(bio);
1163 req->__sector = bio->bi_sector;
1164 req->__data_len += bio->bi_size;
1165 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1167 drive_stat_acct(req, 0);
1172 * Attempts to merge with the plugged list in the current process. Returns
1173 * true if merge was successful, otherwise false.
1175 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1178 struct blk_plug *plug;
1186 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1192 el_ret = elv_try_merge(rq, bio);
1193 if (el_ret == ELEVATOR_BACK_MERGE) {
1194 ret = bio_attempt_back_merge(q, rq, bio);
1197 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1198 ret = bio_attempt_front_merge(q, rq, bio);
1207 void init_request_from_bio(struct request *req, struct bio *bio)
1209 req->cpu = bio->bi_comp_cpu;
1210 req->cmd_type = REQ_TYPE_FS;
1212 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1213 if (bio->bi_rw & REQ_RAHEAD)
1214 req->cmd_flags |= REQ_FAILFAST_MASK;
1217 req->__sector = bio->bi_sector;
1218 req->ioprio = bio_prio(bio);
1219 blk_rq_bio_prep(req->q, req, bio);
1222 static int __make_request(struct request_queue *q, struct bio *bio)
1224 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1225 struct blk_plug *plug;
1226 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1227 struct request *req;
1230 * low level driver can indicate that it wants pages above a
1231 * certain limit bounced to low memory (ie for highmem, or even
1232 * ISA dma in theory)
1234 blk_queue_bounce(q, &bio);
1236 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1237 spin_lock_irq(q->queue_lock);
1238 where = ELEVATOR_INSERT_FLUSH;
1243 * Check if we can merge with the plugged list before grabbing
1246 if (attempt_plug_merge(current, q, bio))
1249 spin_lock_irq(q->queue_lock);
1251 el_ret = elv_merge(q, &req, bio);
1252 if (el_ret == ELEVATOR_BACK_MERGE) {
1253 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1254 if (bio_attempt_back_merge(q, req, bio)) {
1255 if (!attempt_back_merge(q, req))
1256 elv_merged_request(q, req, el_ret);
1259 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1260 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1261 if (bio_attempt_front_merge(q, req, bio)) {
1262 if (!attempt_front_merge(q, req))
1263 elv_merged_request(q, req, el_ret);
1270 * This sync check and mask will be re-done in init_request_from_bio(),
1271 * but we need to set it earlier to expose the sync flag to the
1272 * rq allocator and io schedulers.
1274 rw_flags = bio_data_dir(bio);
1276 rw_flags |= REQ_SYNC;
1279 * Grab a free request. This is might sleep but can not fail.
1280 * Returns with the queue unlocked.
1282 req = get_request_wait(q, rw_flags, bio);
1285 * After dropping the lock and possibly sleeping here, our request
1286 * may now be mergeable after it had proven unmergeable (above).
1287 * We don't worry about that case for efficiency. It won't happen
1288 * often, and the elevators are able to handle it.
1290 init_request_from_bio(req, bio);
1292 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1293 bio_flagged(bio, BIO_CPU_AFFINE)) {
1294 req->cpu = blk_cpu_to_group(get_cpu());
1298 plug = current->plug;
1301 * If this is the first request added after a plug, fire
1302 * of a plug trace. If others have been added before, check
1303 * if we have multiple devices in this plug. If so, make a
1304 * note to sort the list before dispatch.
1306 if (list_empty(&plug->list))
1307 trace_block_plug(q);
1308 else if (!plug->should_sort) {
1309 struct request *__rq;
1311 __rq = list_entry_rq(plug->list.prev);
1313 plug->should_sort = 1;
1316 * Debug flag, kill later
1318 req->cmd_flags |= REQ_ON_PLUG;
1319 list_add_tail(&req->queuelist, &plug->list);
1320 drive_stat_acct(req, 1);
1322 spin_lock_irq(q->queue_lock);
1323 add_acct_request(q, req, where);
1324 __blk_run_queue(q, false);
1326 spin_unlock_irq(q->queue_lock);
1333 * If bio->bi_dev is a partition, remap the location
1335 static inline void blk_partition_remap(struct bio *bio)
1337 struct block_device *bdev = bio->bi_bdev;
1339 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1340 struct hd_struct *p = bdev->bd_part;
1342 bio->bi_sector += p->start_sect;
1343 bio->bi_bdev = bdev->bd_contains;
1345 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1347 bio->bi_sector - p->start_sect);
1351 static void handle_bad_sector(struct bio *bio)
1353 char b[BDEVNAME_SIZE];
1355 printk(KERN_INFO "attempt to access beyond end of device\n");
1356 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1357 bdevname(bio->bi_bdev, b),
1359 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1360 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1362 set_bit(BIO_EOF, &bio->bi_flags);
1365 #ifdef CONFIG_FAIL_MAKE_REQUEST
1367 static DECLARE_FAULT_ATTR(fail_make_request);
1369 static int __init setup_fail_make_request(char *str)
1371 return setup_fault_attr(&fail_make_request, str);
1373 __setup("fail_make_request=", setup_fail_make_request);
1375 static int should_fail_request(struct bio *bio)
1377 struct hd_struct *part = bio->bi_bdev->bd_part;
1379 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1380 return should_fail(&fail_make_request, bio->bi_size);
1385 static int __init fail_make_request_debugfs(void)
1387 return init_fault_attr_dentries(&fail_make_request,
1388 "fail_make_request");
1391 late_initcall(fail_make_request_debugfs);
1393 #else /* CONFIG_FAIL_MAKE_REQUEST */
1395 static inline int should_fail_request(struct bio *bio)
1400 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1403 * Check whether this bio extends beyond the end of the device.
1405 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1412 /* Test device or partition size, when known. */
1413 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1415 sector_t sector = bio->bi_sector;
1417 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1419 * This may well happen - the kernel calls bread()
1420 * without checking the size of the device, e.g., when
1421 * mounting a device.
1423 handle_bad_sector(bio);
1432 * generic_make_request - hand a buffer to its device driver for I/O
1433 * @bio: The bio describing the location in memory and on the device.
1435 * generic_make_request() is used to make I/O requests of block
1436 * devices. It is passed a &struct bio, which describes the I/O that needs
1439 * generic_make_request() does not return any status. The
1440 * success/failure status of the request, along with notification of
1441 * completion, is delivered asynchronously through the bio->bi_end_io
1442 * function described (one day) else where.
1444 * The caller of generic_make_request must make sure that bi_io_vec
1445 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1446 * set to describe the device address, and the
1447 * bi_end_io and optionally bi_private are set to describe how
1448 * completion notification should be signaled.
1450 * generic_make_request and the drivers it calls may use bi_next if this
1451 * bio happens to be merged with someone else, and may change bi_dev and
1452 * bi_sector for remaps as it sees fit. So the values of these fields
1453 * should NOT be depended on after the call to generic_make_request.
1455 static inline void __generic_make_request(struct bio *bio)
1457 struct request_queue *q;
1458 sector_t old_sector;
1459 int ret, nr_sectors = bio_sectors(bio);
1465 if (bio_check_eod(bio, nr_sectors))
1469 * Resolve the mapping until finished. (drivers are
1470 * still free to implement/resolve their own stacking
1471 * by explicitly returning 0)
1473 * NOTE: we don't repeat the blk_size check for each new device.
1474 * Stacking drivers are expected to know what they are doing.
1479 char b[BDEVNAME_SIZE];
1481 q = bdev_get_queue(bio->bi_bdev);
1484 "generic_make_request: Trying to access "
1485 "nonexistent block-device %s (%Lu)\n",
1486 bdevname(bio->bi_bdev, b),
1487 (long long) bio->bi_sector);
1491 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1492 nr_sectors > queue_max_hw_sectors(q))) {
1493 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1494 bdevname(bio->bi_bdev, b),
1496 queue_max_hw_sectors(q));
1500 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1503 if (should_fail_request(bio))
1507 * If this device has partitions, remap block n
1508 * of partition p to block n+start(p) of the disk.
1510 blk_partition_remap(bio);
1512 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1515 if (old_sector != -1)
1516 trace_block_bio_remap(q, bio, old_dev, old_sector);
1518 old_sector = bio->bi_sector;
1519 old_dev = bio->bi_bdev->bd_dev;
1521 if (bio_check_eod(bio, nr_sectors))
1525 * Filter flush bio's early so that make_request based
1526 * drivers without flush support don't have to worry
1529 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1530 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1537 if ((bio->bi_rw & REQ_DISCARD) &&
1538 (!blk_queue_discard(q) ||
1539 ((bio->bi_rw & REQ_SECURE) &&
1540 !blk_queue_secdiscard(q)))) {
1545 blk_throtl_bio(q, &bio);
1548 * If bio = NULL, bio has been throttled and will be submitted
1554 trace_block_bio_queue(q, bio);
1556 ret = q->make_request_fn(q, bio);
1562 bio_endio(bio, err);
1566 * We only want one ->make_request_fn to be active at a time,
1567 * else stack usage with stacked devices could be a problem.
1568 * So use current->bio_list to keep a list of requests
1569 * submited by a make_request_fn function.
1570 * current->bio_list is also used as a flag to say if
1571 * generic_make_request is currently active in this task or not.
1572 * If it is NULL, then no make_request is active. If it is non-NULL,
1573 * then a make_request is active, and new requests should be added
1576 void generic_make_request(struct bio *bio)
1578 struct bio_list bio_list_on_stack;
1580 if (current->bio_list) {
1581 /* make_request is active */
1582 bio_list_add(current->bio_list, bio);
1585 /* following loop may be a bit non-obvious, and so deserves some
1587 * Before entering the loop, bio->bi_next is NULL (as all callers
1588 * ensure that) so we have a list with a single bio.
1589 * We pretend that we have just taken it off a longer list, so
1590 * we assign bio_list to a pointer to the bio_list_on_stack,
1591 * thus initialising the bio_list of new bios to be
1592 * added. __generic_make_request may indeed add some more bios
1593 * through a recursive call to generic_make_request. If it
1594 * did, we find a non-NULL value in bio_list and re-enter the loop
1595 * from the top. In this case we really did just take the bio
1596 * of the top of the list (no pretending) and so remove it from
1597 * bio_list, and call into __generic_make_request again.
1599 * The loop was structured like this to make only one call to
1600 * __generic_make_request (which is important as it is large and
1601 * inlined) and to keep the structure simple.
1603 BUG_ON(bio->bi_next);
1604 bio_list_init(&bio_list_on_stack);
1605 current->bio_list = &bio_list_on_stack;
1607 __generic_make_request(bio);
1608 bio = bio_list_pop(current->bio_list);
1610 current->bio_list = NULL; /* deactivate */
1612 EXPORT_SYMBOL(generic_make_request);
1615 * submit_bio - submit a bio to the block device layer for I/O
1616 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1617 * @bio: The &struct bio which describes the I/O
1619 * submit_bio() is very similar in purpose to generic_make_request(), and
1620 * uses that function to do most of the work. Both are fairly rough
1621 * interfaces; @bio must be presetup and ready for I/O.
1624 void submit_bio(int rw, struct bio *bio)
1626 int count = bio_sectors(bio);
1631 * If it's a regular read/write or a barrier with data attached,
1632 * go through the normal accounting stuff before submission.
1634 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1636 count_vm_events(PGPGOUT, count);
1638 task_io_account_read(bio->bi_size);
1639 count_vm_events(PGPGIN, count);
1642 if (unlikely(block_dump)) {
1643 char b[BDEVNAME_SIZE];
1644 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1645 current->comm, task_pid_nr(current),
1646 (rw & WRITE) ? "WRITE" : "READ",
1647 (unsigned long long)bio->bi_sector,
1648 bdevname(bio->bi_bdev, b),
1653 generic_make_request(bio);
1655 EXPORT_SYMBOL(submit_bio);
1658 * blk_rq_check_limits - Helper function to check a request for the queue limit
1660 * @rq: the request being checked
1663 * @rq may have been made based on weaker limitations of upper-level queues
1664 * in request stacking drivers, and it may violate the limitation of @q.
1665 * Since the block layer and the underlying device driver trust @rq
1666 * after it is inserted to @q, it should be checked against @q before
1667 * the insertion using this generic function.
1669 * This function should also be useful for request stacking drivers
1670 * in some cases below, so export this function.
1671 * Request stacking drivers like request-based dm may change the queue
1672 * limits while requests are in the queue (e.g. dm's table swapping).
1673 * Such request stacking drivers should check those requests agaist
1674 * the new queue limits again when they dispatch those requests,
1675 * although such checkings are also done against the old queue limits
1676 * when submitting requests.
1678 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1680 if (rq->cmd_flags & REQ_DISCARD)
1683 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1684 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1685 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1690 * queue's settings related to segment counting like q->bounce_pfn
1691 * may differ from that of other stacking queues.
1692 * Recalculate it to check the request correctly on this queue's
1695 blk_recalc_rq_segments(rq);
1696 if (rq->nr_phys_segments > queue_max_segments(q)) {
1697 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1703 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1706 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1707 * @q: the queue to submit the request
1708 * @rq: the request being queued
1710 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1712 unsigned long flags;
1714 if (blk_rq_check_limits(q, rq))
1717 #ifdef CONFIG_FAIL_MAKE_REQUEST
1718 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1719 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1723 spin_lock_irqsave(q->queue_lock, flags);
1726 * Submitting request must be dequeued before calling this function
1727 * because it will be linked to another request_queue
1729 BUG_ON(blk_queued_rq(rq));
1731 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1732 spin_unlock_irqrestore(q->queue_lock, flags);
1736 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1739 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1740 * @rq: request to examine
1743 * A request could be merge of IOs which require different failure
1744 * handling. This function determines the number of bytes which
1745 * can be failed from the beginning of the request without
1746 * crossing into area which need to be retried further.
1749 * The number of bytes to fail.
1752 * queue_lock must be held.
1754 unsigned int blk_rq_err_bytes(const struct request *rq)
1756 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1757 unsigned int bytes = 0;
1760 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1761 return blk_rq_bytes(rq);
1764 * Currently the only 'mixing' which can happen is between
1765 * different fastfail types. We can safely fail portions
1766 * which have all the failfast bits that the first one has -
1767 * the ones which are at least as eager to fail as the first
1770 for (bio = rq->bio; bio; bio = bio->bi_next) {
1771 if ((bio->bi_rw & ff) != ff)
1773 bytes += bio->bi_size;
1776 /* this could lead to infinite loop */
1777 BUG_ON(blk_rq_bytes(rq) && !bytes);
1780 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1782 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1784 if (blk_do_io_stat(req)) {
1785 const int rw = rq_data_dir(req);
1786 struct hd_struct *part;
1789 cpu = part_stat_lock();
1791 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1796 static void blk_account_io_done(struct request *req)
1799 * Account IO completion. flush_rq isn't accounted as a
1800 * normal IO on queueing nor completion. Accounting the
1801 * containing request is enough.
1803 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1804 unsigned long duration = jiffies - req->start_time;
1805 const int rw = rq_data_dir(req);
1806 struct hd_struct *part;
1809 cpu = part_stat_lock();
1812 part_stat_inc(cpu, part, ios[rw]);
1813 part_stat_add(cpu, part, ticks[rw], duration);
1814 part_round_stats(cpu, part);
1815 part_dec_in_flight(part, rw);
1817 hd_struct_put(part);
1823 * blk_peek_request - peek at the top of a request queue
1824 * @q: request queue to peek at
1827 * Return the request at the top of @q. The returned request
1828 * should be started using blk_start_request() before LLD starts
1832 * Pointer to the request at the top of @q if available. Null
1836 * queue_lock must be held.
1838 struct request *blk_peek_request(struct request_queue *q)
1843 while ((rq = __elv_next_request(q)) != NULL) {
1844 if (!(rq->cmd_flags & REQ_STARTED)) {
1846 * This is the first time the device driver
1847 * sees this request (possibly after
1848 * requeueing). Notify IO scheduler.
1850 if (rq->cmd_flags & REQ_SORTED)
1851 elv_activate_rq(q, rq);
1854 * just mark as started even if we don't start
1855 * it, a request that has been delayed should
1856 * not be passed by new incoming requests
1858 rq->cmd_flags |= REQ_STARTED;
1859 trace_block_rq_issue(q, rq);
1862 if (!q->boundary_rq || q->boundary_rq == rq) {
1863 q->end_sector = rq_end_sector(rq);
1864 q->boundary_rq = NULL;
1867 if (rq->cmd_flags & REQ_DONTPREP)
1870 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1872 * make sure space for the drain appears we
1873 * know we can do this because max_hw_segments
1874 * has been adjusted to be one fewer than the
1877 rq->nr_phys_segments++;
1883 ret = q->prep_rq_fn(q, rq);
1884 if (ret == BLKPREP_OK) {
1886 } else if (ret == BLKPREP_DEFER) {
1888 * the request may have been (partially) prepped.
1889 * we need to keep this request in the front to
1890 * avoid resource deadlock. REQ_STARTED will
1891 * prevent other fs requests from passing this one.
1893 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1894 !(rq->cmd_flags & REQ_DONTPREP)) {
1896 * remove the space for the drain we added
1897 * so that we don't add it again
1899 --rq->nr_phys_segments;
1904 } else if (ret == BLKPREP_KILL) {
1905 rq->cmd_flags |= REQ_QUIET;
1907 * Mark this request as started so we don't trigger
1908 * any debug logic in the end I/O path.
1910 blk_start_request(rq);
1911 __blk_end_request_all(rq, -EIO);
1913 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1920 EXPORT_SYMBOL(blk_peek_request);
1922 void blk_dequeue_request(struct request *rq)
1924 struct request_queue *q = rq->q;
1926 BUG_ON(list_empty(&rq->queuelist));
1927 BUG_ON(ELV_ON_HASH(rq));
1929 list_del_init(&rq->queuelist);
1932 * the time frame between a request being removed from the lists
1933 * and to it is freed is accounted as io that is in progress at
1936 if (blk_account_rq(rq)) {
1937 q->in_flight[rq_is_sync(rq)]++;
1938 set_io_start_time_ns(rq);
1943 * blk_start_request - start request processing on the driver
1944 * @req: request to dequeue
1947 * Dequeue @req and start timeout timer on it. This hands off the
1948 * request to the driver.
1950 * Block internal functions which don't want to start timer should
1951 * call blk_dequeue_request().
1954 * queue_lock must be held.
1956 void blk_start_request(struct request *req)
1958 blk_dequeue_request(req);
1961 * We are now handing the request to the hardware, initialize
1962 * resid_len to full count and add the timeout handler.
1964 req->resid_len = blk_rq_bytes(req);
1965 if (unlikely(blk_bidi_rq(req)))
1966 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1970 EXPORT_SYMBOL(blk_start_request);
1973 * blk_fetch_request - fetch a request from a request queue
1974 * @q: request queue to fetch a request from
1977 * Return the request at the top of @q. The request is started on
1978 * return and LLD can start processing it immediately.
1981 * Pointer to the request at the top of @q if available. Null
1985 * queue_lock must be held.
1987 struct request *blk_fetch_request(struct request_queue *q)
1991 rq = blk_peek_request(q);
1993 blk_start_request(rq);
1996 EXPORT_SYMBOL(blk_fetch_request);
1999 * blk_update_request - Special helper function for request stacking drivers
2000 * @req: the request being processed
2001 * @error: %0 for success, < %0 for error
2002 * @nr_bytes: number of bytes to complete @req
2005 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2006 * the request structure even if @req doesn't have leftover.
2007 * If @req has leftover, sets it up for the next range of segments.
2009 * This special helper function is only for request stacking drivers
2010 * (e.g. request-based dm) so that they can handle partial completion.
2011 * Actual device drivers should use blk_end_request instead.
2013 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2014 * %false return from this function.
2017 * %false - this request doesn't have any more data
2018 * %true - this request has more data
2020 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2022 int total_bytes, bio_nbytes, next_idx = 0;
2028 trace_block_rq_complete(req->q, req);
2031 * For fs requests, rq is just carrier of independent bio's
2032 * and each partial completion should be handled separately.
2033 * Reset per-request error on each partial completion.
2035 * TODO: tj: This is too subtle. It would be better to let
2036 * low level drivers do what they see fit.
2038 if (req->cmd_type == REQ_TYPE_FS)
2041 if (error && req->cmd_type == REQ_TYPE_FS &&
2042 !(req->cmd_flags & REQ_QUIET)) {
2047 error_type = "recoverable transport";
2050 error_type = "critical target";
2053 error_type = "critical nexus";
2060 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2061 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2062 (unsigned long long)blk_rq_pos(req));
2065 blk_account_io_completion(req, nr_bytes);
2067 total_bytes = bio_nbytes = 0;
2068 while ((bio = req->bio) != NULL) {
2071 if (nr_bytes >= bio->bi_size) {
2072 req->bio = bio->bi_next;
2073 nbytes = bio->bi_size;
2074 req_bio_endio(req, bio, nbytes, error);
2078 int idx = bio->bi_idx + next_idx;
2080 if (unlikely(idx >= bio->bi_vcnt)) {
2081 blk_dump_rq_flags(req, "__end_that");
2082 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2083 __func__, idx, bio->bi_vcnt);
2087 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2088 BIO_BUG_ON(nbytes > bio->bi_size);
2091 * not a complete bvec done
2093 if (unlikely(nbytes > nr_bytes)) {
2094 bio_nbytes += nr_bytes;
2095 total_bytes += nr_bytes;
2100 * advance to the next vector
2103 bio_nbytes += nbytes;
2106 total_bytes += nbytes;
2112 * end more in this run, or just return 'not-done'
2114 if (unlikely(nr_bytes <= 0))
2124 * Reset counters so that the request stacking driver
2125 * can find how many bytes remain in the request
2128 req->__data_len = 0;
2133 * if the request wasn't completed, update state
2136 req_bio_endio(req, bio, bio_nbytes, error);
2137 bio->bi_idx += next_idx;
2138 bio_iovec(bio)->bv_offset += nr_bytes;
2139 bio_iovec(bio)->bv_len -= nr_bytes;
2142 req->__data_len -= total_bytes;
2143 req->buffer = bio_data(req->bio);
2145 /* update sector only for requests with clear definition of sector */
2146 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2147 req->__sector += total_bytes >> 9;
2149 /* mixed attributes always follow the first bio */
2150 if (req->cmd_flags & REQ_MIXED_MERGE) {
2151 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2152 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2156 * If total number of sectors is less than the first segment
2157 * size, something has gone terribly wrong.
2159 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2160 blk_dump_rq_flags(req, "request botched");
2161 req->__data_len = blk_rq_cur_bytes(req);
2164 /* recalculate the number of segments */
2165 blk_recalc_rq_segments(req);
2169 EXPORT_SYMBOL_GPL(blk_update_request);
2171 static bool blk_update_bidi_request(struct request *rq, int error,
2172 unsigned int nr_bytes,
2173 unsigned int bidi_bytes)
2175 if (blk_update_request(rq, error, nr_bytes))
2178 /* Bidi request must be completed as a whole */
2179 if (unlikely(blk_bidi_rq(rq)) &&
2180 blk_update_request(rq->next_rq, error, bidi_bytes))
2183 if (blk_queue_add_random(rq->q))
2184 add_disk_randomness(rq->rq_disk);
2190 * blk_unprep_request - unprepare a request
2193 * This function makes a request ready for complete resubmission (or
2194 * completion). It happens only after all error handling is complete,
2195 * so represents the appropriate moment to deallocate any resources
2196 * that were allocated to the request in the prep_rq_fn. The queue
2197 * lock is held when calling this.
2199 void blk_unprep_request(struct request *req)
2201 struct request_queue *q = req->q;
2203 req->cmd_flags &= ~REQ_DONTPREP;
2204 if (q->unprep_rq_fn)
2205 q->unprep_rq_fn(q, req);
2207 EXPORT_SYMBOL_GPL(blk_unprep_request);
2210 * queue lock must be held
2212 static void blk_finish_request(struct request *req, int error)
2214 if (blk_rq_tagged(req))
2215 blk_queue_end_tag(req->q, req);
2217 BUG_ON(blk_queued_rq(req));
2219 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2220 laptop_io_completion(&req->q->backing_dev_info);
2222 blk_delete_timer(req);
2224 if (req->cmd_flags & REQ_DONTPREP)
2225 blk_unprep_request(req);
2228 blk_account_io_done(req);
2231 req->end_io(req, error);
2233 if (blk_bidi_rq(req))
2234 __blk_put_request(req->next_rq->q, req->next_rq);
2236 __blk_put_request(req->q, req);
2241 * blk_end_bidi_request - Complete a bidi request
2242 * @rq: the request to complete
2243 * @error: %0 for success, < %0 for error
2244 * @nr_bytes: number of bytes to complete @rq
2245 * @bidi_bytes: number of bytes to complete @rq->next_rq
2248 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2249 * Drivers that supports bidi can safely call this member for any
2250 * type of request, bidi or uni. In the later case @bidi_bytes is
2254 * %false - we are done with this request
2255 * %true - still buffers pending for this request
2257 static bool blk_end_bidi_request(struct request *rq, int error,
2258 unsigned int nr_bytes, unsigned int bidi_bytes)
2260 struct request_queue *q = rq->q;
2261 unsigned long flags;
2263 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2266 spin_lock_irqsave(q->queue_lock, flags);
2267 blk_finish_request(rq, error);
2268 spin_unlock_irqrestore(q->queue_lock, flags);
2274 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2275 * @rq: the request to complete
2276 * @error: %0 for success, < %0 for error
2277 * @nr_bytes: number of bytes to complete @rq
2278 * @bidi_bytes: number of bytes to complete @rq->next_rq
2281 * Identical to blk_end_bidi_request() except that queue lock is
2282 * assumed to be locked on entry and remains so on return.
2285 * %false - we are done with this request
2286 * %true - still buffers pending for this request
2288 static bool __blk_end_bidi_request(struct request *rq, int error,
2289 unsigned int nr_bytes, unsigned int bidi_bytes)
2291 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2294 blk_finish_request(rq, error);
2300 * blk_end_request - Helper function for drivers to complete the request.
2301 * @rq: the request being processed
2302 * @error: %0 for success, < %0 for error
2303 * @nr_bytes: number of bytes to complete
2306 * Ends I/O on a number of bytes attached to @rq.
2307 * If @rq has leftover, sets it up for the next range of segments.
2310 * %false - we are done with this request
2311 * %true - still buffers pending for this request
2313 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2315 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2317 EXPORT_SYMBOL(blk_end_request);
2320 * blk_end_request_all - Helper function for drives to finish the request.
2321 * @rq: the request to finish
2322 * @error: %0 for success, < %0 for error
2325 * Completely finish @rq.
2327 void blk_end_request_all(struct request *rq, int error)
2330 unsigned int bidi_bytes = 0;
2332 if (unlikely(blk_bidi_rq(rq)))
2333 bidi_bytes = blk_rq_bytes(rq->next_rq);
2335 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2338 EXPORT_SYMBOL(blk_end_request_all);
2341 * blk_end_request_cur - Helper function to finish the current request chunk.
2342 * @rq: the request to finish the current chunk for
2343 * @error: %0 for success, < %0 for error
2346 * Complete the current consecutively mapped chunk from @rq.
2349 * %false - we are done with this request
2350 * %true - still buffers pending for this request
2352 bool blk_end_request_cur(struct request *rq, int error)
2354 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2356 EXPORT_SYMBOL(blk_end_request_cur);
2359 * blk_end_request_err - Finish a request till the next failure boundary.
2360 * @rq: the request to finish till the next failure boundary for
2361 * @error: must be negative errno
2364 * Complete @rq till the next failure boundary.
2367 * %false - we are done with this request
2368 * %true - still buffers pending for this request
2370 bool blk_end_request_err(struct request *rq, int error)
2372 WARN_ON(error >= 0);
2373 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2375 EXPORT_SYMBOL_GPL(blk_end_request_err);
2378 * __blk_end_request - Helper function for drivers to complete the request.
2379 * @rq: the request being processed
2380 * @error: %0 for success, < %0 for error
2381 * @nr_bytes: number of bytes to complete
2384 * Must be called with queue lock held unlike blk_end_request().
2387 * %false - we are done with this request
2388 * %true - still buffers pending for this request
2390 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2392 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2394 EXPORT_SYMBOL(__blk_end_request);
2397 * __blk_end_request_all - Helper function for drives to finish the request.
2398 * @rq: the request to finish
2399 * @error: %0 for success, < %0 for error
2402 * Completely finish @rq. Must be called with queue lock held.
2404 void __blk_end_request_all(struct request *rq, int error)
2407 unsigned int bidi_bytes = 0;
2409 if (unlikely(blk_bidi_rq(rq)))
2410 bidi_bytes = blk_rq_bytes(rq->next_rq);
2412 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2415 EXPORT_SYMBOL(__blk_end_request_all);
2418 * __blk_end_request_cur - Helper function to finish the current request chunk.
2419 * @rq: the request to finish the current chunk for
2420 * @error: %0 for success, < %0 for error
2423 * Complete the current consecutively mapped chunk from @rq. Must
2424 * be called with queue lock held.
2427 * %false - we are done with this request
2428 * %true - still buffers pending for this request
2430 bool __blk_end_request_cur(struct request *rq, int error)
2432 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2434 EXPORT_SYMBOL(__blk_end_request_cur);
2437 * __blk_end_request_err - Finish a request till the next failure boundary.
2438 * @rq: the request to finish till the next failure boundary for
2439 * @error: must be negative errno
2442 * Complete @rq till the next failure boundary. Must be called
2443 * with queue lock held.
2446 * %false - we are done with this request
2447 * %true - still buffers pending for this request
2449 bool __blk_end_request_err(struct request *rq, int error)
2451 WARN_ON(error >= 0);
2452 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2454 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2456 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2459 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2460 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2462 if (bio_has_data(bio)) {
2463 rq->nr_phys_segments = bio_phys_segments(q, bio);
2464 rq->buffer = bio_data(bio);
2466 rq->__data_len = bio->bi_size;
2467 rq->bio = rq->biotail = bio;
2470 rq->rq_disk = bio->bi_bdev->bd_disk;
2473 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2475 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2476 * @rq: the request to be flushed
2479 * Flush all pages in @rq.
2481 void rq_flush_dcache_pages(struct request *rq)
2483 struct req_iterator iter;
2484 struct bio_vec *bvec;
2486 rq_for_each_segment(bvec, rq, iter)
2487 flush_dcache_page(bvec->bv_page);
2489 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2493 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2494 * @q : the queue of the device being checked
2497 * Check if underlying low-level drivers of a device are busy.
2498 * If the drivers want to export their busy state, they must set own
2499 * exporting function using blk_queue_lld_busy() first.
2501 * Basically, this function is used only by request stacking drivers
2502 * to stop dispatching requests to underlying devices when underlying
2503 * devices are busy. This behavior helps more I/O merging on the queue
2504 * of the request stacking driver and prevents I/O throughput regression
2505 * on burst I/O load.
2508 * 0 - Not busy (The request stacking driver should dispatch request)
2509 * 1 - Busy (The request stacking driver should stop dispatching request)
2511 int blk_lld_busy(struct request_queue *q)
2514 return q->lld_busy_fn(q);
2518 EXPORT_SYMBOL_GPL(blk_lld_busy);
2521 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2522 * @rq: the clone request to be cleaned up
2525 * Free all bios in @rq for a cloned request.
2527 void blk_rq_unprep_clone(struct request *rq)
2531 while ((bio = rq->bio) != NULL) {
2532 rq->bio = bio->bi_next;
2537 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2540 * Copy attributes of the original request to the clone request.
2541 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2543 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2545 dst->cpu = src->cpu;
2546 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2547 dst->cmd_type = src->cmd_type;
2548 dst->__sector = blk_rq_pos(src);
2549 dst->__data_len = blk_rq_bytes(src);
2550 dst->nr_phys_segments = src->nr_phys_segments;
2551 dst->ioprio = src->ioprio;
2552 dst->extra_len = src->extra_len;
2556 * blk_rq_prep_clone - Helper function to setup clone request
2557 * @rq: the request to be setup
2558 * @rq_src: original request to be cloned
2559 * @bs: bio_set that bios for clone are allocated from
2560 * @gfp_mask: memory allocation mask for bio
2561 * @bio_ctr: setup function to be called for each clone bio.
2562 * Returns %0 for success, non %0 for failure.
2563 * @data: private data to be passed to @bio_ctr
2566 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2567 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2568 * are not copied, and copying such parts is the caller's responsibility.
2569 * Also, pages which the original bios are pointing to are not copied
2570 * and the cloned bios just point same pages.
2571 * So cloned bios must be completed before original bios, which means
2572 * the caller must complete @rq before @rq_src.
2574 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2575 struct bio_set *bs, gfp_t gfp_mask,
2576 int (*bio_ctr)(struct bio *, struct bio *, void *),
2579 struct bio *bio, *bio_src;
2584 blk_rq_init(NULL, rq);
2586 __rq_for_each_bio(bio_src, rq_src) {
2587 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2591 __bio_clone(bio, bio_src);
2593 if (bio_integrity(bio_src) &&
2594 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2597 if (bio_ctr && bio_ctr(bio, bio_src, data))
2601 rq->biotail->bi_next = bio;
2604 rq->bio = rq->biotail = bio;
2607 __blk_rq_prep_clone(rq, rq_src);
2614 blk_rq_unprep_clone(rq);
2618 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2620 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2622 return queue_work(kblockd_workqueue, work);
2624 EXPORT_SYMBOL(kblockd_schedule_work);
2626 int kblockd_schedule_delayed_work(struct request_queue *q,
2627 struct delayed_work *dwork, unsigned long delay)
2629 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2631 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2633 #define PLUG_MAGIC 0x91827364
2635 void blk_start_plug(struct blk_plug *plug)
2637 struct task_struct *tsk = current;
2639 plug->magic = PLUG_MAGIC;
2640 INIT_LIST_HEAD(&plug->list);
2641 plug->should_sort = 0;
2644 * If this is a nested plug, don't actually assign it. It will be
2645 * flushed on its own.
2649 * Store ordering should not be needed here, since a potential
2650 * preempt will imply a full memory barrier
2655 EXPORT_SYMBOL(blk_start_plug);
2657 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2659 struct request *rqa = container_of(a, struct request, queuelist);
2660 struct request *rqb = container_of(b, struct request, queuelist);
2662 return !(rqa->q <= rqb->q);
2665 static void queue_unplugged(struct request_queue *q, unsigned int depth)
2667 trace_block_unplug_io(q, depth);
2668 __blk_run_queue(q, true);
2670 if (q->unplugged_fn)
2674 static void flush_plug_list(struct blk_plug *plug)
2676 struct request_queue *q;
2677 unsigned long flags;
2682 BUG_ON(plug->magic != PLUG_MAGIC);
2684 if (list_empty(&plug->list))
2687 list_splice_init(&plug->list, &list);
2689 if (plug->should_sort) {
2690 list_sort(NULL, &list, plug_rq_cmp);
2691 plug->should_sort = 0;
2698 * Save and disable interrupts here, to avoid doing it for every
2699 * queue lock we have to take.
2701 local_irq_save(flags);
2702 while (!list_empty(&list)) {
2703 rq = list_entry_rq(list.next);
2704 list_del_init(&rq->queuelist);
2705 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2709 queue_unplugged(q, depth);
2710 spin_unlock(q->queue_lock);
2714 spin_lock(q->queue_lock);
2716 rq->cmd_flags &= ~REQ_ON_PLUG;
2719 * rq is already accounted, so use raw insert
2721 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2722 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2724 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2730 queue_unplugged(q, depth);
2731 spin_unlock(q->queue_lock);
2734 local_irq_restore(flags);
2737 static void __blk_finish_plug(struct task_struct *tsk, struct blk_plug *plug)
2739 flush_plug_list(plug);
2741 if (plug == tsk->plug)
2745 void blk_finish_plug(struct blk_plug *plug)
2748 __blk_finish_plug(current, plug);
2750 EXPORT_SYMBOL(blk_finish_plug);
2752 void __blk_flush_plug(struct task_struct *tsk, struct blk_plug *plug)
2754 __blk_finish_plug(tsk, plug);
2757 EXPORT_SYMBOL(__blk_flush_plug);
2759 int __init blk_dev_init(void)
2761 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2762 sizeof(((struct request *)0)->cmd_flags));
2764 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2765 kblockd_workqueue = alloc_workqueue("kblockd",
2766 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2767 if (!kblockd_workqueue)
2768 panic("Failed to create kblockd\n");
2770 request_cachep = kmem_cache_create("blkdev_requests",
2771 sizeof(struct request), 0, SLAB_PANIC, NULL);
2773 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2774 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);