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_bio_complete);
39 static int __make_request(struct request_queue *q, struct bio *bio);
42 * For the allocated request tables
44 static struct kmem_cache *request_cachep;
47 * For queue allocation
49 struct kmem_cache *blk_requestq_cachep;
52 * Controlling structure to kblockd
54 static struct workqueue_struct *kblockd_workqueue;
56 static void drive_stat_acct(struct request *rq, int new_io)
58 struct hd_struct *part;
59 int rw = rq_data_dir(rq);
62 if (!blk_do_io_stat(rq))
65 cpu = part_stat_lock();
66 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
69 part_stat_inc(cpu, part, merges[rw]);
71 part_round_stats(cpu, part);
72 part_inc_in_flight(part, rw);
78 void blk_queue_congestion_threshold(struct request_queue *q)
82 nr = q->nr_requests - (q->nr_requests / 8) + 1;
83 if (nr > q->nr_requests)
85 q->nr_congestion_on = nr;
87 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
90 q->nr_congestion_off = nr;
94 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
97 * Locates the passed device's request queue and returns the address of its
100 * Will return NULL if the request queue cannot be located.
102 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
104 struct backing_dev_info *ret = NULL;
105 struct request_queue *q = bdev_get_queue(bdev);
108 ret = &q->backing_dev_info;
111 EXPORT_SYMBOL(blk_get_backing_dev_info);
113 void blk_rq_init(struct request_queue *q, struct request *rq)
115 memset(rq, 0, sizeof(*rq));
117 INIT_LIST_HEAD(&rq->queuelist);
118 INIT_LIST_HEAD(&rq->timeout_list);
121 rq->__sector = (sector_t) -1;
122 INIT_HLIST_NODE(&rq->hash);
123 RB_CLEAR_NODE(&rq->rb_node);
125 rq->cmd_len = BLK_MAX_CDB;
128 rq->start_time = jiffies;
130 EXPORT_SYMBOL(blk_rq_init);
132 static void req_bio_endio(struct request *rq, struct bio *bio,
133 unsigned int nbytes, int error)
135 struct request_queue *q = rq->q;
137 if (&q->bar_rq != rq) {
139 clear_bit(BIO_UPTODATE, &bio->bi_flags);
140 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
143 if (unlikely(nbytes > bio->bi_size)) {
144 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
145 __func__, nbytes, bio->bi_size);
146 nbytes = bio->bi_size;
149 if (unlikely(rq->cmd_flags & REQ_QUIET))
150 set_bit(BIO_QUIET, &bio->bi_flags);
152 bio->bi_size -= nbytes;
153 bio->bi_sector += (nbytes >> 9);
155 if (bio_integrity(bio))
156 bio_integrity_advance(bio, nbytes);
158 if (bio->bi_size == 0)
159 bio_endio(bio, error);
163 * Okay, this is the barrier request in progress, just
166 if (error && !q->orderr)
171 void blk_dump_rq_flags(struct request *rq, char *msg)
175 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
176 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
179 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
180 (unsigned long long)blk_rq_pos(rq),
181 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
182 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
183 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
185 if (blk_pc_request(rq)) {
186 printk(KERN_INFO " cdb: ");
187 for (bit = 0; bit < BLK_MAX_CDB; bit++)
188 printk("%02x ", rq->cmd[bit]);
192 EXPORT_SYMBOL(blk_dump_rq_flags);
195 * "plug" the device if there are no outstanding requests: this will
196 * force the transfer to start only after we have put all the requests
199 * This is called with interrupts off and no requests on the queue and
200 * with the queue lock held.
202 void blk_plug_device(struct request_queue *q)
204 WARN_ON(!irqs_disabled());
207 * don't plug a stopped queue, it must be paired with blk_start_queue()
208 * which will restart the queueing
210 if (blk_queue_stopped(q))
213 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
214 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
218 EXPORT_SYMBOL(blk_plug_device);
221 * blk_plug_device_unlocked - plug a device without queue lock held
222 * @q: The &struct request_queue to plug
225 * Like @blk_plug_device(), but grabs the queue lock and disables
228 void blk_plug_device_unlocked(struct request_queue *q)
232 spin_lock_irqsave(q->queue_lock, flags);
234 spin_unlock_irqrestore(q->queue_lock, flags);
236 EXPORT_SYMBOL(blk_plug_device_unlocked);
239 * remove the queue from the plugged list, if present. called with
240 * queue lock held and interrupts disabled.
242 int blk_remove_plug(struct request_queue *q)
244 WARN_ON(!irqs_disabled());
246 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
249 del_timer(&q->unplug_timer);
252 EXPORT_SYMBOL(blk_remove_plug);
255 * remove the plug and let it rip..
257 void __generic_unplug_device(struct request_queue *q)
259 if (unlikely(blk_queue_stopped(q)))
261 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
268 * generic_unplug_device - fire a request queue
269 * @q: The &struct request_queue in question
272 * Linux uses plugging to build bigger requests queues before letting
273 * the device have at them. If a queue is plugged, the I/O scheduler
274 * is still adding and merging requests on the queue. Once the queue
275 * gets unplugged, the request_fn defined for the queue is invoked and
278 void generic_unplug_device(struct request_queue *q)
280 if (blk_queue_plugged(q)) {
281 spin_lock_irq(q->queue_lock);
282 __generic_unplug_device(q);
283 spin_unlock_irq(q->queue_lock);
286 EXPORT_SYMBOL(generic_unplug_device);
288 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
291 struct request_queue *q = bdi->unplug_io_data;
296 void blk_unplug_work(struct work_struct *work)
298 struct request_queue *q =
299 container_of(work, struct request_queue, unplug_work);
301 trace_block_unplug_io(q);
305 void blk_unplug_timeout(unsigned long data)
307 struct request_queue *q = (struct request_queue *)data;
309 trace_block_unplug_timer(q);
310 kblockd_schedule_work(q, &q->unplug_work);
313 void blk_unplug(struct request_queue *q)
316 * devices don't necessarily have an ->unplug_fn defined
319 trace_block_unplug_io(q);
323 EXPORT_SYMBOL(blk_unplug);
326 * blk_start_queue - restart a previously stopped queue
327 * @q: The &struct request_queue in question
330 * blk_start_queue() will clear the stop flag on the queue, and call
331 * the request_fn for the queue if it was in a stopped state when
332 * entered. Also see blk_stop_queue(). Queue lock must be held.
334 void blk_start_queue(struct request_queue *q)
336 WARN_ON(!irqs_disabled());
338 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
341 EXPORT_SYMBOL(blk_start_queue);
344 * blk_stop_queue - stop a queue
345 * @q: The &struct request_queue in question
348 * The Linux block layer assumes that a block driver will consume all
349 * entries on the request queue when the request_fn strategy is called.
350 * Often this will not happen, because of hardware limitations (queue
351 * depth settings). If a device driver gets a 'queue full' response,
352 * or if it simply chooses not to queue more I/O at one point, it can
353 * call this function to prevent the request_fn from being called until
354 * the driver has signalled it's ready to go again. This happens by calling
355 * blk_start_queue() to restart queue operations. Queue lock must be held.
357 void blk_stop_queue(struct request_queue *q)
360 queue_flag_set(QUEUE_FLAG_STOPPED, q);
362 EXPORT_SYMBOL(blk_stop_queue);
365 * blk_sync_queue - cancel any pending callbacks on a queue
369 * The block layer may perform asynchronous callback activity
370 * on a queue, such as calling the unplug function after a timeout.
371 * A block device may call blk_sync_queue to ensure that any
372 * such activity is cancelled, thus allowing it to release resources
373 * that the callbacks might use. The caller must already have made sure
374 * that its ->make_request_fn will not re-add plugging prior to calling
378 void blk_sync_queue(struct request_queue *q)
380 del_timer_sync(&q->unplug_timer);
381 del_timer_sync(&q->timeout);
382 cancel_work_sync(&q->unplug_work);
384 EXPORT_SYMBOL(blk_sync_queue);
387 * __blk_run_queue - run a single device queue
388 * @q: The queue to run
391 * See @blk_run_queue. This variant must be called with the queue lock
392 * held and interrupts disabled.
395 void __blk_run_queue(struct request_queue *q)
399 if (unlikely(blk_queue_stopped(q)))
402 if (elv_queue_empty(q))
406 * Only recurse once to avoid overrunning the stack, let the unplug
407 * handling reinvoke the handler shortly if we already got there.
409 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
411 queue_flag_clear(QUEUE_FLAG_REENTER, q);
413 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
414 kblockd_schedule_work(q, &q->unplug_work);
417 EXPORT_SYMBOL(__blk_run_queue);
420 * blk_run_queue - run a single device queue
421 * @q: The queue to run
424 * Invoke request handling on this queue, if it has pending work to do.
425 * May be used to restart queueing when a request has completed.
427 void blk_run_queue(struct request_queue *q)
431 spin_lock_irqsave(q->queue_lock, flags);
433 spin_unlock_irqrestore(q->queue_lock, flags);
435 EXPORT_SYMBOL(blk_run_queue);
437 void blk_put_queue(struct request_queue *q)
439 kobject_put(&q->kobj);
442 void blk_cleanup_queue(struct request_queue *q)
445 * We know we have process context here, so we can be a little
446 * cautious and ensure that pending block actions on this device
447 * are done before moving on. Going into this function, we should
448 * not have processes doing IO to this device.
452 mutex_lock(&q->sysfs_lock);
453 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
454 mutex_unlock(&q->sysfs_lock);
457 elevator_exit(q->elevator);
461 EXPORT_SYMBOL(blk_cleanup_queue);
463 static int blk_init_free_list(struct request_queue *q)
465 struct request_list *rl = &q->rq;
467 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
468 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
470 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
471 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
473 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
474 mempool_free_slab, request_cachep, q->node);
482 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
484 return blk_alloc_queue_node(gfp_mask, -1);
486 EXPORT_SYMBOL(blk_alloc_queue);
488 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
490 struct request_queue *q;
493 q = kmem_cache_alloc_node(blk_requestq_cachep,
494 gfp_mask | __GFP_ZERO, node_id);
498 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
499 q->backing_dev_info.unplug_io_data = q;
500 q->backing_dev_info.ra_pages =
501 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
502 q->backing_dev_info.state = 0;
503 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
505 err = bdi_init(&q->backing_dev_info);
507 kmem_cache_free(blk_requestq_cachep, q);
511 init_timer(&q->unplug_timer);
512 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
513 INIT_LIST_HEAD(&q->timeout_list);
514 INIT_WORK(&q->unplug_work, blk_unplug_work);
516 kobject_init(&q->kobj, &blk_queue_ktype);
518 mutex_init(&q->sysfs_lock);
519 spin_lock_init(&q->__queue_lock);
523 EXPORT_SYMBOL(blk_alloc_queue_node);
526 * blk_init_queue - prepare a request queue for use with a block device
527 * @rfn: The function to be called to process requests that have been
528 * placed on the queue.
529 * @lock: Request queue spin lock
532 * If a block device wishes to use the standard request handling procedures,
533 * which sorts requests and coalesces adjacent requests, then it must
534 * call blk_init_queue(). The function @rfn will be called when there
535 * are requests on the queue that need to be processed. If the device
536 * supports plugging, then @rfn may not be called immediately when requests
537 * are available on the queue, but may be called at some time later instead.
538 * Plugged queues are generally unplugged when a buffer belonging to one
539 * of the requests on the queue is needed, or due to memory pressure.
541 * @rfn is not required, or even expected, to remove all requests off the
542 * queue, but only as many as it can handle at a time. If it does leave
543 * requests on the queue, it is responsible for arranging that the requests
544 * get dealt with eventually.
546 * The queue spin lock must be held while manipulating the requests on the
547 * request queue; this lock will be taken also from interrupt context, so irq
548 * disabling is needed for it.
550 * Function returns a pointer to the initialized request queue, or %NULL if
554 * blk_init_queue() must be paired with a blk_cleanup_queue() call
555 * when the block device is deactivated (such as at module unload).
558 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
560 return blk_init_queue_node(rfn, lock, -1);
562 EXPORT_SYMBOL(blk_init_queue);
564 struct request_queue *
565 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
567 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
573 if (blk_init_free_list(q)) {
574 kmem_cache_free(blk_requestq_cachep, q);
579 q->prep_rq_fn = NULL;
580 q->unplug_fn = generic_unplug_device;
581 q->queue_flags = QUEUE_FLAG_DEFAULT;
582 q->queue_lock = lock;
585 * This also sets hw/phys segments, boundary and size
587 blk_queue_make_request(q, __make_request);
589 q->sg_reserved_size = INT_MAX;
594 if (!elevator_init(q, NULL)) {
595 blk_queue_congestion_threshold(q);
602 EXPORT_SYMBOL(blk_init_queue_node);
604 int blk_get_queue(struct request_queue *q)
606 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
607 kobject_get(&q->kobj);
614 static inline void blk_free_request(struct request_queue *q, struct request *rq)
616 if (rq->cmd_flags & REQ_ELVPRIV)
617 elv_put_request(q, rq);
618 mempool_free(rq, q->rq.rq_pool);
621 static struct request *
622 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
624 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
631 rq->cmd_flags = flags | REQ_ALLOCED;
634 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
635 mempool_free(rq, q->rq.rq_pool);
638 rq->cmd_flags |= REQ_ELVPRIV;
645 * ioc_batching returns true if the ioc is a valid batching request and
646 * should be given priority access to a request.
648 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
654 * Make sure the process is able to allocate at least 1 request
655 * even if the batch times out, otherwise we could theoretically
658 return ioc->nr_batch_requests == q->nr_batching ||
659 (ioc->nr_batch_requests > 0
660 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
664 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
665 * will cause the process to be a "batcher" on all queues in the system. This
666 * is the behaviour we want though - once it gets a wakeup it should be given
669 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
671 if (!ioc || ioc_batching(q, ioc))
674 ioc->nr_batch_requests = q->nr_batching;
675 ioc->last_waited = jiffies;
678 static void __freed_request(struct request_queue *q, int sync)
680 struct request_list *rl = &q->rq;
682 if (rl->count[sync] < queue_congestion_off_threshold(q))
683 blk_clear_queue_congested(q, sync);
685 if (rl->count[sync] + 1 <= q->nr_requests) {
686 if (waitqueue_active(&rl->wait[sync]))
687 wake_up(&rl->wait[sync]);
689 blk_clear_queue_full(q, sync);
694 * A request has just been released. Account for it, update the full and
695 * congestion status, wake up any waiters. Called under q->queue_lock.
697 static void freed_request(struct request_queue *q, int sync, int priv)
699 struct request_list *rl = &q->rq;
705 __freed_request(q, sync);
707 if (unlikely(rl->starved[sync ^ 1]))
708 __freed_request(q, sync ^ 1);
712 * Get a free request, queue_lock must be held.
713 * Returns NULL on failure, with queue_lock held.
714 * Returns !NULL on success, with queue_lock *not held*.
716 static struct request *get_request(struct request_queue *q, int rw_flags,
717 struct bio *bio, gfp_t gfp_mask)
719 struct request *rq = NULL;
720 struct request_list *rl = &q->rq;
721 struct io_context *ioc = NULL;
722 const bool is_sync = rw_is_sync(rw_flags) != 0;
725 may_queue = elv_may_queue(q, rw_flags);
726 if (may_queue == ELV_MQUEUE_NO)
729 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
730 if (rl->count[is_sync]+1 >= q->nr_requests) {
731 ioc = current_io_context(GFP_ATOMIC, q->node);
733 * The queue will fill after this allocation, so set
734 * it as full, and mark this process as "batching".
735 * This process will be allowed to complete a batch of
736 * requests, others will be blocked.
738 if (!blk_queue_full(q, is_sync)) {
739 ioc_set_batching(q, ioc);
740 blk_set_queue_full(q, is_sync);
742 if (may_queue != ELV_MQUEUE_MUST
743 && !ioc_batching(q, ioc)) {
745 * The queue is full and the allocating
746 * process is not a "batcher", and not
747 * exempted by the IO scheduler
753 blk_set_queue_congested(q, is_sync);
757 * Only allow batching queuers to allocate up to 50% over the defined
758 * limit of requests, otherwise we could have thousands of requests
759 * allocated with any setting of ->nr_requests
761 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
764 rl->count[is_sync]++;
765 rl->starved[is_sync] = 0;
767 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
771 if (blk_queue_io_stat(q))
772 rw_flags |= REQ_IO_STAT;
773 spin_unlock_irq(q->queue_lock);
775 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
778 * Allocation failed presumably due to memory. Undo anything
779 * we might have messed up.
781 * Allocating task should really be put onto the front of the
782 * wait queue, but this is pretty rare.
784 spin_lock_irq(q->queue_lock);
785 freed_request(q, is_sync, priv);
788 * in the very unlikely event that allocation failed and no
789 * requests for this direction was pending, mark us starved
790 * so that freeing of a request in the other direction will
791 * notice us. another possible fix would be to split the
792 * rq mempool into READ and WRITE
795 if (unlikely(rl->count[is_sync] == 0))
796 rl->starved[is_sync] = 1;
802 * ioc may be NULL here, and ioc_batching will be false. That's
803 * OK, if the queue is under the request limit then requests need
804 * not count toward the nr_batch_requests limit. There will always
805 * be some limit enforced by BLK_BATCH_TIME.
807 if (ioc_batching(q, ioc))
808 ioc->nr_batch_requests--;
810 trace_block_getrq(q, bio, rw_flags & 1);
816 * No available requests for this queue, unplug the device and wait for some
817 * requests to become available.
819 * Called with q->queue_lock held, and returns with it unlocked.
821 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
824 const bool is_sync = rw_is_sync(rw_flags) != 0;
827 rq = get_request(q, rw_flags, bio, GFP_NOIO);
830 struct io_context *ioc;
831 struct request_list *rl = &q->rq;
833 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
834 TASK_UNINTERRUPTIBLE);
836 trace_block_sleeprq(q, bio, rw_flags & 1);
838 __generic_unplug_device(q);
839 spin_unlock_irq(q->queue_lock);
843 * After sleeping, we become a "batching" process and
844 * will be able to allocate at least one request, and
845 * up to a big batch of them for a small period time.
846 * See ioc_batching, ioc_set_batching
848 ioc = current_io_context(GFP_NOIO, q->node);
849 ioc_set_batching(q, ioc);
851 spin_lock_irq(q->queue_lock);
852 finish_wait(&rl->wait[is_sync], &wait);
854 rq = get_request(q, rw_flags, bio, GFP_NOIO);
860 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
864 BUG_ON(rw != READ && rw != WRITE);
866 spin_lock_irq(q->queue_lock);
867 if (gfp_mask & __GFP_WAIT) {
868 rq = get_request_wait(q, rw, NULL);
870 rq = get_request(q, rw, NULL, gfp_mask);
872 spin_unlock_irq(q->queue_lock);
874 /* q->queue_lock is unlocked at this point */
878 EXPORT_SYMBOL(blk_get_request);
881 * blk_make_request - given a bio, allocate a corresponding struct request.
882 * @q: target request queue
883 * @bio: The bio describing the memory mappings that will be submitted for IO.
884 * It may be a chained-bio properly constructed by block/bio layer.
885 * @gfp_mask: gfp flags to be used for memory allocation
887 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
888 * type commands. Where the struct request needs to be farther initialized by
889 * the caller. It is passed a &struct bio, which describes the memory info of
892 * The caller of blk_make_request must make sure that bi_io_vec
893 * are set to describe the memory buffers. That bio_data_dir() will return
894 * the needed direction of the request. (And all bio's in the passed bio-chain
895 * are properly set accordingly)
897 * If called under none-sleepable conditions, mapped bio buffers must not
898 * need bouncing, by calling the appropriate masked or flagged allocator,
899 * suitable for the target device. Otherwise the call to blk_queue_bounce will
902 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
903 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
904 * anything but the first bio in the chain. Otherwise you risk waiting for IO
905 * completion of a bio that hasn't been submitted yet, thus resulting in a
906 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
907 * of bio_alloc(), as that avoids the mempool deadlock.
908 * If possible a big IO should be split into smaller parts when allocation
909 * fails. Partial allocation should not be an error, or you risk a live-lock.
911 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
914 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
917 return ERR_PTR(-ENOMEM);
920 struct bio *bounce_bio = bio;
923 blk_queue_bounce(q, &bounce_bio);
924 ret = blk_rq_append_bio(q, rq, bounce_bio);
933 EXPORT_SYMBOL(blk_make_request);
936 * blk_requeue_request - put a request back on queue
937 * @q: request queue where request should be inserted
938 * @rq: request to be inserted
941 * Drivers often keep queueing requests until the hardware cannot accept
942 * more, when that condition happens we need to put the request back
943 * on the queue. Must be called with queue lock held.
945 void blk_requeue_request(struct request_queue *q, struct request *rq)
947 blk_delete_timer(rq);
948 blk_clear_rq_complete(rq);
949 trace_block_rq_requeue(q, rq);
951 if (blk_rq_tagged(rq))
952 blk_queue_end_tag(q, rq);
954 BUG_ON(blk_queued_rq(rq));
956 elv_requeue_request(q, rq);
958 EXPORT_SYMBOL(blk_requeue_request);
961 * blk_insert_request - insert a special request into a request queue
962 * @q: request queue where request should be inserted
963 * @rq: request to be inserted
964 * @at_head: insert request at head or tail of queue
965 * @data: private data
968 * Many block devices need to execute commands asynchronously, so they don't
969 * block the whole kernel from preemption during request execution. This is
970 * accomplished normally by inserting aritficial requests tagged as
971 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
972 * be scheduled for actual execution by the request queue.
974 * We have the option of inserting the head or the tail of the queue.
975 * Typically we use the tail for new ioctls and so forth. We use the head
976 * of the queue for things like a QUEUE_FULL message from a device, or a
977 * host that is unable to accept a particular command.
979 void blk_insert_request(struct request_queue *q, struct request *rq,
980 int at_head, void *data)
982 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
986 * tell I/O scheduler that this isn't a regular read/write (ie it
987 * must not attempt merges on this) and that it acts as a soft
990 rq->cmd_type = REQ_TYPE_SPECIAL;
994 spin_lock_irqsave(q->queue_lock, flags);
997 * If command is tagged, release the tag
999 if (blk_rq_tagged(rq))
1000 blk_queue_end_tag(q, rq);
1002 drive_stat_acct(rq, 1);
1003 __elv_add_request(q, rq, where, 0);
1005 spin_unlock_irqrestore(q->queue_lock, flags);
1007 EXPORT_SYMBOL(blk_insert_request);
1010 * add-request adds a request to the linked list.
1011 * queue lock is held and interrupts disabled, as we muck with the
1012 * request queue list.
1014 static inline void add_request(struct request_queue *q, struct request *req)
1016 drive_stat_acct(req, 1);
1019 * elevator indicated where it wants this request to be
1020 * inserted at elevator_merge time
1022 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1025 static void part_round_stats_single(int cpu, struct hd_struct *part,
1028 if (now == part->stamp)
1031 if (part->in_flight) {
1032 __part_stat_add(cpu, part, time_in_queue,
1033 part_in_flight(part) * (now - part->stamp));
1034 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1040 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1041 * @cpu: cpu number for stats access
1042 * @part: target partition
1044 * The average IO queue length and utilisation statistics are maintained
1045 * by observing the current state of the queue length and the amount of
1046 * time it has been in this state for.
1048 * Normally, that accounting is done on IO completion, but that can result
1049 * in more than a second's worth of IO being accounted for within any one
1050 * second, leading to >100% utilisation. To deal with that, we call this
1051 * function to do a round-off before returning the results when reading
1052 * /proc/diskstats. This accounts immediately for all queue usage up to
1053 * the current jiffies and restarts the counters again.
1055 void part_round_stats(int cpu, struct hd_struct *part)
1057 unsigned long now = jiffies;
1060 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1061 part_round_stats_single(cpu, part, now);
1063 EXPORT_SYMBOL_GPL(part_round_stats);
1066 * queue lock must be held
1068 void __blk_put_request(struct request_queue *q, struct request *req)
1072 if (unlikely(--req->ref_count))
1075 elv_completed_request(q, req);
1077 /* this is a bio leak */
1078 WARN_ON(req->bio != NULL);
1081 * Request may not have originated from ll_rw_blk. if not,
1082 * it didn't come out of our reserved rq pools
1084 if (req->cmd_flags & REQ_ALLOCED) {
1085 int is_sync = rq_is_sync(req) != 0;
1086 int priv = req->cmd_flags & REQ_ELVPRIV;
1088 BUG_ON(!list_empty(&req->queuelist));
1089 BUG_ON(!hlist_unhashed(&req->hash));
1091 blk_free_request(q, req);
1092 freed_request(q, is_sync, priv);
1095 EXPORT_SYMBOL_GPL(__blk_put_request);
1097 void blk_put_request(struct request *req)
1099 unsigned long flags;
1100 struct request_queue *q = req->q;
1102 spin_lock_irqsave(q->queue_lock, flags);
1103 __blk_put_request(q, req);
1104 spin_unlock_irqrestore(q->queue_lock, flags);
1106 EXPORT_SYMBOL(blk_put_request);
1108 void init_request_from_bio(struct request *req, struct bio *bio)
1110 req->cpu = bio->bi_comp_cpu;
1111 req->cmd_type = REQ_TYPE_FS;
1114 * Inherit FAILFAST from bio (for read-ahead, and explicit
1115 * FAILFAST). FAILFAST flags are identical for req and bio.
1117 if (bio_rw_flagged(bio, BIO_RW_AHEAD))
1118 req->cmd_flags |= REQ_FAILFAST_MASK;
1120 req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK;
1122 if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) {
1123 req->cmd_flags |= REQ_DISCARD;
1124 if (bio_rw_flagged(bio, BIO_RW_BARRIER))
1125 req->cmd_flags |= REQ_SOFTBARRIER;
1126 req->q->prepare_discard_fn(req->q, req);
1127 } else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
1128 req->cmd_flags |= REQ_HARDBARRIER;
1130 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
1131 req->cmd_flags |= REQ_RW_SYNC;
1132 if (bio_rw_flagged(bio, BIO_RW_META))
1133 req->cmd_flags |= REQ_RW_META;
1134 if (bio_rw_flagged(bio, BIO_RW_NOIDLE))
1135 req->cmd_flags |= REQ_NOIDLE;
1138 req->__sector = bio->bi_sector;
1139 req->ioprio = bio_prio(bio);
1140 blk_rq_bio_prep(req->q, req, bio);
1144 * Only disabling plugging for non-rotational devices if it does tagging
1145 * as well, otherwise we do need the proper merging
1147 static inline bool queue_should_plug(struct request_queue *q)
1149 return !(blk_queue_nonrot(q) && blk_queue_queuing(q));
1152 static int __make_request(struct request_queue *q, struct bio *bio)
1154 struct request *req;
1156 unsigned int bytes = bio->bi_size;
1157 const unsigned short prio = bio_prio(bio);
1158 const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
1159 const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG);
1160 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1163 if (bio_rw_flagged(bio, BIO_RW_BARRIER) && bio_has_data(bio) &&
1164 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1165 bio_endio(bio, -EOPNOTSUPP);
1169 * low level driver can indicate that it wants pages above a
1170 * certain limit bounced to low memory (ie for highmem, or even
1171 * ISA dma in theory)
1173 blk_queue_bounce(q, &bio);
1175 spin_lock_irq(q->queue_lock);
1177 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q))
1180 el_ret = elv_merge(q, &req, bio);
1182 case ELEVATOR_BACK_MERGE:
1183 BUG_ON(!rq_mergeable(req));
1185 if (!ll_back_merge_fn(q, req, bio))
1188 trace_block_bio_backmerge(q, bio);
1190 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1191 blk_rq_set_mixed_merge(req);
1193 req->biotail->bi_next = bio;
1195 req->__data_len += bytes;
1196 req->ioprio = ioprio_best(req->ioprio, prio);
1197 if (!blk_rq_cpu_valid(req))
1198 req->cpu = bio->bi_comp_cpu;
1199 drive_stat_acct(req, 0);
1200 if (!attempt_back_merge(q, req))
1201 elv_merged_request(q, req, el_ret);
1204 case ELEVATOR_FRONT_MERGE:
1205 BUG_ON(!rq_mergeable(req));
1207 if (!ll_front_merge_fn(q, req, bio))
1210 trace_block_bio_frontmerge(q, bio);
1212 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1213 blk_rq_set_mixed_merge(req);
1214 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1215 req->cmd_flags |= ff;
1218 bio->bi_next = req->bio;
1222 * may not be valid. if the low level driver said
1223 * it didn't need a bounce buffer then it better
1224 * not touch req->buffer either...
1226 req->buffer = bio_data(bio);
1227 req->__sector = bio->bi_sector;
1228 req->__data_len += bytes;
1229 req->ioprio = ioprio_best(req->ioprio, prio);
1230 if (!blk_rq_cpu_valid(req))
1231 req->cpu = bio->bi_comp_cpu;
1232 drive_stat_acct(req, 0);
1233 if (!attempt_front_merge(q, req))
1234 elv_merged_request(q, req, el_ret);
1237 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1244 * This sync check and mask will be re-done in init_request_from_bio(),
1245 * but we need to set it earlier to expose the sync flag to the
1246 * rq allocator and io schedulers.
1248 rw_flags = bio_data_dir(bio);
1250 rw_flags |= REQ_RW_SYNC;
1253 * Grab a free request. This is might sleep but can not fail.
1254 * Returns with the queue unlocked.
1256 req = get_request_wait(q, rw_flags, bio);
1259 * After dropping the lock and possibly sleeping here, our request
1260 * may now be mergeable after it had proven unmergeable (above).
1261 * We don't worry about that case for efficiency. It won't happen
1262 * often, and the elevators are able to handle it.
1264 init_request_from_bio(req, bio);
1266 spin_lock_irq(q->queue_lock);
1267 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1268 bio_flagged(bio, BIO_CPU_AFFINE))
1269 req->cpu = blk_cpu_to_group(smp_processor_id());
1270 if (queue_should_plug(q) && elv_queue_empty(q))
1272 add_request(q, req);
1274 if (unplug || !queue_should_plug(q))
1275 __generic_unplug_device(q);
1276 spin_unlock_irq(q->queue_lock);
1281 * If bio->bi_dev is a partition, remap the location
1283 static inline void blk_partition_remap(struct bio *bio)
1285 struct block_device *bdev = bio->bi_bdev;
1287 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1288 struct hd_struct *p = bdev->bd_part;
1290 bio->bi_sector += p->start_sect;
1291 bio->bi_bdev = bdev->bd_contains;
1293 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1295 bio->bi_sector - p->start_sect);
1299 static void handle_bad_sector(struct bio *bio)
1301 char b[BDEVNAME_SIZE];
1303 printk(KERN_INFO "attempt to access beyond end of device\n");
1304 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1305 bdevname(bio->bi_bdev, b),
1307 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1308 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1310 set_bit(BIO_EOF, &bio->bi_flags);
1313 #ifdef CONFIG_FAIL_MAKE_REQUEST
1315 static DECLARE_FAULT_ATTR(fail_make_request);
1317 static int __init setup_fail_make_request(char *str)
1319 return setup_fault_attr(&fail_make_request, str);
1321 __setup("fail_make_request=", setup_fail_make_request);
1323 static int should_fail_request(struct bio *bio)
1325 struct hd_struct *part = bio->bi_bdev->bd_part;
1327 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1328 return should_fail(&fail_make_request, bio->bi_size);
1333 static int __init fail_make_request_debugfs(void)
1335 return init_fault_attr_dentries(&fail_make_request,
1336 "fail_make_request");
1339 late_initcall(fail_make_request_debugfs);
1341 #else /* CONFIG_FAIL_MAKE_REQUEST */
1343 static inline int should_fail_request(struct bio *bio)
1348 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1351 * Check whether this bio extends beyond the end of the device.
1353 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1360 /* Test device or partition size, when known. */
1361 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1363 sector_t sector = bio->bi_sector;
1365 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1367 * This may well happen - the kernel calls bread()
1368 * without checking the size of the device, e.g., when
1369 * mounting a device.
1371 handle_bad_sector(bio);
1380 * generic_make_request - hand a buffer to its device driver for I/O
1381 * @bio: The bio describing the location in memory and on the device.
1383 * generic_make_request() is used to make I/O requests of block
1384 * devices. It is passed a &struct bio, which describes the I/O that needs
1387 * generic_make_request() does not return any status. The
1388 * success/failure status of the request, along with notification of
1389 * completion, is delivered asynchronously through the bio->bi_end_io
1390 * function described (one day) else where.
1392 * The caller of generic_make_request must make sure that bi_io_vec
1393 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1394 * set to describe the device address, and the
1395 * bi_end_io and optionally bi_private are set to describe how
1396 * completion notification should be signaled.
1398 * generic_make_request and the drivers it calls may use bi_next if this
1399 * bio happens to be merged with someone else, and may change bi_dev and
1400 * bi_sector for remaps as it sees fit. So the values of these fields
1401 * should NOT be depended on after the call to generic_make_request.
1403 static inline void __generic_make_request(struct bio *bio)
1405 struct request_queue *q;
1406 sector_t old_sector;
1407 int ret, nr_sectors = bio_sectors(bio);
1413 if (bio_check_eod(bio, nr_sectors))
1417 * Resolve the mapping until finished. (drivers are
1418 * still free to implement/resolve their own stacking
1419 * by explicitly returning 0)
1421 * NOTE: we don't repeat the blk_size check for each new device.
1422 * Stacking drivers are expected to know what they are doing.
1427 char b[BDEVNAME_SIZE];
1429 q = bdev_get_queue(bio->bi_bdev);
1432 "generic_make_request: Trying to access "
1433 "nonexistent block-device %s (%Lu)\n",
1434 bdevname(bio->bi_bdev, b),
1435 (long long) bio->bi_sector);
1439 if (unlikely(nr_sectors > queue_max_hw_sectors(q))) {
1440 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1441 bdevname(bio->bi_bdev, b),
1443 queue_max_hw_sectors(q));
1447 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1450 if (should_fail_request(bio))
1454 * If this device has partitions, remap block n
1455 * of partition p to block n+start(p) of the disk.
1457 blk_partition_remap(bio);
1459 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1462 if (old_sector != -1)
1463 trace_block_remap(q, bio, old_dev, old_sector);
1465 old_sector = bio->bi_sector;
1466 old_dev = bio->bi_bdev->bd_dev;
1468 if (bio_check_eod(bio, nr_sectors))
1471 if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1472 !q->prepare_discard_fn) {
1477 trace_block_bio_queue(q, bio);
1479 ret = q->make_request_fn(q, bio);
1485 bio_endio(bio, err);
1489 * We only want one ->make_request_fn to be active at a time,
1490 * else stack usage with stacked devices could be a problem.
1491 * So use current->bio_{list,tail} to keep a list of requests
1492 * submited by a make_request_fn function.
1493 * current->bio_tail is also used as a flag to say if
1494 * generic_make_request is currently active in this task or not.
1495 * If it is NULL, then no make_request is active. If it is non-NULL,
1496 * then a make_request is active, and new requests should be added
1499 void generic_make_request(struct bio *bio)
1501 if (current->bio_tail) {
1502 /* make_request is active */
1503 *(current->bio_tail) = bio;
1504 bio->bi_next = NULL;
1505 current->bio_tail = &bio->bi_next;
1508 /* following loop may be a bit non-obvious, and so deserves some
1510 * Before entering the loop, bio->bi_next is NULL (as all callers
1511 * ensure that) so we have a list with a single bio.
1512 * We pretend that we have just taken it off a longer list, so
1513 * we assign bio_list to the next (which is NULL) and bio_tail
1514 * to &bio_list, thus initialising the bio_list of new bios to be
1515 * added. __generic_make_request may indeed add some more bios
1516 * through a recursive call to generic_make_request. If it
1517 * did, we find a non-NULL value in bio_list and re-enter the loop
1518 * from the top. In this case we really did just take the bio
1519 * of the top of the list (no pretending) and so fixup bio_list and
1520 * bio_tail or bi_next, and call into __generic_make_request again.
1522 * The loop was structured like this to make only one call to
1523 * __generic_make_request (which is important as it is large and
1524 * inlined) and to keep the structure simple.
1526 BUG_ON(bio->bi_next);
1528 current->bio_list = bio->bi_next;
1529 if (bio->bi_next == NULL)
1530 current->bio_tail = ¤t->bio_list;
1532 bio->bi_next = NULL;
1533 __generic_make_request(bio);
1534 bio = current->bio_list;
1536 current->bio_tail = NULL; /* deactivate */
1538 EXPORT_SYMBOL(generic_make_request);
1541 * submit_bio - submit a bio to the block device layer for I/O
1542 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1543 * @bio: The &struct bio which describes the I/O
1545 * submit_bio() is very similar in purpose to generic_make_request(), and
1546 * uses that function to do most of the work. Both are fairly rough
1547 * interfaces; @bio must be presetup and ready for I/O.
1550 void submit_bio(int rw, struct bio *bio)
1552 int count = bio_sectors(bio);
1557 * If it's a regular read/write or a barrier with data attached,
1558 * go through the normal accounting stuff before submission.
1560 if (bio_has_data(bio)) {
1562 count_vm_events(PGPGOUT, count);
1564 task_io_account_read(bio->bi_size);
1565 count_vm_events(PGPGIN, count);
1568 if (unlikely(block_dump)) {
1569 char b[BDEVNAME_SIZE];
1570 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1571 current->comm, task_pid_nr(current),
1572 (rw & WRITE) ? "WRITE" : "READ",
1573 (unsigned long long)bio->bi_sector,
1574 bdevname(bio->bi_bdev, b));
1578 generic_make_request(bio);
1580 EXPORT_SYMBOL(submit_bio);
1583 * blk_rq_check_limits - Helper function to check a request for the queue limit
1585 * @rq: the request being checked
1588 * @rq may have been made based on weaker limitations of upper-level queues
1589 * in request stacking drivers, and it may violate the limitation of @q.
1590 * Since the block layer and the underlying device driver trust @rq
1591 * after it is inserted to @q, it should be checked against @q before
1592 * the insertion using this generic function.
1594 * This function should also be useful for request stacking drivers
1595 * in some cases below, so export this fuction.
1596 * Request stacking drivers like request-based dm may change the queue
1597 * limits while requests are in the queue (e.g. dm's table swapping).
1598 * Such request stacking drivers should check those requests agaist
1599 * the new queue limits again when they dispatch those requests,
1600 * although such checkings are also done against the old queue limits
1601 * when submitting requests.
1603 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1605 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1606 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1607 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1612 * queue's settings related to segment counting like q->bounce_pfn
1613 * may differ from that of other stacking queues.
1614 * Recalculate it to check the request correctly on this queue's
1617 blk_recalc_rq_segments(rq);
1618 if (rq->nr_phys_segments > queue_max_phys_segments(q) ||
1619 rq->nr_phys_segments > queue_max_hw_segments(q)) {
1620 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1626 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1629 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1630 * @q: the queue to submit the request
1631 * @rq: the request being queued
1633 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1635 unsigned long flags;
1637 if (blk_rq_check_limits(q, rq))
1640 #ifdef CONFIG_FAIL_MAKE_REQUEST
1641 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1642 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1646 spin_lock_irqsave(q->queue_lock, flags);
1649 * Submitting request must be dequeued before calling this function
1650 * because it will be linked to another request_queue
1652 BUG_ON(blk_queued_rq(rq));
1654 drive_stat_acct(rq, 1);
1655 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1657 spin_unlock_irqrestore(q->queue_lock, flags);
1661 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1664 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1665 * @rq: request to examine
1668 * A request could be merge of IOs which require different failure
1669 * handling. This function determines the number of bytes which
1670 * can be failed from the beginning of the request without
1671 * crossing into area which need to be retried further.
1674 * The number of bytes to fail.
1677 * queue_lock must be held.
1679 unsigned int blk_rq_err_bytes(const struct request *rq)
1681 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1682 unsigned int bytes = 0;
1685 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1686 return blk_rq_bytes(rq);
1689 * Currently the only 'mixing' which can happen is between
1690 * different fastfail types. We can safely fail portions
1691 * which have all the failfast bits that the first one has -
1692 * the ones which are at least as eager to fail as the first
1695 for (bio = rq->bio; bio; bio = bio->bi_next) {
1696 if ((bio->bi_rw & ff) != ff)
1698 bytes += bio->bi_size;
1701 /* this could lead to infinite loop */
1702 BUG_ON(blk_rq_bytes(rq) && !bytes);
1705 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1707 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1709 if (blk_do_io_stat(req)) {
1710 const int rw = rq_data_dir(req);
1711 struct hd_struct *part;
1714 cpu = part_stat_lock();
1715 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1716 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1721 static void blk_account_io_done(struct request *req)
1724 * Account IO completion. bar_rq isn't accounted as a normal
1725 * IO on queueing nor completion. Accounting the containing
1726 * request is enough.
1728 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
1729 unsigned long duration = jiffies - req->start_time;
1730 const int rw = rq_data_dir(req);
1731 struct hd_struct *part;
1734 cpu = part_stat_lock();
1735 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1737 part_stat_inc(cpu, part, ios[rw]);
1738 part_stat_add(cpu, part, ticks[rw], duration);
1739 part_round_stats(cpu, part);
1740 part_dec_in_flight(part, rw);
1747 * blk_peek_request - peek at the top of a request queue
1748 * @q: request queue to peek at
1751 * Return the request at the top of @q. The returned request
1752 * should be started using blk_start_request() before LLD starts
1756 * Pointer to the request at the top of @q if available. Null
1760 * queue_lock must be held.
1762 struct request *blk_peek_request(struct request_queue *q)
1767 while ((rq = __elv_next_request(q)) != NULL) {
1768 if (!(rq->cmd_flags & REQ_STARTED)) {
1770 * This is the first time the device driver
1771 * sees this request (possibly after
1772 * requeueing). Notify IO scheduler.
1774 if (blk_sorted_rq(rq))
1775 elv_activate_rq(q, rq);
1778 * just mark as started even if we don't start
1779 * it, a request that has been delayed should
1780 * not be passed by new incoming requests
1782 rq->cmd_flags |= REQ_STARTED;
1783 trace_block_rq_issue(q, rq);
1786 if (!q->boundary_rq || q->boundary_rq == rq) {
1787 q->end_sector = rq_end_sector(rq);
1788 q->boundary_rq = NULL;
1791 if (rq->cmd_flags & REQ_DONTPREP)
1794 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1796 * make sure space for the drain appears we
1797 * know we can do this because max_hw_segments
1798 * has been adjusted to be one fewer than the
1801 rq->nr_phys_segments++;
1807 ret = q->prep_rq_fn(q, rq);
1808 if (ret == BLKPREP_OK) {
1810 } else if (ret == BLKPREP_DEFER) {
1812 * the request may have been (partially) prepped.
1813 * we need to keep this request in the front to
1814 * avoid resource deadlock. REQ_STARTED will
1815 * prevent other fs requests from passing this one.
1817 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1818 !(rq->cmd_flags & REQ_DONTPREP)) {
1820 * remove the space for the drain we added
1821 * so that we don't add it again
1823 --rq->nr_phys_segments;
1828 } else if (ret == BLKPREP_KILL) {
1829 rq->cmd_flags |= REQ_QUIET;
1831 * Mark this request as started so we don't trigger
1832 * any debug logic in the end I/O path.
1834 blk_start_request(rq);
1835 __blk_end_request_all(rq, -EIO);
1837 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1844 EXPORT_SYMBOL(blk_peek_request);
1846 void blk_dequeue_request(struct request *rq)
1848 struct request_queue *q = rq->q;
1850 BUG_ON(list_empty(&rq->queuelist));
1851 BUG_ON(ELV_ON_HASH(rq));
1853 list_del_init(&rq->queuelist);
1856 * the time frame between a request being removed from the lists
1857 * and to it is freed is accounted as io that is in progress at
1860 if (blk_account_rq(rq)) {
1861 q->in_flight[rq_is_sync(rq)]++;
1863 * Mark this device as supporting hardware queuing, if
1864 * we have more IOs in flight than 4.
1866 if (!blk_queue_queuing(q) && queue_in_flight(q) > 4)
1867 set_bit(QUEUE_FLAG_CQ, &q->queue_flags);
1872 * blk_start_request - start request processing on the driver
1873 * @req: request to dequeue
1876 * Dequeue @req and start timeout timer on it. This hands off the
1877 * request to the driver.
1879 * Block internal functions which don't want to start timer should
1880 * call blk_dequeue_request().
1883 * queue_lock must be held.
1885 void blk_start_request(struct request *req)
1887 blk_dequeue_request(req);
1890 * We are now handing the request to the hardware, initialize
1891 * resid_len to full count and add the timeout handler.
1893 req->resid_len = blk_rq_bytes(req);
1894 if (unlikely(blk_bidi_rq(req)))
1895 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1899 EXPORT_SYMBOL(blk_start_request);
1902 * blk_fetch_request - fetch a request from a request queue
1903 * @q: request queue to fetch a request from
1906 * Return the request at the top of @q. The request is started on
1907 * return and LLD can start processing it immediately.
1910 * Pointer to the request at the top of @q if available. Null
1914 * queue_lock must be held.
1916 struct request *blk_fetch_request(struct request_queue *q)
1920 rq = blk_peek_request(q);
1922 blk_start_request(rq);
1925 EXPORT_SYMBOL(blk_fetch_request);
1928 * blk_update_request - Special helper function for request stacking drivers
1929 * @req: the request being processed
1930 * @error: %0 for success, < %0 for error
1931 * @nr_bytes: number of bytes to complete @req
1934 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1935 * the request structure even if @req doesn't have leftover.
1936 * If @req has leftover, sets it up for the next range of segments.
1938 * This special helper function is only for request stacking drivers
1939 * (e.g. request-based dm) so that they can handle partial completion.
1940 * Actual device drivers should use blk_end_request instead.
1942 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1943 * %false return from this function.
1946 * %false - this request doesn't have any more data
1947 * %true - this request has more data
1949 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
1951 int total_bytes, bio_nbytes, next_idx = 0;
1957 trace_block_rq_complete(req->q, req);
1960 * For fs requests, rq is just carrier of independent bio's
1961 * and each partial completion should be handled separately.
1962 * Reset per-request error on each partial completion.
1964 * TODO: tj: This is too subtle. It would be better to let
1965 * low level drivers do what they see fit.
1967 if (blk_fs_request(req))
1970 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1971 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1972 req->rq_disk ? req->rq_disk->disk_name : "?",
1973 (unsigned long long)blk_rq_pos(req));
1976 blk_account_io_completion(req, nr_bytes);
1978 total_bytes = bio_nbytes = 0;
1979 while ((bio = req->bio) != NULL) {
1982 if (nr_bytes >= bio->bi_size) {
1983 req->bio = bio->bi_next;
1984 nbytes = bio->bi_size;
1985 req_bio_endio(req, bio, nbytes, error);
1989 int idx = bio->bi_idx + next_idx;
1991 if (unlikely(idx >= bio->bi_vcnt)) {
1992 blk_dump_rq_flags(req, "__end_that");
1993 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1994 __func__, idx, bio->bi_vcnt);
1998 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1999 BIO_BUG_ON(nbytes > bio->bi_size);
2002 * not a complete bvec done
2004 if (unlikely(nbytes > nr_bytes)) {
2005 bio_nbytes += nr_bytes;
2006 total_bytes += nr_bytes;
2011 * advance to the next vector
2014 bio_nbytes += nbytes;
2017 total_bytes += nbytes;
2023 * end more in this run, or just return 'not-done'
2025 if (unlikely(nr_bytes <= 0))
2035 * Reset counters so that the request stacking driver
2036 * can find how many bytes remain in the request
2039 req->__data_len = 0;
2044 * if the request wasn't completed, update state
2047 req_bio_endio(req, bio, bio_nbytes, error);
2048 bio->bi_idx += next_idx;
2049 bio_iovec(bio)->bv_offset += nr_bytes;
2050 bio_iovec(bio)->bv_len -= nr_bytes;
2053 req->__data_len -= total_bytes;
2054 req->buffer = bio_data(req->bio);
2056 /* update sector only for requests with clear definition of sector */
2057 if (blk_fs_request(req) || blk_discard_rq(req))
2058 req->__sector += total_bytes >> 9;
2060 /* mixed attributes always follow the first bio */
2061 if (req->cmd_flags & REQ_MIXED_MERGE) {
2062 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2063 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2067 * If total number of sectors is less than the first segment
2068 * size, something has gone terribly wrong.
2070 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2071 printk(KERN_ERR "blk: request botched\n");
2072 req->__data_len = blk_rq_cur_bytes(req);
2075 /* recalculate the number of segments */
2076 blk_recalc_rq_segments(req);
2080 EXPORT_SYMBOL_GPL(blk_update_request);
2082 static bool blk_update_bidi_request(struct request *rq, int error,
2083 unsigned int nr_bytes,
2084 unsigned int bidi_bytes)
2086 if (blk_update_request(rq, error, nr_bytes))
2089 /* Bidi request must be completed as a whole */
2090 if (unlikely(blk_bidi_rq(rq)) &&
2091 blk_update_request(rq->next_rq, error, bidi_bytes))
2094 add_disk_randomness(rq->rq_disk);
2100 * queue lock must be held
2102 static void blk_finish_request(struct request *req, int error)
2104 if (blk_rq_tagged(req))
2105 blk_queue_end_tag(req->q, req);
2107 BUG_ON(blk_queued_rq(req));
2109 if (unlikely(laptop_mode) && blk_fs_request(req))
2110 laptop_io_completion();
2112 blk_delete_timer(req);
2114 blk_account_io_done(req);
2117 req->end_io(req, error);
2119 if (blk_bidi_rq(req))
2120 __blk_put_request(req->next_rq->q, req->next_rq);
2122 __blk_put_request(req->q, req);
2127 * blk_end_bidi_request - Complete a bidi request
2128 * @rq: the request to complete
2129 * @error: %0 for success, < %0 for error
2130 * @nr_bytes: number of bytes to complete @rq
2131 * @bidi_bytes: number of bytes to complete @rq->next_rq
2134 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2135 * Drivers that supports bidi can safely call this member for any
2136 * type of request, bidi or uni. In the later case @bidi_bytes is
2140 * %false - we are done with this request
2141 * %true - still buffers pending for this request
2143 static bool blk_end_bidi_request(struct request *rq, int error,
2144 unsigned int nr_bytes, unsigned int bidi_bytes)
2146 struct request_queue *q = rq->q;
2147 unsigned long flags;
2149 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2152 spin_lock_irqsave(q->queue_lock, flags);
2153 blk_finish_request(rq, error);
2154 spin_unlock_irqrestore(q->queue_lock, flags);
2160 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2161 * @rq: the request to complete
2162 * @error: %0 for success, < %0 for error
2163 * @nr_bytes: number of bytes to complete @rq
2164 * @bidi_bytes: number of bytes to complete @rq->next_rq
2167 * Identical to blk_end_bidi_request() except that queue lock is
2168 * assumed to be locked on entry and remains so on return.
2171 * %false - we are done with this request
2172 * %true - still buffers pending for this request
2174 static bool __blk_end_bidi_request(struct request *rq, int error,
2175 unsigned int nr_bytes, unsigned int bidi_bytes)
2177 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2180 blk_finish_request(rq, error);
2186 * blk_end_request - Helper function for drivers to complete the request.
2187 * @rq: the request being processed
2188 * @error: %0 for success, < %0 for error
2189 * @nr_bytes: number of bytes to complete
2192 * Ends I/O on a number of bytes attached to @rq.
2193 * If @rq has leftover, sets it up for the next range of segments.
2196 * %false - we are done with this request
2197 * %true - still buffers pending for this request
2199 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2201 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2203 EXPORT_SYMBOL(blk_end_request);
2206 * blk_end_request_all - Helper function for drives to finish the request.
2207 * @rq: the request to finish
2208 * @error: %0 for success, < %0 for error
2211 * Completely finish @rq.
2213 void blk_end_request_all(struct request *rq, int error)
2216 unsigned int bidi_bytes = 0;
2218 if (unlikely(blk_bidi_rq(rq)))
2219 bidi_bytes = blk_rq_bytes(rq->next_rq);
2221 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2224 EXPORT_SYMBOL(blk_end_request_all);
2227 * blk_end_request_cur - Helper function to finish the current request chunk.
2228 * @rq: the request to finish the current chunk for
2229 * @error: %0 for success, < %0 for error
2232 * Complete the current consecutively mapped chunk from @rq.
2235 * %false - we are done with this request
2236 * %true - still buffers pending for this request
2238 bool blk_end_request_cur(struct request *rq, int error)
2240 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2242 EXPORT_SYMBOL(blk_end_request_cur);
2245 * blk_end_request_err - Finish a request till the next failure boundary.
2246 * @rq: the request to finish till the next failure boundary for
2247 * @error: must be negative errno
2250 * Complete @rq till the next failure boundary.
2253 * %false - we are done with this request
2254 * %true - still buffers pending for this request
2256 bool blk_end_request_err(struct request *rq, int error)
2258 WARN_ON(error >= 0);
2259 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2261 EXPORT_SYMBOL_GPL(blk_end_request_err);
2264 * __blk_end_request - Helper function for drivers to complete the request.
2265 * @rq: the request being processed
2266 * @error: %0 for success, < %0 for error
2267 * @nr_bytes: number of bytes to complete
2270 * Must be called with queue lock held unlike blk_end_request().
2273 * %false - we are done with this request
2274 * %true - still buffers pending for this request
2276 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2278 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2280 EXPORT_SYMBOL(__blk_end_request);
2283 * __blk_end_request_all - Helper function for drives to finish the request.
2284 * @rq: the request to finish
2285 * @error: %0 for success, < %0 for error
2288 * Completely finish @rq. Must be called with queue lock held.
2290 void __blk_end_request_all(struct request *rq, int error)
2293 unsigned int bidi_bytes = 0;
2295 if (unlikely(blk_bidi_rq(rq)))
2296 bidi_bytes = blk_rq_bytes(rq->next_rq);
2298 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2301 EXPORT_SYMBOL(__blk_end_request_all);
2304 * __blk_end_request_cur - Helper function to finish the current request chunk.
2305 * @rq: the request to finish the current chunk for
2306 * @error: %0 for success, < %0 for error
2309 * Complete the current consecutively mapped chunk from @rq. Must
2310 * be called with queue lock held.
2313 * %false - we are done with this request
2314 * %true - still buffers pending for this request
2316 bool __blk_end_request_cur(struct request *rq, int error)
2318 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2320 EXPORT_SYMBOL(__blk_end_request_cur);
2323 * __blk_end_request_err - Finish a request till the next failure boundary.
2324 * @rq: the request to finish till the next failure boundary for
2325 * @error: must be negative errno
2328 * Complete @rq till the next failure boundary. Must be called
2329 * with queue lock held.
2332 * %false - we are done with this request
2333 * %true - still buffers pending for this request
2335 bool __blk_end_request_err(struct request *rq, int error)
2337 WARN_ON(error >= 0);
2338 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2340 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2342 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2345 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2346 rq->cmd_flags |= bio->bi_rw & REQ_RW;
2348 if (bio_has_data(bio)) {
2349 rq->nr_phys_segments = bio_phys_segments(q, bio);
2350 rq->buffer = bio_data(bio);
2352 rq->__data_len = bio->bi_size;
2353 rq->bio = rq->biotail = bio;
2356 rq->rq_disk = bio->bi_bdev->bd_disk;
2360 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2361 * @q : the queue of the device being checked
2364 * Check if underlying low-level drivers of a device are busy.
2365 * If the drivers want to export their busy state, they must set own
2366 * exporting function using blk_queue_lld_busy() first.
2368 * Basically, this function is used only by request stacking drivers
2369 * to stop dispatching requests to underlying devices when underlying
2370 * devices are busy. This behavior helps more I/O merging on the queue
2371 * of the request stacking driver and prevents I/O throughput regression
2372 * on burst I/O load.
2375 * 0 - Not busy (The request stacking driver should dispatch request)
2376 * 1 - Busy (The request stacking driver should stop dispatching request)
2378 int blk_lld_busy(struct request_queue *q)
2381 return q->lld_busy_fn(q);
2385 EXPORT_SYMBOL_GPL(blk_lld_busy);
2388 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2389 * @rq: the clone request to be cleaned up
2392 * Free all bios in @rq for a cloned request.
2394 void blk_rq_unprep_clone(struct request *rq)
2398 while ((bio = rq->bio) != NULL) {
2399 rq->bio = bio->bi_next;
2404 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2407 * Copy attributes of the original request to the clone request.
2408 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2410 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2412 dst->cpu = src->cpu;
2413 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2414 dst->cmd_type = src->cmd_type;
2415 dst->__sector = blk_rq_pos(src);
2416 dst->__data_len = blk_rq_bytes(src);
2417 dst->nr_phys_segments = src->nr_phys_segments;
2418 dst->ioprio = src->ioprio;
2419 dst->extra_len = src->extra_len;
2423 * blk_rq_prep_clone - Helper function to setup clone request
2424 * @rq: the request to be setup
2425 * @rq_src: original request to be cloned
2426 * @bs: bio_set that bios for clone are allocated from
2427 * @gfp_mask: memory allocation mask for bio
2428 * @bio_ctr: setup function to be called for each clone bio.
2429 * Returns %0 for success, non %0 for failure.
2430 * @data: private data to be passed to @bio_ctr
2433 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2434 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2435 * are not copied, and copying such parts is the caller's responsibility.
2436 * Also, pages which the original bios are pointing to are not copied
2437 * and the cloned bios just point same pages.
2438 * So cloned bios must be completed before original bios, which means
2439 * the caller must complete @rq before @rq_src.
2441 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2442 struct bio_set *bs, gfp_t gfp_mask,
2443 int (*bio_ctr)(struct bio *, struct bio *, void *),
2446 struct bio *bio, *bio_src;
2451 blk_rq_init(NULL, rq);
2453 __rq_for_each_bio(bio_src, rq_src) {
2454 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2458 __bio_clone(bio, bio_src);
2460 if (bio_integrity(bio_src) &&
2461 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2464 if (bio_ctr && bio_ctr(bio, bio_src, data))
2468 rq->biotail->bi_next = bio;
2471 rq->bio = rq->biotail = bio;
2474 __blk_rq_prep_clone(rq, rq_src);
2481 blk_rq_unprep_clone(rq);
2485 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2487 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2489 return queue_work(kblockd_workqueue, work);
2491 EXPORT_SYMBOL(kblockd_schedule_work);
2493 int __init blk_dev_init(void)
2495 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2496 sizeof(((struct request *)0)->cmd_flags));
2498 kblockd_workqueue = create_workqueue("kblockd");
2499 if (!kblockd_workqueue)
2500 panic("Failed to create kblockd\n");
2502 request_cachep = kmem_cache_create("blkdev_requests",
2503 sizeof(struct request), 0, SLAB_PANIC, NULL);
2505 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2506 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);