2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/block.h>
36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
40 static int __make_request(struct request_queue *q, struct bio *bio);
43 * For the allocated request tables
45 static struct kmem_cache *request_cachep;
48 * For queue allocation
50 struct kmem_cache *blk_requestq_cachep;
53 * Controlling structure to kblockd
55 static struct workqueue_struct *kblockd_workqueue;
57 static void drive_stat_acct(struct request *rq, int new_io)
59 struct hd_struct *part;
60 int rw = rq_data_dir(rq);
63 if (!blk_do_io_stat(rq))
66 cpu = part_stat_lock();
67 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
70 part_stat_inc(cpu, part, merges[rw]);
72 part_round_stats(cpu, part);
73 part_inc_in_flight(part, rw);
79 void blk_queue_congestion_threshold(struct request_queue *q)
83 nr = q->nr_requests - (q->nr_requests / 8) + 1;
84 if (nr > q->nr_requests)
86 q->nr_congestion_on = nr;
88 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
91 q->nr_congestion_off = nr;
95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
98 * Locates the passed device's request queue and returns the address of its
101 * Will return NULL if the request queue cannot be located.
103 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
105 struct backing_dev_info *ret = NULL;
106 struct request_queue *q = bdev_get_queue(bdev);
109 ret = &q->backing_dev_info;
112 EXPORT_SYMBOL(blk_get_backing_dev_info);
114 void blk_rq_init(struct request_queue *q, struct request *rq)
116 memset(rq, 0, sizeof(*rq));
118 INIT_LIST_HEAD(&rq->queuelist);
119 INIT_LIST_HEAD(&rq->timeout_list);
122 rq->__sector = (sector_t) -1;
123 INIT_HLIST_NODE(&rq->hash);
124 RB_CLEAR_NODE(&rq->rb_node);
126 rq->cmd_len = BLK_MAX_CDB;
129 rq->start_time = jiffies;
130 set_start_time_ns(rq);
132 EXPORT_SYMBOL(blk_rq_init);
134 static void req_bio_endio(struct request *rq, struct bio *bio,
135 unsigned int nbytes, int error)
137 struct request_queue *q = rq->q;
139 if (&q->bar_rq != rq) {
141 clear_bit(BIO_UPTODATE, &bio->bi_flags);
142 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
145 if (unlikely(nbytes > bio->bi_size)) {
146 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
147 __func__, nbytes, bio->bi_size);
148 nbytes = bio->bi_size;
151 if (unlikely(rq->cmd_flags & REQ_QUIET))
152 set_bit(BIO_QUIET, &bio->bi_flags);
154 bio->bi_size -= nbytes;
155 bio->bi_sector += (nbytes >> 9);
157 if (bio_integrity(bio))
158 bio_integrity_advance(bio, nbytes);
160 if (bio->bi_size == 0)
161 bio_endio(bio, error);
165 * Okay, this is the barrier request in progress, just
168 if (error && !q->orderr)
173 void blk_dump_rq_flags(struct request *rq, char *msg)
177 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
178 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
181 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
182 (unsigned long long)blk_rq_pos(rq),
183 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
184 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
185 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
187 if (blk_pc_request(rq)) {
188 printk(KERN_INFO " cdb: ");
189 for (bit = 0; bit < BLK_MAX_CDB; bit++)
190 printk("%02x ", rq->cmd[bit]);
194 EXPORT_SYMBOL(blk_dump_rq_flags);
197 * "plug" the device if there are no outstanding requests: this will
198 * force the transfer to start only after we have put all the requests
201 * This is called with interrupts off and no requests on the queue and
202 * with the queue lock held.
204 void blk_plug_device(struct request_queue *q)
206 WARN_ON(!irqs_disabled());
209 * don't plug a stopped queue, it must be paired with blk_start_queue()
210 * which will restart the queueing
212 if (blk_queue_stopped(q))
215 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
216 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
220 EXPORT_SYMBOL(blk_plug_device);
223 * blk_plug_device_unlocked - plug a device without queue lock held
224 * @q: The &struct request_queue to plug
227 * Like @blk_plug_device(), but grabs the queue lock and disables
230 void blk_plug_device_unlocked(struct request_queue *q)
234 spin_lock_irqsave(q->queue_lock, flags);
236 spin_unlock_irqrestore(q->queue_lock, flags);
238 EXPORT_SYMBOL(blk_plug_device_unlocked);
241 * remove the queue from the plugged list, if present. called with
242 * queue lock held and interrupts disabled.
244 int blk_remove_plug(struct request_queue *q)
246 WARN_ON(!irqs_disabled());
248 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
251 del_timer(&q->unplug_timer);
254 EXPORT_SYMBOL(blk_remove_plug);
257 * remove the plug and let it rip..
259 void __generic_unplug_device(struct request_queue *q)
261 if (unlikely(blk_queue_stopped(q)))
263 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
270 * generic_unplug_device - fire a request queue
271 * @q: The &struct request_queue in question
274 * Linux uses plugging to build bigger requests queues before letting
275 * the device have at them. If a queue is plugged, the I/O scheduler
276 * is still adding and merging requests on the queue. Once the queue
277 * gets unplugged, the request_fn defined for the queue is invoked and
280 void generic_unplug_device(struct request_queue *q)
282 if (blk_queue_plugged(q)) {
283 spin_lock_irq(q->queue_lock);
284 __generic_unplug_device(q);
285 spin_unlock_irq(q->queue_lock);
288 EXPORT_SYMBOL(generic_unplug_device);
290 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
293 struct request_queue *q = bdi->unplug_io_data;
298 void blk_unplug_work(struct work_struct *work)
300 struct request_queue *q =
301 container_of(work, struct request_queue, unplug_work);
303 trace_block_unplug_io(q);
307 void blk_unplug_timeout(unsigned long data)
309 struct request_queue *q = (struct request_queue *)data;
311 trace_block_unplug_timer(q);
312 kblockd_schedule_work(q, &q->unplug_work);
315 void blk_unplug(struct request_queue *q)
318 * devices don't necessarily have an ->unplug_fn defined
321 trace_block_unplug_io(q);
325 EXPORT_SYMBOL(blk_unplug);
328 * blk_start_queue - restart a previously stopped queue
329 * @q: The &struct request_queue in question
332 * blk_start_queue() will clear the stop flag on the queue, and call
333 * the request_fn for the queue if it was in a stopped state when
334 * entered. Also see blk_stop_queue(). Queue lock must be held.
336 void blk_start_queue(struct request_queue *q)
338 WARN_ON(!irqs_disabled());
340 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
343 EXPORT_SYMBOL(blk_start_queue);
346 * blk_stop_queue - stop a queue
347 * @q: The &struct request_queue in question
350 * The Linux block layer assumes that a block driver will consume all
351 * entries on the request queue when the request_fn strategy is called.
352 * Often this will not happen, because of hardware limitations (queue
353 * depth settings). If a device driver gets a 'queue full' response,
354 * or if it simply chooses not to queue more I/O at one point, it can
355 * call this function to prevent the request_fn from being called until
356 * the driver has signalled it's ready to go again. This happens by calling
357 * blk_start_queue() to restart queue operations. Queue lock must be held.
359 void blk_stop_queue(struct request_queue *q)
362 queue_flag_set(QUEUE_FLAG_STOPPED, q);
364 EXPORT_SYMBOL(blk_stop_queue);
367 * blk_sync_queue - cancel any pending callbacks on a queue
371 * The block layer may perform asynchronous callback activity
372 * on a queue, such as calling the unplug function after a timeout.
373 * A block device may call blk_sync_queue to ensure that any
374 * such activity is cancelled, thus allowing it to release resources
375 * that the callbacks might use. The caller must already have made sure
376 * that its ->make_request_fn will not re-add plugging prior to calling
380 void blk_sync_queue(struct request_queue *q)
382 del_timer_sync(&q->unplug_timer);
383 del_timer_sync(&q->timeout);
384 cancel_work_sync(&q->unplug_work);
386 EXPORT_SYMBOL(blk_sync_queue);
389 * __blk_run_queue - run a single device queue
390 * @q: The queue to run
393 * See @blk_run_queue. This variant must be called with the queue lock
394 * held and interrupts disabled.
397 void __blk_run_queue(struct request_queue *q)
401 if (unlikely(blk_queue_stopped(q)))
404 if (elv_queue_empty(q))
408 * Only recurse once to avoid overrunning the stack, let the unplug
409 * handling reinvoke the handler shortly if we already got there.
411 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
413 queue_flag_clear(QUEUE_FLAG_REENTER, q);
415 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
416 kblockd_schedule_work(q, &q->unplug_work);
419 EXPORT_SYMBOL(__blk_run_queue);
422 * blk_run_queue - run a single device queue
423 * @q: The queue to run
426 * Invoke request handling on this queue, if it has pending work to do.
427 * May be used to restart queueing when a request has completed.
429 void blk_run_queue(struct request_queue *q)
433 spin_lock_irqsave(q->queue_lock, flags);
435 spin_unlock_irqrestore(q->queue_lock, flags);
437 EXPORT_SYMBOL(blk_run_queue);
439 void blk_put_queue(struct request_queue *q)
441 kobject_put(&q->kobj);
444 void blk_cleanup_queue(struct request_queue *q)
447 * We know we have process context here, so we can be a little
448 * cautious and ensure that pending block actions on this device
449 * are done before moving on. Going into this function, we should
450 * not have processes doing IO to this device.
454 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
455 mutex_lock(&q->sysfs_lock);
456 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
457 mutex_unlock(&q->sysfs_lock);
460 elevator_exit(q->elevator);
464 EXPORT_SYMBOL(blk_cleanup_queue);
466 static int blk_init_free_list(struct request_queue *q)
468 struct request_list *rl = &q->rq;
470 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
471 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
473 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
474 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
476 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
477 mempool_free_slab, request_cachep, q->node);
485 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
487 return blk_alloc_queue_node(gfp_mask, -1);
489 EXPORT_SYMBOL(blk_alloc_queue);
491 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
493 struct request_queue *q;
496 q = kmem_cache_alloc_node(blk_requestq_cachep,
497 gfp_mask | __GFP_ZERO, node_id);
501 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
502 q->backing_dev_info.unplug_io_data = q;
503 q->backing_dev_info.ra_pages =
504 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
505 q->backing_dev_info.state = 0;
506 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
507 q->backing_dev_info.name = "block";
509 err = bdi_init(&q->backing_dev_info);
511 kmem_cache_free(blk_requestq_cachep, q);
515 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
516 laptop_mode_timer_fn, (unsigned long) q);
517 init_timer(&q->unplug_timer);
518 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
519 INIT_LIST_HEAD(&q->timeout_list);
520 INIT_WORK(&q->unplug_work, blk_unplug_work);
522 kobject_init(&q->kobj, &blk_queue_ktype);
524 mutex_init(&q->sysfs_lock);
525 spin_lock_init(&q->__queue_lock);
529 EXPORT_SYMBOL(blk_alloc_queue_node);
532 * blk_init_queue - prepare a request queue for use with a block device
533 * @rfn: The function to be called to process requests that have been
534 * placed on the queue.
535 * @lock: Request queue spin lock
538 * If a block device wishes to use the standard request handling procedures,
539 * which sorts requests and coalesces adjacent requests, then it must
540 * call blk_init_queue(). The function @rfn will be called when there
541 * are requests on the queue that need to be processed. If the device
542 * supports plugging, then @rfn may not be called immediately when requests
543 * are available on the queue, but may be called at some time later instead.
544 * Plugged queues are generally unplugged when a buffer belonging to one
545 * of the requests on the queue is needed, or due to memory pressure.
547 * @rfn is not required, or even expected, to remove all requests off the
548 * queue, but only as many as it can handle at a time. If it does leave
549 * requests on the queue, it is responsible for arranging that the requests
550 * get dealt with eventually.
552 * The queue spin lock must be held while manipulating the requests on the
553 * request queue; this lock will be taken also from interrupt context, so irq
554 * disabling is needed for it.
556 * Function returns a pointer to the initialized request queue, or %NULL if
560 * blk_init_queue() must be paired with a blk_cleanup_queue() call
561 * when the block device is deactivated (such as at module unload).
564 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
566 return blk_init_queue_node(rfn, lock, -1);
568 EXPORT_SYMBOL(blk_init_queue);
570 struct request_queue *
571 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
573 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
579 if (blk_init_free_list(q)) {
580 kmem_cache_free(blk_requestq_cachep, q);
585 q->prep_rq_fn = NULL;
586 q->unplug_fn = generic_unplug_device;
587 q->queue_flags = QUEUE_FLAG_DEFAULT;
588 q->queue_lock = lock;
591 * This also sets hw/phys segments, boundary and size
593 blk_queue_make_request(q, __make_request);
595 q->sg_reserved_size = INT_MAX;
600 if (!elevator_init(q, NULL)) {
601 blk_queue_congestion_threshold(q);
608 EXPORT_SYMBOL(blk_init_queue_node);
610 int blk_get_queue(struct request_queue *q)
612 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
613 kobject_get(&q->kobj);
620 static inline void blk_free_request(struct request_queue *q, struct request *rq)
622 if (rq->cmd_flags & REQ_ELVPRIV)
623 elv_put_request(q, rq);
624 mempool_free(rq, q->rq.rq_pool);
627 static struct request *
628 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
630 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
637 rq->cmd_flags = flags | REQ_ALLOCED;
640 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
641 mempool_free(rq, q->rq.rq_pool);
644 rq->cmd_flags |= REQ_ELVPRIV;
651 * ioc_batching returns true if the ioc is a valid batching request and
652 * should be given priority access to a request.
654 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
660 * Make sure the process is able to allocate at least 1 request
661 * even if the batch times out, otherwise we could theoretically
664 return ioc->nr_batch_requests == q->nr_batching ||
665 (ioc->nr_batch_requests > 0
666 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
670 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
671 * will cause the process to be a "batcher" on all queues in the system. This
672 * is the behaviour we want though - once it gets a wakeup it should be given
675 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
677 if (!ioc || ioc_batching(q, ioc))
680 ioc->nr_batch_requests = q->nr_batching;
681 ioc->last_waited = jiffies;
684 static void __freed_request(struct request_queue *q, int sync)
686 struct request_list *rl = &q->rq;
688 if (rl->count[sync] < queue_congestion_off_threshold(q))
689 blk_clear_queue_congested(q, sync);
691 if (rl->count[sync] + 1 <= q->nr_requests) {
692 if (waitqueue_active(&rl->wait[sync]))
693 wake_up(&rl->wait[sync]);
695 blk_clear_queue_full(q, sync);
700 * A request has just been released. Account for it, update the full and
701 * congestion status, wake up any waiters. Called under q->queue_lock.
703 static void freed_request(struct request_queue *q, int sync, int priv)
705 struct request_list *rl = &q->rq;
711 __freed_request(q, sync);
713 if (unlikely(rl->starved[sync ^ 1]))
714 __freed_request(q, sync ^ 1);
718 * Get a free request, queue_lock must be held.
719 * Returns NULL on failure, with queue_lock held.
720 * Returns !NULL on success, with queue_lock *not held*.
722 static struct request *get_request(struct request_queue *q, int rw_flags,
723 struct bio *bio, gfp_t gfp_mask)
725 struct request *rq = NULL;
726 struct request_list *rl = &q->rq;
727 struct io_context *ioc = NULL;
728 const bool is_sync = rw_is_sync(rw_flags) != 0;
731 may_queue = elv_may_queue(q, rw_flags);
732 if (may_queue == ELV_MQUEUE_NO)
735 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
736 if (rl->count[is_sync]+1 >= q->nr_requests) {
737 ioc = current_io_context(GFP_ATOMIC, q->node);
739 * The queue will fill after this allocation, so set
740 * it as full, and mark this process as "batching".
741 * This process will be allowed to complete a batch of
742 * requests, others will be blocked.
744 if (!blk_queue_full(q, is_sync)) {
745 ioc_set_batching(q, ioc);
746 blk_set_queue_full(q, is_sync);
748 if (may_queue != ELV_MQUEUE_MUST
749 && !ioc_batching(q, ioc)) {
751 * The queue is full and the allocating
752 * process is not a "batcher", and not
753 * exempted by the IO scheduler
759 blk_set_queue_congested(q, is_sync);
763 * Only allow batching queuers to allocate up to 50% over the defined
764 * limit of requests, otherwise we could have thousands of requests
765 * allocated with any setting of ->nr_requests
767 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
770 rl->count[is_sync]++;
771 rl->starved[is_sync] = 0;
773 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
777 if (blk_queue_io_stat(q))
778 rw_flags |= REQ_IO_STAT;
779 spin_unlock_irq(q->queue_lock);
781 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
784 * Allocation failed presumably due to memory. Undo anything
785 * we might have messed up.
787 * Allocating task should really be put onto the front of the
788 * wait queue, but this is pretty rare.
790 spin_lock_irq(q->queue_lock);
791 freed_request(q, is_sync, priv);
794 * in the very unlikely event that allocation failed and no
795 * requests for this direction was pending, mark us starved
796 * so that freeing of a request in the other direction will
797 * notice us. another possible fix would be to split the
798 * rq mempool into READ and WRITE
801 if (unlikely(rl->count[is_sync] == 0))
802 rl->starved[is_sync] = 1;
808 * ioc may be NULL here, and ioc_batching will be false. That's
809 * OK, if the queue is under the request limit then requests need
810 * not count toward the nr_batch_requests limit. There will always
811 * be some limit enforced by BLK_BATCH_TIME.
813 if (ioc_batching(q, ioc))
814 ioc->nr_batch_requests--;
816 trace_block_getrq(q, bio, rw_flags & 1);
822 * No available requests for this queue, unplug the device and wait for some
823 * requests to become available.
825 * Called with q->queue_lock held, and returns with it unlocked.
827 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
830 const bool is_sync = rw_is_sync(rw_flags) != 0;
833 rq = get_request(q, rw_flags, bio, GFP_NOIO);
836 struct io_context *ioc;
837 struct request_list *rl = &q->rq;
839 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
840 TASK_UNINTERRUPTIBLE);
842 trace_block_sleeprq(q, bio, rw_flags & 1);
844 __generic_unplug_device(q);
845 spin_unlock_irq(q->queue_lock);
849 * After sleeping, we become a "batching" process and
850 * will be able to allocate at least one request, and
851 * up to a big batch of them for a small period time.
852 * See ioc_batching, ioc_set_batching
854 ioc = current_io_context(GFP_NOIO, q->node);
855 ioc_set_batching(q, ioc);
857 spin_lock_irq(q->queue_lock);
858 finish_wait(&rl->wait[is_sync], &wait);
860 rq = get_request(q, rw_flags, bio, GFP_NOIO);
866 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
870 BUG_ON(rw != READ && rw != WRITE);
872 spin_lock_irq(q->queue_lock);
873 if (gfp_mask & __GFP_WAIT) {
874 rq = get_request_wait(q, rw, NULL);
876 rq = get_request(q, rw, NULL, gfp_mask);
878 spin_unlock_irq(q->queue_lock);
880 /* q->queue_lock is unlocked at this point */
884 EXPORT_SYMBOL(blk_get_request);
887 * blk_make_request - given a bio, allocate a corresponding struct request.
888 * @q: target request queue
889 * @bio: The bio describing the memory mappings that will be submitted for IO.
890 * It may be a chained-bio properly constructed by block/bio layer.
891 * @gfp_mask: gfp flags to be used for memory allocation
893 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
894 * type commands. Where the struct request needs to be farther initialized by
895 * the caller. It is passed a &struct bio, which describes the memory info of
898 * The caller of blk_make_request must make sure that bi_io_vec
899 * are set to describe the memory buffers. That bio_data_dir() will return
900 * the needed direction of the request. (And all bio's in the passed bio-chain
901 * are properly set accordingly)
903 * If called under none-sleepable conditions, mapped bio buffers must not
904 * need bouncing, by calling the appropriate masked or flagged allocator,
905 * suitable for the target device. Otherwise the call to blk_queue_bounce will
908 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
909 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
910 * anything but the first bio in the chain. Otherwise you risk waiting for IO
911 * completion of a bio that hasn't been submitted yet, thus resulting in a
912 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
913 * of bio_alloc(), as that avoids the mempool deadlock.
914 * If possible a big IO should be split into smaller parts when allocation
915 * fails. Partial allocation should not be an error, or you risk a live-lock.
917 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
920 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
923 return ERR_PTR(-ENOMEM);
926 struct bio *bounce_bio = bio;
929 blk_queue_bounce(q, &bounce_bio);
930 ret = blk_rq_append_bio(q, rq, bounce_bio);
939 EXPORT_SYMBOL(blk_make_request);
942 * blk_requeue_request - put a request back on queue
943 * @q: request queue where request should be inserted
944 * @rq: request to be inserted
947 * Drivers often keep queueing requests until the hardware cannot accept
948 * more, when that condition happens we need to put the request back
949 * on the queue. Must be called with queue lock held.
951 void blk_requeue_request(struct request_queue *q, struct request *rq)
953 blk_delete_timer(rq);
954 blk_clear_rq_complete(rq);
955 trace_block_rq_requeue(q, rq);
957 if (blk_rq_tagged(rq))
958 blk_queue_end_tag(q, rq);
960 BUG_ON(blk_queued_rq(rq));
962 elv_requeue_request(q, rq);
964 EXPORT_SYMBOL(blk_requeue_request);
967 * blk_insert_request - insert a special request into a request queue
968 * @q: request queue where request should be inserted
969 * @rq: request to be inserted
970 * @at_head: insert request at head or tail of queue
971 * @data: private data
974 * Many block devices need to execute commands asynchronously, so they don't
975 * block the whole kernel from preemption during request execution. This is
976 * accomplished normally by inserting aritficial requests tagged as
977 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
978 * be scheduled for actual execution by the request queue.
980 * We have the option of inserting the head or the tail of the queue.
981 * Typically we use the tail for new ioctls and so forth. We use the head
982 * of the queue for things like a QUEUE_FULL message from a device, or a
983 * host that is unable to accept a particular command.
985 void blk_insert_request(struct request_queue *q, struct request *rq,
986 int at_head, void *data)
988 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
992 * tell I/O scheduler that this isn't a regular read/write (ie it
993 * must not attempt merges on this) and that it acts as a soft
996 rq->cmd_type = REQ_TYPE_SPECIAL;
1000 spin_lock_irqsave(q->queue_lock, flags);
1003 * If command is tagged, release the tag
1005 if (blk_rq_tagged(rq))
1006 blk_queue_end_tag(q, rq);
1008 drive_stat_acct(rq, 1);
1009 __elv_add_request(q, rq, where, 0);
1011 spin_unlock_irqrestore(q->queue_lock, flags);
1013 EXPORT_SYMBOL(blk_insert_request);
1016 * add-request adds a request to the linked list.
1017 * queue lock is held and interrupts disabled, as we muck with the
1018 * request queue list.
1020 static inline void add_request(struct request_queue *q, struct request *req)
1022 drive_stat_acct(req, 1);
1025 * elevator indicated where it wants this request to be
1026 * inserted at elevator_merge time
1028 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1031 static void part_round_stats_single(int cpu, struct hd_struct *part,
1034 if (now == part->stamp)
1037 if (part_in_flight(part)) {
1038 __part_stat_add(cpu, part, time_in_queue,
1039 part_in_flight(part) * (now - part->stamp));
1040 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1046 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1047 * @cpu: cpu number for stats access
1048 * @part: target partition
1050 * The average IO queue length and utilisation statistics are maintained
1051 * by observing the current state of the queue length and the amount of
1052 * time it has been in this state for.
1054 * Normally, that accounting is done on IO completion, but that can result
1055 * in more than a second's worth of IO being accounted for within any one
1056 * second, leading to >100% utilisation. To deal with that, we call this
1057 * function to do a round-off before returning the results when reading
1058 * /proc/diskstats. This accounts immediately for all queue usage up to
1059 * the current jiffies and restarts the counters again.
1061 void part_round_stats(int cpu, struct hd_struct *part)
1063 unsigned long now = jiffies;
1066 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1067 part_round_stats_single(cpu, part, now);
1069 EXPORT_SYMBOL_GPL(part_round_stats);
1072 * queue lock must be held
1074 void __blk_put_request(struct request_queue *q, struct request *req)
1078 if (unlikely(--req->ref_count))
1081 elv_completed_request(q, req);
1083 /* this is a bio leak */
1084 WARN_ON(req->bio != NULL);
1087 * Request may not have originated from ll_rw_blk. if not,
1088 * it didn't come out of our reserved rq pools
1090 if (req->cmd_flags & REQ_ALLOCED) {
1091 int is_sync = rq_is_sync(req) != 0;
1092 int priv = req->cmd_flags & REQ_ELVPRIV;
1094 BUG_ON(!list_empty(&req->queuelist));
1095 BUG_ON(!hlist_unhashed(&req->hash));
1097 blk_free_request(q, req);
1098 freed_request(q, is_sync, priv);
1101 EXPORT_SYMBOL_GPL(__blk_put_request);
1103 void blk_put_request(struct request *req)
1105 unsigned long flags;
1106 struct request_queue *q = req->q;
1108 spin_lock_irqsave(q->queue_lock, flags);
1109 __blk_put_request(q, req);
1110 spin_unlock_irqrestore(q->queue_lock, flags);
1112 EXPORT_SYMBOL(blk_put_request);
1114 void init_request_from_bio(struct request *req, struct bio *bio)
1116 req->cpu = bio->bi_comp_cpu;
1117 req->cmd_type = REQ_TYPE_FS;
1120 * Inherit FAILFAST from bio (for read-ahead, and explicit
1121 * FAILFAST). FAILFAST flags are identical for req and bio.
1123 if (bio_rw_flagged(bio, BIO_RW_AHEAD))
1124 req->cmd_flags |= REQ_FAILFAST_MASK;
1126 req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK;
1128 if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) {
1129 req->cmd_flags |= REQ_DISCARD;
1130 if (bio_rw_flagged(bio, BIO_RW_BARRIER))
1131 req->cmd_flags |= REQ_SOFTBARRIER;
1132 } else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
1133 req->cmd_flags |= REQ_HARDBARRIER;
1135 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
1136 req->cmd_flags |= REQ_RW_SYNC;
1137 if (bio_rw_flagged(bio, BIO_RW_META))
1138 req->cmd_flags |= REQ_RW_META;
1139 if (bio_rw_flagged(bio, BIO_RW_NOIDLE))
1140 req->cmd_flags |= REQ_NOIDLE;
1143 req->__sector = bio->bi_sector;
1144 req->ioprio = bio_prio(bio);
1145 blk_rq_bio_prep(req->q, req, bio);
1149 * Only disabling plugging for non-rotational devices if it does tagging
1150 * as well, otherwise we do need the proper merging
1152 static inline bool queue_should_plug(struct request_queue *q)
1154 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1157 static int __make_request(struct request_queue *q, struct bio *bio)
1159 struct request *req;
1161 unsigned int bytes = bio->bi_size;
1162 const unsigned short prio = bio_prio(bio);
1163 const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
1164 const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG);
1165 const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1168 if (bio_rw_flagged(bio, BIO_RW_BARRIER) &&
1169 (q->next_ordered == QUEUE_ORDERED_NONE)) {
1170 bio_endio(bio, -EOPNOTSUPP);
1174 * low level driver can indicate that it wants pages above a
1175 * certain limit bounced to low memory (ie for highmem, or even
1176 * ISA dma in theory)
1178 blk_queue_bounce(q, &bio);
1180 spin_lock_irq(q->queue_lock);
1182 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q))
1185 el_ret = elv_merge(q, &req, bio);
1187 case ELEVATOR_BACK_MERGE:
1188 BUG_ON(!rq_mergeable(req));
1190 if (!ll_back_merge_fn(q, req, bio))
1193 trace_block_bio_backmerge(q, bio);
1195 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1196 blk_rq_set_mixed_merge(req);
1198 req->biotail->bi_next = bio;
1200 req->__data_len += bytes;
1201 req->ioprio = ioprio_best(req->ioprio, prio);
1202 if (!blk_rq_cpu_valid(req))
1203 req->cpu = bio->bi_comp_cpu;
1204 drive_stat_acct(req, 0);
1205 if (!attempt_back_merge(q, req))
1206 elv_merged_request(q, req, el_ret);
1209 case ELEVATOR_FRONT_MERGE:
1210 BUG_ON(!rq_mergeable(req));
1212 if (!ll_front_merge_fn(q, req, bio))
1215 trace_block_bio_frontmerge(q, bio);
1217 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1218 blk_rq_set_mixed_merge(req);
1219 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1220 req->cmd_flags |= ff;
1223 bio->bi_next = req->bio;
1227 * may not be valid. if the low level driver said
1228 * it didn't need a bounce buffer then it better
1229 * not touch req->buffer either...
1231 req->buffer = bio_data(bio);
1232 req->__sector = bio->bi_sector;
1233 req->__data_len += bytes;
1234 req->ioprio = ioprio_best(req->ioprio, prio);
1235 if (!blk_rq_cpu_valid(req))
1236 req->cpu = bio->bi_comp_cpu;
1237 drive_stat_acct(req, 0);
1238 if (!attempt_front_merge(q, req))
1239 elv_merged_request(q, req, el_ret);
1242 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1249 * This sync check and mask will be re-done in init_request_from_bio(),
1250 * but we need to set it earlier to expose the sync flag to the
1251 * rq allocator and io schedulers.
1253 rw_flags = bio_data_dir(bio);
1255 rw_flags |= REQ_RW_SYNC;
1258 * Grab a free request. This is might sleep but can not fail.
1259 * Returns with the queue unlocked.
1261 req = get_request_wait(q, rw_flags, bio);
1264 * After dropping the lock and possibly sleeping here, our request
1265 * may now be mergeable after it had proven unmergeable (above).
1266 * We don't worry about that case for efficiency. It won't happen
1267 * often, and the elevators are able to handle it.
1269 init_request_from_bio(req, bio);
1271 spin_lock_irq(q->queue_lock);
1272 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1273 bio_flagged(bio, BIO_CPU_AFFINE))
1274 req->cpu = blk_cpu_to_group(smp_processor_id());
1275 if (queue_should_plug(q) && elv_queue_empty(q))
1277 add_request(q, req);
1279 if (unplug || !queue_should_plug(q))
1280 __generic_unplug_device(q);
1281 spin_unlock_irq(q->queue_lock);
1286 * If bio->bi_dev is a partition, remap the location
1288 static inline void blk_partition_remap(struct bio *bio)
1290 struct block_device *bdev = bio->bi_bdev;
1292 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1293 struct hd_struct *p = bdev->bd_part;
1295 bio->bi_sector += p->start_sect;
1296 bio->bi_bdev = bdev->bd_contains;
1298 trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1300 bio->bi_sector - p->start_sect);
1304 static void handle_bad_sector(struct bio *bio)
1306 char b[BDEVNAME_SIZE];
1308 printk(KERN_INFO "attempt to access beyond end of device\n");
1309 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1310 bdevname(bio->bi_bdev, b),
1312 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1313 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1315 set_bit(BIO_EOF, &bio->bi_flags);
1318 #ifdef CONFIG_FAIL_MAKE_REQUEST
1320 static DECLARE_FAULT_ATTR(fail_make_request);
1322 static int __init setup_fail_make_request(char *str)
1324 return setup_fault_attr(&fail_make_request, str);
1326 __setup("fail_make_request=", setup_fail_make_request);
1328 static int should_fail_request(struct bio *bio)
1330 struct hd_struct *part = bio->bi_bdev->bd_part;
1332 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1333 return should_fail(&fail_make_request, bio->bi_size);
1338 static int __init fail_make_request_debugfs(void)
1340 return init_fault_attr_dentries(&fail_make_request,
1341 "fail_make_request");
1344 late_initcall(fail_make_request_debugfs);
1346 #else /* CONFIG_FAIL_MAKE_REQUEST */
1348 static inline int should_fail_request(struct bio *bio)
1353 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1356 * Check whether this bio extends beyond the end of the device.
1358 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1365 /* Test device or partition size, when known. */
1366 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1368 sector_t sector = bio->bi_sector;
1370 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1372 * This may well happen - the kernel calls bread()
1373 * without checking the size of the device, e.g., when
1374 * mounting a device.
1376 handle_bad_sector(bio);
1385 * generic_make_request - hand a buffer to its device driver for I/O
1386 * @bio: The bio describing the location in memory and on the device.
1388 * generic_make_request() is used to make I/O requests of block
1389 * devices. It is passed a &struct bio, which describes the I/O that needs
1392 * generic_make_request() does not return any status. The
1393 * success/failure status of the request, along with notification of
1394 * completion, is delivered asynchronously through the bio->bi_end_io
1395 * function described (one day) else where.
1397 * The caller of generic_make_request must make sure that bi_io_vec
1398 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1399 * set to describe the device address, and the
1400 * bi_end_io and optionally bi_private are set to describe how
1401 * completion notification should be signaled.
1403 * generic_make_request and the drivers it calls may use bi_next if this
1404 * bio happens to be merged with someone else, and may change bi_dev and
1405 * bi_sector for remaps as it sees fit. So the values of these fields
1406 * should NOT be depended on after the call to generic_make_request.
1408 static inline void __generic_make_request(struct bio *bio)
1410 struct request_queue *q;
1411 sector_t old_sector;
1412 int ret, nr_sectors = bio_sectors(bio);
1418 if (bio_check_eod(bio, nr_sectors))
1422 * Resolve the mapping until finished. (drivers are
1423 * still free to implement/resolve their own stacking
1424 * by explicitly returning 0)
1426 * NOTE: we don't repeat the blk_size check for each new device.
1427 * Stacking drivers are expected to know what they are doing.
1432 char b[BDEVNAME_SIZE];
1434 q = bdev_get_queue(bio->bi_bdev);
1437 "generic_make_request: Trying to access "
1438 "nonexistent block-device %s (%Lu)\n",
1439 bdevname(bio->bi_bdev, b),
1440 (long long) bio->bi_sector);
1444 if (unlikely(!bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1445 nr_sectors > queue_max_hw_sectors(q))) {
1446 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1447 bdevname(bio->bi_bdev, b),
1449 queue_max_hw_sectors(q));
1453 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1456 if (should_fail_request(bio))
1460 * If this device has partitions, remap block n
1461 * of partition p to block n+start(p) of the disk.
1463 blk_partition_remap(bio);
1465 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1468 if (old_sector != -1)
1469 trace_block_remap(q, bio, old_dev, old_sector);
1471 old_sector = bio->bi_sector;
1472 old_dev = bio->bi_bdev->bd_dev;
1474 if (bio_check_eod(bio, nr_sectors))
1477 if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1478 !blk_queue_discard(q)) {
1483 trace_block_bio_queue(q, bio);
1485 ret = q->make_request_fn(q, bio);
1491 bio_endio(bio, err);
1495 * We only want one ->make_request_fn to be active at a time,
1496 * else stack usage with stacked devices could be a problem.
1497 * So use current->bio_list to keep a list of requests
1498 * submited by a make_request_fn function.
1499 * current->bio_list is also used as a flag to say if
1500 * generic_make_request is currently active in this task or not.
1501 * If it is NULL, then no make_request is active. If it is non-NULL,
1502 * then a make_request is active, and new requests should be added
1505 void generic_make_request(struct bio *bio)
1507 struct bio_list bio_list_on_stack;
1509 if (current->bio_list) {
1510 /* make_request is active */
1511 bio_list_add(current->bio_list, bio);
1514 /* following loop may be a bit non-obvious, and so deserves some
1516 * Before entering the loop, bio->bi_next is NULL (as all callers
1517 * ensure that) so we have a list with a single bio.
1518 * We pretend that we have just taken it off a longer list, so
1519 * we assign bio_list to a pointer to the bio_list_on_stack,
1520 * thus initialising the bio_list of new bios to be
1521 * added. __generic_make_request may indeed add some more bios
1522 * through a recursive call to generic_make_request. If it
1523 * did, we find a non-NULL value in bio_list and re-enter the loop
1524 * from the top. In this case we really did just take the bio
1525 * of the top of the list (no pretending) and so remove it from
1526 * bio_list, and call into __generic_make_request again.
1528 * The loop was structured like this to make only one call to
1529 * __generic_make_request (which is important as it is large and
1530 * inlined) and to keep the structure simple.
1532 BUG_ON(bio->bi_next);
1533 bio_list_init(&bio_list_on_stack);
1534 current->bio_list = &bio_list_on_stack;
1536 __generic_make_request(bio);
1537 bio = bio_list_pop(current->bio_list);
1539 current->bio_list = NULL; /* deactivate */
1541 EXPORT_SYMBOL(generic_make_request);
1544 * submit_bio - submit a bio to the block device layer for I/O
1545 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1546 * @bio: The &struct bio which describes the I/O
1548 * submit_bio() is very similar in purpose to generic_make_request(), and
1549 * uses that function to do most of the work. Both are fairly rough
1550 * interfaces; @bio must be presetup and ready for I/O.
1553 void submit_bio(int rw, struct bio *bio)
1555 int count = bio_sectors(bio);
1560 * If it's a regular read/write or a barrier with data attached,
1561 * go through the normal accounting stuff before submission.
1563 if (bio_has_data(bio)) {
1565 count_vm_events(PGPGOUT, count);
1567 task_io_account_read(bio->bi_size);
1568 count_vm_events(PGPGIN, count);
1571 if (unlikely(block_dump)) {
1572 char b[BDEVNAME_SIZE];
1573 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1574 current->comm, task_pid_nr(current),
1575 (rw & WRITE) ? "WRITE" : "READ",
1576 (unsigned long long)bio->bi_sector,
1577 bdevname(bio->bi_bdev, b));
1581 generic_make_request(bio);
1583 EXPORT_SYMBOL(submit_bio);
1586 * blk_rq_check_limits - Helper function to check a request for the queue limit
1588 * @rq: the request being checked
1591 * @rq may have been made based on weaker limitations of upper-level queues
1592 * in request stacking drivers, and it may violate the limitation of @q.
1593 * Since the block layer and the underlying device driver trust @rq
1594 * after it is inserted to @q, it should be checked against @q before
1595 * the insertion using this generic function.
1597 * This function should also be useful for request stacking drivers
1598 * in some cases below, so export this fuction.
1599 * Request stacking drivers like request-based dm may change the queue
1600 * limits while requests are in the queue (e.g. dm's table swapping).
1601 * Such request stacking drivers should check those requests agaist
1602 * the new queue limits again when they dispatch those requests,
1603 * although such checkings are also done against the old queue limits
1604 * when submitting requests.
1606 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1608 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1609 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1610 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1615 * queue's settings related to segment counting like q->bounce_pfn
1616 * may differ from that of other stacking queues.
1617 * Recalculate it to check the request correctly on this queue's
1620 blk_recalc_rq_segments(rq);
1621 if (rq->nr_phys_segments > queue_max_segments(q)) {
1622 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1628 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1631 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1632 * @q: the queue to submit the request
1633 * @rq: the request being queued
1635 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1637 unsigned long flags;
1639 if (blk_rq_check_limits(q, rq))
1642 #ifdef CONFIG_FAIL_MAKE_REQUEST
1643 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1644 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1648 spin_lock_irqsave(q->queue_lock, flags);
1651 * Submitting request must be dequeued before calling this function
1652 * because it will be linked to another request_queue
1654 BUG_ON(blk_queued_rq(rq));
1656 drive_stat_acct(rq, 1);
1657 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1659 spin_unlock_irqrestore(q->queue_lock, flags);
1663 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1666 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1667 * @rq: request to examine
1670 * A request could be merge of IOs which require different failure
1671 * handling. This function determines the number of bytes which
1672 * can be failed from the beginning of the request without
1673 * crossing into area which need to be retried further.
1676 * The number of bytes to fail.
1679 * queue_lock must be held.
1681 unsigned int blk_rq_err_bytes(const struct request *rq)
1683 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1684 unsigned int bytes = 0;
1687 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1688 return blk_rq_bytes(rq);
1691 * Currently the only 'mixing' which can happen is between
1692 * different fastfail types. We can safely fail portions
1693 * which have all the failfast bits that the first one has -
1694 * the ones which are at least as eager to fail as the first
1697 for (bio = rq->bio; bio; bio = bio->bi_next) {
1698 if ((bio->bi_rw & ff) != ff)
1700 bytes += bio->bi_size;
1703 /* this could lead to infinite loop */
1704 BUG_ON(blk_rq_bytes(rq) && !bytes);
1707 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1709 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1711 if (blk_do_io_stat(req)) {
1712 const int rw = rq_data_dir(req);
1713 struct hd_struct *part;
1716 cpu = part_stat_lock();
1717 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1718 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1723 static void blk_account_io_done(struct request *req)
1726 * Account IO completion. bar_rq isn't accounted as a normal
1727 * IO on queueing nor completion. Accounting the containing
1728 * request is enough.
1730 if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
1731 unsigned long duration = jiffies - req->start_time;
1732 const int rw = rq_data_dir(req);
1733 struct hd_struct *part;
1736 cpu = part_stat_lock();
1737 part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1739 part_stat_inc(cpu, part, ios[rw]);
1740 part_stat_add(cpu, part, ticks[rw], duration);
1741 part_round_stats(cpu, part);
1742 part_dec_in_flight(part, rw);
1749 * blk_peek_request - peek at the top of a request queue
1750 * @q: request queue to peek at
1753 * Return the request at the top of @q. The returned request
1754 * should be started using blk_start_request() before LLD starts
1758 * Pointer to the request at the top of @q if available. Null
1762 * queue_lock must be held.
1764 struct request *blk_peek_request(struct request_queue *q)
1769 while ((rq = __elv_next_request(q)) != NULL) {
1770 if (!(rq->cmd_flags & REQ_STARTED)) {
1772 * This is the first time the device driver
1773 * sees this request (possibly after
1774 * requeueing). Notify IO scheduler.
1776 if (blk_sorted_rq(rq))
1777 elv_activate_rq(q, rq);
1780 * just mark as started even if we don't start
1781 * it, a request that has been delayed should
1782 * not be passed by new incoming requests
1784 rq->cmd_flags |= REQ_STARTED;
1785 trace_block_rq_issue(q, rq);
1788 if (!q->boundary_rq || q->boundary_rq == rq) {
1789 q->end_sector = rq_end_sector(rq);
1790 q->boundary_rq = NULL;
1793 if (rq->cmd_flags & REQ_DONTPREP)
1796 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1798 * make sure space for the drain appears we
1799 * know we can do this because max_hw_segments
1800 * has been adjusted to be one fewer than the
1803 rq->nr_phys_segments++;
1809 ret = q->prep_rq_fn(q, rq);
1810 if (ret == BLKPREP_OK) {
1812 } else if (ret == BLKPREP_DEFER) {
1814 * the request may have been (partially) prepped.
1815 * we need to keep this request in the front to
1816 * avoid resource deadlock. REQ_STARTED will
1817 * prevent other fs requests from passing this one.
1819 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1820 !(rq->cmd_flags & REQ_DONTPREP)) {
1822 * remove the space for the drain we added
1823 * so that we don't add it again
1825 --rq->nr_phys_segments;
1830 } else if (ret == BLKPREP_KILL) {
1831 rq->cmd_flags |= REQ_QUIET;
1833 * Mark this request as started so we don't trigger
1834 * any debug logic in the end I/O path.
1836 blk_start_request(rq);
1837 __blk_end_request_all(rq, -EIO);
1839 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1846 EXPORT_SYMBOL(blk_peek_request);
1848 void blk_dequeue_request(struct request *rq)
1850 struct request_queue *q = rq->q;
1852 BUG_ON(list_empty(&rq->queuelist));
1853 BUG_ON(ELV_ON_HASH(rq));
1855 list_del_init(&rq->queuelist);
1858 * the time frame between a request being removed from the lists
1859 * and to it is freed is accounted as io that is in progress at
1862 if (blk_account_rq(rq)) {
1863 q->in_flight[rq_is_sync(rq)]++;
1864 set_io_start_time_ns(rq);
1869 * blk_start_request - start request processing on the driver
1870 * @req: request to dequeue
1873 * Dequeue @req and start timeout timer on it. This hands off the
1874 * request to the driver.
1876 * Block internal functions which don't want to start timer should
1877 * call blk_dequeue_request().
1880 * queue_lock must be held.
1882 void blk_start_request(struct request *req)
1884 blk_dequeue_request(req);
1887 * We are now handing the request to the hardware, initialize
1888 * resid_len to full count and add the timeout handler.
1890 req->resid_len = blk_rq_bytes(req);
1891 if (unlikely(blk_bidi_rq(req)))
1892 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1896 EXPORT_SYMBOL(blk_start_request);
1899 * blk_fetch_request - fetch a request from a request queue
1900 * @q: request queue to fetch a request from
1903 * Return the request at the top of @q. The request is started on
1904 * return and LLD can start processing it immediately.
1907 * Pointer to the request at the top of @q if available. Null
1911 * queue_lock must be held.
1913 struct request *blk_fetch_request(struct request_queue *q)
1917 rq = blk_peek_request(q);
1919 blk_start_request(rq);
1922 EXPORT_SYMBOL(blk_fetch_request);
1925 * blk_update_request - Special helper function for request stacking drivers
1926 * @req: the request being processed
1927 * @error: %0 for success, < %0 for error
1928 * @nr_bytes: number of bytes to complete @req
1931 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1932 * the request structure even if @req doesn't have leftover.
1933 * If @req has leftover, sets it up for the next range of segments.
1935 * This special helper function is only for request stacking drivers
1936 * (e.g. request-based dm) so that they can handle partial completion.
1937 * Actual device drivers should use blk_end_request instead.
1939 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1940 * %false return from this function.
1943 * %false - this request doesn't have any more data
1944 * %true - this request has more data
1946 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
1948 int total_bytes, bio_nbytes, next_idx = 0;
1954 trace_block_rq_complete(req->q, req);
1957 * For fs requests, rq is just carrier of independent bio's
1958 * and each partial completion should be handled separately.
1959 * Reset per-request error on each partial completion.
1961 * TODO: tj: This is too subtle. It would be better to let
1962 * low level drivers do what they see fit.
1964 if (blk_fs_request(req))
1967 if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1968 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1969 req->rq_disk ? req->rq_disk->disk_name : "?",
1970 (unsigned long long)blk_rq_pos(req));
1973 blk_account_io_completion(req, nr_bytes);
1975 total_bytes = bio_nbytes = 0;
1976 while ((bio = req->bio) != NULL) {
1979 if (nr_bytes >= bio->bi_size) {
1980 req->bio = bio->bi_next;
1981 nbytes = bio->bi_size;
1982 req_bio_endio(req, bio, nbytes, error);
1986 int idx = bio->bi_idx + next_idx;
1988 if (unlikely(idx >= bio->bi_vcnt)) {
1989 blk_dump_rq_flags(req, "__end_that");
1990 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1991 __func__, idx, bio->bi_vcnt);
1995 nbytes = bio_iovec_idx(bio, idx)->bv_len;
1996 BIO_BUG_ON(nbytes > bio->bi_size);
1999 * not a complete bvec done
2001 if (unlikely(nbytes > nr_bytes)) {
2002 bio_nbytes += nr_bytes;
2003 total_bytes += nr_bytes;
2008 * advance to the next vector
2011 bio_nbytes += nbytes;
2014 total_bytes += nbytes;
2020 * end more in this run, or just return 'not-done'
2022 if (unlikely(nr_bytes <= 0))
2032 * Reset counters so that the request stacking driver
2033 * can find how many bytes remain in the request
2036 req->__data_len = 0;
2041 * if the request wasn't completed, update state
2044 req_bio_endio(req, bio, bio_nbytes, error);
2045 bio->bi_idx += next_idx;
2046 bio_iovec(bio)->bv_offset += nr_bytes;
2047 bio_iovec(bio)->bv_len -= nr_bytes;
2050 req->__data_len -= total_bytes;
2051 req->buffer = bio_data(req->bio);
2053 /* update sector only for requests with clear definition of sector */
2054 if (blk_fs_request(req) || blk_discard_rq(req))
2055 req->__sector += total_bytes >> 9;
2057 /* mixed attributes always follow the first bio */
2058 if (req->cmd_flags & REQ_MIXED_MERGE) {
2059 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2060 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2064 * If total number of sectors is less than the first segment
2065 * size, something has gone terribly wrong.
2067 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2068 printk(KERN_ERR "blk: request botched\n");
2069 req->__data_len = blk_rq_cur_bytes(req);
2072 /* recalculate the number of segments */
2073 blk_recalc_rq_segments(req);
2077 EXPORT_SYMBOL_GPL(blk_update_request);
2079 static bool blk_update_bidi_request(struct request *rq, int error,
2080 unsigned int nr_bytes,
2081 unsigned int bidi_bytes)
2083 if (blk_update_request(rq, error, nr_bytes))
2086 /* Bidi request must be completed as a whole */
2087 if (unlikely(blk_bidi_rq(rq)) &&
2088 blk_update_request(rq->next_rq, error, bidi_bytes))
2091 add_disk_randomness(rq->rq_disk);
2097 * queue lock must be held
2099 static void blk_finish_request(struct request *req, int error)
2101 if (blk_rq_tagged(req))
2102 blk_queue_end_tag(req->q, req);
2104 BUG_ON(blk_queued_rq(req));
2106 if (unlikely(laptop_mode) && blk_fs_request(req))
2107 laptop_io_completion(&req->q->backing_dev_info);
2109 blk_delete_timer(req);
2111 blk_account_io_done(req);
2114 req->end_io(req, error);
2116 if (blk_bidi_rq(req))
2117 __blk_put_request(req->next_rq->q, req->next_rq);
2119 __blk_put_request(req->q, req);
2124 * blk_end_bidi_request - Complete a bidi request
2125 * @rq: the request to complete
2126 * @error: %0 for success, < %0 for error
2127 * @nr_bytes: number of bytes to complete @rq
2128 * @bidi_bytes: number of bytes to complete @rq->next_rq
2131 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2132 * Drivers that supports bidi can safely call this member for any
2133 * type of request, bidi or uni. In the later case @bidi_bytes is
2137 * %false - we are done with this request
2138 * %true - still buffers pending for this request
2140 static bool blk_end_bidi_request(struct request *rq, int error,
2141 unsigned int nr_bytes, unsigned int bidi_bytes)
2143 struct request_queue *q = rq->q;
2144 unsigned long flags;
2146 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2149 spin_lock_irqsave(q->queue_lock, flags);
2150 blk_finish_request(rq, error);
2151 spin_unlock_irqrestore(q->queue_lock, flags);
2157 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2158 * @rq: the request to complete
2159 * @error: %0 for success, < %0 for error
2160 * @nr_bytes: number of bytes to complete @rq
2161 * @bidi_bytes: number of bytes to complete @rq->next_rq
2164 * Identical to blk_end_bidi_request() except that queue lock is
2165 * assumed to be locked on entry and remains so on return.
2168 * %false - we are done with this request
2169 * %true - still buffers pending for this request
2171 static bool __blk_end_bidi_request(struct request *rq, int error,
2172 unsigned int nr_bytes, unsigned int bidi_bytes)
2174 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2177 blk_finish_request(rq, error);
2183 * blk_end_request - Helper function for drivers to complete the request.
2184 * @rq: the request being processed
2185 * @error: %0 for success, < %0 for error
2186 * @nr_bytes: number of bytes to complete
2189 * Ends I/O on a number of bytes attached to @rq.
2190 * If @rq has leftover, sets it up for the next range of segments.
2193 * %false - we are done with this request
2194 * %true - still buffers pending for this request
2196 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2198 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2200 EXPORT_SYMBOL(blk_end_request);
2203 * blk_end_request_all - Helper function for drives to finish the request.
2204 * @rq: the request to finish
2205 * @error: %0 for success, < %0 for error
2208 * Completely finish @rq.
2210 void blk_end_request_all(struct request *rq, int error)
2213 unsigned int bidi_bytes = 0;
2215 if (unlikely(blk_bidi_rq(rq)))
2216 bidi_bytes = blk_rq_bytes(rq->next_rq);
2218 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2221 EXPORT_SYMBOL(blk_end_request_all);
2224 * blk_end_request_cur - Helper function to finish the current request chunk.
2225 * @rq: the request to finish the current chunk for
2226 * @error: %0 for success, < %0 for error
2229 * Complete the current consecutively mapped chunk from @rq.
2232 * %false - we are done with this request
2233 * %true - still buffers pending for this request
2235 bool blk_end_request_cur(struct request *rq, int error)
2237 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2239 EXPORT_SYMBOL(blk_end_request_cur);
2242 * blk_end_request_err - Finish a request till the next failure boundary.
2243 * @rq: the request to finish till the next failure boundary for
2244 * @error: must be negative errno
2247 * Complete @rq till the next failure boundary.
2250 * %false - we are done with this request
2251 * %true - still buffers pending for this request
2253 bool blk_end_request_err(struct request *rq, int error)
2255 WARN_ON(error >= 0);
2256 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2258 EXPORT_SYMBOL_GPL(blk_end_request_err);
2261 * __blk_end_request - Helper function for drivers to complete the request.
2262 * @rq: the request being processed
2263 * @error: %0 for success, < %0 for error
2264 * @nr_bytes: number of bytes to complete
2267 * Must be called with queue lock held unlike blk_end_request().
2270 * %false - we are done with this request
2271 * %true - still buffers pending for this request
2273 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2275 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2277 EXPORT_SYMBOL(__blk_end_request);
2280 * __blk_end_request_all - Helper function for drives to finish the request.
2281 * @rq: the request to finish
2282 * @error: %0 for success, < %0 for error
2285 * Completely finish @rq. Must be called with queue lock held.
2287 void __blk_end_request_all(struct request *rq, int error)
2290 unsigned int bidi_bytes = 0;
2292 if (unlikely(blk_bidi_rq(rq)))
2293 bidi_bytes = blk_rq_bytes(rq->next_rq);
2295 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2298 EXPORT_SYMBOL(__blk_end_request_all);
2301 * __blk_end_request_cur - Helper function to finish the current request chunk.
2302 * @rq: the request to finish the current chunk for
2303 * @error: %0 for success, < %0 for error
2306 * Complete the current consecutively mapped chunk from @rq. Must
2307 * be called with queue lock held.
2310 * %false - we are done with this request
2311 * %true - still buffers pending for this request
2313 bool __blk_end_request_cur(struct request *rq, int error)
2315 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2317 EXPORT_SYMBOL(__blk_end_request_cur);
2320 * __blk_end_request_err - Finish a request till the next failure boundary.
2321 * @rq: the request to finish till the next failure boundary for
2322 * @error: must be negative errno
2325 * Complete @rq till the next failure boundary. Must be called
2326 * with queue lock held.
2329 * %false - we are done with this request
2330 * %true - still buffers pending for this request
2332 bool __blk_end_request_err(struct request *rq, int error)
2334 WARN_ON(error >= 0);
2335 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2337 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2339 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2342 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2343 rq->cmd_flags |= bio->bi_rw & REQ_RW;
2345 if (bio_has_data(bio)) {
2346 rq->nr_phys_segments = bio_phys_segments(q, bio);
2347 rq->buffer = bio_data(bio);
2349 rq->__data_len = bio->bi_size;
2350 rq->bio = rq->biotail = bio;
2353 rq->rq_disk = bio->bi_bdev->bd_disk;
2356 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2358 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2359 * @rq: the request to be flushed
2362 * Flush all pages in @rq.
2364 void rq_flush_dcache_pages(struct request *rq)
2366 struct req_iterator iter;
2367 struct bio_vec *bvec;
2369 rq_for_each_segment(bvec, rq, iter)
2370 flush_dcache_page(bvec->bv_page);
2372 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2376 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2377 * @q : the queue of the device being checked
2380 * Check if underlying low-level drivers of a device are busy.
2381 * If the drivers want to export their busy state, they must set own
2382 * exporting function using blk_queue_lld_busy() first.
2384 * Basically, this function is used only by request stacking drivers
2385 * to stop dispatching requests to underlying devices when underlying
2386 * devices are busy. This behavior helps more I/O merging on the queue
2387 * of the request stacking driver and prevents I/O throughput regression
2388 * on burst I/O load.
2391 * 0 - Not busy (The request stacking driver should dispatch request)
2392 * 1 - Busy (The request stacking driver should stop dispatching request)
2394 int blk_lld_busy(struct request_queue *q)
2397 return q->lld_busy_fn(q);
2401 EXPORT_SYMBOL_GPL(blk_lld_busy);
2404 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2405 * @rq: the clone request to be cleaned up
2408 * Free all bios in @rq for a cloned request.
2410 void blk_rq_unprep_clone(struct request *rq)
2414 while ((bio = rq->bio) != NULL) {
2415 rq->bio = bio->bi_next;
2420 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2423 * Copy attributes of the original request to the clone request.
2424 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2426 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2428 dst->cpu = src->cpu;
2429 dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2430 dst->cmd_type = src->cmd_type;
2431 dst->__sector = blk_rq_pos(src);
2432 dst->__data_len = blk_rq_bytes(src);
2433 dst->nr_phys_segments = src->nr_phys_segments;
2434 dst->ioprio = src->ioprio;
2435 dst->extra_len = src->extra_len;
2439 * blk_rq_prep_clone - Helper function to setup clone request
2440 * @rq: the request to be setup
2441 * @rq_src: original request to be cloned
2442 * @bs: bio_set that bios for clone are allocated from
2443 * @gfp_mask: memory allocation mask for bio
2444 * @bio_ctr: setup function to be called for each clone bio.
2445 * Returns %0 for success, non %0 for failure.
2446 * @data: private data to be passed to @bio_ctr
2449 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2450 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2451 * are not copied, and copying such parts is the caller's responsibility.
2452 * Also, pages which the original bios are pointing to are not copied
2453 * and the cloned bios just point same pages.
2454 * So cloned bios must be completed before original bios, which means
2455 * the caller must complete @rq before @rq_src.
2457 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2458 struct bio_set *bs, gfp_t gfp_mask,
2459 int (*bio_ctr)(struct bio *, struct bio *, void *),
2462 struct bio *bio, *bio_src;
2467 blk_rq_init(NULL, rq);
2469 __rq_for_each_bio(bio_src, rq_src) {
2470 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2474 __bio_clone(bio, bio_src);
2476 if (bio_integrity(bio_src) &&
2477 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2480 if (bio_ctr && bio_ctr(bio, bio_src, data))
2484 rq->biotail->bi_next = bio;
2487 rq->bio = rq->biotail = bio;
2490 __blk_rq_prep_clone(rq, rq_src);
2497 blk_rq_unprep_clone(rq);
2501 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2503 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2505 return queue_work(kblockd_workqueue, work);
2507 EXPORT_SYMBOL(kblockd_schedule_work);
2509 int __init blk_dev_init(void)
2511 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2512 sizeof(((struct request *)0)->cmd_flags));
2514 kblockd_workqueue = create_workqueue("kblockd");
2515 if (!kblockd_workqueue)
2516 panic("Failed to create kblockd\n");
2518 request_cachep = kmem_cache_create("blkdev_requests",
2519 sizeof(struct request), 0, SLAB_PANIC, NULL);
2521 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2522 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);