2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/block.h>
37 #include "blk-cgroup.h"
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
43 DEFINE_IDA(blk_queue_ida);
46 * For the allocated request tables
48 static struct kmem_cache *request_cachep;
51 * For queue allocation
53 struct kmem_cache *blk_requestq_cachep;
56 * Controlling structure to kblockd
58 static struct workqueue_struct *kblockd_workqueue;
60 static void drive_stat_acct(struct request *rq, int new_io)
62 struct hd_struct *part;
63 int rw = rq_data_dir(rq);
66 if (!blk_do_io_stat(rq))
69 cpu = part_stat_lock();
73 part_stat_inc(cpu, part, merges[rw]);
75 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
76 if (!hd_struct_try_get(part)) {
78 * The partition is already being removed,
79 * the request will be accounted on the disk only
81 * We take a reference on disk->part0 although that
82 * partition will never be deleted, so we can treat
83 * it as any other partition.
85 part = &rq->rq_disk->part0;
88 part_round_stats(cpu, part);
89 part_inc_in_flight(part, rw);
96 void blk_queue_congestion_threshold(struct request_queue *q)
100 nr = q->nr_requests - (q->nr_requests / 8) + 1;
101 if (nr > q->nr_requests)
103 q->nr_congestion_on = nr;
105 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
108 q->nr_congestion_off = nr;
112 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
115 * Locates the passed device's request queue and returns the address of its
118 * Will return NULL if the request queue cannot be located.
120 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
122 struct backing_dev_info *ret = NULL;
123 struct request_queue *q = bdev_get_queue(bdev);
126 ret = &q->backing_dev_info;
129 EXPORT_SYMBOL(blk_get_backing_dev_info);
131 void blk_rq_init(struct request_queue *q, struct request *rq)
133 memset(rq, 0, sizeof(*rq));
135 INIT_LIST_HEAD(&rq->queuelist);
136 INIT_LIST_HEAD(&rq->timeout_list);
139 rq->__sector = (sector_t) -1;
140 INIT_HLIST_NODE(&rq->hash);
141 RB_CLEAR_NODE(&rq->rb_node);
143 rq->cmd_len = BLK_MAX_CDB;
146 rq->start_time = jiffies;
147 set_start_time_ns(rq);
150 EXPORT_SYMBOL(blk_rq_init);
152 static void req_bio_endio(struct request *rq, struct bio *bio,
153 unsigned int nbytes, int error)
156 clear_bit(BIO_UPTODATE, &bio->bi_flags);
157 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
160 if (unlikely(nbytes > bio->bi_size)) {
161 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
162 __func__, nbytes, bio->bi_size);
163 nbytes = bio->bi_size;
166 if (unlikely(rq->cmd_flags & REQ_QUIET))
167 set_bit(BIO_QUIET, &bio->bi_flags);
169 bio->bi_size -= nbytes;
170 bio->bi_sector += (nbytes >> 9);
172 if (bio_integrity(bio))
173 bio_integrity_advance(bio, nbytes);
175 /* don't actually finish bio if it's part of flush sequence */
176 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
177 bio_endio(bio, error);
180 void blk_dump_rq_flags(struct request *rq, char *msg)
184 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
185 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
188 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
189 (unsigned long long)blk_rq_pos(rq),
190 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
191 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
192 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
194 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
195 printk(KERN_INFO " cdb: ");
196 for (bit = 0; bit < BLK_MAX_CDB; bit++)
197 printk("%02x ", rq->cmd[bit]);
201 EXPORT_SYMBOL(blk_dump_rq_flags);
203 static void blk_delay_work(struct work_struct *work)
205 struct request_queue *q;
207 q = container_of(work, struct request_queue, delay_work.work);
208 spin_lock_irq(q->queue_lock);
210 spin_unlock_irq(q->queue_lock);
214 * blk_delay_queue - restart queueing after defined interval
215 * @q: The &struct request_queue in question
216 * @msecs: Delay in msecs
219 * Sometimes queueing needs to be postponed for a little while, to allow
220 * resources to come back. This function will make sure that queueing is
221 * restarted around the specified time.
223 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
225 queue_delayed_work(kblockd_workqueue, &q->delay_work,
226 msecs_to_jiffies(msecs));
228 EXPORT_SYMBOL(blk_delay_queue);
231 * blk_start_queue - restart a previously stopped queue
232 * @q: The &struct request_queue in question
235 * blk_start_queue() will clear the stop flag on the queue, and call
236 * the request_fn for the queue if it was in a stopped state when
237 * entered. Also see blk_stop_queue(). Queue lock must be held.
239 void blk_start_queue(struct request_queue *q)
241 WARN_ON(!irqs_disabled());
243 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
246 EXPORT_SYMBOL(blk_start_queue);
249 * blk_stop_queue - stop a queue
250 * @q: The &struct request_queue in question
253 * The Linux block layer assumes that a block driver will consume all
254 * entries on the request queue when the request_fn strategy is called.
255 * Often this will not happen, because of hardware limitations (queue
256 * depth settings). If a device driver gets a 'queue full' response,
257 * or if it simply chooses not to queue more I/O at one point, it can
258 * call this function to prevent the request_fn from being called until
259 * the driver has signalled it's ready to go again. This happens by calling
260 * blk_start_queue() to restart queue operations. Queue lock must be held.
262 void blk_stop_queue(struct request_queue *q)
264 __cancel_delayed_work(&q->delay_work);
265 queue_flag_set(QUEUE_FLAG_STOPPED, q);
267 EXPORT_SYMBOL(blk_stop_queue);
270 * blk_sync_queue - cancel any pending callbacks on a queue
274 * The block layer may perform asynchronous callback activity
275 * on a queue, such as calling the unplug function after a timeout.
276 * A block device may call blk_sync_queue to ensure that any
277 * such activity is cancelled, thus allowing it to release resources
278 * that the callbacks might use. The caller must already have made sure
279 * that its ->make_request_fn will not re-add plugging prior to calling
282 * This function does not cancel any asynchronous activity arising
283 * out of elevator or throttling code. That would require elevaotor_exit()
284 * and blkcg_exit_queue() to be called with queue lock initialized.
287 void blk_sync_queue(struct request_queue *q)
289 del_timer_sync(&q->timeout);
290 cancel_delayed_work_sync(&q->delay_work);
292 EXPORT_SYMBOL(blk_sync_queue);
295 * __blk_run_queue - run a single device queue
296 * @q: The queue to run
299 * See @blk_run_queue. This variant must be called with the queue lock
300 * held and interrupts disabled.
302 void __blk_run_queue(struct request_queue *q)
304 if (unlikely(blk_queue_stopped(q)))
309 EXPORT_SYMBOL(__blk_run_queue);
312 * blk_run_queue_async - run a single device queue in workqueue context
313 * @q: The queue to run
316 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
319 void blk_run_queue_async(struct request_queue *q)
321 if (likely(!blk_queue_stopped(q))) {
322 __cancel_delayed_work(&q->delay_work);
323 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
326 EXPORT_SYMBOL(blk_run_queue_async);
329 * blk_run_queue - run a single device queue
330 * @q: The queue to run
333 * Invoke request handling on this queue, if it has pending work to do.
334 * May be used to restart queueing when a request has completed.
336 void blk_run_queue(struct request_queue *q)
340 spin_lock_irqsave(q->queue_lock, flags);
342 spin_unlock_irqrestore(q->queue_lock, flags);
344 EXPORT_SYMBOL(blk_run_queue);
346 void blk_put_queue(struct request_queue *q)
348 kobject_put(&q->kobj);
350 EXPORT_SYMBOL(blk_put_queue);
353 * blk_drain_queue - drain requests from request_queue
355 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
357 * Drain requests from @q. If @drain_all is set, all requests are drained.
358 * If not, only ELVPRIV requests are drained. The caller is responsible
359 * for ensuring that no new requests which need to be drained are queued.
361 void blk_drain_queue(struct request_queue *q, bool drain_all)
367 spin_lock_irq(q->queue_lock);
370 * The caller might be trying to drain @q before its
371 * elevator is initialized.
374 elv_drain_elevator(q);
376 blkcg_drain_queue(q);
379 * This function might be called on a queue which failed
380 * driver init after queue creation or is not yet fully
381 * active yet. Some drivers (e.g. fd and loop) get unhappy
382 * in such cases. Kick queue iff dispatch queue has
383 * something on it and @q has request_fn set.
385 if (!list_empty(&q->queue_head) && q->request_fn)
388 drain |= q->rq.elvpriv;
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
396 drain |= !list_empty(&q->queue_head);
397 for (i = 0; i < 2; i++) {
398 drain |= q->rq.count[i];
399 drain |= q->in_flight[i];
400 drain |= !list_empty(&q->flush_queue[i]);
404 spin_unlock_irq(q->queue_lock);
413 * blk_queue_bypass_start - enter queue bypass mode
414 * @q: queue of interest
416 * In bypass mode, only the dispatch FIFO queue of @q is used. This
417 * function makes @q enter bypass mode and drains all requests which were
418 * throttled or issued before. On return, it's guaranteed that no request
419 * is being throttled or has ELVPRIV set.
421 void blk_queue_bypass_start(struct request_queue *q)
423 spin_lock_irq(q->queue_lock);
425 queue_flag_set(QUEUE_FLAG_BYPASS, q);
426 spin_unlock_irq(q->queue_lock);
428 blk_drain_queue(q, false);
430 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
433 * blk_queue_bypass_end - leave queue bypass mode
434 * @q: queue of interest
436 * Leave bypass mode and restore the normal queueing behavior.
438 void blk_queue_bypass_end(struct request_queue *q)
440 spin_lock_irq(q->queue_lock);
441 if (!--q->bypass_depth)
442 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
443 WARN_ON_ONCE(q->bypass_depth < 0);
444 spin_unlock_irq(q->queue_lock);
446 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
449 * blk_cleanup_queue - shutdown a request queue
450 * @q: request queue to shutdown
452 * Mark @q DEAD, drain all pending requests, destroy and put it. All
453 * future requests will be failed immediately with -ENODEV.
455 void blk_cleanup_queue(struct request_queue *q)
457 spinlock_t *lock = q->queue_lock;
459 /* mark @q DEAD, no new request or merges will be allowed afterwards */
460 mutex_lock(&q->sysfs_lock);
461 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
465 /* dead queue is permanently in bypass mode till released */
467 queue_flag_set(QUEUE_FLAG_BYPASS, q);
469 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
470 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
471 queue_flag_set(QUEUE_FLAG_DEAD, q);
473 if (q->queue_lock != &q->__queue_lock)
474 q->queue_lock = &q->__queue_lock;
476 spin_unlock_irq(lock);
477 mutex_unlock(&q->sysfs_lock);
479 /* drain all requests queued before DEAD marking */
480 blk_drain_queue(q, true);
482 /* @q won't process any more request, flush async actions */
483 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
486 /* @q is and will stay empty, shutdown and put */
489 EXPORT_SYMBOL(blk_cleanup_queue);
491 static int blk_init_free_list(struct request_queue *q)
493 struct request_list *rl = &q->rq;
495 if (unlikely(rl->rq_pool))
498 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
499 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
501 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
502 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
504 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
505 mempool_free_slab, request_cachep, q->node);
513 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
515 return blk_alloc_queue_node(gfp_mask, -1);
517 EXPORT_SYMBOL(blk_alloc_queue);
519 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
521 struct request_queue *q;
524 q = kmem_cache_alloc_node(blk_requestq_cachep,
525 gfp_mask | __GFP_ZERO, node_id);
529 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
533 q->backing_dev_info.ra_pages =
534 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
535 q->backing_dev_info.state = 0;
536 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
537 q->backing_dev_info.name = "block";
540 err = bdi_init(&q->backing_dev_info);
544 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
545 laptop_mode_timer_fn, (unsigned long) q);
546 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
547 INIT_LIST_HEAD(&q->queue_head);
548 INIT_LIST_HEAD(&q->timeout_list);
549 INIT_LIST_HEAD(&q->icq_list);
550 #ifdef CONFIG_BLK_CGROUP
551 INIT_LIST_HEAD(&q->blkg_list);
553 INIT_LIST_HEAD(&q->flush_queue[0]);
554 INIT_LIST_HEAD(&q->flush_queue[1]);
555 INIT_LIST_HEAD(&q->flush_data_in_flight);
556 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
558 kobject_init(&q->kobj, &blk_queue_ktype);
560 mutex_init(&q->sysfs_lock);
561 spin_lock_init(&q->__queue_lock);
564 * By default initialize queue_lock to internal lock and driver can
565 * override it later if need be.
567 q->queue_lock = &q->__queue_lock;
569 if (blkcg_init_queue(q))
575 ida_simple_remove(&blk_queue_ida, q->id);
577 kmem_cache_free(blk_requestq_cachep, q);
580 EXPORT_SYMBOL(blk_alloc_queue_node);
583 * blk_init_queue - prepare a request queue for use with a block device
584 * @rfn: The function to be called to process requests that have been
585 * placed on the queue.
586 * @lock: Request queue spin lock
589 * If a block device wishes to use the standard request handling procedures,
590 * which sorts requests and coalesces adjacent requests, then it must
591 * call blk_init_queue(). The function @rfn will be called when there
592 * are requests on the queue that need to be processed. If the device
593 * supports plugging, then @rfn may not be called immediately when requests
594 * are available on the queue, but may be called at some time later instead.
595 * Plugged queues are generally unplugged when a buffer belonging to one
596 * of the requests on the queue is needed, or due to memory pressure.
598 * @rfn is not required, or even expected, to remove all requests off the
599 * queue, but only as many as it can handle at a time. If it does leave
600 * requests on the queue, it is responsible for arranging that the requests
601 * get dealt with eventually.
603 * The queue spin lock must be held while manipulating the requests on the
604 * request queue; this lock will be taken also from interrupt context, so irq
605 * disabling is needed for it.
607 * Function returns a pointer to the initialized request queue, or %NULL if
611 * blk_init_queue() must be paired with a blk_cleanup_queue() call
612 * when the block device is deactivated (such as at module unload).
615 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
617 return blk_init_queue_node(rfn, lock, -1);
619 EXPORT_SYMBOL(blk_init_queue);
621 struct request_queue *
622 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
624 struct request_queue *uninit_q, *q;
626 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
630 q = blk_init_allocated_queue(uninit_q, rfn, lock);
632 blk_cleanup_queue(uninit_q);
636 EXPORT_SYMBOL(blk_init_queue_node);
638 struct request_queue *
639 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
645 if (blk_init_free_list(q))
649 q->prep_rq_fn = NULL;
650 q->unprep_rq_fn = NULL;
651 q->queue_flags = QUEUE_FLAG_DEFAULT;
653 /* Override internal queue lock with supplied lock pointer */
655 q->queue_lock = lock;
658 * This also sets hw/phys segments, boundary and size
660 blk_queue_make_request(q, blk_queue_bio);
662 q->sg_reserved_size = INT_MAX;
667 if (!elevator_init(q, NULL)) {
668 blk_queue_congestion_threshold(q);
674 EXPORT_SYMBOL(blk_init_allocated_queue);
676 bool blk_get_queue(struct request_queue *q)
678 if (likely(!blk_queue_dead(q))) {
685 EXPORT_SYMBOL(blk_get_queue);
687 static inline void blk_free_request(struct request_queue *q, struct request *rq)
689 if (rq->cmd_flags & REQ_ELVPRIV) {
690 elv_put_request(q, rq);
692 put_io_context(rq->elv.icq->ioc);
695 mempool_free(rq, q->rq.rq_pool);
698 static struct request *
699 blk_alloc_request(struct request_queue *q, struct io_cq *icq,
700 unsigned int flags, gfp_t gfp_mask)
702 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
709 rq->cmd_flags = flags | REQ_ALLOCED;
711 if (flags & REQ_ELVPRIV) {
713 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
714 mempool_free(rq, q->rq.rq_pool);
717 /* @rq->elv.icq holds on to io_context until @rq is freed */
719 get_io_context(icq->ioc);
726 * ioc_batching returns true if the ioc is a valid batching request and
727 * should be given priority access to a request.
729 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
735 * Make sure the process is able to allocate at least 1 request
736 * even if the batch times out, otherwise we could theoretically
739 return ioc->nr_batch_requests == q->nr_batching ||
740 (ioc->nr_batch_requests > 0
741 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
745 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
746 * will cause the process to be a "batcher" on all queues in the system. This
747 * is the behaviour we want though - once it gets a wakeup it should be given
750 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
752 if (!ioc || ioc_batching(q, ioc))
755 ioc->nr_batch_requests = q->nr_batching;
756 ioc->last_waited = jiffies;
759 static void __freed_request(struct request_queue *q, int sync)
761 struct request_list *rl = &q->rq;
763 if (rl->count[sync] < queue_congestion_off_threshold(q))
764 blk_clear_queue_congested(q, sync);
766 if (rl->count[sync] + 1 <= q->nr_requests) {
767 if (waitqueue_active(&rl->wait[sync]))
768 wake_up(&rl->wait[sync]);
770 blk_clear_queue_full(q, sync);
775 * A request has just been released. Account for it, update the full and
776 * congestion status, wake up any waiters. Called under q->queue_lock.
778 static void freed_request(struct request_queue *q, unsigned int flags)
780 struct request_list *rl = &q->rq;
781 int sync = rw_is_sync(flags);
784 if (flags & REQ_ELVPRIV)
787 __freed_request(q, sync);
789 if (unlikely(rl->starved[sync ^ 1]))
790 __freed_request(q, sync ^ 1);
794 * Determine if elevator data should be initialized when allocating the
795 * request associated with @bio.
797 static bool blk_rq_should_init_elevator(struct bio *bio)
803 * Flush requests do not use the elevator so skip initialization.
804 * This allows a request to share the flush and elevator data.
806 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
813 * get_request - get a free request
814 * @q: request_queue to allocate request from
815 * @rw_flags: RW and SYNC flags
816 * @bio: bio to allocate request for (can be %NULL)
817 * @gfp_mask: allocation mask
819 * Get a free request from @q. This function may fail under memory
820 * pressure or if @q is dead.
822 * Must be callled with @q->queue_lock held and,
823 * Returns %NULL on failure, with @q->queue_lock held.
824 * Returns !%NULL on success, with @q->queue_lock *not held*.
826 static struct request *get_request(struct request_queue *q, int rw_flags,
827 struct bio *bio, gfp_t gfp_mask)
829 struct request *rq = NULL;
830 struct request_list *rl = &q->rq;
831 struct elevator_type *et;
832 struct io_context *ioc;
833 struct io_cq *icq = NULL;
834 const bool is_sync = rw_is_sync(rw_flags) != 0;
835 bool retried = false;
838 et = q->elevator->type;
839 ioc = current->io_context;
841 if (unlikely(blk_queue_dead(q)))
844 may_queue = elv_may_queue(q, rw_flags);
845 if (may_queue == ELV_MQUEUE_NO)
848 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
849 if (rl->count[is_sync]+1 >= q->nr_requests) {
851 * We want ioc to record batching state. If it's
852 * not already there, creating a new one requires
853 * dropping queue_lock, which in turn requires
854 * retesting conditions to avoid queue hang.
856 if (!ioc && !retried) {
857 spin_unlock_irq(q->queue_lock);
858 create_io_context(current, gfp_mask, q->node);
859 spin_lock_irq(q->queue_lock);
865 * The queue will fill after this allocation, so set
866 * it as full, and mark this process as "batching".
867 * This process will be allowed to complete a batch of
868 * requests, others will be blocked.
870 if (!blk_queue_full(q, is_sync)) {
871 ioc_set_batching(q, ioc);
872 blk_set_queue_full(q, is_sync);
874 if (may_queue != ELV_MQUEUE_MUST
875 && !ioc_batching(q, ioc)) {
877 * The queue is full and the allocating
878 * process is not a "batcher", and not
879 * exempted by the IO scheduler
885 blk_set_queue_congested(q, is_sync);
889 * Only allow batching queuers to allocate up to 50% over the defined
890 * limit of requests, otherwise we could have thousands of requests
891 * allocated with any setting of ->nr_requests
893 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
896 rl->count[is_sync]++;
897 rl->starved[is_sync] = 0;
900 * Decide whether the new request will be managed by elevator. If
901 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
902 * prevent the current elevator from being destroyed until the new
903 * request is freed. This guarantees icq's won't be destroyed and
904 * makes creating new ones safe.
906 * Also, lookup icq while holding queue_lock. If it doesn't exist,
907 * it will be created after releasing queue_lock.
909 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
910 rw_flags |= REQ_ELVPRIV;
912 if (et->icq_cache && ioc)
913 icq = ioc_lookup_icq(ioc, q);
916 if (blk_queue_io_stat(q))
917 rw_flags |= REQ_IO_STAT;
918 spin_unlock_irq(q->queue_lock);
920 /* create icq if missing */
921 if ((rw_flags & REQ_ELVPRIV) && unlikely(et->icq_cache && !icq)) {
922 icq = ioc_create_icq(q, gfp_mask);
927 rq = blk_alloc_request(q, icq, rw_flags, gfp_mask);
932 * Allocation failed presumably due to memory. Undo anything
933 * we might have messed up.
935 * Allocating task should really be put onto the front of the
936 * wait queue, but this is pretty rare.
938 spin_lock_irq(q->queue_lock);
939 freed_request(q, rw_flags);
942 * in the very unlikely event that allocation failed and no
943 * requests for this direction was pending, mark us starved
944 * so that freeing of a request in the other direction will
945 * notice us. another possible fix would be to split the
946 * rq mempool into READ and WRITE
949 if (unlikely(rl->count[is_sync] == 0))
950 rl->starved[is_sync] = 1;
956 * ioc may be NULL here, and ioc_batching will be false. That's
957 * OK, if the queue is under the request limit then requests need
958 * not count toward the nr_batch_requests limit. There will always
959 * be some limit enforced by BLK_BATCH_TIME.
961 if (ioc_batching(q, ioc))
962 ioc->nr_batch_requests--;
964 trace_block_getrq(q, bio, rw_flags & 1);
970 * get_request_wait - get a free request with retry
971 * @q: request_queue to allocate request from
972 * @rw_flags: RW and SYNC flags
973 * @bio: bio to allocate request for (can be %NULL)
975 * Get a free request from @q. This function keeps retrying under memory
976 * pressure and fails iff @q is dead.
978 * Must be callled with @q->queue_lock held and,
979 * Returns %NULL on failure, with @q->queue_lock held.
980 * Returns !%NULL on success, with @q->queue_lock *not held*.
982 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
985 const bool is_sync = rw_is_sync(rw_flags) != 0;
988 rq = get_request(q, rw_flags, bio, GFP_NOIO);
991 struct request_list *rl = &q->rq;
993 if (unlikely(blk_queue_dead(q)))
996 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
997 TASK_UNINTERRUPTIBLE);
999 trace_block_sleeprq(q, bio, rw_flags & 1);
1001 spin_unlock_irq(q->queue_lock);
1005 * After sleeping, we become a "batching" process and
1006 * will be able to allocate at least one request, and
1007 * up to a big batch of them for a small period time.
1008 * See ioc_batching, ioc_set_batching
1010 create_io_context(current, GFP_NOIO, q->node);
1011 ioc_set_batching(q, current->io_context);
1013 spin_lock_irq(q->queue_lock);
1014 finish_wait(&rl->wait[is_sync], &wait);
1016 rq = get_request(q, rw_flags, bio, GFP_NOIO);
1022 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1026 BUG_ON(rw != READ && rw != WRITE);
1028 spin_lock_irq(q->queue_lock);
1029 if (gfp_mask & __GFP_WAIT)
1030 rq = get_request_wait(q, rw, NULL);
1032 rq = get_request(q, rw, NULL, gfp_mask);
1034 spin_unlock_irq(q->queue_lock);
1035 /* q->queue_lock is unlocked at this point */
1039 EXPORT_SYMBOL(blk_get_request);
1042 * blk_make_request - given a bio, allocate a corresponding struct request.
1043 * @q: target request queue
1044 * @bio: The bio describing the memory mappings that will be submitted for IO.
1045 * It may be a chained-bio properly constructed by block/bio layer.
1046 * @gfp_mask: gfp flags to be used for memory allocation
1048 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1049 * type commands. Where the struct request needs to be farther initialized by
1050 * the caller. It is passed a &struct bio, which describes the memory info of
1053 * The caller of blk_make_request must make sure that bi_io_vec
1054 * are set to describe the memory buffers. That bio_data_dir() will return
1055 * the needed direction of the request. (And all bio's in the passed bio-chain
1056 * are properly set accordingly)
1058 * If called under none-sleepable conditions, mapped bio buffers must not
1059 * need bouncing, by calling the appropriate masked or flagged allocator,
1060 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1063 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1064 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1065 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1066 * completion of a bio that hasn't been submitted yet, thus resulting in a
1067 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1068 * of bio_alloc(), as that avoids the mempool deadlock.
1069 * If possible a big IO should be split into smaller parts when allocation
1070 * fails. Partial allocation should not be an error, or you risk a live-lock.
1072 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1075 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1078 return ERR_PTR(-ENOMEM);
1081 struct bio *bounce_bio = bio;
1084 blk_queue_bounce(q, &bounce_bio);
1085 ret = blk_rq_append_bio(q, rq, bounce_bio);
1086 if (unlikely(ret)) {
1087 blk_put_request(rq);
1088 return ERR_PTR(ret);
1094 EXPORT_SYMBOL(blk_make_request);
1097 * blk_requeue_request - put a request back on queue
1098 * @q: request queue where request should be inserted
1099 * @rq: request to be inserted
1102 * Drivers often keep queueing requests until the hardware cannot accept
1103 * more, when that condition happens we need to put the request back
1104 * on the queue. Must be called with queue lock held.
1106 void blk_requeue_request(struct request_queue *q, struct request *rq)
1108 blk_delete_timer(rq);
1109 blk_clear_rq_complete(rq);
1110 trace_block_rq_requeue(q, rq);
1112 if (blk_rq_tagged(rq))
1113 blk_queue_end_tag(q, rq);
1115 BUG_ON(blk_queued_rq(rq));
1117 elv_requeue_request(q, rq);
1119 EXPORT_SYMBOL(blk_requeue_request);
1121 static void add_acct_request(struct request_queue *q, struct request *rq,
1124 drive_stat_acct(rq, 1);
1125 __elv_add_request(q, rq, where);
1128 static void part_round_stats_single(int cpu, struct hd_struct *part,
1131 if (now == part->stamp)
1134 if (part_in_flight(part)) {
1135 __part_stat_add(cpu, part, time_in_queue,
1136 part_in_flight(part) * (now - part->stamp));
1137 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1143 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1144 * @cpu: cpu number for stats access
1145 * @part: target partition
1147 * The average IO queue length and utilisation statistics are maintained
1148 * by observing the current state of the queue length and the amount of
1149 * time it has been in this state for.
1151 * Normally, that accounting is done on IO completion, but that can result
1152 * in more than a second's worth of IO being accounted for within any one
1153 * second, leading to >100% utilisation. To deal with that, we call this
1154 * function to do a round-off before returning the results when reading
1155 * /proc/diskstats. This accounts immediately for all queue usage up to
1156 * the current jiffies and restarts the counters again.
1158 void part_round_stats(int cpu, struct hd_struct *part)
1160 unsigned long now = jiffies;
1163 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1164 part_round_stats_single(cpu, part, now);
1166 EXPORT_SYMBOL_GPL(part_round_stats);
1169 * queue lock must be held
1171 void __blk_put_request(struct request_queue *q, struct request *req)
1175 if (unlikely(--req->ref_count))
1178 elv_completed_request(q, req);
1180 /* this is a bio leak */
1181 WARN_ON(req->bio != NULL);
1184 * Request may not have originated from ll_rw_blk. if not,
1185 * it didn't come out of our reserved rq pools
1187 if (req->cmd_flags & REQ_ALLOCED) {
1188 unsigned int flags = req->cmd_flags;
1190 BUG_ON(!list_empty(&req->queuelist));
1191 BUG_ON(!hlist_unhashed(&req->hash));
1193 blk_free_request(q, req);
1194 freed_request(q, flags);
1197 EXPORT_SYMBOL_GPL(__blk_put_request);
1199 void blk_put_request(struct request *req)
1201 unsigned long flags;
1202 struct request_queue *q = req->q;
1204 spin_lock_irqsave(q->queue_lock, flags);
1205 __blk_put_request(q, req);
1206 spin_unlock_irqrestore(q->queue_lock, flags);
1208 EXPORT_SYMBOL(blk_put_request);
1211 * blk_add_request_payload - add a payload to a request
1212 * @rq: request to update
1213 * @page: page backing the payload
1214 * @len: length of the payload.
1216 * This allows to later add a payload to an already submitted request by
1217 * a block driver. The driver needs to take care of freeing the payload
1220 * Note that this is a quite horrible hack and nothing but handling of
1221 * discard requests should ever use it.
1223 void blk_add_request_payload(struct request *rq, struct page *page,
1226 struct bio *bio = rq->bio;
1228 bio->bi_io_vec->bv_page = page;
1229 bio->bi_io_vec->bv_offset = 0;
1230 bio->bi_io_vec->bv_len = len;
1234 bio->bi_phys_segments = 1;
1236 rq->__data_len = rq->resid_len = len;
1237 rq->nr_phys_segments = 1;
1238 rq->buffer = bio_data(bio);
1240 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1242 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1245 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1247 if (!ll_back_merge_fn(q, req, bio))
1250 trace_block_bio_backmerge(q, bio);
1252 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1253 blk_rq_set_mixed_merge(req);
1255 req->biotail->bi_next = bio;
1257 req->__data_len += bio->bi_size;
1258 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1260 drive_stat_acct(req, 0);
1264 static bool bio_attempt_front_merge(struct request_queue *q,
1265 struct request *req, struct bio *bio)
1267 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1269 if (!ll_front_merge_fn(q, req, bio))
1272 trace_block_bio_frontmerge(q, bio);
1274 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1275 blk_rq_set_mixed_merge(req);
1277 bio->bi_next = req->bio;
1281 * may not be valid. if the low level driver said
1282 * it didn't need a bounce buffer then it better
1283 * not touch req->buffer either...
1285 req->buffer = bio_data(bio);
1286 req->__sector = bio->bi_sector;
1287 req->__data_len += bio->bi_size;
1288 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1290 drive_stat_acct(req, 0);
1295 * attempt_plug_merge - try to merge with %current's plugged list
1296 * @q: request_queue new bio is being queued at
1297 * @bio: new bio being queued
1298 * @request_count: out parameter for number of traversed plugged requests
1300 * Determine whether @bio being queued on @q can be merged with a request
1301 * on %current's plugged list. Returns %true if merge was successful,
1304 * Plugging coalesces IOs from the same issuer for the same purpose without
1305 * going through @q->queue_lock. As such it's more of an issuing mechanism
1306 * than scheduling, and the request, while may have elvpriv data, is not
1307 * added on the elevator at this point. In addition, we don't have
1308 * reliable access to the elevator outside queue lock. Only check basic
1309 * merging parameters without querying the elevator.
1311 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1312 unsigned int *request_count)
1314 struct blk_plug *plug;
1318 plug = current->plug;
1323 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1328 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1331 el_ret = blk_try_merge(rq, bio);
1332 if (el_ret == ELEVATOR_BACK_MERGE) {
1333 ret = bio_attempt_back_merge(q, rq, bio);
1336 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1337 ret = bio_attempt_front_merge(q, rq, bio);
1346 void init_request_from_bio(struct request *req, struct bio *bio)
1348 req->cmd_type = REQ_TYPE_FS;
1350 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1351 if (bio->bi_rw & REQ_RAHEAD)
1352 req->cmd_flags |= REQ_FAILFAST_MASK;
1355 req->__sector = bio->bi_sector;
1356 req->ioprio = bio_prio(bio);
1357 blk_rq_bio_prep(req->q, req, bio);
1360 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1362 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1363 struct blk_plug *plug;
1364 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1365 struct request *req;
1366 unsigned int request_count = 0;
1369 * low level driver can indicate that it wants pages above a
1370 * certain limit bounced to low memory (ie for highmem, or even
1371 * ISA dma in theory)
1373 blk_queue_bounce(q, &bio);
1375 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1376 spin_lock_irq(q->queue_lock);
1377 where = ELEVATOR_INSERT_FLUSH;
1382 * Check if we can merge with the plugged list before grabbing
1385 if (attempt_plug_merge(q, bio, &request_count))
1388 spin_lock_irq(q->queue_lock);
1390 el_ret = elv_merge(q, &req, bio);
1391 if (el_ret == ELEVATOR_BACK_MERGE) {
1392 if (bio_attempt_back_merge(q, req, bio)) {
1393 elv_bio_merged(q, req, bio);
1394 if (!attempt_back_merge(q, req))
1395 elv_merged_request(q, req, el_ret);
1398 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1399 if (bio_attempt_front_merge(q, req, bio)) {
1400 elv_bio_merged(q, req, bio);
1401 if (!attempt_front_merge(q, req))
1402 elv_merged_request(q, req, el_ret);
1409 * This sync check and mask will be re-done in init_request_from_bio(),
1410 * but we need to set it earlier to expose the sync flag to the
1411 * rq allocator and io schedulers.
1413 rw_flags = bio_data_dir(bio);
1415 rw_flags |= REQ_SYNC;
1418 * Grab a free request. This is might sleep but can not fail.
1419 * Returns with the queue unlocked.
1421 req = get_request_wait(q, rw_flags, bio);
1422 if (unlikely(!req)) {
1423 bio_endio(bio, -ENODEV); /* @q is dead */
1428 * After dropping the lock and possibly sleeping here, our request
1429 * may now be mergeable after it had proven unmergeable (above).
1430 * We don't worry about that case for efficiency. It won't happen
1431 * often, and the elevators are able to handle it.
1433 init_request_from_bio(req, bio);
1435 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1436 req->cpu = raw_smp_processor_id();
1438 plug = current->plug;
1441 * If this is the first request added after a plug, fire
1442 * of a plug trace. If others have been added before, check
1443 * if we have multiple devices in this plug. If so, make a
1444 * note to sort the list before dispatch.
1446 if (list_empty(&plug->list))
1447 trace_block_plug(q);
1449 if (!plug->should_sort) {
1450 struct request *__rq;
1452 __rq = list_entry_rq(plug->list.prev);
1454 plug->should_sort = 1;
1456 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1457 blk_flush_plug_list(plug, false);
1458 trace_block_plug(q);
1461 list_add_tail(&req->queuelist, &plug->list);
1462 drive_stat_acct(req, 1);
1464 spin_lock_irq(q->queue_lock);
1465 add_acct_request(q, req, where);
1468 spin_unlock_irq(q->queue_lock);
1471 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1474 * If bio->bi_dev is a partition, remap the location
1476 static inline void blk_partition_remap(struct bio *bio)
1478 struct block_device *bdev = bio->bi_bdev;
1480 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1481 struct hd_struct *p = bdev->bd_part;
1483 bio->bi_sector += p->start_sect;
1484 bio->bi_bdev = bdev->bd_contains;
1486 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1488 bio->bi_sector - p->start_sect);
1492 static void handle_bad_sector(struct bio *bio)
1494 char b[BDEVNAME_SIZE];
1496 printk(KERN_INFO "attempt to access beyond end of device\n");
1497 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1498 bdevname(bio->bi_bdev, b),
1500 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1501 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1503 set_bit(BIO_EOF, &bio->bi_flags);
1506 #ifdef CONFIG_FAIL_MAKE_REQUEST
1508 static DECLARE_FAULT_ATTR(fail_make_request);
1510 static int __init setup_fail_make_request(char *str)
1512 return setup_fault_attr(&fail_make_request, str);
1514 __setup("fail_make_request=", setup_fail_make_request);
1516 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1518 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1521 static int __init fail_make_request_debugfs(void)
1523 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1524 NULL, &fail_make_request);
1526 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1529 late_initcall(fail_make_request_debugfs);
1531 #else /* CONFIG_FAIL_MAKE_REQUEST */
1533 static inline bool should_fail_request(struct hd_struct *part,
1539 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1542 * Check whether this bio extends beyond the end of the device.
1544 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1551 /* Test device or partition size, when known. */
1552 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1554 sector_t sector = bio->bi_sector;
1556 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1558 * This may well happen - the kernel calls bread()
1559 * without checking the size of the device, e.g., when
1560 * mounting a device.
1562 handle_bad_sector(bio);
1570 static noinline_for_stack bool
1571 generic_make_request_checks(struct bio *bio)
1573 struct request_queue *q;
1574 int nr_sectors = bio_sectors(bio);
1576 char b[BDEVNAME_SIZE];
1577 struct hd_struct *part;
1581 if (bio_check_eod(bio, nr_sectors))
1584 q = bdev_get_queue(bio->bi_bdev);
1587 "generic_make_request: Trying to access "
1588 "nonexistent block-device %s (%Lu)\n",
1589 bdevname(bio->bi_bdev, b),
1590 (long long) bio->bi_sector);
1594 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1595 nr_sectors > queue_max_hw_sectors(q))) {
1596 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1597 bdevname(bio->bi_bdev, b),
1599 queue_max_hw_sectors(q));
1603 part = bio->bi_bdev->bd_part;
1604 if (should_fail_request(part, bio->bi_size) ||
1605 should_fail_request(&part_to_disk(part)->part0,
1610 * If this device has partitions, remap block n
1611 * of partition p to block n+start(p) of the disk.
1613 blk_partition_remap(bio);
1615 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1618 if (bio_check_eod(bio, nr_sectors))
1622 * Filter flush bio's early so that make_request based
1623 * drivers without flush support don't have to worry
1626 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1627 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1634 if ((bio->bi_rw & REQ_DISCARD) &&
1635 (!blk_queue_discard(q) ||
1636 ((bio->bi_rw & REQ_SECURE) &&
1637 !blk_queue_secdiscard(q)))) {
1642 if (blk_throtl_bio(q, bio))
1643 return false; /* throttled, will be resubmitted later */
1645 trace_block_bio_queue(q, bio);
1649 bio_endio(bio, err);
1654 * generic_make_request - hand a buffer to its device driver for I/O
1655 * @bio: The bio describing the location in memory and on the device.
1657 * generic_make_request() is used to make I/O requests of block
1658 * devices. It is passed a &struct bio, which describes the I/O that needs
1661 * generic_make_request() does not return any status. The
1662 * success/failure status of the request, along with notification of
1663 * completion, is delivered asynchronously through the bio->bi_end_io
1664 * function described (one day) else where.
1666 * The caller of generic_make_request must make sure that bi_io_vec
1667 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1668 * set to describe the device address, and the
1669 * bi_end_io and optionally bi_private are set to describe how
1670 * completion notification should be signaled.
1672 * generic_make_request and the drivers it calls may use bi_next if this
1673 * bio happens to be merged with someone else, and may resubmit the bio to
1674 * a lower device by calling into generic_make_request recursively, which
1675 * means the bio should NOT be touched after the call to ->make_request_fn.
1677 void generic_make_request(struct bio *bio)
1679 struct bio_list bio_list_on_stack;
1681 if (!generic_make_request_checks(bio))
1685 * We only want one ->make_request_fn to be active at a time, else
1686 * stack usage with stacked devices could be a problem. So use
1687 * current->bio_list to keep a list of requests submited by a
1688 * make_request_fn function. current->bio_list is also used as a
1689 * flag to say if generic_make_request is currently active in this
1690 * task or not. If it is NULL, then no make_request is active. If
1691 * it is non-NULL, then a make_request is active, and new requests
1692 * should be added at the tail
1694 if (current->bio_list) {
1695 bio_list_add(current->bio_list, bio);
1699 /* following loop may be a bit non-obvious, and so deserves some
1701 * Before entering the loop, bio->bi_next is NULL (as all callers
1702 * ensure that) so we have a list with a single bio.
1703 * We pretend that we have just taken it off a longer list, so
1704 * we assign bio_list to a pointer to the bio_list_on_stack,
1705 * thus initialising the bio_list of new bios to be
1706 * added. ->make_request() may indeed add some more bios
1707 * through a recursive call to generic_make_request. If it
1708 * did, we find a non-NULL value in bio_list and re-enter the loop
1709 * from the top. In this case we really did just take the bio
1710 * of the top of the list (no pretending) and so remove it from
1711 * bio_list, and call into ->make_request() again.
1713 BUG_ON(bio->bi_next);
1714 bio_list_init(&bio_list_on_stack);
1715 current->bio_list = &bio_list_on_stack;
1717 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1719 q->make_request_fn(q, bio);
1721 bio = bio_list_pop(current->bio_list);
1723 current->bio_list = NULL; /* deactivate */
1725 EXPORT_SYMBOL(generic_make_request);
1728 * submit_bio - submit a bio to the block device layer for I/O
1729 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1730 * @bio: The &struct bio which describes the I/O
1732 * submit_bio() is very similar in purpose to generic_make_request(), and
1733 * uses that function to do most of the work. Both are fairly rough
1734 * interfaces; @bio must be presetup and ready for I/O.
1737 void submit_bio(int rw, struct bio *bio)
1739 int count = bio_sectors(bio);
1744 * If it's a regular read/write or a barrier with data attached,
1745 * go through the normal accounting stuff before submission.
1747 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1749 count_vm_events(PGPGOUT, count);
1751 task_io_account_read(bio->bi_size);
1752 count_vm_events(PGPGIN, count);
1755 if (unlikely(block_dump)) {
1756 char b[BDEVNAME_SIZE];
1757 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1758 current->comm, task_pid_nr(current),
1759 (rw & WRITE) ? "WRITE" : "READ",
1760 (unsigned long long)bio->bi_sector,
1761 bdevname(bio->bi_bdev, b),
1766 generic_make_request(bio);
1768 EXPORT_SYMBOL(submit_bio);
1771 * blk_rq_check_limits - Helper function to check a request for the queue limit
1773 * @rq: the request being checked
1776 * @rq may have been made based on weaker limitations of upper-level queues
1777 * in request stacking drivers, and it may violate the limitation of @q.
1778 * Since the block layer and the underlying device driver trust @rq
1779 * after it is inserted to @q, it should be checked against @q before
1780 * the insertion using this generic function.
1782 * This function should also be useful for request stacking drivers
1783 * in some cases below, so export this function.
1784 * Request stacking drivers like request-based dm may change the queue
1785 * limits while requests are in the queue (e.g. dm's table swapping).
1786 * Such request stacking drivers should check those requests agaist
1787 * the new queue limits again when they dispatch those requests,
1788 * although such checkings are also done against the old queue limits
1789 * when submitting requests.
1791 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1793 if (rq->cmd_flags & REQ_DISCARD)
1796 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1797 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1798 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1803 * queue's settings related to segment counting like q->bounce_pfn
1804 * may differ from that of other stacking queues.
1805 * Recalculate it to check the request correctly on this queue's
1808 blk_recalc_rq_segments(rq);
1809 if (rq->nr_phys_segments > queue_max_segments(q)) {
1810 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1816 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1819 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1820 * @q: the queue to submit the request
1821 * @rq: the request being queued
1823 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1825 unsigned long flags;
1826 int where = ELEVATOR_INSERT_BACK;
1828 if (blk_rq_check_limits(q, rq))
1832 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1835 spin_lock_irqsave(q->queue_lock, flags);
1836 if (unlikely(blk_queue_dead(q))) {
1837 spin_unlock_irqrestore(q->queue_lock, flags);
1842 * Submitting request must be dequeued before calling this function
1843 * because it will be linked to another request_queue
1845 BUG_ON(blk_queued_rq(rq));
1847 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1848 where = ELEVATOR_INSERT_FLUSH;
1850 add_acct_request(q, rq, where);
1851 if (where == ELEVATOR_INSERT_FLUSH)
1853 spin_unlock_irqrestore(q->queue_lock, flags);
1857 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1860 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1861 * @rq: request to examine
1864 * A request could be merge of IOs which require different failure
1865 * handling. This function determines the number of bytes which
1866 * can be failed from the beginning of the request without
1867 * crossing into area which need to be retried further.
1870 * The number of bytes to fail.
1873 * queue_lock must be held.
1875 unsigned int blk_rq_err_bytes(const struct request *rq)
1877 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1878 unsigned int bytes = 0;
1881 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1882 return blk_rq_bytes(rq);
1885 * Currently the only 'mixing' which can happen is between
1886 * different fastfail types. We can safely fail portions
1887 * which have all the failfast bits that the first one has -
1888 * the ones which are at least as eager to fail as the first
1891 for (bio = rq->bio; bio; bio = bio->bi_next) {
1892 if ((bio->bi_rw & ff) != ff)
1894 bytes += bio->bi_size;
1897 /* this could lead to infinite loop */
1898 BUG_ON(blk_rq_bytes(rq) && !bytes);
1901 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1903 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1905 if (blk_do_io_stat(req)) {
1906 const int rw = rq_data_dir(req);
1907 struct hd_struct *part;
1910 cpu = part_stat_lock();
1912 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1917 static void blk_account_io_done(struct request *req)
1920 * Account IO completion. flush_rq isn't accounted as a
1921 * normal IO on queueing nor completion. Accounting the
1922 * containing request is enough.
1924 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1925 unsigned long duration = jiffies - req->start_time;
1926 const int rw = rq_data_dir(req);
1927 struct hd_struct *part;
1930 cpu = part_stat_lock();
1933 part_stat_inc(cpu, part, ios[rw]);
1934 part_stat_add(cpu, part, ticks[rw], duration);
1935 part_round_stats(cpu, part);
1936 part_dec_in_flight(part, rw);
1938 hd_struct_put(part);
1944 * blk_peek_request - peek at the top of a request queue
1945 * @q: request queue to peek at
1948 * Return the request at the top of @q. The returned request
1949 * should be started using blk_start_request() before LLD starts
1953 * Pointer to the request at the top of @q if available. Null
1957 * queue_lock must be held.
1959 struct request *blk_peek_request(struct request_queue *q)
1964 while ((rq = __elv_next_request(q)) != NULL) {
1965 if (!(rq->cmd_flags & REQ_STARTED)) {
1967 * This is the first time the device driver
1968 * sees this request (possibly after
1969 * requeueing). Notify IO scheduler.
1971 if (rq->cmd_flags & REQ_SORTED)
1972 elv_activate_rq(q, rq);
1975 * just mark as started even if we don't start
1976 * it, a request that has been delayed should
1977 * not be passed by new incoming requests
1979 rq->cmd_flags |= REQ_STARTED;
1980 trace_block_rq_issue(q, rq);
1983 if (!q->boundary_rq || q->boundary_rq == rq) {
1984 q->end_sector = rq_end_sector(rq);
1985 q->boundary_rq = NULL;
1988 if (rq->cmd_flags & REQ_DONTPREP)
1991 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1993 * make sure space for the drain appears we
1994 * know we can do this because max_hw_segments
1995 * has been adjusted to be one fewer than the
1998 rq->nr_phys_segments++;
2004 ret = q->prep_rq_fn(q, rq);
2005 if (ret == BLKPREP_OK) {
2007 } else if (ret == BLKPREP_DEFER) {
2009 * the request may have been (partially) prepped.
2010 * we need to keep this request in the front to
2011 * avoid resource deadlock. REQ_STARTED will
2012 * prevent other fs requests from passing this one.
2014 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2015 !(rq->cmd_flags & REQ_DONTPREP)) {
2017 * remove the space for the drain we added
2018 * so that we don't add it again
2020 --rq->nr_phys_segments;
2025 } else if (ret == BLKPREP_KILL) {
2026 rq->cmd_flags |= REQ_QUIET;
2028 * Mark this request as started so we don't trigger
2029 * any debug logic in the end I/O path.
2031 blk_start_request(rq);
2032 __blk_end_request_all(rq, -EIO);
2034 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2041 EXPORT_SYMBOL(blk_peek_request);
2043 void blk_dequeue_request(struct request *rq)
2045 struct request_queue *q = rq->q;
2047 BUG_ON(list_empty(&rq->queuelist));
2048 BUG_ON(ELV_ON_HASH(rq));
2050 list_del_init(&rq->queuelist);
2053 * the time frame between a request being removed from the lists
2054 * and to it is freed is accounted as io that is in progress at
2057 if (blk_account_rq(rq)) {
2058 q->in_flight[rq_is_sync(rq)]++;
2059 set_io_start_time_ns(rq);
2064 * blk_start_request - start request processing on the driver
2065 * @req: request to dequeue
2068 * Dequeue @req and start timeout timer on it. This hands off the
2069 * request to the driver.
2071 * Block internal functions which don't want to start timer should
2072 * call blk_dequeue_request().
2075 * queue_lock must be held.
2077 void blk_start_request(struct request *req)
2079 blk_dequeue_request(req);
2082 * We are now handing the request to the hardware, initialize
2083 * resid_len to full count and add the timeout handler.
2085 req->resid_len = blk_rq_bytes(req);
2086 if (unlikely(blk_bidi_rq(req)))
2087 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2091 EXPORT_SYMBOL(blk_start_request);
2094 * blk_fetch_request - fetch a request from a request queue
2095 * @q: request queue to fetch a request from
2098 * Return the request at the top of @q. The request is started on
2099 * return and LLD can start processing it immediately.
2102 * Pointer to the request at the top of @q if available. Null
2106 * queue_lock must be held.
2108 struct request *blk_fetch_request(struct request_queue *q)
2112 rq = blk_peek_request(q);
2114 blk_start_request(rq);
2117 EXPORT_SYMBOL(blk_fetch_request);
2120 * blk_update_request - Special helper function for request stacking drivers
2121 * @req: the request being processed
2122 * @error: %0 for success, < %0 for error
2123 * @nr_bytes: number of bytes to complete @req
2126 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2127 * the request structure even if @req doesn't have leftover.
2128 * If @req has leftover, sets it up for the next range of segments.
2130 * This special helper function is only for request stacking drivers
2131 * (e.g. request-based dm) so that they can handle partial completion.
2132 * Actual device drivers should use blk_end_request instead.
2134 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2135 * %false return from this function.
2138 * %false - this request doesn't have any more data
2139 * %true - this request has more data
2141 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2143 int total_bytes, bio_nbytes, next_idx = 0;
2149 trace_block_rq_complete(req->q, req);
2152 * For fs requests, rq is just carrier of independent bio's
2153 * and each partial completion should be handled separately.
2154 * Reset per-request error on each partial completion.
2156 * TODO: tj: This is too subtle. It would be better to let
2157 * low level drivers do what they see fit.
2159 if (req->cmd_type == REQ_TYPE_FS)
2162 if (error && req->cmd_type == REQ_TYPE_FS &&
2163 !(req->cmd_flags & REQ_QUIET)) {
2168 error_type = "recoverable transport";
2171 error_type = "critical target";
2174 error_type = "critical nexus";
2181 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2182 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2183 (unsigned long long)blk_rq_pos(req));
2186 blk_account_io_completion(req, nr_bytes);
2188 total_bytes = bio_nbytes = 0;
2189 while ((bio = req->bio) != NULL) {
2192 if (nr_bytes >= bio->bi_size) {
2193 req->bio = bio->bi_next;
2194 nbytes = bio->bi_size;
2195 req_bio_endio(req, bio, nbytes, error);
2199 int idx = bio->bi_idx + next_idx;
2201 if (unlikely(idx >= bio->bi_vcnt)) {
2202 blk_dump_rq_flags(req, "__end_that");
2203 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2204 __func__, idx, bio->bi_vcnt);
2208 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2209 BIO_BUG_ON(nbytes > bio->bi_size);
2212 * not a complete bvec done
2214 if (unlikely(nbytes > nr_bytes)) {
2215 bio_nbytes += nr_bytes;
2216 total_bytes += nr_bytes;
2221 * advance to the next vector
2224 bio_nbytes += nbytes;
2227 total_bytes += nbytes;
2233 * end more in this run, or just return 'not-done'
2235 if (unlikely(nr_bytes <= 0))
2245 * Reset counters so that the request stacking driver
2246 * can find how many bytes remain in the request
2249 req->__data_len = 0;
2254 * if the request wasn't completed, update state
2257 req_bio_endio(req, bio, bio_nbytes, error);
2258 bio->bi_idx += next_idx;
2259 bio_iovec(bio)->bv_offset += nr_bytes;
2260 bio_iovec(bio)->bv_len -= nr_bytes;
2263 req->__data_len -= total_bytes;
2264 req->buffer = bio_data(req->bio);
2266 /* update sector only for requests with clear definition of sector */
2267 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2268 req->__sector += total_bytes >> 9;
2270 /* mixed attributes always follow the first bio */
2271 if (req->cmd_flags & REQ_MIXED_MERGE) {
2272 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2273 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2277 * If total number of sectors is less than the first segment
2278 * size, something has gone terribly wrong.
2280 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2281 blk_dump_rq_flags(req, "request botched");
2282 req->__data_len = blk_rq_cur_bytes(req);
2285 /* recalculate the number of segments */
2286 blk_recalc_rq_segments(req);
2290 EXPORT_SYMBOL_GPL(blk_update_request);
2292 static bool blk_update_bidi_request(struct request *rq, int error,
2293 unsigned int nr_bytes,
2294 unsigned int bidi_bytes)
2296 if (blk_update_request(rq, error, nr_bytes))
2299 /* Bidi request must be completed as a whole */
2300 if (unlikely(blk_bidi_rq(rq)) &&
2301 blk_update_request(rq->next_rq, error, bidi_bytes))
2304 if (blk_queue_add_random(rq->q))
2305 add_disk_randomness(rq->rq_disk);
2311 * blk_unprep_request - unprepare a request
2314 * This function makes a request ready for complete resubmission (or
2315 * completion). It happens only after all error handling is complete,
2316 * so represents the appropriate moment to deallocate any resources
2317 * that were allocated to the request in the prep_rq_fn. The queue
2318 * lock is held when calling this.
2320 void blk_unprep_request(struct request *req)
2322 struct request_queue *q = req->q;
2324 req->cmd_flags &= ~REQ_DONTPREP;
2325 if (q->unprep_rq_fn)
2326 q->unprep_rq_fn(q, req);
2328 EXPORT_SYMBOL_GPL(blk_unprep_request);
2331 * queue lock must be held
2333 static void blk_finish_request(struct request *req, int error)
2335 if (blk_rq_tagged(req))
2336 blk_queue_end_tag(req->q, req);
2338 BUG_ON(blk_queued_rq(req));
2340 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2341 laptop_io_completion(&req->q->backing_dev_info);
2343 blk_delete_timer(req);
2345 if (req->cmd_flags & REQ_DONTPREP)
2346 blk_unprep_request(req);
2349 blk_account_io_done(req);
2352 req->end_io(req, error);
2354 if (blk_bidi_rq(req))
2355 __blk_put_request(req->next_rq->q, req->next_rq);
2357 __blk_put_request(req->q, req);
2362 * blk_end_bidi_request - Complete a bidi request
2363 * @rq: the request to complete
2364 * @error: %0 for success, < %0 for error
2365 * @nr_bytes: number of bytes to complete @rq
2366 * @bidi_bytes: number of bytes to complete @rq->next_rq
2369 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2370 * Drivers that supports bidi can safely call this member for any
2371 * type of request, bidi or uni. In the later case @bidi_bytes is
2375 * %false - we are done with this request
2376 * %true - still buffers pending for this request
2378 static bool blk_end_bidi_request(struct request *rq, int error,
2379 unsigned int nr_bytes, unsigned int bidi_bytes)
2381 struct request_queue *q = rq->q;
2382 unsigned long flags;
2384 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2387 spin_lock_irqsave(q->queue_lock, flags);
2388 blk_finish_request(rq, error);
2389 spin_unlock_irqrestore(q->queue_lock, flags);
2395 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2396 * @rq: the request to complete
2397 * @error: %0 for success, < %0 for error
2398 * @nr_bytes: number of bytes to complete @rq
2399 * @bidi_bytes: number of bytes to complete @rq->next_rq
2402 * Identical to blk_end_bidi_request() except that queue lock is
2403 * assumed to be locked on entry and remains so on return.
2406 * %false - we are done with this request
2407 * %true - still buffers pending for this request
2409 bool __blk_end_bidi_request(struct request *rq, int error,
2410 unsigned int nr_bytes, unsigned int bidi_bytes)
2412 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2415 blk_finish_request(rq, error);
2421 * blk_end_request - Helper function for drivers to complete the request.
2422 * @rq: the request being processed
2423 * @error: %0 for success, < %0 for error
2424 * @nr_bytes: number of bytes to complete
2427 * Ends I/O on a number of bytes attached to @rq.
2428 * If @rq has leftover, sets it up for the next range of segments.
2431 * %false - we are done with this request
2432 * %true - still buffers pending for this request
2434 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2436 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2438 EXPORT_SYMBOL(blk_end_request);
2441 * blk_end_request_all - Helper function for drives to finish the request.
2442 * @rq: the request to finish
2443 * @error: %0 for success, < %0 for error
2446 * Completely finish @rq.
2448 void blk_end_request_all(struct request *rq, int error)
2451 unsigned int bidi_bytes = 0;
2453 if (unlikely(blk_bidi_rq(rq)))
2454 bidi_bytes = blk_rq_bytes(rq->next_rq);
2456 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2459 EXPORT_SYMBOL(blk_end_request_all);
2462 * blk_end_request_cur - Helper function to finish the current request chunk.
2463 * @rq: the request to finish the current chunk for
2464 * @error: %0 for success, < %0 for error
2467 * Complete the current consecutively mapped chunk from @rq.
2470 * %false - we are done with this request
2471 * %true - still buffers pending for this request
2473 bool blk_end_request_cur(struct request *rq, int error)
2475 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2477 EXPORT_SYMBOL(blk_end_request_cur);
2480 * blk_end_request_err - Finish a request till the next failure boundary.
2481 * @rq: the request to finish till the next failure boundary for
2482 * @error: must be negative errno
2485 * Complete @rq till the next failure boundary.
2488 * %false - we are done with this request
2489 * %true - still buffers pending for this request
2491 bool blk_end_request_err(struct request *rq, int error)
2493 WARN_ON(error >= 0);
2494 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2496 EXPORT_SYMBOL_GPL(blk_end_request_err);
2499 * __blk_end_request - Helper function for drivers to complete the request.
2500 * @rq: the request being processed
2501 * @error: %0 for success, < %0 for error
2502 * @nr_bytes: number of bytes to complete
2505 * Must be called with queue lock held unlike blk_end_request().
2508 * %false - we are done with this request
2509 * %true - still buffers pending for this request
2511 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2513 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2515 EXPORT_SYMBOL(__blk_end_request);
2518 * __blk_end_request_all - Helper function for drives to finish the request.
2519 * @rq: the request to finish
2520 * @error: %0 for success, < %0 for error
2523 * Completely finish @rq. Must be called with queue lock held.
2525 void __blk_end_request_all(struct request *rq, int error)
2528 unsigned int bidi_bytes = 0;
2530 if (unlikely(blk_bidi_rq(rq)))
2531 bidi_bytes = blk_rq_bytes(rq->next_rq);
2533 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2536 EXPORT_SYMBOL(__blk_end_request_all);
2539 * __blk_end_request_cur - Helper function to finish the current request chunk.
2540 * @rq: the request to finish the current chunk for
2541 * @error: %0 for success, < %0 for error
2544 * Complete the current consecutively mapped chunk from @rq. Must
2545 * be called with queue lock held.
2548 * %false - we are done with this request
2549 * %true - still buffers pending for this request
2551 bool __blk_end_request_cur(struct request *rq, int error)
2553 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2555 EXPORT_SYMBOL(__blk_end_request_cur);
2558 * __blk_end_request_err - Finish a request till the next failure boundary.
2559 * @rq: the request to finish till the next failure boundary for
2560 * @error: must be negative errno
2563 * Complete @rq till the next failure boundary. Must be called
2564 * with queue lock held.
2567 * %false - we are done with this request
2568 * %true - still buffers pending for this request
2570 bool __blk_end_request_err(struct request *rq, int error)
2572 WARN_ON(error >= 0);
2573 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2575 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2577 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2580 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2581 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2583 if (bio_has_data(bio)) {
2584 rq->nr_phys_segments = bio_phys_segments(q, bio);
2585 rq->buffer = bio_data(bio);
2587 rq->__data_len = bio->bi_size;
2588 rq->bio = rq->biotail = bio;
2591 rq->rq_disk = bio->bi_bdev->bd_disk;
2594 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2596 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2597 * @rq: the request to be flushed
2600 * Flush all pages in @rq.
2602 void rq_flush_dcache_pages(struct request *rq)
2604 struct req_iterator iter;
2605 struct bio_vec *bvec;
2607 rq_for_each_segment(bvec, rq, iter)
2608 flush_dcache_page(bvec->bv_page);
2610 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2614 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2615 * @q : the queue of the device being checked
2618 * Check if underlying low-level drivers of a device are busy.
2619 * If the drivers want to export their busy state, they must set own
2620 * exporting function using blk_queue_lld_busy() first.
2622 * Basically, this function is used only by request stacking drivers
2623 * to stop dispatching requests to underlying devices when underlying
2624 * devices are busy. This behavior helps more I/O merging on the queue
2625 * of the request stacking driver and prevents I/O throughput regression
2626 * on burst I/O load.
2629 * 0 - Not busy (The request stacking driver should dispatch request)
2630 * 1 - Busy (The request stacking driver should stop dispatching request)
2632 int blk_lld_busy(struct request_queue *q)
2635 return q->lld_busy_fn(q);
2639 EXPORT_SYMBOL_GPL(blk_lld_busy);
2642 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2643 * @rq: the clone request to be cleaned up
2646 * Free all bios in @rq for a cloned request.
2648 void blk_rq_unprep_clone(struct request *rq)
2652 while ((bio = rq->bio) != NULL) {
2653 rq->bio = bio->bi_next;
2658 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2661 * Copy attributes of the original request to the clone request.
2662 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2664 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2666 dst->cpu = src->cpu;
2667 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2668 dst->cmd_type = src->cmd_type;
2669 dst->__sector = blk_rq_pos(src);
2670 dst->__data_len = blk_rq_bytes(src);
2671 dst->nr_phys_segments = src->nr_phys_segments;
2672 dst->ioprio = src->ioprio;
2673 dst->extra_len = src->extra_len;
2677 * blk_rq_prep_clone - Helper function to setup clone request
2678 * @rq: the request to be setup
2679 * @rq_src: original request to be cloned
2680 * @bs: bio_set that bios for clone are allocated from
2681 * @gfp_mask: memory allocation mask for bio
2682 * @bio_ctr: setup function to be called for each clone bio.
2683 * Returns %0 for success, non %0 for failure.
2684 * @data: private data to be passed to @bio_ctr
2687 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2688 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2689 * are not copied, and copying such parts is the caller's responsibility.
2690 * Also, pages which the original bios are pointing to are not copied
2691 * and the cloned bios just point same pages.
2692 * So cloned bios must be completed before original bios, which means
2693 * the caller must complete @rq before @rq_src.
2695 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2696 struct bio_set *bs, gfp_t gfp_mask,
2697 int (*bio_ctr)(struct bio *, struct bio *, void *),
2700 struct bio *bio, *bio_src;
2705 blk_rq_init(NULL, rq);
2707 __rq_for_each_bio(bio_src, rq_src) {
2708 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2712 __bio_clone(bio, bio_src);
2714 if (bio_integrity(bio_src) &&
2715 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2718 if (bio_ctr && bio_ctr(bio, bio_src, data))
2722 rq->biotail->bi_next = bio;
2725 rq->bio = rq->biotail = bio;
2728 __blk_rq_prep_clone(rq, rq_src);
2735 blk_rq_unprep_clone(rq);
2739 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2741 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2743 return queue_work(kblockd_workqueue, work);
2745 EXPORT_SYMBOL(kblockd_schedule_work);
2747 int kblockd_schedule_delayed_work(struct request_queue *q,
2748 struct delayed_work *dwork, unsigned long delay)
2750 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2752 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2754 #define PLUG_MAGIC 0x91827364
2757 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2758 * @plug: The &struct blk_plug that needs to be initialized
2761 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2762 * pending I/O should the task end up blocking between blk_start_plug() and
2763 * blk_finish_plug(). This is important from a performance perspective, but
2764 * also ensures that we don't deadlock. For instance, if the task is blocking
2765 * for a memory allocation, memory reclaim could end up wanting to free a
2766 * page belonging to that request that is currently residing in our private
2767 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2768 * this kind of deadlock.
2770 void blk_start_plug(struct blk_plug *plug)
2772 struct task_struct *tsk = current;
2774 plug->magic = PLUG_MAGIC;
2775 INIT_LIST_HEAD(&plug->list);
2776 INIT_LIST_HEAD(&plug->cb_list);
2777 plug->should_sort = 0;
2780 * If this is a nested plug, don't actually assign it. It will be
2781 * flushed on its own.
2785 * Store ordering should not be needed here, since a potential
2786 * preempt will imply a full memory barrier
2791 EXPORT_SYMBOL(blk_start_plug);
2793 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2795 struct request *rqa = container_of(a, struct request, queuelist);
2796 struct request *rqb = container_of(b, struct request, queuelist);
2798 return !(rqa->q <= rqb->q);
2802 * If 'from_schedule' is true, then postpone the dispatch of requests
2803 * until a safe kblockd context. We due this to avoid accidental big
2804 * additional stack usage in driver dispatch, in places where the originally
2805 * plugger did not intend it.
2807 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2809 __releases(q->queue_lock)
2811 trace_block_unplug(q, depth, !from_schedule);
2814 * Don't mess with dead queue.
2816 if (unlikely(blk_queue_dead(q))) {
2817 spin_unlock(q->queue_lock);
2822 * If we are punting this to kblockd, then we can safely drop
2823 * the queue_lock before waking kblockd (which needs to take
2826 if (from_schedule) {
2827 spin_unlock(q->queue_lock);
2828 blk_run_queue_async(q);
2831 spin_unlock(q->queue_lock);
2836 static void flush_plug_callbacks(struct blk_plug *plug)
2838 LIST_HEAD(callbacks);
2840 if (list_empty(&plug->cb_list))
2843 list_splice_init(&plug->cb_list, &callbacks);
2845 while (!list_empty(&callbacks)) {
2846 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2849 list_del(&cb->list);
2854 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2856 struct request_queue *q;
2857 unsigned long flags;
2862 BUG_ON(plug->magic != PLUG_MAGIC);
2864 flush_plug_callbacks(plug);
2865 if (list_empty(&plug->list))
2868 list_splice_init(&plug->list, &list);
2870 if (plug->should_sort) {
2871 list_sort(NULL, &list, plug_rq_cmp);
2872 plug->should_sort = 0;
2879 * Save and disable interrupts here, to avoid doing it for every
2880 * queue lock we have to take.
2882 local_irq_save(flags);
2883 while (!list_empty(&list)) {
2884 rq = list_entry_rq(list.next);
2885 list_del_init(&rq->queuelist);
2889 * This drops the queue lock
2892 queue_unplugged(q, depth, from_schedule);
2895 spin_lock(q->queue_lock);
2899 * Short-circuit if @q is dead
2901 if (unlikely(blk_queue_dead(q))) {
2902 __blk_end_request_all(rq, -ENODEV);
2907 * rq is already accounted, so use raw insert
2909 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2910 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2912 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2918 * This drops the queue lock
2921 queue_unplugged(q, depth, from_schedule);
2923 local_irq_restore(flags);
2926 void blk_finish_plug(struct blk_plug *plug)
2928 blk_flush_plug_list(plug, false);
2930 if (plug == current->plug)
2931 current->plug = NULL;
2933 EXPORT_SYMBOL(blk_finish_plug);
2935 int __init blk_dev_init(void)
2937 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2938 sizeof(((struct request *)0)->cmd_flags));
2940 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2941 kblockd_workqueue = alloc_workqueue("kblockd",
2942 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2943 if (!kblockd_workqueue)
2944 panic("Failed to create kblockd\n");
2946 request_cachep = kmem_cache_create("blkdev_requests",
2947 sizeof(struct request), 0, SLAB_PANIC, NULL);
2949 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2950 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);