2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
41 static int __make_request(struct request_queue *q, struct bio *bio);
44 * For the allocated request tables
46 static struct kmem_cache *request_cachep;
49 * For queue allocation
51 struct kmem_cache *blk_requestq_cachep;
54 * Controlling structure to kblockd
56 static struct workqueue_struct *kblockd_workqueue;
58 static void drive_stat_acct(struct request *rq, int new_io)
60 struct hd_struct *part;
61 int rw = rq_data_dir(rq);
64 if (!blk_do_io_stat(rq))
67 cpu = part_stat_lock();
71 part_stat_inc(cpu, part, merges[rw]);
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
74 if (!hd_struct_try_get(part)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part = &rq->rq_disk->part0;
86 part_round_stats(cpu, part);
87 part_inc_in_flight(part, rw);
94 void blk_queue_congestion_threshold(struct request_queue *q)
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
101 q->nr_congestion_on = nr;
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
106 q->nr_congestion_off = nr;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120 struct backing_dev_info *ret = NULL;
121 struct request_queue *q = bdev_get_queue(bdev);
124 ret = &q->backing_dev_info;
127 EXPORT_SYMBOL(blk_get_backing_dev_info);
129 void blk_rq_init(struct request_queue *q, struct request *rq)
131 memset(rq, 0, sizeof(*rq));
133 INIT_LIST_HEAD(&rq->queuelist);
134 INIT_LIST_HEAD(&rq->timeout_list);
137 rq->__sector = (sector_t) -1;
138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
141 rq->cmd_len = BLK_MAX_CDB;
144 rq->start_time = jiffies;
145 set_start_time_ns(rq);
148 EXPORT_SYMBOL(blk_rq_init);
150 static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
178 void blk_dump_rq_flags(struct request *rq, char *msg)
182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
193 printk(KERN_INFO " cdb: ");
194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
195 printk("%02x ", rq->cmd[bit]);
199 EXPORT_SYMBOL(blk_dump_rq_flags);
202 * Make sure that plugs that were pending when this function was entered,
203 * are now complete and requests pushed to the queue.
205 static inline void queue_sync_plugs(struct request_queue *q)
208 * If the current process is plugged and has barriers submitted,
209 * we will livelock if we don't unplug first.
211 blk_flush_plug(current);
214 static void blk_delay_work(struct work_struct *work)
216 struct request_queue *q;
218 q = container_of(work, struct request_queue, delay_work.work);
219 spin_lock_irq(q->queue_lock);
220 __blk_run_queue(q, false);
221 spin_unlock_irq(q->queue_lock);
225 * blk_delay_queue - restart queueing after defined interval
226 * @q: The &struct request_queue in question
227 * @msecs: Delay in msecs
230 * Sometimes queueing needs to be postponed for a little while, to allow
231 * resources to come back. This function will make sure that queueing is
232 * restarted around the specified time.
234 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
236 schedule_delayed_work(&q->delay_work, msecs_to_jiffies(msecs));
238 EXPORT_SYMBOL(blk_delay_queue);
241 * blk_start_queue - restart a previously stopped queue
242 * @q: The &struct request_queue in question
245 * blk_start_queue() will clear the stop flag on the queue, and call
246 * the request_fn for the queue if it was in a stopped state when
247 * entered. Also see blk_stop_queue(). Queue lock must be held.
249 void blk_start_queue(struct request_queue *q)
251 WARN_ON(!irqs_disabled());
253 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
254 __blk_run_queue(q, false);
256 EXPORT_SYMBOL(blk_start_queue);
259 * blk_stop_queue - stop a queue
260 * @q: The &struct request_queue in question
263 * The Linux block layer assumes that a block driver will consume all
264 * entries on the request queue when the request_fn strategy is called.
265 * Often this will not happen, because of hardware limitations (queue
266 * depth settings). If a device driver gets a 'queue full' response,
267 * or if it simply chooses not to queue more I/O at one point, it can
268 * call this function to prevent the request_fn from being called until
269 * the driver has signalled it's ready to go again. This happens by calling
270 * blk_start_queue() to restart queue operations. Queue lock must be held.
272 void blk_stop_queue(struct request_queue *q)
274 __cancel_delayed_work(&q->delay_work);
275 queue_flag_set(QUEUE_FLAG_STOPPED, q);
277 EXPORT_SYMBOL(blk_stop_queue);
280 * blk_sync_queue - cancel any pending callbacks on a queue
284 * The block layer may perform asynchronous callback activity
285 * on a queue, such as calling the unplug function after a timeout.
286 * A block device may call blk_sync_queue to ensure that any
287 * such activity is cancelled, thus allowing it to release resources
288 * that the callbacks might use. The caller must already have made sure
289 * that its ->make_request_fn will not re-add plugging prior to calling
292 * This function does not cancel any asynchronous activity arising
293 * out of elevator or throttling code. That would require elevaotor_exit()
294 * and blk_throtl_exit() to be called with queue lock initialized.
297 void blk_sync_queue(struct request_queue *q)
299 del_timer_sync(&q->timeout);
300 cancel_delayed_work_sync(&q->delay_work);
303 EXPORT_SYMBOL(blk_sync_queue);
306 * __blk_run_queue - run a single device queue
307 * @q: The queue to run
308 * @force_kblockd: Don't run @q->request_fn directly. Use kblockd.
311 * See @blk_run_queue. This variant must be called with the queue lock
312 * held and interrupts disabled.
315 void __blk_run_queue(struct request_queue *q, bool force_kblockd)
317 if (unlikely(blk_queue_stopped(q)))
321 * Only recurse once to avoid overrunning the stack, let the unplug
322 * handling reinvoke the handler shortly if we already got there.
324 if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
326 queue_flag_clear(QUEUE_FLAG_REENTER, q);
328 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
330 EXPORT_SYMBOL(__blk_run_queue);
333 * blk_run_queue - run a single device queue
334 * @q: The queue to run
337 * Invoke request handling on this queue, if it has pending work to do.
338 * May be used to restart queueing when a request has completed.
340 void blk_run_queue(struct request_queue *q)
344 spin_lock_irqsave(q->queue_lock, flags);
345 __blk_run_queue(q, false);
346 spin_unlock_irqrestore(q->queue_lock, flags);
348 EXPORT_SYMBOL(blk_run_queue);
350 void blk_put_queue(struct request_queue *q)
352 kobject_put(&q->kobj);
356 * Note: If a driver supplied the queue lock, it should not zap that lock
357 * unexpectedly as some queue cleanup components like elevator_exit() and
358 * blk_throtl_exit() need queue lock.
360 void blk_cleanup_queue(struct request_queue *q)
363 * We know we have process context here, so we can be a little
364 * cautious and ensure that pending block actions on this device
365 * are done before moving on. Going into this function, we should
366 * not have processes doing IO to this device.
370 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
371 mutex_lock(&q->sysfs_lock);
372 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
373 mutex_unlock(&q->sysfs_lock);
376 elevator_exit(q->elevator);
382 EXPORT_SYMBOL(blk_cleanup_queue);
384 static int blk_init_free_list(struct request_queue *q)
386 struct request_list *rl = &q->rq;
388 if (unlikely(rl->rq_pool))
391 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
392 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
394 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
395 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
397 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
398 mempool_free_slab, request_cachep, q->node);
406 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
408 return blk_alloc_queue_node(gfp_mask, -1);
410 EXPORT_SYMBOL(blk_alloc_queue);
412 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
414 struct request_queue *q;
417 q = kmem_cache_alloc_node(blk_requestq_cachep,
418 gfp_mask | __GFP_ZERO, node_id);
422 q->backing_dev_info.ra_pages =
423 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
424 q->backing_dev_info.state = 0;
425 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
426 q->backing_dev_info.name = "block";
428 err = bdi_init(&q->backing_dev_info);
430 kmem_cache_free(blk_requestq_cachep, q);
434 if (blk_throtl_init(q)) {
435 kmem_cache_free(blk_requestq_cachep, q);
439 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
440 laptop_mode_timer_fn, (unsigned long) q);
441 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
442 INIT_LIST_HEAD(&q->timeout_list);
443 INIT_LIST_HEAD(&q->flush_queue[0]);
444 INIT_LIST_HEAD(&q->flush_queue[1]);
445 INIT_LIST_HEAD(&q->flush_data_in_flight);
446 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
448 kobject_init(&q->kobj, &blk_queue_ktype);
450 mutex_init(&q->sysfs_lock);
451 spin_lock_init(&q->__queue_lock);
454 * By default initialize queue_lock to internal lock and driver can
455 * override it later if need be.
457 q->queue_lock = &q->__queue_lock;
461 EXPORT_SYMBOL(blk_alloc_queue_node);
464 * blk_init_queue - prepare a request queue for use with a block device
465 * @rfn: The function to be called to process requests that have been
466 * placed on the queue.
467 * @lock: Request queue spin lock
470 * If a block device wishes to use the standard request handling procedures,
471 * which sorts requests and coalesces adjacent requests, then it must
472 * call blk_init_queue(). The function @rfn will be called when there
473 * are requests on the queue that need to be processed. If the device
474 * supports plugging, then @rfn may not be called immediately when requests
475 * are available on the queue, but may be called at some time later instead.
476 * Plugged queues are generally unplugged when a buffer belonging to one
477 * of the requests on the queue is needed, or due to memory pressure.
479 * @rfn is not required, or even expected, to remove all requests off the
480 * queue, but only as many as it can handle at a time. If it does leave
481 * requests on the queue, it is responsible for arranging that the requests
482 * get dealt with eventually.
484 * The queue spin lock must be held while manipulating the requests on the
485 * request queue; this lock will be taken also from interrupt context, so irq
486 * disabling is needed for it.
488 * Function returns a pointer to the initialized request queue, or %NULL if
492 * blk_init_queue() must be paired with a blk_cleanup_queue() call
493 * when the block device is deactivated (such as at module unload).
496 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
498 return blk_init_queue_node(rfn, lock, -1);
500 EXPORT_SYMBOL(blk_init_queue);
502 struct request_queue *
503 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
505 struct request_queue *uninit_q, *q;
507 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
511 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
513 blk_cleanup_queue(uninit_q);
517 EXPORT_SYMBOL(blk_init_queue_node);
519 struct request_queue *
520 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
523 return blk_init_allocated_queue_node(q, rfn, lock, -1);
525 EXPORT_SYMBOL(blk_init_allocated_queue);
527 struct request_queue *
528 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
529 spinlock_t *lock, int node_id)
535 if (blk_init_free_list(q))
539 q->prep_rq_fn = NULL;
540 q->unprep_rq_fn = NULL;
541 q->queue_flags = QUEUE_FLAG_DEFAULT;
543 /* Override internal queue lock with supplied lock pointer */
545 q->queue_lock = lock;
548 * This also sets hw/phys segments, boundary and size
550 blk_queue_make_request(q, __make_request);
552 q->sg_reserved_size = INT_MAX;
557 if (!elevator_init(q, NULL)) {
558 blk_queue_congestion_threshold(q);
564 EXPORT_SYMBOL(blk_init_allocated_queue_node);
566 int blk_get_queue(struct request_queue *q)
568 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
569 kobject_get(&q->kobj);
576 static inline void blk_free_request(struct request_queue *q, struct request *rq)
578 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
580 if (rq->cmd_flags & REQ_ELVPRIV)
581 elv_put_request(q, rq);
582 mempool_free(rq, q->rq.rq_pool);
585 static struct request *
586 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
588 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
595 rq->cmd_flags = flags | REQ_ALLOCED;
598 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
599 mempool_free(rq, q->rq.rq_pool);
602 rq->cmd_flags |= REQ_ELVPRIV;
609 * ioc_batching returns true if the ioc is a valid batching request and
610 * should be given priority access to a request.
612 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
618 * Make sure the process is able to allocate at least 1 request
619 * even if the batch times out, otherwise we could theoretically
622 return ioc->nr_batch_requests == q->nr_batching ||
623 (ioc->nr_batch_requests > 0
624 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
628 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
629 * will cause the process to be a "batcher" on all queues in the system. This
630 * is the behaviour we want though - once it gets a wakeup it should be given
633 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
635 if (!ioc || ioc_batching(q, ioc))
638 ioc->nr_batch_requests = q->nr_batching;
639 ioc->last_waited = jiffies;
642 static void __freed_request(struct request_queue *q, int sync)
644 struct request_list *rl = &q->rq;
646 if (rl->count[sync] < queue_congestion_off_threshold(q))
647 blk_clear_queue_congested(q, sync);
649 if (rl->count[sync] + 1 <= q->nr_requests) {
650 if (waitqueue_active(&rl->wait[sync]))
651 wake_up(&rl->wait[sync]);
653 blk_clear_queue_full(q, sync);
658 * A request has just been released. Account for it, update the full and
659 * congestion status, wake up any waiters. Called under q->queue_lock.
661 static void freed_request(struct request_queue *q, int sync, int priv)
663 struct request_list *rl = &q->rq;
669 __freed_request(q, sync);
671 if (unlikely(rl->starved[sync ^ 1]))
672 __freed_request(q, sync ^ 1);
676 * Determine if elevator data should be initialized when allocating the
677 * request associated with @bio.
679 static bool blk_rq_should_init_elevator(struct bio *bio)
685 * Flush requests do not use the elevator so skip initialization.
686 * This allows a request to share the flush and elevator data.
688 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
695 * Get a free request, queue_lock must be held.
696 * Returns NULL on failure, with queue_lock held.
697 * Returns !NULL on success, with queue_lock *not held*.
699 static struct request *get_request(struct request_queue *q, int rw_flags,
700 struct bio *bio, gfp_t gfp_mask)
702 struct request *rq = NULL;
703 struct request_list *rl = &q->rq;
704 struct io_context *ioc = NULL;
705 const bool is_sync = rw_is_sync(rw_flags) != 0;
706 int may_queue, priv = 0;
708 may_queue = elv_may_queue(q, rw_flags);
709 if (may_queue == ELV_MQUEUE_NO)
712 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
713 if (rl->count[is_sync]+1 >= q->nr_requests) {
714 ioc = current_io_context(GFP_ATOMIC, q->node);
716 * The queue will fill after this allocation, so set
717 * it as full, and mark this process as "batching".
718 * This process will be allowed to complete a batch of
719 * requests, others will be blocked.
721 if (!blk_queue_full(q, is_sync)) {
722 ioc_set_batching(q, ioc);
723 blk_set_queue_full(q, is_sync);
725 if (may_queue != ELV_MQUEUE_MUST
726 && !ioc_batching(q, ioc)) {
728 * The queue is full and the allocating
729 * process is not a "batcher", and not
730 * exempted by the IO scheduler
736 blk_set_queue_congested(q, is_sync);
740 * Only allow batching queuers to allocate up to 50% over the defined
741 * limit of requests, otherwise we could have thousands of requests
742 * allocated with any setting of ->nr_requests
744 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
747 rl->count[is_sync]++;
748 rl->starved[is_sync] = 0;
750 if (blk_rq_should_init_elevator(bio)) {
751 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
756 if (blk_queue_io_stat(q))
757 rw_flags |= REQ_IO_STAT;
758 spin_unlock_irq(q->queue_lock);
760 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
763 * Allocation failed presumably due to memory. Undo anything
764 * we might have messed up.
766 * Allocating task should really be put onto the front of the
767 * wait queue, but this is pretty rare.
769 spin_lock_irq(q->queue_lock);
770 freed_request(q, is_sync, priv);
773 * in the very unlikely event that allocation failed and no
774 * requests for this direction was pending, mark us starved
775 * so that freeing of a request in the other direction will
776 * notice us. another possible fix would be to split the
777 * rq mempool into READ and WRITE
780 if (unlikely(rl->count[is_sync] == 0))
781 rl->starved[is_sync] = 1;
787 * ioc may be NULL here, and ioc_batching will be false. That's
788 * OK, if the queue is under the request limit then requests need
789 * not count toward the nr_batch_requests limit. There will always
790 * be some limit enforced by BLK_BATCH_TIME.
792 if (ioc_batching(q, ioc))
793 ioc->nr_batch_requests--;
795 trace_block_getrq(q, bio, rw_flags & 1);
801 * No available requests for this queue, wait for some requests to become
804 * Called with q->queue_lock held, and returns with it unlocked.
806 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
809 const bool is_sync = rw_is_sync(rw_flags) != 0;
812 rq = get_request(q, rw_flags, bio, GFP_NOIO);
815 struct io_context *ioc;
816 struct request_list *rl = &q->rq;
818 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
819 TASK_UNINTERRUPTIBLE);
821 trace_block_sleeprq(q, bio, rw_flags & 1);
823 spin_unlock_irq(q->queue_lock);
827 * After sleeping, we become a "batching" process and
828 * will be able to allocate at least one request, and
829 * up to a big batch of them for a small period time.
830 * See ioc_batching, ioc_set_batching
832 ioc = current_io_context(GFP_NOIO, q->node);
833 ioc_set_batching(q, ioc);
835 spin_lock_irq(q->queue_lock);
836 finish_wait(&rl->wait[is_sync], &wait);
838 rq = get_request(q, rw_flags, bio, GFP_NOIO);
844 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
848 BUG_ON(rw != READ && rw != WRITE);
850 spin_lock_irq(q->queue_lock);
851 if (gfp_mask & __GFP_WAIT) {
852 rq = get_request_wait(q, rw, NULL);
854 rq = get_request(q, rw, NULL, gfp_mask);
856 spin_unlock_irq(q->queue_lock);
858 /* q->queue_lock is unlocked at this point */
862 EXPORT_SYMBOL(blk_get_request);
865 * blk_make_request - given a bio, allocate a corresponding struct request.
866 * @q: target request queue
867 * @bio: The bio describing the memory mappings that will be submitted for IO.
868 * It may be a chained-bio properly constructed by block/bio layer.
869 * @gfp_mask: gfp flags to be used for memory allocation
871 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
872 * type commands. Where the struct request needs to be farther initialized by
873 * the caller. It is passed a &struct bio, which describes the memory info of
876 * The caller of blk_make_request must make sure that bi_io_vec
877 * are set to describe the memory buffers. That bio_data_dir() will return
878 * the needed direction of the request. (And all bio's in the passed bio-chain
879 * are properly set accordingly)
881 * If called under none-sleepable conditions, mapped bio buffers must not
882 * need bouncing, by calling the appropriate masked or flagged allocator,
883 * suitable for the target device. Otherwise the call to blk_queue_bounce will
886 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
887 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
888 * anything but the first bio in the chain. Otherwise you risk waiting for IO
889 * completion of a bio that hasn't been submitted yet, thus resulting in a
890 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
891 * of bio_alloc(), as that avoids the mempool deadlock.
892 * If possible a big IO should be split into smaller parts when allocation
893 * fails. Partial allocation should not be an error, or you risk a live-lock.
895 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
898 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
901 return ERR_PTR(-ENOMEM);
904 struct bio *bounce_bio = bio;
907 blk_queue_bounce(q, &bounce_bio);
908 ret = blk_rq_append_bio(q, rq, bounce_bio);
917 EXPORT_SYMBOL(blk_make_request);
920 * blk_requeue_request - put a request back on queue
921 * @q: request queue where request should be inserted
922 * @rq: request to be inserted
925 * Drivers often keep queueing requests until the hardware cannot accept
926 * more, when that condition happens we need to put the request back
927 * on the queue. Must be called with queue lock held.
929 void blk_requeue_request(struct request_queue *q, struct request *rq)
931 blk_delete_timer(rq);
932 blk_clear_rq_complete(rq);
933 trace_block_rq_requeue(q, rq);
935 if (blk_rq_tagged(rq))
936 blk_queue_end_tag(q, rq);
938 BUG_ON(blk_queued_rq(rq));
940 elv_requeue_request(q, rq);
942 EXPORT_SYMBOL(blk_requeue_request);
944 static void add_acct_request(struct request_queue *q, struct request *rq,
947 drive_stat_acct(rq, 1);
948 __elv_add_request(q, rq, where);
952 * blk_insert_request - insert a special request into a request queue
953 * @q: request queue where request should be inserted
954 * @rq: request to be inserted
955 * @at_head: insert request at head or tail of queue
956 * @data: private data
959 * Many block devices need to execute commands asynchronously, so they don't
960 * block the whole kernel from preemption during request execution. This is
961 * accomplished normally by inserting aritficial requests tagged as
962 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
963 * be scheduled for actual execution by the request queue.
965 * We have the option of inserting the head or the tail of the queue.
966 * Typically we use the tail for new ioctls and so forth. We use the head
967 * of the queue for things like a QUEUE_FULL message from a device, or a
968 * host that is unable to accept a particular command.
970 void blk_insert_request(struct request_queue *q, struct request *rq,
971 int at_head, void *data)
973 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
977 * tell I/O scheduler that this isn't a regular read/write (ie it
978 * must not attempt merges on this) and that it acts as a soft
981 rq->cmd_type = REQ_TYPE_SPECIAL;
985 spin_lock_irqsave(q->queue_lock, flags);
988 * If command is tagged, release the tag
990 if (blk_rq_tagged(rq))
991 blk_queue_end_tag(q, rq);
993 add_acct_request(q, rq, where);
994 __blk_run_queue(q, false);
995 spin_unlock_irqrestore(q->queue_lock, flags);
997 EXPORT_SYMBOL(blk_insert_request);
999 static void part_round_stats_single(int cpu, struct hd_struct *part,
1002 if (now == part->stamp)
1005 if (part_in_flight(part)) {
1006 __part_stat_add(cpu, part, time_in_queue,
1007 part_in_flight(part) * (now - part->stamp));
1008 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1014 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1015 * @cpu: cpu number for stats access
1016 * @part: target partition
1018 * The average IO queue length and utilisation statistics are maintained
1019 * by observing the current state of the queue length and the amount of
1020 * time it has been in this state for.
1022 * Normally, that accounting is done on IO completion, but that can result
1023 * in more than a second's worth of IO being accounted for within any one
1024 * second, leading to >100% utilisation. To deal with that, we call this
1025 * function to do a round-off before returning the results when reading
1026 * /proc/diskstats. This accounts immediately for all queue usage up to
1027 * the current jiffies and restarts the counters again.
1029 void part_round_stats(int cpu, struct hd_struct *part)
1031 unsigned long now = jiffies;
1034 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1035 part_round_stats_single(cpu, part, now);
1037 EXPORT_SYMBOL_GPL(part_round_stats);
1040 * queue lock must be held
1042 void __blk_put_request(struct request_queue *q, struct request *req)
1046 if (unlikely(--req->ref_count))
1049 elv_completed_request(q, req);
1051 /* this is a bio leak */
1052 WARN_ON(req->bio != NULL);
1055 * Request may not have originated from ll_rw_blk. if not,
1056 * it didn't come out of our reserved rq pools
1058 if (req->cmd_flags & REQ_ALLOCED) {
1059 int is_sync = rq_is_sync(req) != 0;
1060 int priv = req->cmd_flags & REQ_ELVPRIV;
1062 BUG_ON(!list_empty(&req->queuelist));
1063 BUG_ON(!hlist_unhashed(&req->hash));
1065 blk_free_request(q, req);
1066 freed_request(q, is_sync, priv);
1069 EXPORT_SYMBOL_GPL(__blk_put_request);
1071 void blk_put_request(struct request *req)
1073 unsigned long flags;
1074 struct request_queue *q = req->q;
1076 spin_lock_irqsave(q->queue_lock, flags);
1077 __blk_put_request(q, req);
1078 spin_unlock_irqrestore(q->queue_lock, flags);
1080 EXPORT_SYMBOL(blk_put_request);
1083 * blk_add_request_payload - add a payload to a request
1084 * @rq: request to update
1085 * @page: page backing the payload
1086 * @len: length of the payload.
1088 * This allows to later add a payload to an already submitted request by
1089 * a block driver. The driver needs to take care of freeing the payload
1092 * Note that this is a quite horrible hack and nothing but handling of
1093 * discard requests should ever use it.
1095 void blk_add_request_payload(struct request *rq, struct page *page,
1098 struct bio *bio = rq->bio;
1100 bio->bi_io_vec->bv_page = page;
1101 bio->bi_io_vec->bv_offset = 0;
1102 bio->bi_io_vec->bv_len = len;
1106 bio->bi_phys_segments = 1;
1108 rq->__data_len = rq->resid_len = len;
1109 rq->nr_phys_segments = 1;
1110 rq->buffer = bio_data(bio);
1112 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1114 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1117 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1120 * Debug stuff, kill later
1122 if (!rq_mergeable(req)) {
1123 blk_dump_rq_flags(req, "back");
1127 if (!ll_back_merge_fn(q, req, bio))
1130 trace_block_bio_backmerge(q, bio);
1132 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1133 blk_rq_set_mixed_merge(req);
1135 req->biotail->bi_next = bio;
1137 req->__data_len += bio->bi_size;
1138 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1140 drive_stat_acct(req, 0);
1144 static bool bio_attempt_front_merge(struct request_queue *q,
1145 struct request *req, struct bio *bio)
1147 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1151 * Debug stuff, kill later
1153 if (!rq_mergeable(req)) {
1154 blk_dump_rq_flags(req, "front");
1158 if (!ll_front_merge_fn(q, req, bio))
1161 trace_block_bio_frontmerge(q, bio);
1163 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1164 blk_rq_set_mixed_merge(req);
1166 sector = bio->bi_sector;
1168 bio->bi_next = req->bio;
1172 * may not be valid. if the low level driver said
1173 * it didn't need a bounce buffer then it better
1174 * not touch req->buffer either...
1176 req->buffer = bio_data(bio);
1177 req->__sector = bio->bi_sector;
1178 req->__data_len += bio->bi_size;
1179 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1181 drive_stat_acct(req, 0);
1186 * Attempts to merge with the plugged list in the current process. Returns
1187 * true if merge was succesful, otherwise false.
1189 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1192 struct blk_plug *plug;
1200 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1206 el_ret = elv_try_merge(rq, bio);
1207 if (el_ret == ELEVATOR_BACK_MERGE) {
1208 ret = bio_attempt_back_merge(q, rq, bio);
1211 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1212 ret = bio_attempt_front_merge(q, rq, bio);
1221 void init_request_from_bio(struct request *req, struct bio *bio)
1223 req->cpu = bio->bi_comp_cpu;
1224 req->cmd_type = REQ_TYPE_FS;
1226 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1227 if (bio->bi_rw & REQ_RAHEAD)
1228 req->cmd_flags |= REQ_FAILFAST_MASK;
1231 req->__sector = bio->bi_sector;
1232 req->ioprio = bio_prio(bio);
1233 blk_rq_bio_prep(req->q, req, bio);
1236 static int __make_request(struct request_queue *q, struct bio *bio)
1238 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1239 struct blk_plug *plug;
1240 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1241 struct request *req;
1244 * low level driver can indicate that it wants pages above a
1245 * certain limit bounced to low memory (ie for highmem, or even
1246 * ISA dma in theory)
1248 blk_queue_bounce(q, &bio);
1250 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1251 spin_lock_irq(q->queue_lock);
1252 where = ELEVATOR_INSERT_FLUSH;
1257 * Check if we can merge with the plugged list before grabbing
1260 if (attempt_plug_merge(current, q, bio))
1263 spin_lock_irq(q->queue_lock);
1265 el_ret = elv_merge(q, &req, bio);
1266 if (el_ret == ELEVATOR_BACK_MERGE) {
1267 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1268 if (bio_attempt_back_merge(q, req, bio)) {
1269 if (!attempt_back_merge(q, req))
1270 elv_merged_request(q, req, el_ret);
1273 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1274 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1275 if (bio_attempt_front_merge(q, req, bio)) {
1276 if (!attempt_front_merge(q, req))
1277 elv_merged_request(q, req, el_ret);
1284 * This sync check and mask will be re-done in init_request_from_bio(),
1285 * but we need to set it earlier to expose the sync flag to the
1286 * rq allocator and io schedulers.
1288 rw_flags = bio_data_dir(bio);
1290 rw_flags |= REQ_SYNC;
1293 * Grab a free request. This is might sleep but can not fail.
1294 * Returns with the queue unlocked.
1296 req = get_request_wait(q, rw_flags, bio);
1299 * After dropping the lock and possibly sleeping here, our request
1300 * may now be mergeable after it had proven unmergeable (above).
1301 * We don't worry about that case for efficiency. It won't happen
1302 * often, and the elevators are able to handle it.
1304 init_request_from_bio(req, bio);
1306 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1307 bio_flagged(bio, BIO_CPU_AFFINE)) {
1308 req->cpu = blk_cpu_to_group(get_cpu());
1312 plug = current->plug;
1314 if (!plug->should_sort && !list_empty(&plug->list)) {
1315 struct request *__rq;
1317 __rq = list_entry_rq(plug->list.prev);
1319 plug->should_sort = 1;
1322 * Debug flag, kill later
1324 req->cmd_flags |= REQ_ON_PLUG;
1325 list_add_tail(&req->queuelist, &plug->list);
1326 drive_stat_acct(req, 1);
1328 spin_lock_irq(q->queue_lock);
1329 add_acct_request(q, req, where);
1330 __blk_run_queue(q, false);
1332 spin_unlock_irq(q->queue_lock);
1339 * If bio->bi_dev is a partition, remap the location
1341 static inline void blk_partition_remap(struct bio *bio)
1343 struct block_device *bdev = bio->bi_bdev;
1345 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1346 struct hd_struct *p = bdev->bd_part;
1348 bio->bi_sector += p->start_sect;
1349 bio->bi_bdev = bdev->bd_contains;
1351 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1353 bio->bi_sector - p->start_sect);
1357 static void handle_bad_sector(struct bio *bio)
1359 char b[BDEVNAME_SIZE];
1361 printk(KERN_INFO "attempt to access beyond end of device\n");
1362 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1363 bdevname(bio->bi_bdev, b),
1365 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1366 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1368 set_bit(BIO_EOF, &bio->bi_flags);
1371 #ifdef CONFIG_FAIL_MAKE_REQUEST
1373 static DECLARE_FAULT_ATTR(fail_make_request);
1375 static int __init setup_fail_make_request(char *str)
1377 return setup_fault_attr(&fail_make_request, str);
1379 __setup("fail_make_request=", setup_fail_make_request);
1381 static int should_fail_request(struct bio *bio)
1383 struct hd_struct *part = bio->bi_bdev->bd_part;
1385 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1386 return should_fail(&fail_make_request, bio->bi_size);
1391 static int __init fail_make_request_debugfs(void)
1393 return init_fault_attr_dentries(&fail_make_request,
1394 "fail_make_request");
1397 late_initcall(fail_make_request_debugfs);
1399 #else /* CONFIG_FAIL_MAKE_REQUEST */
1401 static inline int should_fail_request(struct bio *bio)
1406 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1409 * Check whether this bio extends beyond the end of the device.
1411 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1418 /* Test device or partition size, when known. */
1419 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1421 sector_t sector = bio->bi_sector;
1423 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1425 * This may well happen - the kernel calls bread()
1426 * without checking the size of the device, e.g., when
1427 * mounting a device.
1429 handle_bad_sector(bio);
1438 * generic_make_request - hand a buffer to its device driver for I/O
1439 * @bio: The bio describing the location in memory and on the device.
1441 * generic_make_request() is used to make I/O requests of block
1442 * devices. It is passed a &struct bio, which describes the I/O that needs
1445 * generic_make_request() does not return any status. The
1446 * success/failure status of the request, along with notification of
1447 * completion, is delivered asynchronously through the bio->bi_end_io
1448 * function described (one day) else where.
1450 * The caller of generic_make_request must make sure that bi_io_vec
1451 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1452 * set to describe the device address, and the
1453 * bi_end_io and optionally bi_private are set to describe how
1454 * completion notification should be signaled.
1456 * generic_make_request and the drivers it calls may use bi_next if this
1457 * bio happens to be merged with someone else, and may change bi_dev and
1458 * bi_sector for remaps as it sees fit. So the values of these fields
1459 * should NOT be depended on after the call to generic_make_request.
1461 static inline void __generic_make_request(struct bio *bio)
1463 struct request_queue *q;
1464 sector_t old_sector;
1465 int ret, nr_sectors = bio_sectors(bio);
1471 if (bio_check_eod(bio, nr_sectors))
1475 * Resolve the mapping until finished. (drivers are
1476 * still free to implement/resolve their own stacking
1477 * by explicitly returning 0)
1479 * NOTE: we don't repeat the blk_size check for each new device.
1480 * Stacking drivers are expected to know what they are doing.
1485 char b[BDEVNAME_SIZE];
1487 q = bdev_get_queue(bio->bi_bdev);
1490 "generic_make_request: Trying to access "
1491 "nonexistent block-device %s (%Lu)\n",
1492 bdevname(bio->bi_bdev, b),
1493 (long long) bio->bi_sector);
1497 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1498 nr_sectors > queue_max_hw_sectors(q))) {
1499 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1500 bdevname(bio->bi_bdev, b),
1502 queue_max_hw_sectors(q));
1506 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1509 if (should_fail_request(bio))
1513 * If this device has partitions, remap block n
1514 * of partition p to block n+start(p) of the disk.
1516 blk_partition_remap(bio);
1518 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1521 if (old_sector != -1)
1522 trace_block_bio_remap(q, bio, old_dev, old_sector);
1524 old_sector = bio->bi_sector;
1525 old_dev = bio->bi_bdev->bd_dev;
1527 if (bio_check_eod(bio, nr_sectors))
1531 * Filter flush bio's early so that make_request based
1532 * drivers without flush support don't have to worry
1535 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1536 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1543 if ((bio->bi_rw & REQ_DISCARD) &&
1544 (!blk_queue_discard(q) ||
1545 ((bio->bi_rw & REQ_SECURE) &&
1546 !blk_queue_secdiscard(q)))) {
1551 blk_throtl_bio(q, &bio);
1554 * If bio = NULL, bio has been throttled and will be submitted
1560 trace_block_bio_queue(q, bio);
1562 ret = q->make_request_fn(q, bio);
1568 bio_endio(bio, err);
1572 * We only want one ->make_request_fn to be active at a time,
1573 * else stack usage with stacked devices could be a problem.
1574 * So use current->bio_list to keep a list of requests
1575 * submited by a make_request_fn function.
1576 * current->bio_list is also used as a flag to say if
1577 * generic_make_request is currently active in this task or not.
1578 * If it is NULL, then no make_request is active. If it is non-NULL,
1579 * then a make_request is active, and new requests should be added
1582 void generic_make_request(struct bio *bio)
1584 struct bio_list bio_list_on_stack;
1586 if (current->bio_list) {
1587 /* make_request is active */
1588 bio_list_add(current->bio_list, bio);
1591 /* following loop may be a bit non-obvious, and so deserves some
1593 * Before entering the loop, bio->bi_next is NULL (as all callers
1594 * ensure that) so we have a list with a single bio.
1595 * We pretend that we have just taken it off a longer list, so
1596 * we assign bio_list to a pointer to the bio_list_on_stack,
1597 * thus initialising the bio_list of new bios to be
1598 * added. __generic_make_request may indeed add some more bios
1599 * through a recursive call to generic_make_request. If it
1600 * did, we find a non-NULL value in bio_list and re-enter the loop
1601 * from the top. In this case we really did just take the bio
1602 * of the top of the list (no pretending) and so remove it from
1603 * bio_list, and call into __generic_make_request again.
1605 * The loop was structured like this to make only one call to
1606 * __generic_make_request (which is important as it is large and
1607 * inlined) and to keep the structure simple.
1609 BUG_ON(bio->bi_next);
1610 bio_list_init(&bio_list_on_stack);
1611 current->bio_list = &bio_list_on_stack;
1613 __generic_make_request(bio);
1614 bio = bio_list_pop(current->bio_list);
1616 current->bio_list = NULL; /* deactivate */
1618 EXPORT_SYMBOL(generic_make_request);
1621 * submit_bio - submit a bio to the block device layer for I/O
1622 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1623 * @bio: The &struct bio which describes the I/O
1625 * submit_bio() is very similar in purpose to generic_make_request(), and
1626 * uses that function to do most of the work. Both are fairly rough
1627 * interfaces; @bio must be presetup and ready for I/O.
1630 void submit_bio(int rw, struct bio *bio)
1632 int count = bio_sectors(bio);
1637 * If it's a regular read/write or a barrier with data attached,
1638 * go through the normal accounting stuff before submission.
1640 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1642 count_vm_events(PGPGOUT, count);
1644 task_io_account_read(bio->bi_size);
1645 count_vm_events(PGPGIN, count);
1648 if (unlikely(block_dump)) {
1649 char b[BDEVNAME_SIZE];
1650 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1651 current->comm, task_pid_nr(current),
1652 (rw & WRITE) ? "WRITE" : "READ",
1653 (unsigned long long)bio->bi_sector,
1654 bdevname(bio->bi_bdev, b),
1659 generic_make_request(bio);
1661 EXPORT_SYMBOL(submit_bio);
1664 * blk_rq_check_limits - Helper function to check a request for the queue limit
1666 * @rq: the request being checked
1669 * @rq may have been made based on weaker limitations of upper-level queues
1670 * in request stacking drivers, and it may violate the limitation of @q.
1671 * Since the block layer and the underlying device driver trust @rq
1672 * after it is inserted to @q, it should be checked against @q before
1673 * the insertion using this generic function.
1675 * This function should also be useful for request stacking drivers
1676 * in some cases below, so export this function.
1677 * Request stacking drivers like request-based dm may change the queue
1678 * limits while requests are in the queue (e.g. dm's table swapping).
1679 * Such request stacking drivers should check those requests agaist
1680 * the new queue limits again when they dispatch those requests,
1681 * although such checkings are also done against the old queue limits
1682 * when submitting requests.
1684 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1686 if (rq->cmd_flags & REQ_DISCARD)
1689 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1690 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1691 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1696 * queue's settings related to segment counting like q->bounce_pfn
1697 * may differ from that of other stacking queues.
1698 * Recalculate it to check the request correctly on this queue's
1701 blk_recalc_rq_segments(rq);
1702 if (rq->nr_phys_segments > queue_max_segments(q)) {
1703 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1709 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1712 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1713 * @q: the queue to submit the request
1714 * @rq: the request being queued
1716 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1718 unsigned long flags;
1720 if (blk_rq_check_limits(q, rq))
1723 #ifdef CONFIG_FAIL_MAKE_REQUEST
1724 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1725 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1729 spin_lock_irqsave(q->queue_lock, flags);
1732 * Submitting request must be dequeued before calling this function
1733 * because it will be linked to another request_queue
1735 BUG_ON(blk_queued_rq(rq));
1737 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1738 spin_unlock_irqrestore(q->queue_lock, flags);
1742 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1745 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1746 * @rq: request to examine
1749 * A request could be merge of IOs which require different failure
1750 * handling. This function determines the number of bytes which
1751 * can be failed from the beginning of the request without
1752 * crossing into area which need to be retried further.
1755 * The number of bytes to fail.
1758 * queue_lock must be held.
1760 unsigned int blk_rq_err_bytes(const struct request *rq)
1762 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1763 unsigned int bytes = 0;
1766 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1767 return blk_rq_bytes(rq);
1770 * Currently the only 'mixing' which can happen is between
1771 * different fastfail types. We can safely fail portions
1772 * which have all the failfast bits that the first one has -
1773 * the ones which are at least as eager to fail as the first
1776 for (bio = rq->bio; bio; bio = bio->bi_next) {
1777 if ((bio->bi_rw & ff) != ff)
1779 bytes += bio->bi_size;
1782 /* this could lead to infinite loop */
1783 BUG_ON(blk_rq_bytes(rq) && !bytes);
1786 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1788 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1790 if (blk_do_io_stat(req)) {
1791 const int rw = rq_data_dir(req);
1792 struct hd_struct *part;
1795 cpu = part_stat_lock();
1797 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1802 static void blk_account_io_done(struct request *req)
1805 * Account IO completion. flush_rq isn't accounted as a
1806 * normal IO on queueing nor completion. Accounting the
1807 * containing request is enough.
1809 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1810 unsigned long duration = jiffies - req->start_time;
1811 const int rw = rq_data_dir(req);
1812 struct hd_struct *part;
1815 cpu = part_stat_lock();
1818 part_stat_inc(cpu, part, ios[rw]);
1819 part_stat_add(cpu, part, ticks[rw], duration);
1820 part_round_stats(cpu, part);
1821 part_dec_in_flight(part, rw);
1823 hd_struct_put(part);
1829 * blk_peek_request - peek at the top of a request queue
1830 * @q: request queue to peek at
1833 * Return the request at the top of @q. The returned request
1834 * should be started using blk_start_request() before LLD starts
1838 * Pointer to the request at the top of @q if available. Null
1842 * queue_lock must be held.
1844 struct request *blk_peek_request(struct request_queue *q)
1849 while ((rq = __elv_next_request(q)) != NULL) {
1850 if (!(rq->cmd_flags & REQ_STARTED)) {
1852 * This is the first time the device driver
1853 * sees this request (possibly after
1854 * requeueing). Notify IO scheduler.
1856 if (rq->cmd_flags & REQ_SORTED)
1857 elv_activate_rq(q, rq);
1860 * just mark as started even if we don't start
1861 * it, a request that has been delayed should
1862 * not be passed by new incoming requests
1864 rq->cmd_flags |= REQ_STARTED;
1865 trace_block_rq_issue(q, rq);
1868 if (!q->boundary_rq || q->boundary_rq == rq) {
1869 q->end_sector = rq_end_sector(rq);
1870 q->boundary_rq = NULL;
1873 if (rq->cmd_flags & REQ_DONTPREP)
1876 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1878 * make sure space for the drain appears we
1879 * know we can do this because max_hw_segments
1880 * has been adjusted to be one fewer than the
1883 rq->nr_phys_segments++;
1889 ret = q->prep_rq_fn(q, rq);
1890 if (ret == BLKPREP_OK) {
1892 } else if (ret == BLKPREP_DEFER) {
1894 * the request may have been (partially) prepped.
1895 * we need to keep this request in the front to
1896 * avoid resource deadlock. REQ_STARTED will
1897 * prevent other fs requests from passing this one.
1899 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1900 !(rq->cmd_flags & REQ_DONTPREP)) {
1902 * remove the space for the drain we added
1903 * so that we don't add it again
1905 --rq->nr_phys_segments;
1910 } else if (ret == BLKPREP_KILL) {
1911 rq->cmd_flags |= REQ_QUIET;
1913 * Mark this request as started so we don't trigger
1914 * any debug logic in the end I/O path.
1916 blk_start_request(rq);
1917 __blk_end_request_all(rq, -EIO);
1919 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1926 EXPORT_SYMBOL(blk_peek_request);
1928 void blk_dequeue_request(struct request *rq)
1930 struct request_queue *q = rq->q;
1932 BUG_ON(list_empty(&rq->queuelist));
1933 BUG_ON(ELV_ON_HASH(rq));
1935 list_del_init(&rq->queuelist);
1938 * the time frame between a request being removed from the lists
1939 * and to it is freed is accounted as io that is in progress at
1942 if (blk_account_rq(rq)) {
1943 q->in_flight[rq_is_sync(rq)]++;
1944 set_io_start_time_ns(rq);
1949 * blk_start_request - start request processing on the driver
1950 * @req: request to dequeue
1953 * Dequeue @req and start timeout timer on it. This hands off the
1954 * request to the driver.
1956 * Block internal functions which don't want to start timer should
1957 * call blk_dequeue_request().
1960 * queue_lock must be held.
1962 void blk_start_request(struct request *req)
1964 blk_dequeue_request(req);
1967 * We are now handing the request to the hardware, initialize
1968 * resid_len to full count and add the timeout handler.
1970 req->resid_len = blk_rq_bytes(req);
1971 if (unlikely(blk_bidi_rq(req)))
1972 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1976 EXPORT_SYMBOL(blk_start_request);
1979 * blk_fetch_request - fetch a request from a request queue
1980 * @q: request queue to fetch a request from
1983 * Return the request at the top of @q. The request is started on
1984 * return and LLD can start processing it immediately.
1987 * Pointer to the request at the top of @q if available. Null
1991 * queue_lock must be held.
1993 struct request *blk_fetch_request(struct request_queue *q)
1997 rq = blk_peek_request(q);
1999 blk_start_request(rq);
2002 EXPORT_SYMBOL(blk_fetch_request);
2005 * blk_update_request - Special helper function for request stacking drivers
2006 * @req: the request being processed
2007 * @error: %0 for success, < %0 for error
2008 * @nr_bytes: number of bytes to complete @req
2011 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2012 * the request structure even if @req doesn't have leftover.
2013 * If @req has leftover, sets it up for the next range of segments.
2015 * This special helper function is only for request stacking drivers
2016 * (e.g. request-based dm) so that they can handle partial completion.
2017 * Actual device drivers should use blk_end_request instead.
2019 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2020 * %false return from this function.
2023 * %false - this request doesn't have any more data
2024 * %true - this request has more data
2026 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2028 int total_bytes, bio_nbytes, next_idx = 0;
2034 trace_block_rq_complete(req->q, req);
2037 * For fs requests, rq is just carrier of independent bio's
2038 * and each partial completion should be handled separately.
2039 * Reset per-request error on each partial completion.
2041 * TODO: tj: This is too subtle. It would be better to let
2042 * low level drivers do what they see fit.
2044 if (req->cmd_type == REQ_TYPE_FS)
2047 if (error && req->cmd_type == REQ_TYPE_FS &&
2048 !(req->cmd_flags & REQ_QUIET)) {
2053 error_type = "recoverable transport";
2056 error_type = "critical target";
2059 error_type = "critical nexus";
2066 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2067 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2068 (unsigned long long)blk_rq_pos(req));
2071 blk_account_io_completion(req, nr_bytes);
2073 total_bytes = bio_nbytes = 0;
2074 while ((bio = req->bio) != NULL) {
2077 if (nr_bytes >= bio->bi_size) {
2078 req->bio = bio->bi_next;
2079 nbytes = bio->bi_size;
2080 req_bio_endio(req, bio, nbytes, error);
2084 int idx = bio->bi_idx + next_idx;
2086 if (unlikely(idx >= bio->bi_vcnt)) {
2087 blk_dump_rq_flags(req, "__end_that");
2088 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2089 __func__, idx, bio->bi_vcnt);
2093 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2094 BIO_BUG_ON(nbytes > bio->bi_size);
2097 * not a complete bvec done
2099 if (unlikely(nbytes > nr_bytes)) {
2100 bio_nbytes += nr_bytes;
2101 total_bytes += nr_bytes;
2106 * advance to the next vector
2109 bio_nbytes += nbytes;
2112 total_bytes += nbytes;
2118 * end more in this run, or just return 'not-done'
2120 if (unlikely(nr_bytes <= 0))
2130 * Reset counters so that the request stacking driver
2131 * can find how many bytes remain in the request
2134 req->__data_len = 0;
2139 * if the request wasn't completed, update state
2142 req_bio_endio(req, bio, bio_nbytes, error);
2143 bio->bi_idx += next_idx;
2144 bio_iovec(bio)->bv_offset += nr_bytes;
2145 bio_iovec(bio)->bv_len -= nr_bytes;
2148 req->__data_len -= total_bytes;
2149 req->buffer = bio_data(req->bio);
2151 /* update sector only for requests with clear definition of sector */
2152 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2153 req->__sector += total_bytes >> 9;
2155 /* mixed attributes always follow the first bio */
2156 if (req->cmd_flags & REQ_MIXED_MERGE) {
2157 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2158 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2162 * If total number of sectors is less than the first segment
2163 * size, something has gone terribly wrong.
2165 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2166 printk(KERN_ERR "blk: request botched\n");
2167 req->__data_len = blk_rq_cur_bytes(req);
2170 /* recalculate the number of segments */
2171 blk_recalc_rq_segments(req);
2175 EXPORT_SYMBOL_GPL(blk_update_request);
2177 static bool blk_update_bidi_request(struct request *rq, int error,
2178 unsigned int nr_bytes,
2179 unsigned int bidi_bytes)
2181 if (blk_update_request(rq, error, nr_bytes))
2184 /* Bidi request must be completed as a whole */
2185 if (unlikely(blk_bidi_rq(rq)) &&
2186 blk_update_request(rq->next_rq, error, bidi_bytes))
2189 if (blk_queue_add_random(rq->q))
2190 add_disk_randomness(rq->rq_disk);
2196 * blk_unprep_request - unprepare a request
2199 * This function makes a request ready for complete resubmission (or
2200 * completion). It happens only after all error handling is complete,
2201 * so represents the appropriate moment to deallocate any resources
2202 * that were allocated to the request in the prep_rq_fn. The queue
2203 * lock is held when calling this.
2205 void blk_unprep_request(struct request *req)
2207 struct request_queue *q = req->q;
2209 req->cmd_flags &= ~REQ_DONTPREP;
2210 if (q->unprep_rq_fn)
2211 q->unprep_rq_fn(q, req);
2213 EXPORT_SYMBOL_GPL(blk_unprep_request);
2216 * queue lock must be held
2218 static void blk_finish_request(struct request *req, int error)
2220 if (blk_rq_tagged(req))
2221 blk_queue_end_tag(req->q, req);
2223 BUG_ON(blk_queued_rq(req));
2225 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2226 laptop_io_completion(&req->q->backing_dev_info);
2228 blk_delete_timer(req);
2230 if (req->cmd_flags & REQ_DONTPREP)
2231 blk_unprep_request(req);
2234 blk_account_io_done(req);
2237 req->end_io(req, error);
2239 if (blk_bidi_rq(req))
2240 __blk_put_request(req->next_rq->q, req->next_rq);
2242 __blk_put_request(req->q, req);
2247 * blk_end_bidi_request - Complete a bidi request
2248 * @rq: the request to complete
2249 * @error: %0 for success, < %0 for error
2250 * @nr_bytes: number of bytes to complete @rq
2251 * @bidi_bytes: number of bytes to complete @rq->next_rq
2254 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2255 * Drivers that supports bidi can safely call this member for any
2256 * type of request, bidi or uni. In the later case @bidi_bytes is
2260 * %false - we are done with this request
2261 * %true - still buffers pending for this request
2263 static bool blk_end_bidi_request(struct request *rq, int error,
2264 unsigned int nr_bytes, unsigned int bidi_bytes)
2266 struct request_queue *q = rq->q;
2267 unsigned long flags;
2269 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2272 spin_lock_irqsave(q->queue_lock, flags);
2273 blk_finish_request(rq, error);
2274 spin_unlock_irqrestore(q->queue_lock, flags);
2280 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2281 * @rq: the request to complete
2282 * @error: %0 for success, < %0 for error
2283 * @nr_bytes: number of bytes to complete @rq
2284 * @bidi_bytes: number of bytes to complete @rq->next_rq
2287 * Identical to blk_end_bidi_request() except that queue lock is
2288 * assumed to be locked on entry and remains so on return.
2291 * %false - we are done with this request
2292 * %true - still buffers pending for this request
2294 static bool __blk_end_bidi_request(struct request *rq, int error,
2295 unsigned int nr_bytes, unsigned int bidi_bytes)
2297 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2300 blk_finish_request(rq, error);
2306 * blk_end_request - Helper function for drivers to complete the request.
2307 * @rq: the request being processed
2308 * @error: %0 for success, < %0 for error
2309 * @nr_bytes: number of bytes to complete
2312 * Ends I/O on a number of bytes attached to @rq.
2313 * If @rq has leftover, sets it up for the next range of segments.
2316 * %false - we are done with this request
2317 * %true - still buffers pending for this request
2319 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2321 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2323 EXPORT_SYMBOL(blk_end_request);
2326 * blk_end_request_all - Helper function for drives to finish the request.
2327 * @rq: the request to finish
2328 * @error: %0 for success, < %0 for error
2331 * Completely finish @rq.
2333 void blk_end_request_all(struct request *rq, int error)
2336 unsigned int bidi_bytes = 0;
2338 if (unlikely(blk_bidi_rq(rq)))
2339 bidi_bytes = blk_rq_bytes(rq->next_rq);
2341 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2344 EXPORT_SYMBOL(blk_end_request_all);
2347 * blk_end_request_cur - Helper function to finish the current request chunk.
2348 * @rq: the request to finish the current chunk for
2349 * @error: %0 for success, < %0 for error
2352 * Complete the current consecutively mapped chunk from @rq.
2355 * %false - we are done with this request
2356 * %true - still buffers pending for this request
2358 bool blk_end_request_cur(struct request *rq, int error)
2360 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2362 EXPORT_SYMBOL(blk_end_request_cur);
2365 * blk_end_request_err - Finish a request till the next failure boundary.
2366 * @rq: the request to finish till the next failure boundary for
2367 * @error: must be negative errno
2370 * Complete @rq till the next failure boundary.
2373 * %false - we are done with this request
2374 * %true - still buffers pending for this request
2376 bool blk_end_request_err(struct request *rq, int error)
2378 WARN_ON(error >= 0);
2379 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2381 EXPORT_SYMBOL_GPL(blk_end_request_err);
2384 * __blk_end_request - Helper function for drivers to complete the request.
2385 * @rq: the request being processed
2386 * @error: %0 for success, < %0 for error
2387 * @nr_bytes: number of bytes to complete
2390 * Must be called with queue lock held unlike blk_end_request().
2393 * %false - we are done with this request
2394 * %true - still buffers pending for this request
2396 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2398 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2400 EXPORT_SYMBOL(__blk_end_request);
2403 * __blk_end_request_all - Helper function for drives to finish the request.
2404 * @rq: the request to finish
2405 * @error: %0 for success, < %0 for error
2408 * Completely finish @rq. Must be called with queue lock held.
2410 void __blk_end_request_all(struct request *rq, int error)
2413 unsigned int bidi_bytes = 0;
2415 if (unlikely(blk_bidi_rq(rq)))
2416 bidi_bytes = blk_rq_bytes(rq->next_rq);
2418 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2421 EXPORT_SYMBOL(__blk_end_request_all);
2424 * __blk_end_request_cur - Helper function to finish the current request chunk.
2425 * @rq: the request to finish the current chunk for
2426 * @error: %0 for success, < %0 for error
2429 * Complete the current consecutively mapped chunk from @rq. Must
2430 * be called with queue lock held.
2433 * %false - we are done with this request
2434 * %true - still buffers pending for this request
2436 bool __blk_end_request_cur(struct request *rq, int error)
2438 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2440 EXPORT_SYMBOL(__blk_end_request_cur);
2443 * __blk_end_request_err - Finish a request till the next failure boundary.
2444 * @rq: the request to finish till the next failure boundary for
2445 * @error: must be negative errno
2448 * Complete @rq till the next failure boundary. Must be called
2449 * with queue lock held.
2452 * %false - we are done with this request
2453 * %true - still buffers pending for this request
2455 bool __blk_end_request_err(struct request *rq, int error)
2457 WARN_ON(error >= 0);
2458 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2460 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2462 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2465 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2466 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2468 if (bio_has_data(bio)) {
2469 rq->nr_phys_segments = bio_phys_segments(q, bio);
2470 rq->buffer = bio_data(bio);
2472 rq->__data_len = bio->bi_size;
2473 rq->bio = rq->biotail = bio;
2476 rq->rq_disk = bio->bi_bdev->bd_disk;
2479 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2481 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2482 * @rq: the request to be flushed
2485 * Flush all pages in @rq.
2487 void rq_flush_dcache_pages(struct request *rq)
2489 struct req_iterator iter;
2490 struct bio_vec *bvec;
2492 rq_for_each_segment(bvec, rq, iter)
2493 flush_dcache_page(bvec->bv_page);
2495 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2499 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2500 * @q : the queue of the device being checked
2503 * Check if underlying low-level drivers of a device are busy.
2504 * If the drivers want to export their busy state, they must set own
2505 * exporting function using blk_queue_lld_busy() first.
2507 * Basically, this function is used only by request stacking drivers
2508 * to stop dispatching requests to underlying devices when underlying
2509 * devices are busy. This behavior helps more I/O merging on the queue
2510 * of the request stacking driver and prevents I/O throughput regression
2511 * on burst I/O load.
2514 * 0 - Not busy (The request stacking driver should dispatch request)
2515 * 1 - Busy (The request stacking driver should stop dispatching request)
2517 int blk_lld_busy(struct request_queue *q)
2520 return q->lld_busy_fn(q);
2524 EXPORT_SYMBOL_GPL(blk_lld_busy);
2527 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2528 * @rq: the clone request to be cleaned up
2531 * Free all bios in @rq for a cloned request.
2533 void blk_rq_unprep_clone(struct request *rq)
2537 while ((bio = rq->bio) != NULL) {
2538 rq->bio = bio->bi_next;
2543 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2546 * Copy attributes of the original request to the clone request.
2547 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2549 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2551 dst->cpu = src->cpu;
2552 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2553 dst->cmd_type = src->cmd_type;
2554 dst->__sector = blk_rq_pos(src);
2555 dst->__data_len = blk_rq_bytes(src);
2556 dst->nr_phys_segments = src->nr_phys_segments;
2557 dst->ioprio = src->ioprio;
2558 dst->extra_len = src->extra_len;
2562 * blk_rq_prep_clone - Helper function to setup clone request
2563 * @rq: the request to be setup
2564 * @rq_src: original request to be cloned
2565 * @bs: bio_set that bios for clone are allocated from
2566 * @gfp_mask: memory allocation mask for bio
2567 * @bio_ctr: setup function to be called for each clone bio.
2568 * Returns %0 for success, non %0 for failure.
2569 * @data: private data to be passed to @bio_ctr
2572 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2573 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2574 * are not copied, and copying such parts is the caller's responsibility.
2575 * Also, pages which the original bios are pointing to are not copied
2576 * and the cloned bios just point same pages.
2577 * So cloned bios must be completed before original bios, which means
2578 * the caller must complete @rq before @rq_src.
2580 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2581 struct bio_set *bs, gfp_t gfp_mask,
2582 int (*bio_ctr)(struct bio *, struct bio *, void *),
2585 struct bio *bio, *bio_src;
2590 blk_rq_init(NULL, rq);
2592 __rq_for_each_bio(bio_src, rq_src) {
2593 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2597 __bio_clone(bio, bio_src);
2599 if (bio_integrity(bio_src) &&
2600 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2603 if (bio_ctr && bio_ctr(bio, bio_src, data))
2607 rq->biotail->bi_next = bio;
2610 rq->bio = rq->biotail = bio;
2613 __blk_rq_prep_clone(rq, rq_src);
2620 blk_rq_unprep_clone(rq);
2624 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2626 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2628 return queue_work(kblockd_workqueue, work);
2630 EXPORT_SYMBOL(kblockd_schedule_work);
2632 int kblockd_schedule_delayed_work(struct request_queue *q,
2633 struct delayed_work *dwork, unsigned long delay)
2635 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2637 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2639 #define PLUG_MAGIC 0x91827364
2641 void blk_start_plug(struct blk_plug *plug)
2643 struct task_struct *tsk = current;
2645 plug->magic = PLUG_MAGIC;
2646 INIT_LIST_HEAD(&plug->list);
2647 plug->should_sort = 0;
2650 * If this is a nested plug, don't actually assign it. It will be
2651 * flushed on its own.
2655 * Store ordering should not be needed here, since a potential
2656 * preempt will imply a full memory barrier
2661 EXPORT_SYMBOL(blk_start_plug);
2663 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2665 struct request *rqa = container_of(a, struct request, queuelist);
2666 struct request *rqb = container_of(b, struct request, queuelist);
2668 return !(rqa->q == rqb->q);
2671 static void flush_plug_list(struct blk_plug *plug)
2673 struct request_queue *q;
2674 unsigned long flags;
2677 BUG_ON(plug->magic != PLUG_MAGIC);
2679 if (list_empty(&plug->list))
2682 if (plug->should_sort)
2683 list_sort(NULL, &plug->list, plug_rq_cmp);
2686 local_irq_save(flags);
2687 while (!list_empty(&plug->list)) {
2688 rq = list_entry_rq(plug->list.next);
2689 list_del_init(&rq->queuelist);
2690 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2694 __blk_run_queue(q, false);
2695 spin_unlock(q->queue_lock);
2698 spin_lock(q->queue_lock);
2700 rq->cmd_flags &= ~REQ_ON_PLUG;
2703 * rq is already accounted, so use raw insert
2705 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2706 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2708 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2712 __blk_run_queue(q, false);
2713 spin_unlock(q->queue_lock);
2716 BUG_ON(!list_empty(&plug->list));
2717 local_irq_restore(flags);
2720 static void __blk_finish_plug(struct task_struct *tsk, struct blk_plug *plug)
2722 flush_plug_list(plug);
2724 if (plug == tsk->plug)
2728 void blk_finish_plug(struct blk_plug *plug)
2731 __blk_finish_plug(current, plug);
2733 EXPORT_SYMBOL(blk_finish_plug);
2735 void __blk_flush_plug(struct task_struct *tsk, struct blk_plug *plug)
2737 __blk_finish_plug(tsk, plug);
2740 EXPORT_SYMBOL(__blk_flush_plug);
2742 int __init blk_dev_init(void)
2744 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2745 sizeof(((struct request *)0)->cmd_flags));
2747 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2748 kblockd_workqueue = alloc_workqueue("kblockd",
2749 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2750 if (!kblockd_workqueue)
2751 panic("Failed to create kblockd\n");
2753 request_cachep = kmem_cache_create("blkdev_requests",
2754 sizeof(struct request), 0, SLAB_PANIC, NULL);
2756 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2757 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);