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);
42 * For the allocated request tables
44 static struct kmem_cache *request_cachep;
47 * For queue allocation
49 struct kmem_cache *blk_requestq_cachep;
52 * Controlling structure to kblockd
54 static struct workqueue_struct *kblockd_workqueue;
56 static void drive_stat_acct(struct request *rq, int new_io)
58 struct hd_struct *part;
59 int rw = rq_data_dir(rq);
62 if (!blk_do_io_stat(rq))
65 cpu = part_stat_lock();
69 part_stat_inc(cpu, part, merges[rw]);
71 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
72 if (!hd_struct_try_get(part)) {
74 * The partition is already being removed,
75 * the request will be accounted on the disk only
77 * We take a reference on disk->part0 although that
78 * partition will never be deleted, so we can treat
79 * it as any other partition.
81 part = &rq->rq_disk->part0;
84 part_round_stats(cpu, part);
85 part_inc_in_flight(part, rw);
92 void blk_queue_congestion_threshold(struct request_queue *q)
96 nr = q->nr_requests - (q->nr_requests / 8) + 1;
97 if (nr > q->nr_requests)
99 q->nr_congestion_on = nr;
101 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 q->nr_congestion_off = nr;
108 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * Locates the passed device's request queue and returns the address of its
114 * Will return NULL if the request queue cannot be located.
116 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
118 struct backing_dev_info *ret = NULL;
119 struct request_queue *q = bdev_get_queue(bdev);
122 ret = &q->backing_dev_info;
125 EXPORT_SYMBOL(blk_get_backing_dev_info);
127 void blk_rq_init(struct request_queue *q, struct request *rq)
129 memset(rq, 0, sizeof(*rq));
131 INIT_LIST_HEAD(&rq->queuelist);
132 INIT_LIST_HEAD(&rq->timeout_list);
135 rq->__sector = (sector_t) -1;
136 INIT_HLIST_NODE(&rq->hash);
137 RB_CLEAR_NODE(&rq->rb_node);
139 rq->cmd_len = BLK_MAX_CDB;
142 rq->start_time = jiffies;
143 set_start_time_ns(rq);
146 EXPORT_SYMBOL(blk_rq_init);
148 static void req_bio_endio(struct request *rq, struct bio *bio,
149 unsigned int nbytes, int error)
152 clear_bit(BIO_UPTODATE, &bio->bi_flags);
153 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
156 if (unlikely(nbytes > bio->bi_size)) {
157 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
158 __func__, nbytes, bio->bi_size);
159 nbytes = bio->bi_size;
162 if (unlikely(rq->cmd_flags & REQ_QUIET))
163 set_bit(BIO_QUIET, &bio->bi_flags);
165 bio->bi_size -= nbytes;
166 bio->bi_sector += (nbytes >> 9);
168 if (bio_integrity(bio))
169 bio_integrity_advance(bio, nbytes);
171 /* don't actually finish bio if it's part of flush sequence */
172 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
173 bio_endio(bio, error);
176 void blk_dump_rq_flags(struct request *rq, char *msg)
180 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
181 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
185 (unsigned long long)blk_rq_pos(rq),
186 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
187 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
188 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
190 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
191 printk(KERN_INFO " cdb: ");
192 for (bit = 0; bit < BLK_MAX_CDB; bit++)
193 printk("%02x ", rq->cmd[bit]);
197 EXPORT_SYMBOL(blk_dump_rq_flags);
199 static void blk_delay_work(struct work_struct *work)
201 struct request_queue *q;
203 q = container_of(work, struct request_queue, delay_work.work);
204 spin_lock_irq(q->queue_lock);
206 spin_unlock_irq(q->queue_lock);
210 * blk_delay_queue - restart queueing after defined interval
211 * @q: The &struct request_queue in question
212 * @msecs: Delay in msecs
215 * Sometimes queueing needs to be postponed for a little while, to allow
216 * resources to come back. This function will make sure that queueing is
217 * restarted around the specified time.
219 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
221 queue_delayed_work(kblockd_workqueue, &q->delay_work,
222 msecs_to_jiffies(msecs));
224 EXPORT_SYMBOL(blk_delay_queue);
227 * blk_start_queue - restart a previously stopped queue
228 * @q: The &struct request_queue in question
231 * blk_start_queue() will clear the stop flag on the queue, and call
232 * the request_fn for the queue if it was in a stopped state when
233 * entered. Also see blk_stop_queue(). Queue lock must be held.
235 void blk_start_queue(struct request_queue *q)
237 WARN_ON(!irqs_disabled());
239 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
242 EXPORT_SYMBOL(blk_start_queue);
245 * blk_stop_queue - stop a queue
246 * @q: The &struct request_queue in question
249 * The Linux block layer assumes that a block driver will consume all
250 * entries on the request queue when the request_fn strategy is called.
251 * Often this will not happen, because of hardware limitations (queue
252 * depth settings). If a device driver gets a 'queue full' response,
253 * or if it simply chooses not to queue more I/O at one point, it can
254 * call this function to prevent the request_fn from being called until
255 * the driver has signalled it's ready to go again. This happens by calling
256 * blk_start_queue() to restart queue operations. Queue lock must be held.
258 void blk_stop_queue(struct request_queue *q)
260 __cancel_delayed_work(&q->delay_work);
261 queue_flag_set(QUEUE_FLAG_STOPPED, q);
263 EXPORT_SYMBOL(blk_stop_queue);
266 * blk_sync_queue - cancel any pending callbacks on a queue
270 * The block layer may perform asynchronous callback activity
271 * on a queue, such as calling the unplug function after a timeout.
272 * A block device may call blk_sync_queue to ensure that any
273 * such activity is cancelled, thus allowing it to release resources
274 * that the callbacks might use. The caller must already have made sure
275 * that its ->make_request_fn will not re-add plugging prior to calling
278 * This function does not cancel any asynchronous activity arising
279 * out of elevator or throttling code. That would require elevaotor_exit()
280 * and blk_throtl_exit() to be called with queue lock initialized.
283 void blk_sync_queue(struct request_queue *q)
285 del_timer_sync(&q->timeout);
286 cancel_delayed_work_sync(&q->delay_work);
288 EXPORT_SYMBOL(blk_sync_queue);
291 * __blk_run_queue - run a single device queue
292 * @q: The queue to run
295 * See @blk_run_queue. This variant must be called with the queue lock
296 * held and interrupts disabled.
298 void __blk_run_queue(struct request_queue *q)
300 if (unlikely(blk_queue_stopped(q)))
305 EXPORT_SYMBOL(__blk_run_queue);
308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 void blk_run_queue_async(struct request_queue *q)
317 if (likely(!blk_queue_stopped(q))) {
318 __cancel_delayed_work(&q->delay_work);
319 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
322 EXPORT_SYMBOL(blk_run_queue_async);
325 * blk_run_queue - run a single device queue
326 * @q: The queue to run
329 * Invoke request handling on this queue, if it has pending work to do.
330 * May be used to restart queueing when a request has completed.
332 void blk_run_queue(struct request_queue *q)
336 spin_lock_irqsave(q->queue_lock, flags);
338 spin_unlock_irqrestore(q->queue_lock, flags);
340 EXPORT_SYMBOL(blk_run_queue);
342 void blk_put_queue(struct request_queue *q)
344 kobject_put(&q->kobj);
346 EXPORT_SYMBOL(blk_put_queue);
349 * Note: If a driver supplied the queue lock, it is disconnected
350 * by this function. The actual state of the lock doesn't matter
351 * here as the request_queue isn't accessible after this point
352 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
354 void blk_cleanup_queue(struct request_queue *q)
357 * We know we have process context here, so we can be a little
358 * cautious and ensure that pending block actions on this device
359 * are done before moving on. Going into this function, we should
360 * not have processes doing IO to this device.
364 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
365 mutex_lock(&q->sysfs_lock);
366 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
367 mutex_unlock(&q->sysfs_lock);
369 if (q->queue_lock != &q->__queue_lock)
370 q->queue_lock = &q->__queue_lock;
374 EXPORT_SYMBOL(blk_cleanup_queue);
376 static int blk_init_free_list(struct request_queue *q)
378 struct request_list *rl = &q->rq;
380 if (unlikely(rl->rq_pool))
383 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
384 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
386 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
387 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
389 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
390 mempool_free_slab, request_cachep, q->node);
398 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
400 return blk_alloc_queue_node(gfp_mask, -1);
402 EXPORT_SYMBOL(blk_alloc_queue);
404 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
406 struct request_queue *q;
409 q = kmem_cache_alloc_node(blk_requestq_cachep,
410 gfp_mask | __GFP_ZERO, node_id);
414 q->backing_dev_info.ra_pages =
415 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
416 q->backing_dev_info.state = 0;
417 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
418 q->backing_dev_info.name = "block";
420 err = bdi_init(&q->backing_dev_info);
422 kmem_cache_free(blk_requestq_cachep, q);
426 if (blk_throtl_init(q)) {
427 kmem_cache_free(blk_requestq_cachep, q);
431 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
432 laptop_mode_timer_fn, (unsigned long) q);
433 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
434 INIT_LIST_HEAD(&q->timeout_list);
435 INIT_LIST_HEAD(&q->flush_queue[0]);
436 INIT_LIST_HEAD(&q->flush_queue[1]);
437 INIT_LIST_HEAD(&q->flush_data_in_flight);
438 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
440 kobject_init(&q->kobj, &blk_queue_ktype);
442 mutex_init(&q->sysfs_lock);
443 spin_lock_init(&q->__queue_lock);
446 * By default initialize queue_lock to internal lock and driver can
447 * override it later if need be.
449 q->queue_lock = &q->__queue_lock;
453 EXPORT_SYMBOL(blk_alloc_queue_node);
456 * blk_init_queue - prepare a request queue for use with a block device
457 * @rfn: The function to be called to process requests that have been
458 * placed on the queue.
459 * @lock: Request queue spin lock
462 * If a block device wishes to use the standard request handling procedures,
463 * which sorts requests and coalesces adjacent requests, then it must
464 * call blk_init_queue(). The function @rfn will be called when there
465 * are requests on the queue that need to be processed. If the device
466 * supports plugging, then @rfn may not be called immediately when requests
467 * are available on the queue, but may be called at some time later instead.
468 * Plugged queues are generally unplugged when a buffer belonging to one
469 * of the requests on the queue is needed, or due to memory pressure.
471 * @rfn is not required, or even expected, to remove all requests off the
472 * queue, but only as many as it can handle at a time. If it does leave
473 * requests on the queue, it is responsible for arranging that the requests
474 * get dealt with eventually.
476 * The queue spin lock must be held while manipulating the requests on the
477 * request queue; this lock will be taken also from interrupt context, so irq
478 * disabling is needed for it.
480 * Function returns a pointer to the initialized request queue, or %NULL if
484 * blk_init_queue() must be paired with a blk_cleanup_queue() call
485 * when the block device is deactivated (such as at module unload).
488 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
490 return blk_init_queue_node(rfn, lock, -1);
492 EXPORT_SYMBOL(blk_init_queue);
494 struct request_queue *
495 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
497 struct request_queue *uninit_q, *q;
499 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
503 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
505 blk_cleanup_queue(uninit_q);
509 EXPORT_SYMBOL(blk_init_queue_node);
511 struct request_queue *
512 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
515 return blk_init_allocated_queue_node(q, rfn, lock, -1);
517 EXPORT_SYMBOL(blk_init_allocated_queue);
519 struct request_queue *
520 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
521 spinlock_t *lock, int node_id)
527 if (blk_init_free_list(q))
531 q->prep_rq_fn = NULL;
532 q->unprep_rq_fn = NULL;
533 q->queue_flags = QUEUE_FLAG_DEFAULT;
535 /* Override internal queue lock with supplied lock pointer */
537 q->queue_lock = lock;
540 * This also sets hw/phys segments, boundary and size
542 blk_queue_make_request(q, blk_queue_bio);
544 q->sg_reserved_size = INT_MAX;
549 if (!elevator_init(q, NULL)) {
550 blk_queue_congestion_threshold(q);
556 EXPORT_SYMBOL(blk_init_allocated_queue_node);
558 int blk_get_queue(struct request_queue *q)
560 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
561 kobject_get(&q->kobj);
567 EXPORT_SYMBOL(blk_get_queue);
569 static inline void blk_free_request(struct request_queue *q, struct request *rq)
571 if (rq->cmd_flags & REQ_ELVPRIV)
572 elv_put_request(q, rq);
573 mempool_free(rq, q->rq.rq_pool);
576 static struct request *
577 blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
579 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
586 rq->cmd_flags = flags | REQ_ALLOCED;
588 if ((flags & REQ_ELVPRIV) &&
589 unlikely(elv_set_request(q, rq, gfp_mask))) {
590 mempool_free(rq, q->rq.rq_pool);
598 * ioc_batching returns true if the ioc is a valid batching request and
599 * should be given priority access to a request.
601 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
607 * Make sure the process is able to allocate at least 1 request
608 * even if the batch times out, otherwise we could theoretically
611 return ioc->nr_batch_requests == q->nr_batching ||
612 (ioc->nr_batch_requests > 0
613 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
617 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
618 * will cause the process to be a "batcher" on all queues in the system. This
619 * is the behaviour we want though - once it gets a wakeup it should be given
622 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
624 if (!ioc || ioc_batching(q, ioc))
627 ioc->nr_batch_requests = q->nr_batching;
628 ioc->last_waited = jiffies;
631 static void __freed_request(struct request_queue *q, int sync)
633 struct request_list *rl = &q->rq;
635 if (rl->count[sync] < queue_congestion_off_threshold(q))
636 blk_clear_queue_congested(q, sync);
638 if (rl->count[sync] + 1 <= q->nr_requests) {
639 if (waitqueue_active(&rl->wait[sync]))
640 wake_up(&rl->wait[sync]);
642 blk_clear_queue_full(q, sync);
647 * A request has just been released. Account for it, update the full and
648 * congestion status, wake up any waiters. Called under q->queue_lock.
650 static void freed_request(struct request_queue *q, unsigned int flags)
652 struct request_list *rl = &q->rq;
653 int sync = rw_is_sync(flags);
656 if (flags & REQ_ELVPRIV)
659 __freed_request(q, sync);
661 if (unlikely(rl->starved[sync ^ 1]))
662 __freed_request(q, sync ^ 1);
666 * Determine if elevator data should be initialized when allocating the
667 * request associated with @bio.
669 static bool blk_rq_should_init_elevator(struct bio *bio)
675 * Flush requests do not use the elevator so skip initialization.
676 * This allows a request to share the flush and elevator data.
678 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
685 * Get a free request, queue_lock must be held.
686 * Returns NULL on failure, with queue_lock held.
687 * Returns !NULL on success, with queue_lock *not held*.
689 static struct request *get_request(struct request_queue *q, int rw_flags,
690 struct bio *bio, gfp_t gfp_mask)
692 struct request *rq = NULL;
693 struct request_list *rl = &q->rq;
694 struct io_context *ioc = NULL;
695 const bool is_sync = rw_is_sync(rw_flags) != 0;
698 may_queue = elv_may_queue(q, rw_flags);
699 if (may_queue == ELV_MQUEUE_NO)
702 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
703 if (rl->count[is_sync]+1 >= q->nr_requests) {
704 ioc = current_io_context(GFP_ATOMIC, q->node);
706 * The queue will fill after this allocation, so set
707 * it as full, and mark this process as "batching".
708 * This process will be allowed to complete a batch of
709 * requests, others will be blocked.
711 if (!blk_queue_full(q, is_sync)) {
712 ioc_set_batching(q, ioc);
713 blk_set_queue_full(q, is_sync);
715 if (may_queue != ELV_MQUEUE_MUST
716 && !ioc_batching(q, ioc)) {
718 * The queue is full and the allocating
719 * process is not a "batcher", and not
720 * exempted by the IO scheduler
726 blk_set_queue_congested(q, is_sync);
730 * Only allow batching queuers to allocate up to 50% over the defined
731 * limit of requests, otherwise we could have thousands of requests
732 * allocated with any setting of ->nr_requests
734 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
737 rl->count[is_sync]++;
738 rl->starved[is_sync] = 0;
740 if (blk_rq_should_init_elevator(bio) &&
741 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
742 rw_flags |= REQ_ELVPRIV;
746 if (blk_queue_io_stat(q))
747 rw_flags |= REQ_IO_STAT;
748 spin_unlock_irq(q->queue_lock);
750 rq = blk_alloc_request(q, rw_flags, gfp_mask);
753 * Allocation failed presumably due to memory. Undo anything
754 * we might have messed up.
756 * Allocating task should really be put onto the front of the
757 * wait queue, but this is pretty rare.
759 spin_lock_irq(q->queue_lock);
760 freed_request(q, rw_flags);
763 * in the very unlikely event that allocation failed and no
764 * requests for this direction was pending, mark us starved
765 * so that freeing of a request in the other direction will
766 * notice us. another possible fix would be to split the
767 * rq mempool into READ and WRITE
770 if (unlikely(rl->count[is_sync] == 0))
771 rl->starved[is_sync] = 1;
777 * ioc may be NULL here, and ioc_batching will be false. That's
778 * OK, if the queue is under the request limit then requests need
779 * not count toward the nr_batch_requests limit. There will always
780 * be some limit enforced by BLK_BATCH_TIME.
782 if (ioc_batching(q, ioc))
783 ioc->nr_batch_requests--;
785 trace_block_getrq(q, bio, rw_flags & 1);
791 * No available requests for this queue, wait for some requests to become
794 * Called with q->queue_lock held, and returns with it unlocked.
796 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
799 const bool is_sync = rw_is_sync(rw_flags) != 0;
802 rq = get_request(q, rw_flags, bio, GFP_NOIO);
805 struct io_context *ioc;
806 struct request_list *rl = &q->rq;
808 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
809 TASK_UNINTERRUPTIBLE);
811 trace_block_sleeprq(q, bio, rw_flags & 1);
813 spin_unlock_irq(q->queue_lock);
817 * After sleeping, we become a "batching" process and
818 * will be able to allocate at least one request, and
819 * up to a big batch of them for a small period time.
820 * See ioc_batching, ioc_set_batching
822 ioc = current_io_context(GFP_NOIO, q->node);
823 ioc_set_batching(q, ioc);
825 spin_lock_irq(q->queue_lock);
826 finish_wait(&rl->wait[is_sync], &wait);
828 rq = get_request(q, rw_flags, bio, GFP_NOIO);
834 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
838 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
841 BUG_ON(rw != READ && rw != WRITE);
843 spin_lock_irq(q->queue_lock);
844 if (gfp_mask & __GFP_WAIT) {
845 rq = get_request_wait(q, rw, NULL);
847 rq = get_request(q, rw, NULL, gfp_mask);
849 spin_unlock_irq(q->queue_lock);
851 /* q->queue_lock is unlocked at this point */
855 EXPORT_SYMBOL(blk_get_request);
858 * blk_make_request - given a bio, allocate a corresponding struct request.
859 * @q: target request queue
860 * @bio: The bio describing the memory mappings that will be submitted for IO.
861 * It may be a chained-bio properly constructed by block/bio layer.
862 * @gfp_mask: gfp flags to be used for memory allocation
864 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
865 * type commands. Where the struct request needs to be farther initialized by
866 * the caller. It is passed a &struct bio, which describes the memory info of
869 * The caller of blk_make_request must make sure that bi_io_vec
870 * are set to describe the memory buffers. That bio_data_dir() will return
871 * the needed direction of the request. (And all bio's in the passed bio-chain
872 * are properly set accordingly)
874 * If called under none-sleepable conditions, mapped bio buffers must not
875 * need bouncing, by calling the appropriate masked or flagged allocator,
876 * suitable for the target device. Otherwise the call to blk_queue_bounce will
879 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
880 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
881 * anything but the first bio in the chain. Otherwise you risk waiting for IO
882 * completion of a bio that hasn't been submitted yet, thus resulting in a
883 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
884 * of bio_alloc(), as that avoids the mempool deadlock.
885 * If possible a big IO should be split into smaller parts when allocation
886 * fails. Partial allocation should not be an error, or you risk a live-lock.
888 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
891 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
894 return ERR_PTR(-ENOMEM);
897 struct bio *bounce_bio = bio;
900 blk_queue_bounce(q, &bounce_bio);
901 ret = blk_rq_append_bio(q, rq, bounce_bio);
910 EXPORT_SYMBOL(blk_make_request);
913 * blk_requeue_request - put a request back on queue
914 * @q: request queue where request should be inserted
915 * @rq: request to be inserted
918 * Drivers often keep queueing requests until the hardware cannot accept
919 * more, when that condition happens we need to put the request back
920 * on the queue. Must be called with queue lock held.
922 void blk_requeue_request(struct request_queue *q, struct request *rq)
924 blk_delete_timer(rq);
925 blk_clear_rq_complete(rq);
926 trace_block_rq_requeue(q, rq);
928 if (blk_rq_tagged(rq))
929 blk_queue_end_tag(q, rq);
931 BUG_ON(blk_queued_rq(rq));
933 elv_requeue_request(q, rq);
935 EXPORT_SYMBOL(blk_requeue_request);
937 static void add_acct_request(struct request_queue *q, struct request *rq,
940 drive_stat_acct(rq, 1);
941 __elv_add_request(q, rq, where);
945 * blk_insert_request - insert a special request into a request queue
946 * @q: request queue where request should be inserted
947 * @rq: request to be inserted
948 * @at_head: insert request at head or tail of queue
949 * @data: private data
952 * Many block devices need to execute commands asynchronously, so they don't
953 * block the whole kernel from preemption during request execution. This is
954 * accomplished normally by inserting aritficial requests tagged as
955 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
956 * be scheduled for actual execution by the request queue.
958 * We have the option of inserting the head or the tail of the queue.
959 * Typically we use the tail for new ioctls and so forth. We use the head
960 * of the queue for things like a QUEUE_FULL message from a device, or a
961 * host that is unable to accept a particular command.
963 void blk_insert_request(struct request_queue *q, struct request *rq,
964 int at_head, void *data)
966 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
970 * tell I/O scheduler that this isn't a regular read/write (ie it
971 * must not attempt merges on this) and that it acts as a soft
974 rq->cmd_type = REQ_TYPE_SPECIAL;
978 spin_lock_irqsave(q->queue_lock, flags);
981 * If command is tagged, release the tag
983 if (blk_rq_tagged(rq))
984 blk_queue_end_tag(q, rq);
986 add_acct_request(q, rq, where);
988 spin_unlock_irqrestore(q->queue_lock, flags);
990 EXPORT_SYMBOL(blk_insert_request);
992 static void part_round_stats_single(int cpu, struct hd_struct *part,
995 if (now == part->stamp)
998 if (part_in_flight(part)) {
999 __part_stat_add(cpu, part, time_in_queue,
1000 part_in_flight(part) * (now - part->stamp));
1001 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1007 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1008 * @cpu: cpu number for stats access
1009 * @part: target partition
1011 * The average IO queue length and utilisation statistics are maintained
1012 * by observing the current state of the queue length and the amount of
1013 * time it has been in this state for.
1015 * Normally, that accounting is done on IO completion, but that can result
1016 * in more than a second's worth of IO being accounted for within any one
1017 * second, leading to >100% utilisation. To deal with that, we call this
1018 * function to do a round-off before returning the results when reading
1019 * /proc/diskstats. This accounts immediately for all queue usage up to
1020 * the current jiffies and restarts the counters again.
1022 void part_round_stats(int cpu, struct hd_struct *part)
1024 unsigned long now = jiffies;
1027 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1028 part_round_stats_single(cpu, part, now);
1030 EXPORT_SYMBOL_GPL(part_round_stats);
1033 * queue lock must be held
1035 void __blk_put_request(struct request_queue *q, struct request *req)
1039 if (unlikely(--req->ref_count))
1042 elv_completed_request(q, req);
1044 /* this is a bio leak */
1045 WARN_ON(req->bio != NULL);
1048 * Request may not have originated from ll_rw_blk. if not,
1049 * it didn't come out of our reserved rq pools
1051 if (req->cmd_flags & REQ_ALLOCED) {
1052 unsigned int flags = req->cmd_flags;
1054 BUG_ON(!list_empty(&req->queuelist));
1055 BUG_ON(!hlist_unhashed(&req->hash));
1057 blk_free_request(q, req);
1058 freed_request(q, flags);
1061 EXPORT_SYMBOL_GPL(__blk_put_request);
1063 void blk_put_request(struct request *req)
1065 unsigned long flags;
1066 struct request_queue *q = req->q;
1068 spin_lock_irqsave(q->queue_lock, flags);
1069 __blk_put_request(q, req);
1070 spin_unlock_irqrestore(q->queue_lock, flags);
1072 EXPORT_SYMBOL(blk_put_request);
1075 * blk_add_request_payload - add a payload to a request
1076 * @rq: request to update
1077 * @page: page backing the payload
1078 * @len: length of the payload.
1080 * This allows to later add a payload to an already submitted request by
1081 * a block driver. The driver needs to take care of freeing the payload
1084 * Note that this is a quite horrible hack and nothing but handling of
1085 * discard requests should ever use it.
1087 void blk_add_request_payload(struct request *rq, struct page *page,
1090 struct bio *bio = rq->bio;
1092 bio->bi_io_vec->bv_page = page;
1093 bio->bi_io_vec->bv_offset = 0;
1094 bio->bi_io_vec->bv_len = len;
1098 bio->bi_phys_segments = 1;
1100 rq->__data_len = rq->resid_len = len;
1101 rq->nr_phys_segments = 1;
1102 rq->buffer = bio_data(bio);
1104 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1106 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1109 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1111 if (!ll_back_merge_fn(q, req, bio))
1114 trace_block_bio_backmerge(q, bio);
1116 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1117 blk_rq_set_mixed_merge(req);
1119 req->biotail->bi_next = bio;
1121 req->__data_len += bio->bi_size;
1122 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1124 drive_stat_acct(req, 0);
1125 elv_bio_merged(q, req, bio);
1129 static bool bio_attempt_front_merge(struct request_queue *q,
1130 struct request *req, struct bio *bio)
1132 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1134 if (!ll_front_merge_fn(q, req, bio))
1137 trace_block_bio_frontmerge(q, bio);
1139 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1140 blk_rq_set_mixed_merge(req);
1142 bio->bi_next = req->bio;
1146 * may not be valid. if the low level driver said
1147 * it didn't need a bounce buffer then it better
1148 * not touch req->buffer either...
1150 req->buffer = bio_data(bio);
1151 req->__sector = bio->bi_sector;
1152 req->__data_len += bio->bi_size;
1153 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1155 drive_stat_acct(req, 0);
1156 elv_bio_merged(q, req, bio);
1161 * Attempts to merge with the plugged list in the current process. Returns
1162 * true if merge was successful, otherwise false.
1164 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1165 struct bio *bio, unsigned int *request_count)
1167 struct blk_plug *plug;
1176 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1184 el_ret = elv_try_merge(rq, bio);
1185 if (el_ret == ELEVATOR_BACK_MERGE) {
1186 ret = bio_attempt_back_merge(q, rq, bio);
1189 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1190 ret = bio_attempt_front_merge(q, rq, bio);
1199 void init_request_from_bio(struct request *req, struct bio *bio)
1201 req->cpu = bio->bi_comp_cpu;
1202 req->cmd_type = REQ_TYPE_FS;
1204 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1205 if (bio->bi_rw & REQ_RAHEAD)
1206 req->cmd_flags |= REQ_FAILFAST_MASK;
1209 req->__sector = bio->bi_sector;
1210 req->ioprio = bio_prio(bio);
1211 blk_rq_bio_prep(req->q, req, bio);
1214 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1216 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1217 struct blk_plug *plug;
1218 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1219 struct request *req;
1220 unsigned int request_count = 0;
1223 * low level driver can indicate that it wants pages above a
1224 * certain limit bounced to low memory (ie for highmem, or even
1225 * ISA dma in theory)
1227 blk_queue_bounce(q, &bio);
1229 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1230 spin_lock_irq(q->queue_lock);
1231 where = ELEVATOR_INSERT_FLUSH;
1236 * Check if we can merge with the plugged list before grabbing
1239 if (attempt_plug_merge(current, q, bio, &request_count))
1242 spin_lock_irq(q->queue_lock);
1244 el_ret = elv_merge(q, &req, bio);
1245 if (el_ret == ELEVATOR_BACK_MERGE) {
1246 if (bio_attempt_back_merge(q, req, bio)) {
1247 if (!attempt_back_merge(q, req))
1248 elv_merged_request(q, req, el_ret);
1251 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1252 if (bio_attempt_front_merge(q, req, bio)) {
1253 if (!attempt_front_merge(q, req))
1254 elv_merged_request(q, req, el_ret);
1261 * This sync check and mask will be re-done in init_request_from_bio(),
1262 * but we need to set it earlier to expose the sync flag to the
1263 * rq allocator and io schedulers.
1265 rw_flags = bio_data_dir(bio);
1267 rw_flags |= REQ_SYNC;
1270 * Grab a free request. This is might sleep but can not fail.
1271 * Returns with the queue unlocked.
1273 req = get_request_wait(q, rw_flags, bio);
1276 * After dropping the lock and possibly sleeping here, our request
1277 * may now be mergeable after it had proven unmergeable (above).
1278 * We don't worry about that case for efficiency. It won't happen
1279 * often, and the elevators are able to handle it.
1281 init_request_from_bio(req, bio);
1283 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1284 bio_flagged(bio, BIO_CPU_AFFINE))
1285 req->cpu = raw_smp_processor_id();
1287 plug = current->plug;
1290 * If this is the first request added after a plug, fire
1291 * of a plug trace. If others have been added before, check
1292 * if we have multiple devices in this plug. If so, make a
1293 * note to sort the list before dispatch.
1295 if (list_empty(&plug->list))
1296 trace_block_plug(q);
1297 else if (!plug->should_sort) {
1298 struct request *__rq;
1300 __rq = list_entry_rq(plug->list.prev);
1302 plug->should_sort = 1;
1304 if (request_count >= BLK_MAX_REQUEST_COUNT)
1305 blk_flush_plug_list(plug, false);
1306 list_add_tail(&req->queuelist, &plug->list);
1307 drive_stat_acct(req, 1);
1309 spin_lock_irq(q->queue_lock);
1310 add_acct_request(q, req, where);
1313 spin_unlock_irq(q->queue_lock);
1316 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1319 * If bio->bi_dev is a partition, remap the location
1321 static inline void blk_partition_remap(struct bio *bio)
1323 struct block_device *bdev = bio->bi_bdev;
1325 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1326 struct hd_struct *p = bdev->bd_part;
1328 bio->bi_sector += p->start_sect;
1329 bio->bi_bdev = bdev->bd_contains;
1331 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1333 bio->bi_sector - p->start_sect);
1337 static void handle_bad_sector(struct bio *bio)
1339 char b[BDEVNAME_SIZE];
1341 printk(KERN_INFO "attempt to access beyond end of device\n");
1342 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1343 bdevname(bio->bi_bdev, b),
1345 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1346 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1348 set_bit(BIO_EOF, &bio->bi_flags);
1351 #ifdef CONFIG_FAIL_MAKE_REQUEST
1353 static DECLARE_FAULT_ATTR(fail_make_request);
1355 static int __init setup_fail_make_request(char *str)
1357 return setup_fault_attr(&fail_make_request, str);
1359 __setup("fail_make_request=", setup_fail_make_request);
1361 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1363 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1366 static int __init fail_make_request_debugfs(void)
1368 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1369 NULL, &fail_make_request);
1371 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1374 late_initcall(fail_make_request_debugfs);
1376 #else /* CONFIG_FAIL_MAKE_REQUEST */
1378 static inline bool should_fail_request(struct hd_struct *part,
1384 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1387 * Check whether this bio extends beyond the end of the device.
1389 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1396 /* Test device or partition size, when known. */
1397 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1399 sector_t sector = bio->bi_sector;
1401 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1403 * This may well happen - the kernel calls bread()
1404 * without checking the size of the device, e.g., when
1405 * mounting a device.
1407 handle_bad_sector(bio);
1415 static noinline_for_stack bool
1416 generic_make_request_checks(struct bio *bio)
1418 struct request_queue *q;
1419 int nr_sectors = bio_sectors(bio);
1421 char b[BDEVNAME_SIZE];
1422 struct hd_struct *part;
1426 if (bio_check_eod(bio, nr_sectors))
1429 q = bdev_get_queue(bio->bi_bdev);
1432 "generic_make_request: Trying to access "
1433 "nonexistent block-device %s (%Lu)\n",
1434 bdevname(bio->bi_bdev, b),
1435 (long long) bio->bi_sector);
1439 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1440 nr_sectors > queue_max_hw_sectors(q))) {
1441 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1442 bdevname(bio->bi_bdev, b),
1444 queue_max_hw_sectors(q));
1448 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1451 part = bio->bi_bdev->bd_part;
1452 if (should_fail_request(part, bio->bi_size) ||
1453 should_fail_request(&part_to_disk(part)->part0,
1458 * If this device has partitions, remap block n
1459 * of partition p to block n+start(p) of the disk.
1461 blk_partition_remap(bio);
1463 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1466 if (bio_check_eod(bio, nr_sectors))
1470 * Filter flush bio's early so that make_request based
1471 * drivers without flush support don't have to worry
1474 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1475 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1482 if ((bio->bi_rw & REQ_DISCARD) &&
1483 (!blk_queue_discard(q) ||
1484 ((bio->bi_rw & REQ_SECURE) &&
1485 !blk_queue_secdiscard(q)))) {
1490 if (blk_throtl_bio(q, &bio))
1493 /* if bio = NULL, bio has been throttled and will be submitted later. */
1497 trace_block_bio_queue(q, bio);
1501 bio_endio(bio, err);
1506 * generic_make_request - hand a buffer to its device driver for I/O
1507 * @bio: The bio describing the location in memory and on the device.
1509 * generic_make_request() is used to make I/O requests of block
1510 * devices. It is passed a &struct bio, which describes the I/O that needs
1513 * generic_make_request() does not return any status. The
1514 * success/failure status of the request, along with notification of
1515 * completion, is delivered asynchronously through the bio->bi_end_io
1516 * function described (one day) else where.
1518 * The caller of generic_make_request must make sure that bi_io_vec
1519 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1520 * set to describe the device address, and the
1521 * bi_end_io and optionally bi_private are set to describe how
1522 * completion notification should be signaled.
1524 * generic_make_request and the drivers it calls may use bi_next if this
1525 * bio happens to be merged with someone else, and may resubmit the bio to
1526 * a lower device by calling into generic_make_request recursively, which
1527 * means the bio should NOT be touched after the call to ->make_request_fn.
1529 void generic_make_request(struct bio *bio)
1531 struct bio_list bio_list_on_stack;
1533 if (!generic_make_request_checks(bio))
1537 * We only want one ->make_request_fn to be active at a time, else
1538 * stack usage with stacked devices could be a problem. So use
1539 * current->bio_list to keep a list of requests submited by a
1540 * make_request_fn function. current->bio_list is also used as a
1541 * flag to say if generic_make_request is currently active in this
1542 * task or not. If it is NULL, then no make_request is active. If
1543 * it is non-NULL, then a make_request is active, and new requests
1544 * should be added at the tail
1546 if (current->bio_list) {
1547 bio_list_add(current->bio_list, bio);
1551 /* following loop may be a bit non-obvious, and so deserves some
1553 * Before entering the loop, bio->bi_next is NULL (as all callers
1554 * ensure that) so we have a list with a single bio.
1555 * We pretend that we have just taken it off a longer list, so
1556 * we assign bio_list to a pointer to the bio_list_on_stack,
1557 * thus initialising the bio_list of new bios to be
1558 * added. ->make_request() may indeed add some more bios
1559 * through a recursive call to generic_make_request. If it
1560 * did, we find a non-NULL value in bio_list and re-enter the loop
1561 * from the top. In this case we really did just take the bio
1562 * of the top of the list (no pretending) and so remove it from
1563 * bio_list, and call into ->make_request() again.
1565 BUG_ON(bio->bi_next);
1566 bio_list_init(&bio_list_on_stack);
1567 current->bio_list = &bio_list_on_stack;
1569 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1571 q->make_request_fn(q, bio);
1573 bio = bio_list_pop(current->bio_list);
1575 current->bio_list = NULL; /* deactivate */
1577 EXPORT_SYMBOL(generic_make_request);
1580 * submit_bio - submit a bio to the block device layer for I/O
1581 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1582 * @bio: The &struct bio which describes the I/O
1584 * submit_bio() is very similar in purpose to generic_make_request(), and
1585 * uses that function to do most of the work. Both are fairly rough
1586 * interfaces; @bio must be presetup and ready for I/O.
1589 void submit_bio(int rw, struct bio *bio)
1591 int count = bio_sectors(bio);
1596 * If it's a regular read/write or a barrier with data attached,
1597 * go through the normal accounting stuff before submission.
1599 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1601 count_vm_events(PGPGOUT, count);
1603 task_io_account_read(bio->bi_size);
1604 count_vm_events(PGPGIN, count);
1607 if (unlikely(block_dump)) {
1608 char b[BDEVNAME_SIZE];
1609 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1610 current->comm, task_pid_nr(current),
1611 (rw & WRITE) ? "WRITE" : "READ",
1612 (unsigned long long)bio->bi_sector,
1613 bdevname(bio->bi_bdev, b),
1618 generic_make_request(bio);
1620 EXPORT_SYMBOL(submit_bio);
1623 * blk_rq_check_limits - Helper function to check a request for the queue limit
1625 * @rq: the request being checked
1628 * @rq may have been made based on weaker limitations of upper-level queues
1629 * in request stacking drivers, and it may violate the limitation of @q.
1630 * Since the block layer and the underlying device driver trust @rq
1631 * after it is inserted to @q, it should be checked against @q before
1632 * the insertion using this generic function.
1634 * This function should also be useful for request stacking drivers
1635 * in some cases below, so export this function.
1636 * Request stacking drivers like request-based dm may change the queue
1637 * limits while requests are in the queue (e.g. dm's table swapping).
1638 * Such request stacking drivers should check those requests agaist
1639 * the new queue limits again when they dispatch those requests,
1640 * although such checkings are also done against the old queue limits
1641 * when submitting requests.
1643 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1645 if (rq->cmd_flags & REQ_DISCARD)
1648 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1649 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1650 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1655 * queue's settings related to segment counting like q->bounce_pfn
1656 * may differ from that of other stacking queues.
1657 * Recalculate it to check the request correctly on this queue's
1660 blk_recalc_rq_segments(rq);
1661 if (rq->nr_phys_segments > queue_max_segments(q)) {
1662 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1668 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1671 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1672 * @q: the queue to submit the request
1673 * @rq: the request being queued
1675 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1677 unsigned long flags;
1678 int where = ELEVATOR_INSERT_BACK;
1680 if (blk_rq_check_limits(q, rq))
1684 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1687 spin_lock_irqsave(q->queue_lock, flags);
1690 * Submitting request must be dequeued before calling this function
1691 * because it will be linked to another request_queue
1693 BUG_ON(blk_queued_rq(rq));
1695 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1696 where = ELEVATOR_INSERT_FLUSH;
1698 add_acct_request(q, rq, where);
1699 spin_unlock_irqrestore(q->queue_lock, flags);
1703 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1706 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1707 * @rq: request to examine
1710 * A request could be merge of IOs which require different failure
1711 * handling. This function determines the number of bytes which
1712 * can be failed from the beginning of the request without
1713 * crossing into area which need to be retried further.
1716 * The number of bytes to fail.
1719 * queue_lock must be held.
1721 unsigned int blk_rq_err_bytes(const struct request *rq)
1723 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1724 unsigned int bytes = 0;
1727 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1728 return blk_rq_bytes(rq);
1731 * Currently the only 'mixing' which can happen is between
1732 * different fastfail types. We can safely fail portions
1733 * which have all the failfast bits that the first one has -
1734 * the ones which are at least as eager to fail as the first
1737 for (bio = rq->bio; bio; bio = bio->bi_next) {
1738 if ((bio->bi_rw & ff) != ff)
1740 bytes += bio->bi_size;
1743 /* this could lead to infinite loop */
1744 BUG_ON(blk_rq_bytes(rq) && !bytes);
1747 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1749 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1751 if (blk_do_io_stat(req)) {
1752 const int rw = rq_data_dir(req);
1753 struct hd_struct *part;
1756 cpu = part_stat_lock();
1758 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1763 static void blk_account_io_done(struct request *req)
1766 * Account IO completion. flush_rq isn't accounted as a
1767 * normal IO on queueing nor completion. Accounting the
1768 * containing request is enough.
1770 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1771 unsigned long duration = jiffies - req->start_time;
1772 const int rw = rq_data_dir(req);
1773 struct hd_struct *part;
1776 cpu = part_stat_lock();
1779 part_stat_inc(cpu, part, ios[rw]);
1780 part_stat_add(cpu, part, ticks[rw], duration);
1781 part_round_stats(cpu, part);
1782 part_dec_in_flight(part, rw);
1784 hd_struct_put(part);
1790 * blk_peek_request - peek at the top of a request queue
1791 * @q: request queue to peek at
1794 * Return the request at the top of @q. The returned request
1795 * should be started using blk_start_request() before LLD starts
1799 * Pointer to the request at the top of @q if available. Null
1803 * queue_lock must be held.
1805 struct request *blk_peek_request(struct request_queue *q)
1810 while ((rq = __elv_next_request(q)) != NULL) {
1811 if (!(rq->cmd_flags & REQ_STARTED)) {
1813 * This is the first time the device driver
1814 * sees this request (possibly after
1815 * requeueing). Notify IO scheduler.
1817 if (rq->cmd_flags & REQ_SORTED)
1818 elv_activate_rq(q, rq);
1821 * just mark as started even if we don't start
1822 * it, a request that has been delayed should
1823 * not be passed by new incoming requests
1825 rq->cmd_flags |= REQ_STARTED;
1826 trace_block_rq_issue(q, rq);
1829 if (!q->boundary_rq || q->boundary_rq == rq) {
1830 q->end_sector = rq_end_sector(rq);
1831 q->boundary_rq = NULL;
1834 if (rq->cmd_flags & REQ_DONTPREP)
1837 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1839 * make sure space for the drain appears we
1840 * know we can do this because max_hw_segments
1841 * has been adjusted to be one fewer than the
1844 rq->nr_phys_segments++;
1850 ret = q->prep_rq_fn(q, rq);
1851 if (ret == BLKPREP_OK) {
1853 } else if (ret == BLKPREP_DEFER) {
1855 * the request may have been (partially) prepped.
1856 * we need to keep this request in the front to
1857 * avoid resource deadlock. REQ_STARTED will
1858 * prevent other fs requests from passing this one.
1860 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1861 !(rq->cmd_flags & REQ_DONTPREP)) {
1863 * remove the space for the drain we added
1864 * so that we don't add it again
1866 --rq->nr_phys_segments;
1871 } else if (ret == BLKPREP_KILL) {
1872 rq->cmd_flags |= REQ_QUIET;
1874 * Mark this request as started so we don't trigger
1875 * any debug logic in the end I/O path.
1877 blk_start_request(rq);
1878 __blk_end_request_all(rq, -EIO);
1880 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1887 EXPORT_SYMBOL(blk_peek_request);
1889 void blk_dequeue_request(struct request *rq)
1891 struct request_queue *q = rq->q;
1893 BUG_ON(list_empty(&rq->queuelist));
1894 BUG_ON(ELV_ON_HASH(rq));
1896 list_del_init(&rq->queuelist);
1899 * the time frame between a request being removed from the lists
1900 * and to it is freed is accounted as io that is in progress at
1903 if (blk_account_rq(rq)) {
1904 q->in_flight[rq_is_sync(rq)]++;
1905 set_io_start_time_ns(rq);
1910 * blk_start_request - start request processing on the driver
1911 * @req: request to dequeue
1914 * Dequeue @req and start timeout timer on it. This hands off the
1915 * request to the driver.
1917 * Block internal functions which don't want to start timer should
1918 * call blk_dequeue_request().
1921 * queue_lock must be held.
1923 void blk_start_request(struct request *req)
1925 blk_dequeue_request(req);
1928 * We are now handing the request to the hardware, initialize
1929 * resid_len to full count and add the timeout handler.
1931 req->resid_len = blk_rq_bytes(req);
1932 if (unlikely(blk_bidi_rq(req)))
1933 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1937 EXPORT_SYMBOL(blk_start_request);
1940 * blk_fetch_request - fetch a request from a request queue
1941 * @q: request queue to fetch a request from
1944 * Return the request at the top of @q. The request is started on
1945 * return and LLD can start processing it immediately.
1948 * Pointer to the request at the top of @q if available. Null
1952 * queue_lock must be held.
1954 struct request *blk_fetch_request(struct request_queue *q)
1958 rq = blk_peek_request(q);
1960 blk_start_request(rq);
1963 EXPORT_SYMBOL(blk_fetch_request);
1966 * blk_update_request - Special helper function for request stacking drivers
1967 * @req: the request being processed
1968 * @error: %0 for success, < %0 for error
1969 * @nr_bytes: number of bytes to complete @req
1972 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1973 * the request structure even if @req doesn't have leftover.
1974 * If @req has leftover, sets it up for the next range of segments.
1976 * This special helper function is only for request stacking drivers
1977 * (e.g. request-based dm) so that they can handle partial completion.
1978 * Actual device drivers should use blk_end_request instead.
1980 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1981 * %false return from this function.
1984 * %false - this request doesn't have any more data
1985 * %true - this request has more data
1987 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
1989 int total_bytes, bio_nbytes, next_idx = 0;
1995 trace_block_rq_complete(req->q, req);
1998 * For fs requests, rq is just carrier of independent bio's
1999 * and each partial completion should be handled separately.
2000 * Reset per-request error on each partial completion.
2002 * TODO: tj: This is too subtle. It would be better to let
2003 * low level drivers do what they see fit.
2005 if (req->cmd_type == REQ_TYPE_FS)
2008 if (error && req->cmd_type == REQ_TYPE_FS &&
2009 !(req->cmd_flags & REQ_QUIET)) {
2014 error_type = "recoverable transport";
2017 error_type = "critical target";
2020 error_type = "critical nexus";
2027 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2028 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2029 (unsigned long long)blk_rq_pos(req));
2032 blk_account_io_completion(req, nr_bytes);
2034 total_bytes = bio_nbytes = 0;
2035 while ((bio = req->bio) != NULL) {
2038 if (nr_bytes >= bio->bi_size) {
2039 req->bio = bio->bi_next;
2040 nbytes = bio->bi_size;
2041 req_bio_endio(req, bio, nbytes, error);
2045 int idx = bio->bi_idx + next_idx;
2047 if (unlikely(idx >= bio->bi_vcnt)) {
2048 blk_dump_rq_flags(req, "__end_that");
2049 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2050 __func__, idx, bio->bi_vcnt);
2054 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2055 BIO_BUG_ON(nbytes > bio->bi_size);
2058 * not a complete bvec done
2060 if (unlikely(nbytes > nr_bytes)) {
2061 bio_nbytes += nr_bytes;
2062 total_bytes += nr_bytes;
2067 * advance to the next vector
2070 bio_nbytes += nbytes;
2073 total_bytes += nbytes;
2079 * end more in this run, or just return 'not-done'
2081 if (unlikely(nr_bytes <= 0))
2091 * Reset counters so that the request stacking driver
2092 * can find how many bytes remain in the request
2095 req->__data_len = 0;
2100 * if the request wasn't completed, update state
2103 req_bio_endio(req, bio, bio_nbytes, error);
2104 bio->bi_idx += next_idx;
2105 bio_iovec(bio)->bv_offset += nr_bytes;
2106 bio_iovec(bio)->bv_len -= nr_bytes;
2109 req->__data_len -= total_bytes;
2110 req->buffer = bio_data(req->bio);
2112 /* update sector only for requests with clear definition of sector */
2113 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2114 req->__sector += total_bytes >> 9;
2116 /* mixed attributes always follow the first bio */
2117 if (req->cmd_flags & REQ_MIXED_MERGE) {
2118 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2119 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2123 * If total number of sectors is less than the first segment
2124 * size, something has gone terribly wrong.
2126 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2127 blk_dump_rq_flags(req, "request botched");
2128 req->__data_len = blk_rq_cur_bytes(req);
2131 /* recalculate the number of segments */
2132 blk_recalc_rq_segments(req);
2136 EXPORT_SYMBOL_GPL(blk_update_request);
2138 static bool blk_update_bidi_request(struct request *rq, int error,
2139 unsigned int nr_bytes,
2140 unsigned int bidi_bytes)
2142 if (blk_update_request(rq, error, nr_bytes))
2145 /* Bidi request must be completed as a whole */
2146 if (unlikely(blk_bidi_rq(rq)) &&
2147 blk_update_request(rq->next_rq, error, bidi_bytes))
2150 if (blk_queue_add_random(rq->q))
2151 add_disk_randomness(rq->rq_disk);
2157 * blk_unprep_request - unprepare a request
2160 * This function makes a request ready for complete resubmission (or
2161 * completion). It happens only after all error handling is complete,
2162 * so represents the appropriate moment to deallocate any resources
2163 * that were allocated to the request in the prep_rq_fn. The queue
2164 * lock is held when calling this.
2166 void blk_unprep_request(struct request *req)
2168 struct request_queue *q = req->q;
2170 req->cmd_flags &= ~REQ_DONTPREP;
2171 if (q->unprep_rq_fn)
2172 q->unprep_rq_fn(q, req);
2174 EXPORT_SYMBOL_GPL(blk_unprep_request);
2177 * queue lock must be held
2179 static void blk_finish_request(struct request *req, int error)
2181 if (blk_rq_tagged(req))
2182 blk_queue_end_tag(req->q, req);
2184 BUG_ON(blk_queued_rq(req));
2186 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2187 laptop_io_completion(&req->q->backing_dev_info);
2189 blk_delete_timer(req);
2191 if (req->cmd_flags & REQ_DONTPREP)
2192 blk_unprep_request(req);
2195 blk_account_io_done(req);
2198 req->end_io(req, error);
2200 if (blk_bidi_rq(req))
2201 __blk_put_request(req->next_rq->q, req->next_rq);
2203 __blk_put_request(req->q, req);
2208 * blk_end_bidi_request - Complete a bidi request
2209 * @rq: the request to complete
2210 * @error: %0 for success, < %0 for error
2211 * @nr_bytes: number of bytes to complete @rq
2212 * @bidi_bytes: number of bytes to complete @rq->next_rq
2215 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2216 * Drivers that supports bidi can safely call this member for any
2217 * type of request, bidi or uni. In the later case @bidi_bytes is
2221 * %false - we are done with this request
2222 * %true - still buffers pending for this request
2224 static bool blk_end_bidi_request(struct request *rq, int error,
2225 unsigned int nr_bytes, unsigned int bidi_bytes)
2227 struct request_queue *q = rq->q;
2228 unsigned long flags;
2230 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2233 spin_lock_irqsave(q->queue_lock, flags);
2234 blk_finish_request(rq, error);
2235 spin_unlock_irqrestore(q->queue_lock, flags);
2241 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2242 * @rq: the request to complete
2243 * @error: %0 for success, < %0 for error
2244 * @nr_bytes: number of bytes to complete @rq
2245 * @bidi_bytes: number of bytes to complete @rq->next_rq
2248 * Identical to blk_end_bidi_request() except that queue lock is
2249 * assumed to be locked on entry and remains so on return.
2252 * %false - we are done with this request
2253 * %true - still buffers pending for this request
2255 bool __blk_end_bidi_request(struct request *rq, int error,
2256 unsigned int nr_bytes, unsigned int bidi_bytes)
2258 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2261 blk_finish_request(rq, error);
2267 * blk_end_request - Helper function for drivers to complete the request.
2268 * @rq: the request being processed
2269 * @error: %0 for success, < %0 for error
2270 * @nr_bytes: number of bytes to complete
2273 * Ends I/O on a number of bytes attached to @rq.
2274 * If @rq has leftover, sets it up for the next range of segments.
2277 * %false - we are done with this request
2278 * %true - still buffers pending for this request
2280 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2282 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2284 EXPORT_SYMBOL(blk_end_request);
2287 * blk_end_request_all - Helper function for drives to finish the request.
2288 * @rq: the request to finish
2289 * @error: %0 for success, < %0 for error
2292 * Completely finish @rq.
2294 void blk_end_request_all(struct request *rq, int error)
2297 unsigned int bidi_bytes = 0;
2299 if (unlikely(blk_bidi_rq(rq)))
2300 bidi_bytes = blk_rq_bytes(rq->next_rq);
2302 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2305 EXPORT_SYMBOL(blk_end_request_all);
2308 * blk_end_request_cur - Helper function to finish the current request chunk.
2309 * @rq: the request to finish the current chunk for
2310 * @error: %0 for success, < %0 for error
2313 * Complete the current consecutively mapped chunk from @rq.
2316 * %false - we are done with this request
2317 * %true - still buffers pending for this request
2319 bool blk_end_request_cur(struct request *rq, int error)
2321 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2323 EXPORT_SYMBOL(blk_end_request_cur);
2326 * blk_end_request_err - Finish a request till the next failure boundary.
2327 * @rq: the request to finish till the next failure boundary for
2328 * @error: must be negative errno
2331 * Complete @rq till the next failure boundary.
2334 * %false - we are done with this request
2335 * %true - still buffers pending for this request
2337 bool blk_end_request_err(struct request *rq, int error)
2339 WARN_ON(error >= 0);
2340 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2342 EXPORT_SYMBOL_GPL(blk_end_request_err);
2345 * __blk_end_request - Helper function for drivers to complete the request.
2346 * @rq: the request being processed
2347 * @error: %0 for success, < %0 for error
2348 * @nr_bytes: number of bytes to complete
2351 * Must be called with queue lock held unlike blk_end_request().
2354 * %false - we are done with this request
2355 * %true - still buffers pending for this request
2357 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2359 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2361 EXPORT_SYMBOL(__blk_end_request);
2364 * __blk_end_request_all - Helper function for drives to finish the request.
2365 * @rq: the request to finish
2366 * @error: %0 for success, < %0 for error
2369 * Completely finish @rq. Must be called with queue lock held.
2371 void __blk_end_request_all(struct request *rq, int error)
2374 unsigned int bidi_bytes = 0;
2376 if (unlikely(blk_bidi_rq(rq)))
2377 bidi_bytes = blk_rq_bytes(rq->next_rq);
2379 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2382 EXPORT_SYMBOL(__blk_end_request_all);
2385 * __blk_end_request_cur - Helper function to finish the current request chunk.
2386 * @rq: the request to finish the current chunk for
2387 * @error: %0 for success, < %0 for error
2390 * Complete the current consecutively mapped chunk from @rq. Must
2391 * be called with queue lock held.
2394 * %false - we are done with this request
2395 * %true - still buffers pending for this request
2397 bool __blk_end_request_cur(struct request *rq, int error)
2399 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2401 EXPORT_SYMBOL(__blk_end_request_cur);
2404 * __blk_end_request_err - Finish a request till the next failure boundary.
2405 * @rq: the request to finish till the next failure boundary for
2406 * @error: must be negative errno
2409 * Complete @rq till the next failure boundary. Must be called
2410 * with queue lock held.
2413 * %false - we are done with this request
2414 * %true - still buffers pending for this request
2416 bool __blk_end_request_err(struct request *rq, int error)
2418 WARN_ON(error >= 0);
2419 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2421 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2423 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2426 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2427 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2429 if (bio_has_data(bio)) {
2430 rq->nr_phys_segments = bio_phys_segments(q, bio);
2431 rq->buffer = bio_data(bio);
2433 rq->__data_len = bio->bi_size;
2434 rq->bio = rq->biotail = bio;
2437 rq->rq_disk = bio->bi_bdev->bd_disk;
2440 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2442 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2443 * @rq: the request to be flushed
2446 * Flush all pages in @rq.
2448 void rq_flush_dcache_pages(struct request *rq)
2450 struct req_iterator iter;
2451 struct bio_vec *bvec;
2453 rq_for_each_segment(bvec, rq, iter)
2454 flush_dcache_page(bvec->bv_page);
2456 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2460 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2461 * @q : the queue of the device being checked
2464 * Check if underlying low-level drivers of a device are busy.
2465 * If the drivers want to export their busy state, they must set own
2466 * exporting function using blk_queue_lld_busy() first.
2468 * Basically, this function is used only by request stacking drivers
2469 * to stop dispatching requests to underlying devices when underlying
2470 * devices are busy. This behavior helps more I/O merging on the queue
2471 * of the request stacking driver and prevents I/O throughput regression
2472 * on burst I/O load.
2475 * 0 - Not busy (The request stacking driver should dispatch request)
2476 * 1 - Busy (The request stacking driver should stop dispatching request)
2478 int blk_lld_busy(struct request_queue *q)
2481 return q->lld_busy_fn(q);
2485 EXPORT_SYMBOL_GPL(blk_lld_busy);
2488 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2489 * @rq: the clone request to be cleaned up
2492 * Free all bios in @rq for a cloned request.
2494 void blk_rq_unprep_clone(struct request *rq)
2498 while ((bio = rq->bio) != NULL) {
2499 rq->bio = bio->bi_next;
2504 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2507 * Copy attributes of the original request to the clone request.
2508 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2510 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2512 dst->cpu = src->cpu;
2513 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2514 dst->cmd_type = src->cmd_type;
2515 dst->__sector = blk_rq_pos(src);
2516 dst->__data_len = blk_rq_bytes(src);
2517 dst->nr_phys_segments = src->nr_phys_segments;
2518 dst->ioprio = src->ioprio;
2519 dst->extra_len = src->extra_len;
2523 * blk_rq_prep_clone - Helper function to setup clone request
2524 * @rq: the request to be setup
2525 * @rq_src: original request to be cloned
2526 * @bs: bio_set that bios for clone are allocated from
2527 * @gfp_mask: memory allocation mask for bio
2528 * @bio_ctr: setup function to be called for each clone bio.
2529 * Returns %0 for success, non %0 for failure.
2530 * @data: private data to be passed to @bio_ctr
2533 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2534 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2535 * are not copied, and copying such parts is the caller's responsibility.
2536 * Also, pages which the original bios are pointing to are not copied
2537 * and the cloned bios just point same pages.
2538 * So cloned bios must be completed before original bios, which means
2539 * the caller must complete @rq before @rq_src.
2541 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2542 struct bio_set *bs, gfp_t gfp_mask,
2543 int (*bio_ctr)(struct bio *, struct bio *, void *),
2546 struct bio *bio, *bio_src;
2551 blk_rq_init(NULL, rq);
2553 __rq_for_each_bio(bio_src, rq_src) {
2554 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2558 __bio_clone(bio, bio_src);
2560 if (bio_integrity(bio_src) &&
2561 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2564 if (bio_ctr && bio_ctr(bio, bio_src, data))
2568 rq->biotail->bi_next = bio;
2571 rq->bio = rq->biotail = bio;
2574 __blk_rq_prep_clone(rq, rq_src);
2581 blk_rq_unprep_clone(rq);
2585 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2587 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2589 return queue_work(kblockd_workqueue, work);
2591 EXPORT_SYMBOL(kblockd_schedule_work);
2593 int kblockd_schedule_delayed_work(struct request_queue *q,
2594 struct delayed_work *dwork, unsigned long delay)
2596 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2598 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2600 #define PLUG_MAGIC 0x91827364
2603 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2604 * @plug: The &struct blk_plug that needs to be initialized
2607 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2608 * pending I/O should the task end up blocking between blk_start_plug() and
2609 * blk_finish_plug(). This is important from a performance perspective, but
2610 * also ensures that we don't deadlock. For instance, if the task is blocking
2611 * for a memory allocation, memory reclaim could end up wanting to free a
2612 * page belonging to that request that is currently residing in our private
2613 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2614 * this kind of deadlock.
2616 void blk_start_plug(struct blk_plug *plug)
2618 struct task_struct *tsk = current;
2620 plug->magic = PLUG_MAGIC;
2621 INIT_LIST_HEAD(&plug->list);
2622 INIT_LIST_HEAD(&plug->cb_list);
2623 plug->should_sort = 0;
2626 * If this is a nested plug, don't actually assign it. It will be
2627 * flushed on its own.
2631 * Store ordering should not be needed here, since a potential
2632 * preempt will imply a full memory barrier
2637 EXPORT_SYMBOL(blk_start_plug);
2639 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2641 struct request *rqa = container_of(a, struct request, queuelist);
2642 struct request *rqb = container_of(b, struct request, queuelist);
2644 return !(rqa->q <= rqb->q);
2648 * If 'from_schedule' is true, then postpone the dispatch of requests
2649 * until a safe kblockd context. We due this to avoid accidental big
2650 * additional stack usage in driver dispatch, in places where the originally
2651 * plugger did not intend it.
2653 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2655 __releases(q->queue_lock)
2657 trace_block_unplug(q, depth, !from_schedule);
2660 * If we are punting this to kblockd, then we can safely drop
2661 * the queue_lock before waking kblockd (which needs to take
2664 if (from_schedule) {
2665 spin_unlock(q->queue_lock);
2666 blk_run_queue_async(q);
2669 spin_unlock(q->queue_lock);
2674 static void flush_plug_callbacks(struct blk_plug *plug)
2676 LIST_HEAD(callbacks);
2678 if (list_empty(&plug->cb_list))
2681 list_splice_init(&plug->cb_list, &callbacks);
2683 while (!list_empty(&callbacks)) {
2684 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2687 list_del(&cb->list);
2692 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2694 struct request_queue *q;
2695 unsigned long flags;
2700 BUG_ON(plug->magic != PLUG_MAGIC);
2702 flush_plug_callbacks(plug);
2703 if (list_empty(&plug->list))
2706 list_splice_init(&plug->list, &list);
2708 if (plug->should_sort) {
2709 list_sort(NULL, &list, plug_rq_cmp);
2710 plug->should_sort = 0;
2717 * Save and disable interrupts here, to avoid doing it for every
2718 * queue lock we have to take.
2720 local_irq_save(flags);
2721 while (!list_empty(&list)) {
2722 rq = list_entry_rq(list.next);
2723 list_del_init(&rq->queuelist);
2727 * This drops the queue lock
2730 queue_unplugged(q, depth, from_schedule);
2733 spin_lock(q->queue_lock);
2736 * rq is already accounted, so use raw insert
2738 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2739 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2741 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2747 * This drops the queue lock
2750 queue_unplugged(q, depth, from_schedule);
2752 local_irq_restore(flags);
2755 void blk_finish_plug(struct blk_plug *plug)
2757 blk_flush_plug_list(plug, false);
2759 if (plug == current->plug)
2760 current->plug = NULL;
2762 EXPORT_SYMBOL(blk_finish_plug);
2764 int __init blk_dev_init(void)
2766 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2767 sizeof(((struct request *)0)->cmd_flags));
2769 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2770 kblockd_workqueue = alloc_workqueue("kblockd",
2771 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2772 if (!kblockd_workqueue)
2773 panic("Failed to create kblockd\n");
2775 request_cachep = kmem_cache_create("blkdev_requests",
2776 sizeof(struct request), 0, SLAB_PANIC, NULL);
2778 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2779 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);