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);
201 static void blk_delay_work(struct work_struct *work)
203 struct request_queue *q;
205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
208 spin_unlock_irq(q->queue_lock);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
226 EXPORT_SYMBOL(blk_delay_queue);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue *q)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
244 EXPORT_SYMBOL(blk_start_queue);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue *q)
262 __cancel_delayed_work(&q->delay_work);
263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
265 EXPORT_SYMBOL(blk_stop_queue);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue *q)
287 del_timer_sync(&q->timeout);
288 cancel_delayed_work_sync(&q->delay_work);
290 EXPORT_SYMBOL(blk_sync_queue);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
295 * @force_kblockd: Don't run @q->request_fn directly. Use kblockd.
298 * See @blk_run_queue. This variant must be called with the queue lock
299 * held and interrupts disabled.
301 void __blk_run_queue(struct request_queue *q)
303 if (unlikely(blk_queue_stopped(q)))
307 * Only recurse once to avoid overrunning the stack, let the unplug
308 * handling reinvoke the handler shortly if we already got there.
310 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
312 queue_flag_clear(QUEUE_FLAG_REENTER, q);
314 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
316 EXPORT_SYMBOL(__blk_run_queue);
319 * blk_run_queue_async - run a single device queue in workqueue context
320 * @q: The queue to run
323 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
326 void blk_run_queue_async(struct request_queue *q)
328 if (likely(!blk_queue_stopped(q)))
329 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
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);
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);
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 successful, 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;
1315 * If this is the first request added after a plug, fire
1316 * of a plug trace. If others have been added before, check
1317 * if we have multiple devices in this plug. If so, make a
1318 * note to sort the list before dispatch.
1320 if (list_empty(&plug->list))
1321 trace_block_plug(q);
1322 else if (!plug->should_sort) {
1323 struct request *__rq;
1325 __rq = list_entry_rq(plug->list.prev);
1327 plug->should_sort = 1;
1330 * Debug flag, kill later
1332 req->cmd_flags |= REQ_ON_PLUG;
1333 list_add_tail(&req->queuelist, &plug->list);
1334 drive_stat_acct(req, 1);
1336 spin_lock_irq(q->queue_lock);
1337 add_acct_request(q, req, where);
1340 spin_unlock_irq(q->queue_lock);
1347 * If bio->bi_dev is a partition, remap the location
1349 static inline void blk_partition_remap(struct bio *bio)
1351 struct block_device *bdev = bio->bi_bdev;
1353 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1354 struct hd_struct *p = bdev->bd_part;
1356 bio->bi_sector += p->start_sect;
1357 bio->bi_bdev = bdev->bd_contains;
1359 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1361 bio->bi_sector - p->start_sect);
1365 static void handle_bad_sector(struct bio *bio)
1367 char b[BDEVNAME_SIZE];
1369 printk(KERN_INFO "attempt to access beyond end of device\n");
1370 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1371 bdevname(bio->bi_bdev, b),
1373 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1374 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1376 set_bit(BIO_EOF, &bio->bi_flags);
1379 #ifdef CONFIG_FAIL_MAKE_REQUEST
1381 static DECLARE_FAULT_ATTR(fail_make_request);
1383 static int __init setup_fail_make_request(char *str)
1385 return setup_fault_attr(&fail_make_request, str);
1387 __setup("fail_make_request=", setup_fail_make_request);
1389 static int should_fail_request(struct bio *bio)
1391 struct hd_struct *part = bio->bi_bdev->bd_part;
1393 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1394 return should_fail(&fail_make_request, bio->bi_size);
1399 static int __init fail_make_request_debugfs(void)
1401 return init_fault_attr_dentries(&fail_make_request,
1402 "fail_make_request");
1405 late_initcall(fail_make_request_debugfs);
1407 #else /* CONFIG_FAIL_MAKE_REQUEST */
1409 static inline int should_fail_request(struct bio *bio)
1414 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1417 * Check whether this bio extends beyond the end of the device.
1419 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1426 /* Test device or partition size, when known. */
1427 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1429 sector_t sector = bio->bi_sector;
1431 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1433 * This may well happen - the kernel calls bread()
1434 * without checking the size of the device, e.g., when
1435 * mounting a device.
1437 handle_bad_sector(bio);
1446 * generic_make_request - hand a buffer to its device driver for I/O
1447 * @bio: The bio describing the location in memory and on the device.
1449 * generic_make_request() is used to make I/O requests of block
1450 * devices. It is passed a &struct bio, which describes the I/O that needs
1453 * generic_make_request() does not return any status. The
1454 * success/failure status of the request, along with notification of
1455 * completion, is delivered asynchronously through the bio->bi_end_io
1456 * function described (one day) else where.
1458 * The caller of generic_make_request must make sure that bi_io_vec
1459 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1460 * set to describe the device address, and the
1461 * bi_end_io and optionally bi_private are set to describe how
1462 * completion notification should be signaled.
1464 * generic_make_request and the drivers it calls may use bi_next if this
1465 * bio happens to be merged with someone else, and may change bi_dev and
1466 * bi_sector for remaps as it sees fit. So the values of these fields
1467 * should NOT be depended on after the call to generic_make_request.
1469 static inline void __generic_make_request(struct bio *bio)
1471 struct request_queue *q;
1472 sector_t old_sector;
1473 int ret, nr_sectors = bio_sectors(bio);
1479 if (bio_check_eod(bio, nr_sectors))
1483 * Resolve the mapping until finished. (drivers are
1484 * still free to implement/resolve their own stacking
1485 * by explicitly returning 0)
1487 * NOTE: we don't repeat the blk_size check for each new device.
1488 * Stacking drivers are expected to know what they are doing.
1493 char b[BDEVNAME_SIZE];
1495 q = bdev_get_queue(bio->bi_bdev);
1498 "generic_make_request: Trying to access "
1499 "nonexistent block-device %s (%Lu)\n",
1500 bdevname(bio->bi_bdev, b),
1501 (long long) bio->bi_sector);
1505 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1506 nr_sectors > queue_max_hw_sectors(q))) {
1507 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1508 bdevname(bio->bi_bdev, b),
1510 queue_max_hw_sectors(q));
1514 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1517 if (should_fail_request(bio))
1521 * If this device has partitions, remap block n
1522 * of partition p to block n+start(p) of the disk.
1524 blk_partition_remap(bio);
1526 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1529 if (old_sector != -1)
1530 trace_block_bio_remap(q, bio, old_dev, old_sector);
1532 old_sector = bio->bi_sector;
1533 old_dev = bio->bi_bdev->bd_dev;
1535 if (bio_check_eod(bio, nr_sectors))
1539 * Filter flush bio's early so that make_request based
1540 * drivers without flush support don't have to worry
1543 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1544 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1551 if ((bio->bi_rw & REQ_DISCARD) &&
1552 (!blk_queue_discard(q) ||
1553 ((bio->bi_rw & REQ_SECURE) &&
1554 !blk_queue_secdiscard(q)))) {
1559 blk_throtl_bio(q, &bio);
1562 * If bio = NULL, bio has been throttled and will be submitted
1568 trace_block_bio_queue(q, bio);
1570 ret = q->make_request_fn(q, bio);
1576 bio_endio(bio, err);
1580 * We only want one ->make_request_fn to be active at a time,
1581 * else stack usage with stacked devices could be a problem.
1582 * So use current->bio_list to keep a list of requests
1583 * submited by a make_request_fn function.
1584 * current->bio_list is also used as a flag to say if
1585 * generic_make_request is currently active in this task or not.
1586 * If it is NULL, then no make_request is active. If it is non-NULL,
1587 * then a make_request is active, and new requests should be added
1590 void generic_make_request(struct bio *bio)
1592 struct bio_list bio_list_on_stack;
1594 if (current->bio_list) {
1595 /* make_request is active */
1596 bio_list_add(current->bio_list, bio);
1599 /* following loop may be a bit non-obvious, and so deserves some
1601 * Before entering the loop, bio->bi_next is NULL (as all callers
1602 * ensure that) so we have a list with a single bio.
1603 * We pretend that we have just taken it off a longer list, so
1604 * we assign bio_list to a pointer to the bio_list_on_stack,
1605 * thus initialising the bio_list of new bios to be
1606 * added. __generic_make_request may indeed add some more bios
1607 * through a recursive call to generic_make_request. If it
1608 * did, we find a non-NULL value in bio_list and re-enter the loop
1609 * from the top. In this case we really did just take the bio
1610 * of the top of the list (no pretending) and so remove it from
1611 * bio_list, and call into __generic_make_request again.
1613 * The loop was structured like this to make only one call to
1614 * __generic_make_request (which is important as it is large and
1615 * inlined) and to keep the structure simple.
1617 BUG_ON(bio->bi_next);
1618 bio_list_init(&bio_list_on_stack);
1619 current->bio_list = &bio_list_on_stack;
1621 __generic_make_request(bio);
1622 bio = bio_list_pop(current->bio_list);
1624 current->bio_list = NULL; /* deactivate */
1626 EXPORT_SYMBOL(generic_make_request);
1629 * submit_bio - submit a bio to the block device layer for I/O
1630 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1631 * @bio: The &struct bio which describes the I/O
1633 * submit_bio() is very similar in purpose to generic_make_request(), and
1634 * uses that function to do most of the work. Both are fairly rough
1635 * interfaces; @bio must be presetup and ready for I/O.
1638 void submit_bio(int rw, struct bio *bio)
1640 int count = bio_sectors(bio);
1645 * If it's a regular read/write or a barrier with data attached,
1646 * go through the normal accounting stuff before submission.
1648 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1650 count_vm_events(PGPGOUT, count);
1652 task_io_account_read(bio->bi_size);
1653 count_vm_events(PGPGIN, count);
1656 if (unlikely(block_dump)) {
1657 char b[BDEVNAME_SIZE];
1658 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1659 current->comm, task_pid_nr(current),
1660 (rw & WRITE) ? "WRITE" : "READ",
1661 (unsigned long long)bio->bi_sector,
1662 bdevname(bio->bi_bdev, b),
1667 generic_make_request(bio);
1669 EXPORT_SYMBOL(submit_bio);
1672 * blk_rq_check_limits - Helper function to check a request for the queue limit
1674 * @rq: the request being checked
1677 * @rq may have been made based on weaker limitations of upper-level queues
1678 * in request stacking drivers, and it may violate the limitation of @q.
1679 * Since the block layer and the underlying device driver trust @rq
1680 * after it is inserted to @q, it should be checked against @q before
1681 * the insertion using this generic function.
1683 * This function should also be useful for request stacking drivers
1684 * in some cases below, so export this function.
1685 * Request stacking drivers like request-based dm may change the queue
1686 * limits while requests are in the queue (e.g. dm's table swapping).
1687 * Such request stacking drivers should check those requests agaist
1688 * the new queue limits again when they dispatch those requests,
1689 * although such checkings are also done against the old queue limits
1690 * when submitting requests.
1692 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1694 if (rq->cmd_flags & REQ_DISCARD)
1697 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1698 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1699 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1704 * queue's settings related to segment counting like q->bounce_pfn
1705 * may differ from that of other stacking queues.
1706 * Recalculate it to check the request correctly on this queue's
1709 blk_recalc_rq_segments(rq);
1710 if (rq->nr_phys_segments > queue_max_segments(q)) {
1711 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1717 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1720 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1721 * @q: the queue to submit the request
1722 * @rq: the request being queued
1724 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1726 unsigned long flags;
1728 if (blk_rq_check_limits(q, rq))
1731 #ifdef CONFIG_FAIL_MAKE_REQUEST
1732 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1733 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1737 spin_lock_irqsave(q->queue_lock, flags);
1740 * Submitting request must be dequeued before calling this function
1741 * because it will be linked to another request_queue
1743 BUG_ON(blk_queued_rq(rq));
1745 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1746 spin_unlock_irqrestore(q->queue_lock, flags);
1750 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1753 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1754 * @rq: request to examine
1757 * A request could be merge of IOs which require different failure
1758 * handling. This function determines the number of bytes which
1759 * can be failed from the beginning of the request without
1760 * crossing into area which need to be retried further.
1763 * The number of bytes to fail.
1766 * queue_lock must be held.
1768 unsigned int blk_rq_err_bytes(const struct request *rq)
1770 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1771 unsigned int bytes = 0;
1774 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1775 return blk_rq_bytes(rq);
1778 * Currently the only 'mixing' which can happen is between
1779 * different fastfail types. We can safely fail portions
1780 * which have all the failfast bits that the first one has -
1781 * the ones which are at least as eager to fail as the first
1784 for (bio = rq->bio; bio; bio = bio->bi_next) {
1785 if ((bio->bi_rw & ff) != ff)
1787 bytes += bio->bi_size;
1790 /* this could lead to infinite loop */
1791 BUG_ON(blk_rq_bytes(rq) && !bytes);
1794 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1796 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1798 if (blk_do_io_stat(req)) {
1799 const int rw = rq_data_dir(req);
1800 struct hd_struct *part;
1803 cpu = part_stat_lock();
1805 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1810 static void blk_account_io_done(struct request *req)
1813 * Account IO completion. flush_rq isn't accounted as a
1814 * normal IO on queueing nor completion. Accounting the
1815 * containing request is enough.
1817 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1818 unsigned long duration = jiffies - req->start_time;
1819 const int rw = rq_data_dir(req);
1820 struct hd_struct *part;
1823 cpu = part_stat_lock();
1826 part_stat_inc(cpu, part, ios[rw]);
1827 part_stat_add(cpu, part, ticks[rw], duration);
1828 part_round_stats(cpu, part);
1829 part_dec_in_flight(part, rw);
1831 hd_struct_put(part);
1837 * blk_peek_request - peek at the top of a request queue
1838 * @q: request queue to peek at
1841 * Return the request at the top of @q. The returned request
1842 * should be started using blk_start_request() before LLD starts
1846 * Pointer to the request at the top of @q if available. Null
1850 * queue_lock must be held.
1852 struct request *blk_peek_request(struct request_queue *q)
1857 while ((rq = __elv_next_request(q)) != NULL) {
1858 if (!(rq->cmd_flags & REQ_STARTED)) {
1860 * This is the first time the device driver
1861 * sees this request (possibly after
1862 * requeueing). Notify IO scheduler.
1864 if (rq->cmd_flags & REQ_SORTED)
1865 elv_activate_rq(q, rq);
1868 * just mark as started even if we don't start
1869 * it, a request that has been delayed should
1870 * not be passed by new incoming requests
1872 rq->cmd_flags |= REQ_STARTED;
1873 trace_block_rq_issue(q, rq);
1876 if (!q->boundary_rq || q->boundary_rq == rq) {
1877 q->end_sector = rq_end_sector(rq);
1878 q->boundary_rq = NULL;
1881 if (rq->cmd_flags & REQ_DONTPREP)
1884 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1886 * make sure space for the drain appears we
1887 * know we can do this because max_hw_segments
1888 * has been adjusted to be one fewer than the
1891 rq->nr_phys_segments++;
1897 ret = q->prep_rq_fn(q, rq);
1898 if (ret == BLKPREP_OK) {
1900 } else if (ret == BLKPREP_DEFER) {
1902 * the request may have been (partially) prepped.
1903 * we need to keep this request in the front to
1904 * avoid resource deadlock. REQ_STARTED will
1905 * prevent other fs requests from passing this one.
1907 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1908 !(rq->cmd_flags & REQ_DONTPREP)) {
1910 * remove the space for the drain we added
1911 * so that we don't add it again
1913 --rq->nr_phys_segments;
1918 } else if (ret == BLKPREP_KILL) {
1919 rq->cmd_flags |= REQ_QUIET;
1921 * Mark this request as started so we don't trigger
1922 * any debug logic in the end I/O path.
1924 blk_start_request(rq);
1925 __blk_end_request_all(rq, -EIO);
1927 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1934 EXPORT_SYMBOL(blk_peek_request);
1936 void blk_dequeue_request(struct request *rq)
1938 struct request_queue *q = rq->q;
1940 BUG_ON(list_empty(&rq->queuelist));
1941 BUG_ON(ELV_ON_HASH(rq));
1943 list_del_init(&rq->queuelist);
1946 * the time frame between a request being removed from the lists
1947 * and to it is freed is accounted as io that is in progress at
1950 if (blk_account_rq(rq)) {
1951 q->in_flight[rq_is_sync(rq)]++;
1952 set_io_start_time_ns(rq);
1957 * blk_start_request - start request processing on the driver
1958 * @req: request to dequeue
1961 * Dequeue @req and start timeout timer on it. This hands off the
1962 * request to the driver.
1964 * Block internal functions which don't want to start timer should
1965 * call blk_dequeue_request().
1968 * queue_lock must be held.
1970 void blk_start_request(struct request *req)
1972 blk_dequeue_request(req);
1975 * We are now handing the request to the hardware, initialize
1976 * resid_len to full count and add the timeout handler.
1978 req->resid_len = blk_rq_bytes(req);
1979 if (unlikely(blk_bidi_rq(req)))
1980 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1984 EXPORT_SYMBOL(blk_start_request);
1987 * blk_fetch_request - fetch a request from a request queue
1988 * @q: request queue to fetch a request from
1991 * Return the request at the top of @q. The request is started on
1992 * return and LLD can start processing it immediately.
1995 * Pointer to the request at the top of @q if available. Null
1999 * queue_lock must be held.
2001 struct request *blk_fetch_request(struct request_queue *q)
2005 rq = blk_peek_request(q);
2007 blk_start_request(rq);
2010 EXPORT_SYMBOL(blk_fetch_request);
2013 * blk_update_request - Special helper function for request stacking drivers
2014 * @req: the request being processed
2015 * @error: %0 for success, < %0 for error
2016 * @nr_bytes: number of bytes to complete @req
2019 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2020 * the request structure even if @req doesn't have leftover.
2021 * If @req has leftover, sets it up for the next range of segments.
2023 * This special helper function is only for request stacking drivers
2024 * (e.g. request-based dm) so that they can handle partial completion.
2025 * Actual device drivers should use blk_end_request instead.
2027 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2028 * %false return from this function.
2031 * %false - this request doesn't have any more data
2032 * %true - this request has more data
2034 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2036 int total_bytes, bio_nbytes, next_idx = 0;
2042 trace_block_rq_complete(req->q, req);
2045 * For fs requests, rq is just carrier of independent bio's
2046 * and each partial completion should be handled separately.
2047 * Reset per-request error on each partial completion.
2049 * TODO: tj: This is too subtle. It would be better to let
2050 * low level drivers do what they see fit.
2052 if (req->cmd_type == REQ_TYPE_FS)
2055 if (error && req->cmd_type == REQ_TYPE_FS &&
2056 !(req->cmd_flags & REQ_QUIET)) {
2061 error_type = "recoverable transport";
2064 error_type = "critical target";
2067 error_type = "critical nexus";
2074 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2075 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2076 (unsigned long long)blk_rq_pos(req));
2079 blk_account_io_completion(req, nr_bytes);
2081 total_bytes = bio_nbytes = 0;
2082 while ((bio = req->bio) != NULL) {
2085 if (nr_bytes >= bio->bi_size) {
2086 req->bio = bio->bi_next;
2087 nbytes = bio->bi_size;
2088 req_bio_endio(req, bio, nbytes, error);
2092 int idx = bio->bi_idx + next_idx;
2094 if (unlikely(idx >= bio->bi_vcnt)) {
2095 blk_dump_rq_flags(req, "__end_that");
2096 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2097 __func__, idx, bio->bi_vcnt);
2101 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2102 BIO_BUG_ON(nbytes > bio->bi_size);
2105 * not a complete bvec done
2107 if (unlikely(nbytes > nr_bytes)) {
2108 bio_nbytes += nr_bytes;
2109 total_bytes += nr_bytes;
2114 * advance to the next vector
2117 bio_nbytes += nbytes;
2120 total_bytes += nbytes;
2126 * end more in this run, or just return 'not-done'
2128 if (unlikely(nr_bytes <= 0))
2138 * Reset counters so that the request stacking driver
2139 * can find how many bytes remain in the request
2142 req->__data_len = 0;
2147 * if the request wasn't completed, update state
2150 req_bio_endio(req, bio, bio_nbytes, error);
2151 bio->bi_idx += next_idx;
2152 bio_iovec(bio)->bv_offset += nr_bytes;
2153 bio_iovec(bio)->bv_len -= nr_bytes;
2156 req->__data_len -= total_bytes;
2157 req->buffer = bio_data(req->bio);
2159 /* update sector only for requests with clear definition of sector */
2160 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2161 req->__sector += total_bytes >> 9;
2163 /* mixed attributes always follow the first bio */
2164 if (req->cmd_flags & REQ_MIXED_MERGE) {
2165 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2166 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2170 * If total number of sectors is less than the first segment
2171 * size, something has gone terribly wrong.
2173 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2174 blk_dump_rq_flags(req, "request botched");
2175 req->__data_len = blk_rq_cur_bytes(req);
2178 /* recalculate the number of segments */
2179 blk_recalc_rq_segments(req);
2183 EXPORT_SYMBOL_GPL(blk_update_request);
2185 static bool blk_update_bidi_request(struct request *rq, int error,
2186 unsigned int nr_bytes,
2187 unsigned int bidi_bytes)
2189 if (blk_update_request(rq, error, nr_bytes))
2192 /* Bidi request must be completed as a whole */
2193 if (unlikely(blk_bidi_rq(rq)) &&
2194 blk_update_request(rq->next_rq, error, bidi_bytes))
2197 if (blk_queue_add_random(rq->q))
2198 add_disk_randomness(rq->rq_disk);
2204 * blk_unprep_request - unprepare a request
2207 * This function makes a request ready for complete resubmission (or
2208 * completion). It happens only after all error handling is complete,
2209 * so represents the appropriate moment to deallocate any resources
2210 * that were allocated to the request in the prep_rq_fn. The queue
2211 * lock is held when calling this.
2213 void blk_unprep_request(struct request *req)
2215 struct request_queue *q = req->q;
2217 req->cmd_flags &= ~REQ_DONTPREP;
2218 if (q->unprep_rq_fn)
2219 q->unprep_rq_fn(q, req);
2221 EXPORT_SYMBOL_GPL(blk_unprep_request);
2224 * queue lock must be held
2226 static void blk_finish_request(struct request *req, int error)
2228 if (blk_rq_tagged(req))
2229 blk_queue_end_tag(req->q, req);
2231 BUG_ON(blk_queued_rq(req));
2233 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2234 laptop_io_completion(&req->q->backing_dev_info);
2236 blk_delete_timer(req);
2238 if (req->cmd_flags & REQ_DONTPREP)
2239 blk_unprep_request(req);
2242 blk_account_io_done(req);
2245 req->end_io(req, error);
2247 if (blk_bidi_rq(req))
2248 __blk_put_request(req->next_rq->q, req->next_rq);
2250 __blk_put_request(req->q, req);
2255 * blk_end_bidi_request - Complete a bidi request
2256 * @rq: the request to complete
2257 * @error: %0 for success, < %0 for error
2258 * @nr_bytes: number of bytes to complete @rq
2259 * @bidi_bytes: number of bytes to complete @rq->next_rq
2262 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2263 * Drivers that supports bidi can safely call this member for any
2264 * type of request, bidi or uni. In the later case @bidi_bytes is
2268 * %false - we are done with this request
2269 * %true - still buffers pending for this request
2271 static bool blk_end_bidi_request(struct request *rq, int error,
2272 unsigned int nr_bytes, unsigned int bidi_bytes)
2274 struct request_queue *q = rq->q;
2275 unsigned long flags;
2277 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2280 spin_lock_irqsave(q->queue_lock, flags);
2281 blk_finish_request(rq, error);
2282 spin_unlock_irqrestore(q->queue_lock, flags);
2288 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2289 * @rq: the request to complete
2290 * @error: %0 for success, < %0 for error
2291 * @nr_bytes: number of bytes to complete @rq
2292 * @bidi_bytes: number of bytes to complete @rq->next_rq
2295 * Identical to blk_end_bidi_request() except that queue lock is
2296 * assumed to be locked on entry and remains so on return.
2299 * %false - we are done with this request
2300 * %true - still buffers pending for this request
2302 static bool __blk_end_bidi_request(struct request *rq, int error,
2303 unsigned int nr_bytes, unsigned int bidi_bytes)
2305 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2308 blk_finish_request(rq, error);
2314 * blk_end_request - Helper function for drivers to complete the request.
2315 * @rq: the request being processed
2316 * @error: %0 for success, < %0 for error
2317 * @nr_bytes: number of bytes to complete
2320 * Ends I/O on a number of bytes attached to @rq.
2321 * If @rq has leftover, sets it up for the next range of segments.
2324 * %false - we are done with this request
2325 * %true - still buffers pending for this request
2327 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2329 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2331 EXPORT_SYMBOL(blk_end_request);
2334 * blk_end_request_all - Helper function for drives to finish the request.
2335 * @rq: the request to finish
2336 * @error: %0 for success, < %0 for error
2339 * Completely finish @rq.
2341 void blk_end_request_all(struct request *rq, int error)
2344 unsigned int bidi_bytes = 0;
2346 if (unlikely(blk_bidi_rq(rq)))
2347 bidi_bytes = blk_rq_bytes(rq->next_rq);
2349 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2352 EXPORT_SYMBOL(blk_end_request_all);
2355 * blk_end_request_cur - Helper function to finish the current request chunk.
2356 * @rq: the request to finish the current chunk for
2357 * @error: %0 for success, < %0 for error
2360 * Complete the current consecutively mapped chunk from @rq.
2363 * %false - we are done with this request
2364 * %true - still buffers pending for this request
2366 bool blk_end_request_cur(struct request *rq, int error)
2368 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2370 EXPORT_SYMBOL(blk_end_request_cur);
2373 * blk_end_request_err - Finish a request till the next failure boundary.
2374 * @rq: the request to finish till the next failure boundary for
2375 * @error: must be negative errno
2378 * Complete @rq till the next failure boundary.
2381 * %false - we are done with this request
2382 * %true - still buffers pending for this request
2384 bool blk_end_request_err(struct request *rq, int error)
2386 WARN_ON(error >= 0);
2387 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2389 EXPORT_SYMBOL_GPL(blk_end_request_err);
2392 * __blk_end_request - Helper function for drivers to complete the request.
2393 * @rq: the request being processed
2394 * @error: %0 for success, < %0 for error
2395 * @nr_bytes: number of bytes to complete
2398 * Must be called with queue lock held unlike blk_end_request().
2401 * %false - we are done with this request
2402 * %true - still buffers pending for this request
2404 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2406 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2408 EXPORT_SYMBOL(__blk_end_request);
2411 * __blk_end_request_all - Helper function for drives to finish the request.
2412 * @rq: the request to finish
2413 * @error: %0 for success, < %0 for error
2416 * Completely finish @rq. Must be called with queue lock held.
2418 void __blk_end_request_all(struct request *rq, int error)
2421 unsigned int bidi_bytes = 0;
2423 if (unlikely(blk_bidi_rq(rq)))
2424 bidi_bytes = blk_rq_bytes(rq->next_rq);
2426 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2429 EXPORT_SYMBOL(__blk_end_request_all);
2432 * __blk_end_request_cur - Helper function to finish the current request chunk.
2433 * @rq: the request to finish the current chunk for
2434 * @error: %0 for success, < %0 for error
2437 * Complete the current consecutively mapped chunk from @rq. Must
2438 * be called with queue lock held.
2441 * %false - we are done with this request
2442 * %true - still buffers pending for this request
2444 bool __blk_end_request_cur(struct request *rq, int error)
2446 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2448 EXPORT_SYMBOL(__blk_end_request_cur);
2451 * __blk_end_request_err - Finish a request till the next failure boundary.
2452 * @rq: the request to finish till the next failure boundary for
2453 * @error: must be negative errno
2456 * Complete @rq till the next failure boundary. Must be called
2457 * with queue lock held.
2460 * %false - we are done with this request
2461 * %true - still buffers pending for this request
2463 bool __blk_end_request_err(struct request *rq, int error)
2465 WARN_ON(error >= 0);
2466 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2468 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2470 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2473 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2474 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2476 if (bio_has_data(bio)) {
2477 rq->nr_phys_segments = bio_phys_segments(q, bio);
2478 rq->buffer = bio_data(bio);
2480 rq->__data_len = bio->bi_size;
2481 rq->bio = rq->biotail = bio;
2484 rq->rq_disk = bio->bi_bdev->bd_disk;
2487 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2489 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2490 * @rq: the request to be flushed
2493 * Flush all pages in @rq.
2495 void rq_flush_dcache_pages(struct request *rq)
2497 struct req_iterator iter;
2498 struct bio_vec *bvec;
2500 rq_for_each_segment(bvec, rq, iter)
2501 flush_dcache_page(bvec->bv_page);
2503 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2507 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2508 * @q : the queue of the device being checked
2511 * Check if underlying low-level drivers of a device are busy.
2512 * If the drivers want to export their busy state, they must set own
2513 * exporting function using blk_queue_lld_busy() first.
2515 * Basically, this function is used only by request stacking drivers
2516 * to stop dispatching requests to underlying devices when underlying
2517 * devices are busy. This behavior helps more I/O merging on the queue
2518 * of the request stacking driver and prevents I/O throughput regression
2519 * on burst I/O load.
2522 * 0 - Not busy (The request stacking driver should dispatch request)
2523 * 1 - Busy (The request stacking driver should stop dispatching request)
2525 int blk_lld_busy(struct request_queue *q)
2528 return q->lld_busy_fn(q);
2532 EXPORT_SYMBOL_GPL(blk_lld_busy);
2535 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2536 * @rq: the clone request to be cleaned up
2539 * Free all bios in @rq for a cloned request.
2541 void blk_rq_unprep_clone(struct request *rq)
2545 while ((bio = rq->bio) != NULL) {
2546 rq->bio = bio->bi_next;
2551 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2554 * Copy attributes of the original request to the clone request.
2555 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2557 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2559 dst->cpu = src->cpu;
2560 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2561 dst->cmd_type = src->cmd_type;
2562 dst->__sector = blk_rq_pos(src);
2563 dst->__data_len = blk_rq_bytes(src);
2564 dst->nr_phys_segments = src->nr_phys_segments;
2565 dst->ioprio = src->ioprio;
2566 dst->extra_len = src->extra_len;
2570 * blk_rq_prep_clone - Helper function to setup clone request
2571 * @rq: the request to be setup
2572 * @rq_src: original request to be cloned
2573 * @bs: bio_set that bios for clone are allocated from
2574 * @gfp_mask: memory allocation mask for bio
2575 * @bio_ctr: setup function to be called for each clone bio.
2576 * Returns %0 for success, non %0 for failure.
2577 * @data: private data to be passed to @bio_ctr
2580 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2581 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2582 * are not copied, and copying such parts is the caller's responsibility.
2583 * Also, pages which the original bios are pointing to are not copied
2584 * and the cloned bios just point same pages.
2585 * So cloned bios must be completed before original bios, which means
2586 * the caller must complete @rq before @rq_src.
2588 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2589 struct bio_set *bs, gfp_t gfp_mask,
2590 int (*bio_ctr)(struct bio *, struct bio *, void *),
2593 struct bio *bio, *bio_src;
2598 blk_rq_init(NULL, rq);
2600 __rq_for_each_bio(bio_src, rq_src) {
2601 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2605 __bio_clone(bio, bio_src);
2607 if (bio_integrity(bio_src) &&
2608 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2611 if (bio_ctr && bio_ctr(bio, bio_src, data))
2615 rq->biotail->bi_next = bio;
2618 rq->bio = rq->biotail = bio;
2621 __blk_rq_prep_clone(rq, rq_src);
2628 blk_rq_unprep_clone(rq);
2632 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2634 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2636 return queue_work(kblockd_workqueue, work);
2638 EXPORT_SYMBOL(kblockd_schedule_work);
2640 int kblockd_schedule_delayed_work(struct request_queue *q,
2641 struct delayed_work *dwork, unsigned long delay)
2643 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2645 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2647 #define PLUG_MAGIC 0x91827364
2649 void blk_start_plug(struct blk_plug *plug)
2651 struct task_struct *tsk = current;
2653 plug->magic = PLUG_MAGIC;
2654 INIT_LIST_HEAD(&plug->list);
2655 INIT_LIST_HEAD(&plug->cb_list);
2656 plug->should_sort = 0;
2659 * If this is a nested plug, don't actually assign it. It will be
2660 * flushed on its own.
2664 * Store ordering should not be needed here, since a potential
2665 * preempt will imply a full memory barrier
2670 EXPORT_SYMBOL(blk_start_plug);
2672 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2674 struct request *rqa = container_of(a, struct request, queuelist);
2675 struct request *rqb = container_of(b, struct request, queuelist);
2677 return !(rqa->q <= rqb->q);
2681 * If 'from_schedule' is true, then postpone the dispatch of requests
2682 * until a safe kblockd context. We due this to avoid accidental big
2683 * additional stack usage in driver dispatch, in places where the originally
2684 * plugger did not intend it.
2686 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2688 __releases(q->queue_lock)
2690 trace_block_unplug(q, depth, !from_schedule);
2693 * If we are punting this to kblockd, then we can safely drop
2694 * the queue_lock before waking kblockd (which needs to take
2697 if (from_schedule) {
2698 spin_unlock(q->queue_lock);
2699 blk_run_queue_async(q);
2702 spin_unlock(q->queue_lock);
2707 static void flush_plug_callbacks(struct blk_plug *plug)
2709 LIST_HEAD(callbacks);
2711 if (list_empty(&plug->cb_list))
2714 list_splice_init(&plug->cb_list, &callbacks);
2716 while (!list_empty(&callbacks)) {
2717 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2720 list_del(&cb->list);
2725 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2727 struct request_queue *q;
2728 unsigned long flags;
2733 BUG_ON(plug->magic != PLUG_MAGIC);
2735 flush_plug_callbacks(plug);
2736 if (list_empty(&plug->list))
2739 list_splice_init(&plug->list, &list);
2741 if (plug->should_sort) {
2742 list_sort(NULL, &list, plug_rq_cmp);
2743 plug->should_sort = 0;
2750 * Save and disable interrupts here, to avoid doing it for every
2751 * queue lock we have to take.
2753 local_irq_save(flags);
2754 while (!list_empty(&list)) {
2755 rq = list_entry_rq(list.next);
2756 list_del_init(&rq->queuelist);
2757 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2761 * This drops the queue lock
2764 queue_unplugged(q, depth, from_schedule);
2767 spin_lock(q->queue_lock);
2769 rq->cmd_flags &= ~REQ_ON_PLUG;
2772 * rq is already accounted, so use raw insert
2774 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2775 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2777 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2783 * This drops the queue lock
2786 queue_unplugged(q, depth, from_schedule);
2788 local_irq_restore(flags);
2790 EXPORT_SYMBOL(blk_flush_plug_list);
2792 void blk_finish_plug(struct blk_plug *plug)
2794 blk_flush_plug_list(plug, false);
2796 if (plug == current->plug)
2797 current->plug = NULL;
2799 EXPORT_SYMBOL(blk_finish_plug);
2801 int __init blk_dev_init(void)
2803 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2804 sizeof(((struct request *)0)->cmd_flags));
2806 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2807 kblockd_workqueue = alloc_workqueue("kblockd",
2808 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2809 if (!kblockd_workqueue)
2810 panic("Failed to create kblockd\n");
2812 request_cachep = kmem_cache_create("blkdev_requests",
2813 sizeof(struct request), 0, SLAB_PANIC, NULL);
2815 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2816 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);