2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
32 #include <linux/ratelimit.h>
33 #include <linux/pm_runtime.h>
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/block.h>
39 #include "blk-cgroup.h"
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
46 DEFINE_IDA(blk_queue_ida);
49 * For the allocated request tables
51 static struct kmem_cache *request_cachep;
54 * For queue allocation
56 struct kmem_cache *blk_requestq_cachep;
59 * Controlling structure to kblockd
61 static struct workqueue_struct *kblockd_workqueue;
63 static void drive_stat_acct(struct request *rq, int new_io)
65 struct hd_struct *part;
66 int rw = rq_data_dir(rq);
69 if (!blk_do_io_stat(rq))
72 cpu = part_stat_lock();
76 part_stat_inc(cpu, part, merges[rw]);
78 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
79 if (!hd_struct_try_get(part)) {
81 * The partition is already being removed,
82 * the request will be accounted on the disk only
84 * We take a reference on disk->part0 although that
85 * partition will never be deleted, so we can treat
86 * it as any other partition.
88 part = &rq->rq_disk->part0;
91 part_round_stats(cpu, part);
92 part_inc_in_flight(part, rw);
99 void blk_queue_congestion_threshold(struct request_queue *q)
103 nr = q->nr_requests - (q->nr_requests / 8) + 1;
104 if (nr > q->nr_requests)
106 q->nr_congestion_on = nr;
108 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
111 q->nr_congestion_off = nr;
115 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
118 * Locates the passed device's request queue and returns the address of its
121 * Will return NULL if the request queue cannot be located.
123 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
125 struct backing_dev_info *ret = NULL;
126 struct request_queue *q = bdev_get_queue(bdev);
129 ret = &q->backing_dev_info;
132 EXPORT_SYMBOL(blk_get_backing_dev_info);
134 void blk_rq_init(struct request_queue *q, struct request *rq)
136 memset(rq, 0, sizeof(*rq));
138 INIT_LIST_HEAD(&rq->queuelist);
139 INIT_LIST_HEAD(&rq->timeout_list);
142 rq->__sector = (sector_t) -1;
143 INIT_HLIST_NODE(&rq->hash);
144 RB_CLEAR_NODE(&rq->rb_node);
146 rq->cmd_len = BLK_MAX_CDB;
149 rq->start_time = jiffies;
150 set_start_time_ns(rq);
153 EXPORT_SYMBOL(blk_rq_init);
155 static void req_bio_endio(struct request *rq, struct bio *bio,
156 unsigned int nbytes, int error,
157 struct batch_complete *batch)
160 clear_bit(BIO_UPTODATE, &bio->bi_flags);
161 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
167 bio_advance(bio, nbytes);
169 /* don't actually finish bio if it's part of flush sequence */
170 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
171 bio_endio_batch(bio, error, batch);
174 void blk_dump_rq_flags(struct request *rq, char *msg)
178 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
179 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
182 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
183 (unsigned long long)blk_rq_pos(rq),
184 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
185 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
186 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
188 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
189 printk(KERN_INFO " cdb: ");
190 for (bit = 0; bit < BLK_MAX_CDB; bit++)
191 printk("%02x ", rq->cmd[bit]);
195 EXPORT_SYMBOL(blk_dump_rq_flags);
197 static void blk_delay_work(struct work_struct *work)
199 struct request_queue *q;
201 q = container_of(work, struct request_queue, delay_work.work);
202 spin_lock_irq(q->queue_lock);
204 spin_unlock_irq(q->queue_lock);
208 * blk_delay_queue - restart queueing after defined interval
209 * @q: The &struct request_queue in question
210 * @msecs: Delay in msecs
213 * Sometimes queueing needs to be postponed for a little while, to allow
214 * resources to come back. This function will make sure that queueing is
215 * restarted around the specified time. Queue lock must be held.
217 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
219 if (likely(!blk_queue_dead(q)))
220 queue_delayed_work(kblockd_workqueue, &q->delay_work,
221 msecs_to_jiffies(msecs));
223 EXPORT_SYMBOL(blk_delay_queue);
226 * blk_start_queue - restart a previously stopped queue
227 * @q: The &struct request_queue in question
230 * blk_start_queue() will clear the stop flag on the queue, and call
231 * the request_fn for the queue if it was in a stopped state when
232 * entered. Also see blk_stop_queue(). Queue lock must be held.
234 void blk_start_queue(struct request_queue *q)
236 WARN_ON(!irqs_disabled());
238 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
241 EXPORT_SYMBOL(blk_start_queue);
244 * blk_stop_queue - stop a queue
245 * @q: The &struct request_queue in question
248 * The Linux block layer assumes that a block driver will consume all
249 * entries on the request queue when the request_fn strategy is called.
250 * Often this will not happen, because of hardware limitations (queue
251 * depth settings). If a device driver gets a 'queue full' response,
252 * or if it simply chooses not to queue more I/O at one point, it can
253 * call this function to prevent the request_fn from being called until
254 * the driver has signalled it's ready to go again. This happens by calling
255 * blk_start_queue() to restart queue operations. Queue lock must be held.
257 void blk_stop_queue(struct request_queue *q)
259 cancel_delayed_work(&q->delay_work);
260 queue_flag_set(QUEUE_FLAG_STOPPED, q);
262 EXPORT_SYMBOL(blk_stop_queue);
265 * blk_sync_queue - cancel any pending callbacks on a queue
269 * The block layer may perform asynchronous callback activity
270 * on a queue, such as calling the unplug function after a timeout.
271 * A block device may call blk_sync_queue to ensure that any
272 * such activity is cancelled, thus allowing it to release resources
273 * that the callbacks might use. The caller must already have made sure
274 * that its ->make_request_fn will not re-add plugging prior to calling
277 * This function does not cancel any asynchronous activity arising
278 * out of elevator or throttling code. That would require elevaotor_exit()
279 * and blkcg_exit_queue() to be called with queue lock initialized.
282 void blk_sync_queue(struct request_queue *q)
284 del_timer_sync(&q->timeout);
285 cancel_delayed_work_sync(&q->delay_work);
287 EXPORT_SYMBOL(blk_sync_queue);
290 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
291 * @q: The queue to run
294 * Invoke request handling on a queue if there are any pending requests.
295 * May be used to restart request handling after a request has completed.
296 * This variant runs the queue whether or not the queue has been
297 * stopped. Must be called with the queue lock held and interrupts
298 * disabled. See also @blk_run_queue.
300 inline void __blk_run_queue_uncond(struct request_queue *q)
302 if (unlikely(blk_queue_dead(q)))
306 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
307 * the queue lock internally. As a result multiple threads may be
308 * running such a request function concurrently. Keep track of the
309 * number of active request_fn invocations such that blk_drain_queue()
310 * can wait until all these request_fn calls have finished.
312 q->request_fn_active++;
314 q->request_fn_active--;
318 * __blk_run_queue - run a single device queue
319 * @q: The queue to run
322 * See @blk_run_queue. This variant must be called with the queue lock
323 * held and interrupts disabled.
325 void __blk_run_queue(struct request_queue *q)
327 if (unlikely(blk_queue_stopped(q)))
330 __blk_run_queue_uncond(q);
332 EXPORT_SYMBOL(__blk_run_queue);
335 * blk_run_queue_async - run a single device queue in workqueue context
336 * @q: The queue to run
339 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
340 * of us. The caller must hold the queue lock.
342 void blk_run_queue_async(struct request_queue *q)
344 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
345 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
347 EXPORT_SYMBOL(blk_run_queue_async);
350 * blk_run_queue - run a single device queue
351 * @q: The queue to run
354 * Invoke request handling on this queue, if it has pending work to do.
355 * May be used to restart queueing when a request has completed.
357 void blk_run_queue(struct request_queue *q)
361 spin_lock_irqsave(q->queue_lock, flags);
363 spin_unlock_irqrestore(q->queue_lock, flags);
365 EXPORT_SYMBOL(blk_run_queue);
367 void blk_put_queue(struct request_queue *q)
369 kobject_put(&q->kobj);
371 EXPORT_SYMBOL(blk_put_queue);
374 * __blk_drain_queue - drain requests from request_queue
376 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
378 * Drain requests from @q. If @drain_all is set, all requests are drained.
379 * If not, only ELVPRIV requests are drained. The caller is responsible
380 * for ensuring that no new requests which need to be drained are queued.
382 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
383 __releases(q->queue_lock)
384 __acquires(q->queue_lock)
388 lockdep_assert_held(q->queue_lock);
394 * The caller might be trying to drain @q before its
395 * elevator is initialized.
398 elv_drain_elevator(q);
400 blkcg_drain_queue(q);
403 * This function might be called on a queue which failed
404 * driver init after queue creation or is not yet fully
405 * active yet. Some drivers (e.g. fd and loop) get unhappy
406 * in such cases. Kick queue iff dispatch queue has
407 * something on it and @q has request_fn set.
409 if (!list_empty(&q->queue_head) && q->request_fn)
412 drain |= q->nr_rqs_elvpriv;
413 drain |= q->request_fn_active;
416 * Unfortunately, requests are queued at and tracked from
417 * multiple places and there's no single counter which can
418 * be drained. Check all the queues and counters.
421 drain |= !list_empty(&q->queue_head);
422 for (i = 0; i < 2; i++) {
423 drain |= q->nr_rqs[i];
424 drain |= q->in_flight[i];
425 drain |= !list_empty(&q->flush_queue[i]);
432 spin_unlock_irq(q->queue_lock);
436 spin_lock_irq(q->queue_lock);
440 * With queue marked dead, any woken up waiter will fail the
441 * allocation path, so the wakeup chaining is lost and we're
442 * left with hung waiters. We need to wake up those waiters.
445 struct request_list *rl;
447 blk_queue_for_each_rl(rl, q)
448 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
449 wake_up_all(&rl->wait[i]);
454 * blk_queue_bypass_start - enter queue bypass mode
455 * @q: queue of interest
457 * In bypass mode, only the dispatch FIFO queue of @q is used. This
458 * function makes @q enter bypass mode and drains all requests which were
459 * throttled or issued before. On return, it's guaranteed that no request
460 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
461 * inside queue or RCU read lock.
463 void blk_queue_bypass_start(struct request_queue *q)
467 spin_lock_irq(q->queue_lock);
468 drain = !q->bypass_depth++;
469 queue_flag_set(QUEUE_FLAG_BYPASS, q);
470 spin_unlock_irq(q->queue_lock);
473 spin_lock_irq(q->queue_lock);
474 __blk_drain_queue(q, false);
475 spin_unlock_irq(q->queue_lock);
477 /* ensure blk_queue_bypass() is %true inside RCU read lock */
481 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
484 * blk_queue_bypass_end - leave queue bypass mode
485 * @q: queue of interest
487 * Leave bypass mode and restore the normal queueing behavior.
489 void blk_queue_bypass_end(struct request_queue *q)
491 spin_lock_irq(q->queue_lock);
492 if (!--q->bypass_depth)
493 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
494 WARN_ON_ONCE(q->bypass_depth < 0);
495 spin_unlock_irq(q->queue_lock);
497 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
500 * blk_cleanup_queue - shutdown a request queue
501 * @q: request queue to shutdown
503 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
504 * put it. All future requests will be failed immediately with -ENODEV.
506 void blk_cleanup_queue(struct request_queue *q)
508 spinlock_t *lock = q->queue_lock;
510 /* mark @q DYING, no new request or merges will be allowed afterwards */
511 mutex_lock(&q->sysfs_lock);
512 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
516 * A dying queue is permanently in bypass mode till released. Note
517 * that, unlike blk_queue_bypass_start(), we aren't performing
518 * synchronize_rcu() after entering bypass mode to avoid the delay
519 * as some drivers create and destroy a lot of queues while
520 * probing. This is still safe because blk_release_queue() will be
521 * called only after the queue refcnt drops to zero and nothing,
522 * RCU or not, would be traversing the queue by then.
525 queue_flag_set(QUEUE_FLAG_BYPASS, q);
527 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
528 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
529 queue_flag_set(QUEUE_FLAG_DYING, q);
530 spin_unlock_irq(lock);
531 mutex_unlock(&q->sysfs_lock);
534 * Drain all requests queued before DYING marking. Set DEAD flag to
535 * prevent that q->request_fn() gets invoked after draining finished.
538 __blk_drain_queue(q, true);
539 queue_flag_set(QUEUE_FLAG_DEAD, q);
540 spin_unlock_irq(lock);
542 /* @q won't process any more request, flush async actions */
543 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
547 if (q->queue_lock != &q->__queue_lock)
548 q->queue_lock = &q->__queue_lock;
549 spin_unlock_irq(lock);
551 /* @q is and will stay empty, shutdown and put */
554 EXPORT_SYMBOL(blk_cleanup_queue);
556 int blk_init_rl(struct request_list *rl, struct request_queue *q,
559 if (unlikely(rl->rq_pool))
563 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
564 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
565 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
566 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
568 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
569 mempool_free_slab, request_cachep,
577 void blk_exit_rl(struct request_list *rl)
580 mempool_destroy(rl->rq_pool);
583 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
585 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
587 EXPORT_SYMBOL(blk_alloc_queue);
589 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
591 struct request_queue *q;
594 q = kmem_cache_alloc_node(blk_requestq_cachep,
595 gfp_mask | __GFP_ZERO, node_id);
599 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
603 q->backing_dev_info.ra_pages =
604 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
605 q->backing_dev_info.state = 0;
606 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
607 q->backing_dev_info.name = "block";
610 err = bdi_init(&q->backing_dev_info);
614 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
615 laptop_mode_timer_fn, (unsigned long) q);
616 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
617 INIT_LIST_HEAD(&q->queue_head);
618 INIT_LIST_HEAD(&q->timeout_list);
619 INIT_LIST_HEAD(&q->icq_list);
620 #ifdef CONFIG_BLK_CGROUP
621 INIT_LIST_HEAD(&q->blkg_list);
623 INIT_LIST_HEAD(&q->flush_queue[0]);
624 INIT_LIST_HEAD(&q->flush_queue[1]);
625 INIT_LIST_HEAD(&q->flush_data_in_flight);
626 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
628 kobject_init(&q->kobj, &blk_queue_ktype);
630 mutex_init(&q->sysfs_lock);
631 spin_lock_init(&q->__queue_lock);
634 * By default initialize queue_lock to internal lock and driver can
635 * override it later if need be.
637 q->queue_lock = &q->__queue_lock;
640 * A queue starts its life with bypass turned on to avoid
641 * unnecessary bypass on/off overhead and nasty surprises during
642 * init. The initial bypass will be finished when the queue is
643 * registered by blk_register_queue().
646 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
648 if (blkcg_init_queue(q))
654 ida_simple_remove(&blk_queue_ida, q->id);
656 kmem_cache_free(blk_requestq_cachep, q);
659 EXPORT_SYMBOL(blk_alloc_queue_node);
662 * blk_init_queue - prepare a request queue for use with a block device
663 * @rfn: The function to be called to process requests that have been
664 * placed on the queue.
665 * @lock: Request queue spin lock
668 * If a block device wishes to use the standard request handling procedures,
669 * which sorts requests and coalesces adjacent requests, then it must
670 * call blk_init_queue(). The function @rfn will be called when there
671 * are requests on the queue that need to be processed. If the device
672 * supports plugging, then @rfn may not be called immediately when requests
673 * are available on the queue, but may be called at some time later instead.
674 * Plugged queues are generally unplugged when a buffer belonging to one
675 * of the requests on the queue is needed, or due to memory pressure.
677 * @rfn is not required, or even expected, to remove all requests off the
678 * queue, but only as many as it can handle at a time. If it does leave
679 * requests on the queue, it is responsible for arranging that the requests
680 * get dealt with eventually.
682 * The queue spin lock must be held while manipulating the requests on the
683 * request queue; this lock will be taken also from interrupt context, so irq
684 * disabling is needed for it.
686 * Function returns a pointer to the initialized request queue, or %NULL if
690 * blk_init_queue() must be paired with a blk_cleanup_queue() call
691 * when the block device is deactivated (such as at module unload).
694 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
696 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
698 EXPORT_SYMBOL(blk_init_queue);
700 struct request_queue *
701 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
703 struct request_queue *uninit_q, *q;
705 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
709 q = blk_init_allocated_queue(uninit_q, rfn, lock);
711 blk_cleanup_queue(uninit_q);
715 EXPORT_SYMBOL(blk_init_queue_node);
717 struct request_queue *
718 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
724 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
728 q->prep_rq_fn = NULL;
729 q->unprep_rq_fn = NULL;
730 q->queue_flags |= QUEUE_FLAG_DEFAULT;
732 /* Override internal queue lock with supplied lock pointer */
734 q->queue_lock = lock;
737 * This also sets hw/phys segments, boundary and size
739 blk_queue_make_request(q, blk_queue_bio);
741 q->sg_reserved_size = INT_MAX;
744 if (elevator_init(q, NULL))
748 EXPORT_SYMBOL(blk_init_allocated_queue);
750 bool blk_get_queue(struct request_queue *q)
752 if (likely(!blk_queue_dying(q))) {
759 EXPORT_SYMBOL(blk_get_queue);
761 static inline void blk_free_request(struct request_list *rl, struct request *rq)
763 if (rq->cmd_flags & REQ_ELVPRIV) {
764 elv_put_request(rl->q, rq);
766 put_io_context(rq->elv.icq->ioc);
769 mempool_free(rq, rl->rq_pool);
773 * ioc_batching returns true if the ioc is a valid batching request and
774 * should be given priority access to a request.
776 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
782 * Make sure the process is able to allocate at least 1 request
783 * even if the batch times out, otherwise we could theoretically
786 return ioc->nr_batch_requests == q->nr_batching ||
787 (ioc->nr_batch_requests > 0
788 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
792 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
793 * will cause the process to be a "batcher" on all queues in the system. This
794 * is the behaviour we want though - once it gets a wakeup it should be given
797 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
799 if (!ioc || ioc_batching(q, ioc))
802 ioc->nr_batch_requests = q->nr_batching;
803 ioc->last_waited = jiffies;
806 static void __freed_request(struct request_list *rl, int sync)
808 struct request_queue *q = rl->q;
811 * bdi isn't aware of blkcg yet. As all async IOs end up root
812 * blkcg anyway, just use root blkcg state.
814 if (rl == &q->root_rl &&
815 rl->count[sync] < queue_congestion_off_threshold(q))
816 blk_clear_queue_congested(q, sync);
818 if (rl->count[sync] + 1 <= q->nr_requests) {
819 if (waitqueue_active(&rl->wait[sync]))
820 wake_up(&rl->wait[sync]);
822 blk_clear_rl_full(rl, sync);
827 * A request has just been released. Account for it, update the full and
828 * congestion status, wake up any waiters. Called under q->queue_lock.
830 static void freed_request(struct request_list *rl, unsigned int flags)
832 struct request_queue *q = rl->q;
833 int sync = rw_is_sync(flags);
837 if (flags & REQ_ELVPRIV)
840 __freed_request(rl, sync);
842 if (unlikely(rl->starved[sync ^ 1]))
843 __freed_request(rl, sync ^ 1);
847 * Determine if elevator data should be initialized when allocating the
848 * request associated with @bio.
850 static bool blk_rq_should_init_elevator(struct bio *bio)
856 * Flush requests do not use the elevator so skip initialization.
857 * This allows a request to share the flush and elevator data.
859 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
866 * rq_ioc - determine io_context for request allocation
867 * @bio: request being allocated is for this bio (can be %NULL)
869 * Determine io_context to use for request allocation for @bio. May return
870 * %NULL if %current->io_context doesn't exist.
872 static struct io_context *rq_ioc(struct bio *bio)
874 #ifdef CONFIG_BLK_CGROUP
875 if (bio && bio->bi_ioc)
878 return current->io_context;
882 * __get_request - get a free request
883 * @rl: request list to allocate from
884 * @rw_flags: RW and SYNC flags
885 * @bio: bio to allocate request for (can be %NULL)
886 * @gfp_mask: allocation mask
888 * Get a free request from @q. This function may fail under memory
889 * pressure or if @q is dead.
891 * Must be callled with @q->queue_lock held and,
892 * Returns %NULL on failure, with @q->queue_lock held.
893 * Returns !%NULL on success, with @q->queue_lock *not held*.
895 static struct request *__get_request(struct request_list *rl, int rw_flags,
896 struct bio *bio, gfp_t gfp_mask)
898 struct request_queue *q = rl->q;
900 struct elevator_type *et = q->elevator->type;
901 struct io_context *ioc = rq_ioc(bio);
902 struct io_cq *icq = NULL;
903 const bool is_sync = rw_is_sync(rw_flags) != 0;
906 if (unlikely(blk_queue_dying(q)))
909 may_queue = elv_may_queue(q, rw_flags);
910 if (may_queue == ELV_MQUEUE_NO)
913 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
914 if (rl->count[is_sync]+1 >= q->nr_requests) {
916 * The queue will fill after this allocation, so set
917 * it as full, and mark this process as "batching".
918 * This process will be allowed to complete a batch of
919 * requests, others will be blocked.
921 if (!blk_rl_full(rl, is_sync)) {
922 ioc_set_batching(q, ioc);
923 blk_set_rl_full(rl, is_sync);
925 if (may_queue != ELV_MQUEUE_MUST
926 && !ioc_batching(q, ioc)) {
928 * The queue is full and the allocating
929 * process is not a "batcher", and not
930 * exempted by the IO scheduler
937 * bdi isn't aware of blkcg yet. As all async IOs end up
938 * root blkcg anyway, just use root blkcg state.
940 if (rl == &q->root_rl)
941 blk_set_queue_congested(q, is_sync);
945 * Only allow batching queuers to allocate up to 50% over the defined
946 * limit of requests, otherwise we could have thousands of requests
947 * allocated with any setting of ->nr_requests
949 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
952 q->nr_rqs[is_sync]++;
953 rl->count[is_sync]++;
954 rl->starved[is_sync] = 0;
957 * Decide whether the new request will be managed by elevator. If
958 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
959 * prevent the current elevator from being destroyed until the new
960 * request is freed. This guarantees icq's won't be destroyed and
961 * makes creating new ones safe.
963 * Also, lookup icq while holding queue_lock. If it doesn't exist,
964 * it will be created after releasing queue_lock.
966 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
967 rw_flags |= REQ_ELVPRIV;
969 if (et->icq_cache && ioc)
970 icq = ioc_lookup_icq(ioc, q);
973 if (blk_queue_io_stat(q))
974 rw_flags |= REQ_IO_STAT;
975 spin_unlock_irq(q->queue_lock);
977 /* allocate and init request */
978 rq = mempool_alloc(rl->rq_pool, gfp_mask);
983 blk_rq_set_rl(rq, rl);
984 rq->cmd_flags = rw_flags | REQ_ALLOCED;
987 if (rw_flags & REQ_ELVPRIV) {
988 if (unlikely(et->icq_cache && !icq)) {
990 icq = ioc_create_icq(ioc, q, gfp_mask);
996 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
999 /* @rq->elv.icq holds io_context until @rq is freed */
1001 get_io_context(icq->ioc);
1005 * ioc may be NULL here, and ioc_batching will be false. That's
1006 * OK, if the queue is under the request limit then requests need
1007 * not count toward the nr_batch_requests limit. There will always
1008 * be some limit enforced by BLK_BATCH_TIME.
1010 if (ioc_batching(q, ioc))
1011 ioc->nr_batch_requests--;
1013 trace_block_getrq(q, bio, rw_flags & 1);
1018 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1019 * and may fail indefinitely under memory pressure and thus
1020 * shouldn't stall IO. Treat this request as !elvpriv. This will
1021 * disturb iosched and blkcg but weird is bettern than dead.
1023 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1024 dev_name(q->backing_dev_info.dev));
1026 rq->cmd_flags &= ~REQ_ELVPRIV;
1029 spin_lock_irq(q->queue_lock);
1030 q->nr_rqs_elvpriv--;
1031 spin_unlock_irq(q->queue_lock);
1036 * Allocation failed presumably due to memory. Undo anything we
1037 * might have messed up.
1039 * Allocating task should really be put onto the front of the wait
1040 * queue, but this is pretty rare.
1042 spin_lock_irq(q->queue_lock);
1043 freed_request(rl, rw_flags);
1046 * in the very unlikely event that allocation failed and no
1047 * requests for this direction was pending, mark us starved so that
1048 * freeing of a request in the other direction will notice
1049 * us. another possible fix would be to split the rq mempool into
1053 if (unlikely(rl->count[is_sync] == 0))
1054 rl->starved[is_sync] = 1;
1059 * get_request - get a free request
1060 * @q: request_queue to allocate request from
1061 * @rw_flags: RW and SYNC flags
1062 * @bio: bio to allocate request for (can be %NULL)
1063 * @gfp_mask: allocation mask
1065 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1066 * function keeps retrying under memory pressure and fails iff @q is dead.
1068 * Must be callled with @q->queue_lock held and,
1069 * Returns %NULL on failure, with @q->queue_lock held.
1070 * Returns !%NULL on success, with @q->queue_lock *not held*.
1072 static struct request *get_request(struct request_queue *q, int rw_flags,
1073 struct bio *bio, gfp_t gfp_mask)
1075 const bool is_sync = rw_is_sync(rw_flags) != 0;
1077 struct request_list *rl;
1080 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1082 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1086 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1091 /* wait on @rl and retry */
1092 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1093 TASK_UNINTERRUPTIBLE);
1095 trace_block_sleeprq(q, bio, rw_flags & 1);
1097 spin_unlock_irq(q->queue_lock);
1101 * After sleeping, we become a "batching" process and will be able
1102 * to allocate at least one request, and up to a big batch of them
1103 * for a small period time. See ioc_batching, ioc_set_batching
1105 ioc_set_batching(q, current->io_context);
1107 spin_lock_irq(q->queue_lock);
1108 finish_wait(&rl->wait[is_sync], &wait);
1113 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1117 BUG_ON(rw != READ && rw != WRITE);
1119 /* create ioc upfront */
1120 create_io_context(gfp_mask, q->node);
1122 spin_lock_irq(q->queue_lock);
1123 rq = get_request(q, rw, NULL, gfp_mask);
1125 spin_unlock_irq(q->queue_lock);
1126 /* q->queue_lock is unlocked at this point */
1130 EXPORT_SYMBOL(blk_get_request);
1133 * blk_make_request - given a bio, allocate a corresponding struct request.
1134 * @q: target request queue
1135 * @bio: The bio describing the memory mappings that will be submitted for IO.
1136 * It may be a chained-bio properly constructed by block/bio layer.
1137 * @gfp_mask: gfp flags to be used for memory allocation
1139 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1140 * type commands. Where the struct request needs to be farther initialized by
1141 * the caller. It is passed a &struct bio, which describes the memory info of
1144 * The caller of blk_make_request must make sure that bi_io_vec
1145 * are set to describe the memory buffers. That bio_data_dir() will return
1146 * the needed direction of the request. (And all bio's in the passed bio-chain
1147 * are properly set accordingly)
1149 * If called under none-sleepable conditions, mapped bio buffers must not
1150 * need bouncing, by calling the appropriate masked or flagged allocator,
1151 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1154 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1155 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1156 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1157 * completion of a bio that hasn't been submitted yet, thus resulting in a
1158 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1159 * of bio_alloc(), as that avoids the mempool deadlock.
1160 * If possible a big IO should be split into smaller parts when allocation
1161 * fails. Partial allocation should not be an error, or you risk a live-lock.
1163 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1166 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1169 return ERR_PTR(-ENOMEM);
1172 struct bio *bounce_bio = bio;
1175 blk_queue_bounce(q, &bounce_bio);
1176 ret = blk_rq_append_bio(q, rq, bounce_bio);
1177 if (unlikely(ret)) {
1178 blk_put_request(rq);
1179 return ERR_PTR(ret);
1185 EXPORT_SYMBOL(blk_make_request);
1188 * blk_requeue_request - put a request back on queue
1189 * @q: request queue where request should be inserted
1190 * @rq: request to be inserted
1193 * Drivers often keep queueing requests until the hardware cannot accept
1194 * more, when that condition happens we need to put the request back
1195 * on the queue. Must be called with queue lock held.
1197 void blk_requeue_request(struct request_queue *q, struct request *rq)
1199 blk_delete_timer(rq);
1200 blk_clear_rq_complete(rq);
1201 trace_block_rq_requeue(q, rq);
1203 if (blk_rq_tagged(rq))
1204 blk_queue_end_tag(q, rq);
1206 BUG_ON(blk_queued_rq(rq));
1208 elv_requeue_request(q, rq);
1210 EXPORT_SYMBOL(blk_requeue_request);
1212 static void add_acct_request(struct request_queue *q, struct request *rq,
1215 drive_stat_acct(rq, 1);
1216 __elv_add_request(q, rq, where);
1219 static void part_round_stats_single(int cpu, struct hd_struct *part,
1222 if (now == part->stamp)
1225 if (part_in_flight(part)) {
1226 __part_stat_add(cpu, part, time_in_queue,
1227 part_in_flight(part) * (now - part->stamp));
1228 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1234 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1235 * @cpu: cpu number for stats access
1236 * @part: target partition
1238 * The average IO queue length and utilisation statistics are maintained
1239 * by observing the current state of the queue length and the amount of
1240 * time it has been in this state for.
1242 * Normally, that accounting is done on IO completion, but that can result
1243 * in more than a second's worth of IO being accounted for within any one
1244 * second, leading to >100% utilisation. To deal with that, we call this
1245 * function to do a round-off before returning the results when reading
1246 * /proc/diskstats. This accounts immediately for all queue usage up to
1247 * the current jiffies and restarts the counters again.
1249 void part_round_stats(int cpu, struct hd_struct *part)
1251 unsigned long now = jiffies;
1254 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1255 part_round_stats_single(cpu, part, now);
1257 EXPORT_SYMBOL_GPL(part_round_stats);
1259 #ifdef CONFIG_PM_RUNTIME
1260 static void blk_pm_put_request(struct request *rq)
1262 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1263 pm_runtime_mark_last_busy(rq->q->dev);
1266 static inline void blk_pm_put_request(struct request *rq) {}
1270 * queue lock must be held
1272 void __blk_put_request(struct request_queue *q, struct request *req)
1276 if (unlikely(--req->ref_count))
1279 blk_pm_put_request(req);
1281 elv_completed_request(q, req);
1283 /* this is a bio leak */
1284 WARN_ON(req->bio != NULL);
1287 * Request may not have originated from ll_rw_blk. if not,
1288 * it didn't come out of our reserved rq pools
1290 if (req->cmd_flags & REQ_ALLOCED) {
1291 unsigned int flags = req->cmd_flags;
1292 struct request_list *rl = blk_rq_rl(req);
1294 BUG_ON(!list_empty(&req->queuelist));
1295 BUG_ON(!hlist_unhashed(&req->hash));
1297 blk_free_request(rl, req);
1298 freed_request(rl, flags);
1302 EXPORT_SYMBOL_GPL(__blk_put_request);
1304 void blk_put_request(struct request *req)
1306 unsigned long flags;
1307 struct request_queue *q = req->q;
1309 spin_lock_irqsave(q->queue_lock, flags);
1310 __blk_put_request(q, req);
1311 spin_unlock_irqrestore(q->queue_lock, flags);
1313 EXPORT_SYMBOL(blk_put_request);
1316 * blk_add_request_payload - add a payload to a request
1317 * @rq: request to update
1318 * @page: page backing the payload
1319 * @len: length of the payload.
1321 * This allows to later add a payload to an already submitted request by
1322 * a block driver. The driver needs to take care of freeing the payload
1325 * Note that this is a quite horrible hack and nothing but handling of
1326 * discard requests should ever use it.
1328 void blk_add_request_payload(struct request *rq, struct page *page,
1331 struct bio *bio = rq->bio;
1333 bio->bi_io_vec->bv_page = page;
1334 bio->bi_io_vec->bv_offset = 0;
1335 bio->bi_io_vec->bv_len = len;
1339 bio->bi_phys_segments = 1;
1341 rq->__data_len = rq->resid_len = len;
1342 rq->nr_phys_segments = 1;
1343 rq->buffer = bio_data(bio);
1345 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1347 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1350 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1352 if (!ll_back_merge_fn(q, req, bio))
1355 trace_block_bio_backmerge(q, req, bio);
1357 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1358 blk_rq_set_mixed_merge(req);
1360 req->biotail->bi_next = bio;
1362 req->__data_len += bio->bi_size;
1363 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1365 drive_stat_acct(req, 0);
1369 static bool bio_attempt_front_merge(struct request_queue *q,
1370 struct request *req, struct bio *bio)
1372 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1374 if (!ll_front_merge_fn(q, req, bio))
1377 trace_block_bio_frontmerge(q, req, bio);
1379 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1380 blk_rq_set_mixed_merge(req);
1382 bio->bi_next = req->bio;
1386 * may not be valid. if the low level driver said
1387 * it didn't need a bounce buffer then it better
1388 * not touch req->buffer either...
1390 req->buffer = bio_data(bio);
1391 req->__sector = bio->bi_sector;
1392 req->__data_len += bio->bi_size;
1393 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1395 drive_stat_acct(req, 0);
1400 * attempt_plug_merge - try to merge with %current's plugged list
1401 * @q: request_queue new bio is being queued at
1402 * @bio: new bio being queued
1403 * @request_count: out parameter for number of traversed plugged requests
1405 * Determine whether @bio being queued on @q can be merged with a request
1406 * on %current's plugged list. Returns %true if merge was successful,
1409 * Plugging coalesces IOs from the same issuer for the same purpose without
1410 * going through @q->queue_lock. As such it's more of an issuing mechanism
1411 * than scheduling, and the request, while may have elvpriv data, is not
1412 * added on the elevator at this point. In addition, we don't have
1413 * reliable access to the elevator outside queue lock. Only check basic
1414 * merging parameters without querying the elevator.
1416 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1417 unsigned int *request_count)
1419 struct blk_plug *plug;
1423 plug = current->plug;
1428 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1434 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1437 el_ret = blk_try_merge(rq, bio);
1438 if (el_ret == ELEVATOR_BACK_MERGE) {
1439 ret = bio_attempt_back_merge(q, rq, bio);
1442 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1443 ret = bio_attempt_front_merge(q, rq, bio);
1452 void init_request_from_bio(struct request *req, struct bio *bio)
1454 req->cmd_type = REQ_TYPE_FS;
1456 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1457 if (bio->bi_rw & REQ_RAHEAD)
1458 req->cmd_flags |= REQ_FAILFAST_MASK;
1461 req->__sector = bio->bi_sector;
1462 req->ioprio = bio_prio(bio);
1463 blk_rq_bio_prep(req->q, req, bio);
1466 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1468 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1469 struct blk_plug *plug;
1470 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1471 struct request *req;
1472 unsigned int request_count = 0;
1475 * low level driver can indicate that it wants pages above a
1476 * certain limit bounced to low memory (ie for highmem, or even
1477 * ISA dma in theory)
1479 blk_queue_bounce(q, &bio);
1481 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1482 bio_endio(bio, -EIO);
1486 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1487 spin_lock_irq(q->queue_lock);
1488 where = ELEVATOR_INSERT_FLUSH;
1493 * Check if we can merge with the plugged list before grabbing
1496 if (attempt_plug_merge(q, bio, &request_count))
1499 spin_lock_irq(q->queue_lock);
1501 el_ret = elv_merge(q, &req, bio);
1502 if (el_ret == ELEVATOR_BACK_MERGE) {
1503 if (bio_attempt_back_merge(q, req, bio)) {
1504 elv_bio_merged(q, req, bio);
1505 if (!attempt_back_merge(q, req))
1506 elv_merged_request(q, req, el_ret);
1509 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1510 if (bio_attempt_front_merge(q, req, bio)) {
1511 elv_bio_merged(q, req, bio);
1512 if (!attempt_front_merge(q, req))
1513 elv_merged_request(q, req, el_ret);
1520 * This sync check and mask will be re-done in init_request_from_bio(),
1521 * but we need to set it earlier to expose the sync flag to the
1522 * rq allocator and io schedulers.
1524 rw_flags = bio_data_dir(bio);
1526 rw_flags |= REQ_SYNC;
1529 * Grab a free request. This is might sleep but can not fail.
1530 * Returns with the queue unlocked.
1532 req = get_request(q, rw_flags, bio, GFP_NOIO);
1533 if (unlikely(!req)) {
1534 bio_endio(bio, -ENODEV); /* @q is dead */
1539 * After dropping the lock and possibly sleeping here, our request
1540 * may now be mergeable after it had proven unmergeable (above).
1541 * We don't worry about that case for efficiency. It won't happen
1542 * often, and the elevators are able to handle it.
1544 init_request_from_bio(req, bio);
1546 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1547 req->cpu = raw_smp_processor_id();
1549 plug = current->plug;
1552 * If this is the first request added after a plug, fire
1553 * of a plug trace. If others have been added before, check
1554 * if we have multiple devices in this plug. If so, make a
1555 * note to sort the list before dispatch.
1557 if (list_empty(&plug->list))
1558 trace_block_plug(q);
1560 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1561 blk_flush_plug_list(plug, false);
1562 trace_block_plug(q);
1565 list_add_tail(&req->queuelist, &plug->list);
1566 drive_stat_acct(req, 1);
1568 spin_lock_irq(q->queue_lock);
1569 add_acct_request(q, req, where);
1572 spin_unlock_irq(q->queue_lock);
1575 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1578 * If bio->bi_dev is a partition, remap the location
1580 static inline void blk_partition_remap(struct bio *bio)
1582 struct block_device *bdev = bio->bi_bdev;
1584 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1585 struct hd_struct *p = bdev->bd_part;
1587 bio->bi_sector += p->start_sect;
1588 bio->bi_bdev = bdev->bd_contains;
1590 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1592 bio->bi_sector - p->start_sect);
1596 static void handle_bad_sector(struct bio *bio)
1598 char b[BDEVNAME_SIZE];
1600 printk(KERN_INFO "attempt to access beyond end of device\n");
1601 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1602 bdevname(bio->bi_bdev, b),
1604 (unsigned long long)bio_end_sector(bio),
1605 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1607 set_bit(BIO_EOF, &bio->bi_flags);
1610 #ifdef CONFIG_FAIL_MAKE_REQUEST
1612 static DECLARE_FAULT_ATTR(fail_make_request);
1614 static int __init setup_fail_make_request(char *str)
1616 return setup_fault_attr(&fail_make_request, str);
1618 __setup("fail_make_request=", setup_fail_make_request);
1620 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1622 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1625 static int __init fail_make_request_debugfs(void)
1627 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1628 NULL, &fail_make_request);
1630 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1633 late_initcall(fail_make_request_debugfs);
1635 #else /* CONFIG_FAIL_MAKE_REQUEST */
1637 static inline bool should_fail_request(struct hd_struct *part,
1643 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1646 * Check whether this bio extends beyond the end of the device.
1648 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1655 /* Test device or partition size, when known. */
1656 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1658 sector_t sector = bio->bi_sector;
1660 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1662 * This may well happen - the kernel calls bread()
1663 * without checking the size of the device, e.g., when
1664 * mounting a device.
1666 handle_bad_sector(bio);
1674 static noinline_for_stack bool
1675 generic_make_request_checks(struct bio *bio)
1677 struct request_queue *q;
1678 int nr_sectors = bio_sectors(bio);
1680 char b[BDEVNAME_SIZE];
1681 struct hd_struct *part;
1685 if (bio_check_eod(bio, nr_sectors))
1688 q = bdev_get_queue(bio->bi_bdev);
1691 "generic_make_request: Trying to access "
1692 "nonexistent block-device %s (%Lu)\n",
1693 bdevname(bio->bi_bdev, b),
1694 (long long) bio->bi_sector);
1698 if (likely(bio_is_rw(bio) &&
1699 nr_sectors > queue_max_hw_sectors(q))) {
1700 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1701 bdevname(bio->bi_bdev, b),
1703 queue_max_hw_sectors(q));
1707 part = bio->bi_bdev->bd_part;
1708 if (should_fail_request(part, bio->bi_size) ||
1709 should_fail_request(&part_to_disk(part)->part0,
1714 * If this device has partitions, remap block n
1715 * of partition p to block n+start(p) of the disk.
1717 blk_partition_remap(bio);
1719 if (bio_check_eod(bio, nr_sectors))
1723 * Filter flush bio's early so that make_request based
1724 * drivers without flush support don't have to worry
1727 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1728 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1735 if ((bio->bi_rw & REQ_DISCARD) &&
1736 (!blk_queue_discard(q) ||
1737 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1742 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1748 * Various block parts want %current->io_context and lazy ioc
1749 * allocation ends up trading a lot of pain for a small amount of
1750 * memory. Just allocate it upfront. This may fail and block
1751 * layer knows how to live with it.
1753 create_io_context(GFP_ATOMIC, q->node);
1755 if (blk_throtl_bio(q, bio))
1756 return false; /* throttled, will be resubmitted later */
1758 trace_block_bio_queue(q, bio);
1762 bio_endio(bio, err);
1767 * generic_make_request - hand a buffer to its device driver for I/O
1768 * @bio: The bio describing the location in memory and on the device.
1770 * generic_make_request() is used to make I/O requests of block
1771 * devices. It is passed a &struct bio, which describes the I/O that needs
1774 * generic_make_request() does not return any status. The
1775 * success/failure status of the request, along with notification of
1776 * completion, is delivered asynchronously through the bio->bi_end_io
1777 * function described (one day) else where.
1779 * The caller of generic_make_request must make sure that bi_io_vec
1780 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1781 * set to describe the device address, and the
1782 * bi_end_io and optionally bi_private are set to describe how
1783 * completion notification should be signaled.
1785 * generic_make_request and the drivers it calls may use bi_next if this
1786 * bio happens to be merged with someone else, and may resubmit the bio to
1787 * a lower device by calling into generic_make_request recursively, which
1788 * means the bio should NOT be touched after the call to ->make_request_fn.
1790 void generic_make_request(struct bio *bio)
1792 struct bio_list bio_list_on_stack;
1794 if (!generic_make_request_checks(bio))
1798 * We only want one ->make_request_fn to be active at a time, else
1799 * stack usage with stacked devices could be a problem. So use
1800 * current->bio_list to keep a list of requests submited by a
1801 * make_request_fn function. current->bio_list is also used as a
1802 * flag to say if generic_make_request is currently active in this
1803 * task or not. If it is NULL, then no make_request is active. If
1804 * it is non-NULL, then a make_request is active, and new requests
1805 * should be added at the tail
1807 if (current->bio_list) {
1808 bio_list_add(current->bio_list, bio);
1812 /* following loop may be a bit non-obvious, and so deserves some
1814 * Before entering the loop, bio->bi_next is NULL (as all callers
1815 * ensure that) so we have a list with a single bio.
1816 * We pretend that we have just taken it off a longer list, so
1817 * we assign bio_list to a pointer to the bio_list_on_stack,
1818 * thus initialising the bio_list of new bios to be
1819 * added. ->make_request() may indeed add some more bios
1820 * through a recursive call to generic_make_request. If it
1821 * did, we find a non-NULL value in bio_list and re-enter the loop
1822 * from the top. In this case we really did just take the bio
1823 * of the top of the list (no pretending) and so remove it from
1824 * bio_list, and call into ->make_request() again.
1826 BUG_ON(bio->bi_next);
1827 bio_list_init(&bio_list_on_stack);
1828 current->bio_list = &bio_list_on_stack;
1830 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1832 q->make_request_fn(q, bio);
1834 bio = bio_list_pop(current->bio_list);
1836 current->bio_list = NULL; /* deactivate */
1838 EXPORT_SYMBOL(generic_make_request);
1841 * submit_bio - submit a bio to the block device layer for I/O
1842 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1843 * @bio: The &struct bio which describes the I/O
1845 * submit_bio() is very similar in purpose to generic_make_request(), and
1846 * uses that function to do most of the work. Both are fairly rough
1847 * interfaces; @bio must be presetup and ready for I/O.
1850 void submit_bio(int rw, struct bio *bio)
1855 * If it's a regular read/write or a barrier with data attached,
1856 * go through the normal accounting stuff before submission.
1858 if (bio_has_data(bio)) {
1861 if (unlikely(rw & REQ_WRITE_SAME))
1862 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1864 count = bio_sectors(bio);
1867 count_vm_events(PGPGOUT, count);
1869 task_io_account_read(bio->bi_size);
1870 count_vm_events(PGPGIN, count);
1873 if (unlikely(block_dump)) {
1874 char b[BDEVNAME_SIZE];
1875 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1876 current->comm, task_pid_nr(current),
1877 (rw & WRITE) ? "WRITE" : "READ",
1878 (unsigned long long)bio->bi_sector,
1879 bdevname(bio->bi_bdev, b),
1884 generic_make_request(bio);
1886 EXPORT_SYMBOL(submit_bio);
1889 * blk_rq_check_limits - Helper function to check a request for the queue limit
1891 * @rq: the request being checked
1894 * @rq may have been made based on weaker limitations of upper-level queues
1895 * in request stacking drivers, and it may violate the limitation of @q.
1896 * Since the block layer and the underlying device driver trust @rq
1897 * after it is inserted to @q, it should be checked against @q before
1898 * the insertion using this generic function.
1900 * This function should also be useful for request stacking drivers
1901 * in some cases below, so export this function.
1902 * Request stacking drivers like request-based dm may change the queue
1903 * limits while requests are in the queue (e.g. dm's table swapping).
1904 * Such request stacking drivers should check those requests agaist
1905 * the new queue limits again when they dispatch those requests,
1906 * although such checkings are also done against the old queue limits
1907 * when submitting requests.
1909 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1911 if (!rq_mergeable(rq))
1914 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1915 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1920 * queue's settings related to segment counting like q->bounce_pfn
1921 * may differ from that of other stacking queues.
1922 * Recalculate it to check the request correctly on this queue's
1925 blk_recalc_rq_segments(rq);
1926 if (rq->nr_phys_segments > queue_max_segments(q)) {
1927 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1933 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1936 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1937 * @q: the queue to submit the request
1938 * @rq: the request being queued
1940 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1942 unsigned long flags;
1943 int where = ELEVATOR_INSERT_BACK;
1945 if (blk_rq_check_limits(q, rq))
1949 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1952 spin_lock_irqsave(q->queue_lock, flags);
1953 if (unlikely(blk_queue_dying(q))) {
1954 spin_unlock_irqrestore(q->queue_lock, flags);
1959 * Submitting request must be dequeued before calling this function
1960 * because it will be linked to another request_queue
1962 BUG_ON(blk_queued_rq(rq));
1964 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1965 where = ELEVATOR_INSERT_FLUSH;
1967 add_acct_request(q, rq, where);
1968 if (where == ELEVATOR_INSERT_FLUSH)
1970 spin_unlock_irqrestore(q->queue_lock, flags);
1974 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1977 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1978 * @rq: request to examine
1981 * A request could be merge of IOs which require different failure
1982 * handling. This function determines the number of bytes which
1983 * can be failed from the beginning of the request without
1984 * crossing into area which need to be retried further.
1987 * The number of bytes to fail.
1990 * queue_lock must be held.
1992 unsigned int blk_rq_err_bytes(const struct request *rq)
1994 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1995 unsigned int bytes = 0;
1998 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1999 return blk_rq_bytes(rq);
2002 * Currently the only 'mixing' which can happen is between
2003 * different fastfail types. We can safely fail portions
2004 * which have all the failfast bits that the first one has -
2005 * the ones which are at least as eager to fail as the first
2008 for (bio = rq->bio; bio; bio = bio->bi_next) {
2009 if ((bio->bi_rw & ff) != ff)
2011 bytes += bio->bi_size;
2014 /* this could lead to infinite loop */
2015 BUG_ON(blk_rq_bytes(rq) && !bytes);
2018 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2020 static void blk_account_io_completion(struct request *req, unsigned int bytes)
2022 if (blk_do_io_stat(req)) {
2023 const int rw = rq_data_dir(req);
2024 struct hd_struct *part;
2027 cpu = part_stat_lock();
2029 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2034 static void blk_account_io_done(struct request *req)
2037 * Account IO completion. flush_rq isn't accounted as a
2038 * normal IO on queueing nor completion. Accounting the
2039 * containing request is enough.
2041 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2042 unsigned long duration = jiffies - req->start_time;
2043 const int rw = rq_data_dir(req);
2044 struct hd_struct *part;
2047 cpu = part_stat_lock();
2050 part_stat_inc(cpu, part, ios[rw]);
2051 part_stat_add(cpu, part, ticks[rw], duration);
2052 part_round_stats(cpu, part);
2053 part_dec_in_flight(part, rw);
2055 hd_struct_put(part);
2060 #ifdef CONFIG_PM_RUNTIME
2062 * Don't process normal requests when queue is suspended
2063 * or in the process of suspending/resuming
2065 static struct request *blk_pm_peek_request(struct request_queue *q,
2068 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2069 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2075 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2083 * blk_peek_request - peek at the top of a request queue
2084 * @q: request queue to peek at
2087 * Return the request at the top of @q. The returned request
2088 * should be started using blk_start_request() before LLD starts
2092 * Pointer to the request at the top of @q if available. Null
2096 * queue_lock must be held.
2098 struct request *blk_peek_request(struct request_queue *q)
2103 while ((rq = __elv_next_request(q)) != NULL) {
2105 rq = blk_pm_peek_request(q, rq);
2109 if (!(rq->cmd_flags & REQ_STARTED)) {
2111 * This is the first time the device driver
2112 * sees this request (possibly after
2113 * requeueing). Notify IO scheduler.
2115 if (rq->cmd_flags & REQ_SORTED)
2116 elv_activate_rq(q, rq);
2119 * just mark as started even if we don't start
2120 * it, a request that has been delayed should
2121 * not be passed by new incoming requests
2123 rq->cmd_flags |= REQ_STARTED;
2124 trace_block_rq_issue(q, rq);
2127 if (!q->boundary_rq || q->boundary_rq == rq) {
2128 q->end_sector = rq_end_sector(rq);
2129 q->boundary_rq = NULL;
2132 if (rq->cmd_flags & REQ_DONTPREP)
2135 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2137 * make sure space for the drain appears we
2138 * know we can do this because max_hw_segments
2139 * has been adjusted to be one fewer than the
2142 rq->nr_phys_segments++;
2148 ret = q->prep_rq_fn(q, rq);
2149 if (ret == BLKPREP_OK) {
2151 } else if (ret == BLKPREP_DEFER) {
2153 * the request may have been (partially) prepped.
2154 * we need to keep this request in the front to
2155 * avoid resource deadlock. REQ_STARTED will
2156 * prevent other fs requests from passing this one.
2158 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2159 !(rq->cmd_flags & REQ_DONTPREP)) {
2161 * remove the space for the drain we added
2162 * so that we don't add it again
2164 --rq->nr_phys_segments;
2169 } else if (ret == BLKPREP_KILL) {
2170 rq->cmd_flags |= REQ_QUIET;
2172 * Mark this request as started so we don't trigger
2173 * any debug logic in the end I/O path.
2175 blk_start_request(rq);
2176 __blk_end_request_all(rq, -EIO);
2178 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2185 EXPORT_SYMBOL(blk_peek_request);
2187 void blk_dequeue_request(struct request *rq)
2189 struct request_queue *q = rq->q;
2191 BUG_ON(list_empty(&rq->queuelist));
2192 BUG_ON(ELV_ON_HASH(rq));
2194 list_del_init(&rq->queuelist);
2197 * the time frame between a request being removed from the lists
2198 * and to it is freed is accounted as io that is in progress at
2201 if (blk_account_rq(rq)) {
2202 q->in_flight[rq_is_sync(rq)]++;
2203 set_io_start_time_ns(rq);
2208 * blk_start_request - start request processing on the driver
2209 * @req: request to dequeue
2212 * Dequeue @req and start timeout timer on it. This hands off the
2213 * request to the driver.
2215 * Block internal functions which don't want to start timer should
2216 * call blk_dequeue_request().
2219 * queue_lock must be held.
2221 void blk_start_request(struct request *req)
2223 blk_dequeue_request(req);
2226 * We are now handing the request to the hardware, initialize
2227 * resid_len to full count and add the timeout handler.
2229 req->resid_len = blk_rq_bytes(req);
2230 if (unlikely(blk_bidi_rq(req)))
2231 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2235 EXPORT_SYMBOL(blk_start_request);
2238 * blk_fetch_request - fetch a request from a request queue
2239 * @q: request queue to fetch a request from
2242 * Return the request at the top of @q. The request is started on
2243 * return and LLD can start processing it immediately.
2246 * Pointer to the request at the top of @q if available. Null
2250 * queue_lock must be held.
2252 struct request *blk_fetch_request(struct request_queue *q)
2256 rq = blk_peek_request(q);
2258 blk_start_request(rq);
2261 EXPORT_SYMBOL(blk_fetch_request);
2264 * blk_update_request - Special helper function for request stacking drivers
2265 * @req: the request being processed
2266 * @error: %0 for success, < %0 for error
2267 * @nr_bytes: number of bytes to complete @req
2270 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2271 * the request structure even if @req doesn't have leftover.
2272 * If @req has leftover, sets it up for the next range of segments.
2274 * This special helper function is only for request stacking drivers
2275 * (e.g. request-based dm) so that they can handle partial completion.
2276 * Actual device drivers should use blk_end_request instead.
2278 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2279 * %false return from this function.
2282 * %false - this request doesn't have any more data
2283 * %true - this request has more data
2285 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes,
2286 struct batch_complete *batch)
2293 trace_block_rq_complete(req->q, req);
2296 * For fs requests, rq is just carrier of independent bio's
2297 * and each partial completion should be handled separately.
2298 * Reset per-request error on each partial completion.
2300 * TODO: tj: This is too subtle. It would be better to let
2301 * low level drivers do what they see fit.
2303 if (req->cmd_type == REQ_TYPE_FS)
2306 if (error && req->cmd_type == REQ_TYPE_FS &&
2307 !(req->cmd_flags & REQ_QUIET)) {
2312 error_type = "recoverable transport";
2315 error_type = "critical target";
2318 error_type = "critical nexus";
2325 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2326 error_type, req->rq_disk ?
2327 req->rq_disk->disk_name : "?",
2328 (unsigned long long)blk_rq_pos(req));
2332 blk_account_io_completion(req, nr_bytes);
2336 struct bio *bio = req->bio;
2337 unsigned bio_bytes = min(bio->bi_size, nr_bytes);
2339 if (bio_bytes == bio->bi_size)
2340 req->bio = bio->bi_next;
2342 req_bio_endio(req, bio, bio_bytes, error, batch);
2344 total_bytes += bio_bytes;
2345 nr_bytes -= bio_bytes;
2356 * Reset counters so that the request stacking driver
2357 * can find how many bytes remain in the request
2360 req->__data_len = 0;
2364 req->__data_len -= total_bytes;
2365 req->buffer = bio_data(req->bio);
2367 /* update sector only for requests with clear definition of sector */
2368 if (req->cmd_type == REQ_TYPE_FS)
2369 req->__sector += total_bytes >> 9;
2371 /* mixed attributes always follow the first bio */
2372 if (req->cmd_flags & REQ_MIXED_MERGE) {
2373 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2374 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2378 * If total number of sectors is less than the first segment
2379 * size, something has gone terribly wrong.
2381 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2382 blk_dump_rq_flags(req, "request botched");
2383 req->__data_len = blk_rq_cur_bytes(req);
2386 /* recalculate the number of segments */
2387 blk_recalc_rq_segments(req);
2391 EXPORT_SYMBOL_GPL(blk_update_request);
2393 static bool blk_update_bidi_request(struct request *rq, int error,
2394 unsigned int nr_bytes,
2395 unsigned int bidi_bytes,
2396 struct batch_complete *batch)
2398 if (blk_update_request(rq, error, nr_bytes, batch))
2401 /* Bidi request must be completed as a whole */
2402 if (unlikely(blk_bidi_rq(rq)) &&
2403 blk_update_request(rq->next_rq, error, bidi_bytes, batch))
2406 if (blk_queue_add_random(rq->q))
2407 add_disk_randomness(rq->rq_disk);
2413 * blk_unprep_request - unprepare a request
2416 * This function makes a request ready for complete resubmission (or
2417 * completion). It happens only after all error handling is complete,
2418 * so represents the appropriate moment to deallocate any resources
2419 * that were allocated to the request in the prep_rq_fn. The queue
2420 * lock is held when calling this.
2422 void blk_unprep_request(struct request *req)
2424 struct request_queue *q = req->q;
2426 req->cmd_flags &= ~REQ_DONTPREP;
2427 if (q->unprep_rq_fn)
2428 q->unprep_rq_fn(q, req);
2430 EXPORT_SYMBOL_GPL(blk_unprep_request);
2433 * queue lock must be held
2435 static void blk_finish_request(struct request *req, int error)
2437 if (blk_rq_tagged(req))
2438 blk_queue_end_tag(req->q, req);
2440 BUG_ON(blk_queued_rq(req));
2442 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2443 laptop_io_completion(&req->q->backing_dev_info);
2445 blk_delete_timer(req);
2447 if (req->cmd_flags & REQ_DONTPREP)
2448 blk_unprep_request(req);
2451 blk_account_io_done(req);
2454 req->end_io(req, error);
2456 if (blk_bidi_rq(req))
2457 __blk_put_request(req->next_rq->q, req->next_rq);
2459 __blk_put_request(req->q, req);
2464 * blk_end_bidi_request - Complete a bidi request
2465 * @rq: the request to complete
2466 * @error: %0 for success, < %0 for error
2467 * @nr_bytes: number of bytes to complete @rq
2468 * @bidi_bytes: number of bytes to complete @rq->next_rq
2471 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2472 * Drivers that supports bidi can safely call this member for any
2473 * type of request, bidi or uni. In the later case @bidi_bytes is
2477 * %false - we are done with this request
2478 * %true - still buffers pending for this request
2480 static bool blk_end_bidi_request(struct request *rq, int error,
2481 unsigned int nr_bytes, unsigned int bidi_bytes)
2483 struct request_queue *q = rq->q;
2484 unsigned long flags;
2486 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes, NULL))
2489 spin_lock_irqsave(q->queue_lock, flags);
2490 blk_finish_request(rq, error);
2491 spin_unlock_irqrestore(q->queue_lock, flags);
2497 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2498 * @rq: the request to complete
2499 * @error: %0 for success, < %0 for error
2500 * @nr_bytes: number of bytes to complete @rq
2501 * @bidi_bytes: number of bytes to complete @rq->next_rq
2504 * Identical to blk_end_bidi_request() except that queue lock is
2505 * assumed to be locked on entry and remains so on return.
2508 * %false - we are done with this request
2509 * %true - still buffers pending for this request
2511 bool __blk_end_bidi_request(struct request *rq, int error,
2512 unsigned int nr_bytes, unsigned int bidi_bytes,
2513 struct batch_complete *batch)
2515 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes, batch))
2518 blk_finish_request(rq, error);
2524 * blk_end_request - Helper function for drivers to complete the request.
2525 * @rq: the request being processed
2526 * @error: %0 for success, < %0 for error
2527 * @nr_bytes: number of bytes to complete
2530 * Ends I/O on a number of bytes attached to @rq.
2531 * If @rq has leftover, sets it up for the next range of segments.
2534 * %false - we are done with this request
2535 * %true - still buffers pending for this request
2537 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2539 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2541 EXPORT_SYMBOL(blk_end_request);
2544 * blk_end_request_all - Helper function for drives to finish the request.
2545 * @rq: the request to finish
2546 * @error: %0 for success, < %0 for error
2549 * Completely finish @rq.
2551 void blk_end_request_all(struct request *rq, int error)
2554 unsigned int bidi_bytes = 0;
2556 if (unlikely(blk_bidi_rq(rq)))
2557 bidi_bytes = blk_rq_bytes(rq->next_rq);
2559 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2562 EXPORT_SYMBOL(blk_end_request_all);
2565 * blk_end_request_cur - Helper function to finish the current request chunk.
2566 * @rq: the request to finish the current chunk for
2567 * @error: %0 for success, < %0 for error
2570 * Complete the current consecutively mapped chunk from @rq.
2573 * %false - we are done with this request
2574 * %true - still buffers pending for this request
2576 bool blk_end_request_cur(struct request *rq, int error)
2578 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2580 EXPORT_SYMBOL(blk_end_request_cur);
2583 * blk_end_request_err - Finish a request till the next failure boundary.
2584 * @rq: the request to finish till the next failure boundary for
2585 * @error: must be negative errno
2588 * Complete @rq till the next failure boundary.
2591 * %false - we are done with this request
2592 * %true - still buffers pending for this request
2594 bool blk_end_request_err(struct request *rq, int error)
2596 WARN_ON(error >= 0);
2597 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2599 EXPORT_SYMBOL_GPL(blk_end_request_err);
2602 * __blk_end_request - Helper function for drivers to complete the request.
2603 * @rq: the request being processed
2604 * @error: %0 for success, < %0 for error
2605 * @nr_bytes: number of bytes to complete
2608 * Must be called with queue lock held unlike blk_end_request().
2611 * %false - we are done with this request
2612 * %true - still buffers pending for this request
2614 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2616 return __blk_end_bidi_request(rq, error, nr_bytes, 0, NULL);
2618 EXPORT_SYMBOL(__blk_end_request);
2621 * __blk_end_request_all - Helper function for drives to finish the request.
2622 * @rq: the request to finish
2623 * @error: %0 for success, < %0 for error
2626 * Completely finish @rq. Must be called with queue lock held.
2628 void blk_end_request_all_batch(struct request *rq, int error, struct batch_complete *batch)
2631 unsigned int bidi_bytes = 0;
2633 if (unlikely(blk_bidi_rq(rq)))
2634 bidi_bytes = blk_rq_bytes(rq->next_rq);
2636 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes, batch);
2639 EXPORT_SYMBOL(blk_end_request_all_batch);
2642 * __blk_end_request_cur - Helper function to finish the current request chunk.
2643 * @rq: the request to finish the current chunk for
2644 * @error: %0 for success, < %0 for error
2647 * Complete the current consecutively mapped chunk from @rq. Must
2648 * be called with queue lock held.
2651 * %false - we are done with this request
2652 * %true - still buffers pending for this request
2654 bool __blk_end_request_cur(struct request *rq, int error)
2656 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2658 EXPORT_SYMBOL(__blk_end_request_cur);
2661 * __blk_end_request_err - Finish a request till the next failure boundary.
2662 * @rq: the request to finish till the next failure boundary for
2663 * @error: must be negative errno
2666 * Complete @rq till the next failure boundary. Must be called
2667 * with queue lock held.
2670 * %false - we are done with this request
2671 * %true - still buffers pending for this request
2673 bool __blk_end_request_err(struct request *rq, int error)
2675 WARN_ON(error >= 0);
2676 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2678 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2680 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2683 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2684 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2686 if (bio_has_data(bio)) {
2687 rq->nr_phys_segments = bio_phys_segments(q, bio);
2688 rq->buffer = bio_data(bio);
2690 rq->__data_len = bio->bi_size;
2691 rq->bio = rq->biotail = bio;
2694 rq->rq_disk = bio->bi_bdev->bd_disk;
2697 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2699 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2700 * @rq: the request to be flushed
2703 * Flush all pages in @rq.
2705 void rq_flush_dcache_pages(struct request *rq)
2707 struct req_iterator iter;
2708 struct bio_vec *bvec;
2710 rq_for_each_segment(bvec, rq, iter)
2711 flush_dcache_page(bvec->bv_page);
2713 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2717 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2718 * @q : the queue of the device being checked
2721 * Check if underlying low-level drivers of a device are busy.
2722 * If the drivers want to export their busy state, they must set own
2723 * exporting function using blk_queue_lld_busy() first.
2725 * Basically, this function is used only by request stacking drivers
2726 * to stop dispatching requests to underlying devices when underlying
2727 * devices are busy. This behavior helps more I/O merging on the queue
2728 * of the request stacking driver and prevents I/O throughput regression
2729 * on burst I/O load.
2732 * 0 - Not busy (The request stacking driver should dispatch request)
2733 * 1 - Busy (The request stacking driver should stop dispatching request)
2735 int blk_lld_busy(struct request_queue *q)
2738 return q->lld_busy_fn(q);
2742 EXPORT_SYMBOL_GPL(blk_lld_busy);
2745 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2746 * @rq: the clone request to be cleaned up
2749 * Free all bios in @rq for a cloned request.
2751 void blk_rq_unprep_clone(struct request *rq)
2755 while ((bio = rq->bio) != NULL) {
2756 rq->bio = bio->bi_next;
2761 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2764 * Copy attributes of the original request to the clone request.
2765 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2767 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2769 dst->cpu = src->cpu;
2770 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2771 dst->cmd_type = src->cmd_type;
2772 dst->__sector = blk_rq_pos(src);
2773 dst->__data_len = blk_rq_bytes(src);
2774 dst->nr_phys_segments = src->nr_phys_segments;
2775 dst->ioprio = src->ioprio;
2776 dst->extra_len = src->extra_len;
2780 * blk_rq_prep_clone - Helper function to setup clone request
2781 * @rq: the request to be setup
2782 * @rq_src: original request to be cloned
2783 * @bs: bio_set that bios for clone are allocated from
2784 * @gfp_mask: memory allocation mask for bio
2785 * @bio_ctr: setup function to be called for each clone bio.
2786 * Returns %0 for success, non %0 for failure.
2787 * @data: private data to be passed to @bio_ctr
2790 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2791 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2792 * are not copied, and copying such parts is the caller's responsibility.
2793 * Also, pages which the original bios are pointing to are not copied
2794 * and the cloned bios just point same pages.
2795 * So cloned bios must be completed before original bios, which means
2796 * the caller must complete @rq before @rq_src.
2798 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2799 struct bio_set *bs, gfp_t gfp_mask,
2800 int (*bio_ctr)(struct bio *, struct bio *, void *),
2803 struct bio *bio, *bio_src;
2808 blk_rq_init(NULL, rq);
2810 __rq_for_each_bio(bio_src, rq_src) {
2811 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2815 if (bio_ctr && bio_ctr(bio, bio_src, data))
2819 rq->biotail->bi_next = bio;
2822 rq->bio = rq->biotail = bio;
2825 __blk_rq_prep_clone(rq, rq_src);
2832 blk_rq_unprep_clone(rq);
2836 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2838 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2840 return queue_work(kblockd_workqueue, work);
2842 EXPORT_SYMBOL(kblockd_schedule_work);
2844 int kblockd_schedule_delayed_work(struct request_queue *q,
2845 struct delayed_work *dwork, unsigned long delay)
2847 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2849 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2851 #define PLUG_MAGIC 0x91827364
2854 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2855 * @plug: The &struct blk_plug that needs to be initialized
2858 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2859 * pending I/O should the task end up blocking between blk_start_plug() and
2860 * blk_finish_plug(). This is important from a performance perspective, but
2861 * also ensures that we don't deadlock. For instance, if the task is blocking
2862 * for a memory allocation, memory reclaim could end up wanting to free a
2863 * page belonging to that request that is currently residing in our private
2864 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2865 * this kind of deadlock.
2867 void blk_start_plug(struct blk_plug *plug)
2869 struct task_struct *tsk = current;
2871 plug->magic = PLUG_MAGIC;
2872 INIT_LIST_HEAD(&plug->list);
2873 INIT_LIST_HEAD(&plug->cb_list);
2876 * If this is a nested plug, don't actually assign it. It will be
2877 * flushed on its own.
2881 * Store ordering should not be needed here, since a potential
2882 * preempt will imply a full memory barrier
2887 EXPORT_SYMBOL(blk_start_plug);
2889 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2891 struct request *rqa = container_of(a, struct request, queuelist);
2892 struct request *rqb = container_of(b, struct request, queuelist);
2894 return !(rqa->q < rqb->q ||
2895 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2899 * If 'from_schedule' is true, then postpone the dispatch of requests
2900 * until a safe kblockd context. We due this to avoid accidental big
2901 * additional stack usage in driver dispatch, in places where the originally
2902 * plugger did not intend it.
2904 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2906 __releases(q->queue_lock)
2908 trace_block_unplug(q, depth, !from_schedule);
2911 blk_run_queue_async(q);
2914 spin_unlock(q->queue_lock);
2917 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2919 LIST_HEAD(callbacks);
2921 while (!list_empty(&plug->cb_list)) {
2922 list_splice_init(&plug->cb_list, &callbacks);
2924 while (!list_empty(&callbacks)) {
2925 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2928 list_del(&cb->list);
2929 cb->callback(cb, from_schedule);
2934 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2937 struct blk_plug *plug = current->plug;
2938 struct blk_plug_cb *cb;
2943 list_for_each_entry(cb, &plug->cb_list, list)
2944 if (cb->callback == unplug && cb->data == data)
2947 /* Not currently on the callback list */
2948 BUG_ON(size < sizeof(*cb));
2949 cb = kzalloc(size, GFP_ATOMIC);
2952 cb->callback = unplug;
2953 list_add(&cb->list, &plug->cb_list);
2957 EXPORT_SYMBOL(blk_check_plugged);
2959 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2961 struct request_queue *q;
2962 unsigned long flags;
2967 BUG_ON(plug->magic != PLUG_MAGIC);
2969 flush_plug_callbacks(plug, from_schedule);
2970 if (list_empty(&plug->list))
2973 list_splice_init(&plug->list, &list);
2975 list_sort(NULL, &list, plug_rq_cmp);
2981 * Save and disable interrupts here, to avoid doing it for every
2982 * queue lock we have to take.
2984 local_irq_save(flags);
2985 while (!list_empty(&list)) {
2986 rq = list_entry_rq(list.next);
2987 list_del_init(&rq->queuelist);
2991 * This drops the queue lock
2994 queue_unplugged(q, depth, from_schedule);
2997 spin_lock(q->queue_lock);
3001 * Short-circuit if @q is dead
3003 if (unlikely(blk_queue_dying(q))) {
3004 __blk_end_request_all(rq, -ENODEV);
3009 * rq is already accounted, so use raw insert
3011 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3012 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3014 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3020 * This drops the queue lock
3023 queue_unplugged(q, depth, from_schedule);
3025 local_irq_restore(flags);
3028 void blk_finish_plug(struct blk_plug *plug)
3030 blk_flush_plug_list(plug, false);
3032 if (plug == current->plug)
3033 current->plug = NULL;
3035 EXPORT_SYMBOL(blk_finish_plug);
3037 #ifdef CONFIG_PM_RUNTIME
3039 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3040 * @q: the queue of the device
3041 * @dev: the device the queue belongs to
3044 * Initialize runtime-PM-related fields for @q and start auto suspend for
3045 * @dev. Drivers that want to take advantage of request-based runtime PM
3046 * should call this function after @dev has been initialized, and its
3047 * request queue @q has been allocated, and runtime PM for it can not happen
3048 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3049 * cases, driver should call this function before any I/O has taken place.
3051 * This function takes care of setting up using auto suspend for the device,
3052 * the autosuspend delay is set to -1 to make runtime suspend impossible
3053 * until an updated value is either set by user or by driver. Drivers do
3054 * not need to touch other autosuspend settings.
3056 * The block layer runtime PM is request based, so only works for drivers
3057 * that use request as their IO unit instead of those directly use bio's.
3059 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3062 q->rpm_status = RPM_ACTIVE;
3063 pm_runtime_set_autosuspend_delay(q->dev, -1);
3064 pm_runtime_use_autosuspend(q->dev);
3066 EXPORT_SYMBOL(blk_pm_runtime_init);
3069 * blk_pre_runtime_suspend - Pre runtime suspend check
3070 * @q: the queue of the device
3073 * This function will check if runtime suspend is allowed for the device
3074 * by examining if there are any requests pending in the queue. If there
3075 * are requests pending, the device can not be runtime suspended; otherwise,
3076 * the queue's status will be updated to SUSPENDING and the driver can
3077 * proceed to suspend the device.
3079 * For the not allowed case, we mark last busy for the device so that
3080 * runtime PM core will try to autosuspend it some time later.
3082 * This function should be called near the start of the device's
3083 * runtime_suspend callback.
3086 * 0 - OK to runtime suspend the device
3087 * -EBUSY - Device should not be runtime suspended
3089 int blk_pre_runtime_suspend(struct request_queue *q)
3093 spin_lock_irq(q->queue_lock);
3094 if (q->nr_pending) {
3096 pm_runtime_mark_last_busy(q->dev);
3098 q->rpm_status = RPM_SUSPENDING;
3100 spin_unlock_irq(q->queue_lock);
3103 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3106 * blk_post_runtime_suspend - Post runtime suspend processing
3107 * @q: the queue of the device
3108 * @err: return value of the device's runtime_suspend function
3111 * Update the queue's runtime status according to the return value of the
3112 * device's runtime suspend function and mark last busy for the device so
3113 * that PM core will try to auto suspend the device at a later time.
3115 * This function should be called near the end of the device's
3116 * runtime_suspend callback.
3118 void blk_post_runtime_suspend(struct request_queue *q, int err)
3120 spin_lock_irq(q->queue_lock);
3122 q->rpm_status = RPM_SUSPENDED;
3124 q->rpm_status = RPM_ACTIVE;
3125 pm_runtime_mark_last_busy(q->dev);
3127 spin_unlock_irq(q->queue_lock);
3129 EXPORT_SYMBOL(blk_post_runtime_suspend);
3132 * blk_pre_runtime_resume - Pre runtime resume processing
3133 * @q: the queue of the device
3136 * Update the queue's runtime status to RESUMING in preparation for the
3137 * runtime resume of the device.
3139 * This function should be called near the start of the device's
3140 * runtime_resume callback.
3142 void blk_pre_runtime_resume(struct request_queue *q)
3144 spin_lock_irq(q->queue_lock);
3145 q->rpm_status = RPM_RESUMING;
3146 spin_unlock_irq(q->queue_lock);
3148 EXPORT_SYMBOL(blk_pre_runtime_resume);
3151 * blk_post_runtime_resume - Post runtime resume processing
3152 * @q: the queue of the device
3153 * @err: return value of the device's runtime_resume function
3156 * Update the queue's runtime status according to the return value of the
3157 * device's runtime_resume function. If it is successfully resumed, process
3158 * the requests that are queued into the device's queue when it is resuming
3159 * and then mark last busy and initiate autosuspend for it.
3161 * This function should be called near the end of the device's
3162 * runtime_resume callback.
3164 void blk_post_runtime_resume(struct request_queue *q, int err)
3166 spin_lock_irq(q->queue_lock);
3168 q->rpm_status = RPM_ACTIVE;
3170 pm_runtime_mark_last_busy(q->dev);
3171 pm_request_autosuspend(q->dev);
3173 q->rpm_status = RPM_SUSPENDED;
3175 spin_unlock_irq(q->queue_lock);
3177 EXPORT_SYMBOL(blk_post_runtime_resume);
3180 int __init blk_dev_init(void)
3182 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3183 sizeof(((struct request *)0)->cmd_flags));
3185 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3186 kblockd_workqueue = alloc_workqueue("kblockd",
3187 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3188 WQ_POWER_EFFICIENT, 0);
3189 if (!kblockd_workqueue)
3190 panic("Failed to create kblockd\n");
3192 request_cachep = kmem_cache_create("blkdev_requests",
3193 sizeof(struct request), 0, SLAB_PANIC, NULL);
3195 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3196 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);