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>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/block.h>
38 #include "blk-cgroup.h"
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
44 DEFINE_IDA(blk_queue_ida);
47 * For the allocated request tables
49 static struct kmem_cache *request_cachep;
52 * For queue allocation
54 struct kmem_cache *blk_requestq_cachep;
57 * Controlling structure to kblockd
59 static struct workqueue_struct *kblockd_workqueue;
61 static void drive_stat_acct(struct request *rq, int new_io)
63 struct hd_struct *part;
64 int rw = rq_data_dir(rq);
67 if (!blk_do_io_stat(rq))
70 cpu = part_stat_lock();
74 part_stat_inc(cpu, part, merges[rw]);
76 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
77 if (!hd_struct_try_get(part)) {
79 * The partition is already being removed,
80 * the request will be accounted on the disk only
82 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat
84 * it as any other partition.
86 part = &rq->rq_disk->part0;
89 part_round_stats(cpu, part);
90 part_inc_in_flight(part, rw);
97 void blk_queue_congestion_threshold(struct request_queue *q)
101 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests)
104 q->nr_congestion_on = nr;
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
109 q->nr_congestion_off = nr;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
123 struct backing_dev_info *ret = NULL;
124 struct request_queue *q = bdev_get_queue(bdev);
127 ret = &q->backing_dev_info;
130 EXPORT_SYMBOL(blk_get_backing_dev_info);
132 void blk_rq_init(struct request_queue *q, struct request *rq)
134 memset(rq, 0, sizeof(*rq));
136 INIT_LIST_HEAD(&rq->queuelist);
137 INIT_LIST_HEAD(&rq->timeout_list);
140 rq->__sector = (sector_t) -1;
141 INIT_HLIST_NODE(&rq->hash);
142 RB_CLEAR_NODE(&rq->rb_node);
144 rq->cmd_len = BLK_MAX_CDB;
147 rq->start_time = jiffies;
148 set_start_time_ns(rq);
151 EXPORT_SYMBOL(blk_rq_init);
153 static void req_bio_endio(struct request *rq, struct bio *bio,
154 unsigned int nbytes, int error)
157 clear_bit(BIO_UPTODATE, &bio->bi_flags);
158 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
161 if (unlikely(nbytes > bio->bi_size)) {
162 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
163 __func__, nbytes, bio->bi_size);
164 nbytes = bio->bi_size;
167 if (unlikely(rq->cmd_flags & REQ_QUIET))
168 set_bit(BIO_QUIET, &bio->bi_flags);
170 bio->bi_size -= nbytes;
171 bio->bi_sector += (nbytes >> 9);
173 if (bio_integrity(bio))
174 bio_integrity_advance(bio, nbytes);
176 /* don't actually finish bio if it's part of flush sequence */
177 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
178 bio_endio(bio, error);
181 void blk_dump_rq_flags(struct request *rq, char *msg)
185 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
186 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
189 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq),
191 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
192 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
193 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
195 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
196 printk(KERN_INFO " cdb: ");
197 for (bit = 0; bit < BLK_MAX_CDB; bit++)
198 printk("%02x ", rq->cmd[bit]);
202 EXPORT_SYMBOL(blk_dump_rq_flags);
204 static void blk_delay_work(struct work_struct *work)
206 struct request_queue *q;
208 q = container_of(work, struct request_queue, delay_work.work);
209 spin_lock_irq(q->queue_lock);
211 spin_unlock_irq(q->queue_lock);
215 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs
220 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is
222 * restarted around the specified time.
224 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
226 queue_delayed_work(kblockd_workqueue, &q->delay_work,
227 msecs_to_jiffies(msecs));
229 EXPORT_SYMBOL(blk_delay_queue);
232 * blk_start_queue - restart a previously stopped queue
233 * @q: The &struct request_queue in question
236 * blk_start_queue() will clear the stop flag on the queue, and call
237 * the request_fn for the queue if it was in a stopped state when
238 * entered. Also see blk_stop_queue(). Queue lock must be held.
240 void blk_start_queue(struct request_queue *q)
242 WARN_ON(!irqs_disabled());
244 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
247 EXPORT_SYMBOL(blk_start_queue);
250 * blk_stop_queue - stop a queue
251 * @q: The &struct request_queue in question
254 * The Linux block layer assumes that a block driver will consume all
255 * entries on the request queue when the request_fn strategy is called.
256 * Often this will not happen, because of hardware limitations (queue
257 * depth settings). If a device driver gets a 'queue full' response,
258 * or if it simply chooses not to queue more I/O at one point, it can
259 * call this function to prevent the request_fn from being called until
260 * the driver has signalled it's ready to go again. This happens by calling
261 * blk_start_queue() to restart queue operations. Queue lock must be held.
263 void blk_stop_queue(struct request_queue *q)
265 cancel_delayed_work(&q->delay_work);
266 queue_flag_set(QUEUE_FLAG_STOPPED, q);
268 EXPORT_SYMBOL(blk_stop_queue);
271 * blk_sync_queue - cancel any pending callbacks on a queue
275 * The block layer may perform asynchronous callback activity
276 * on a queue, such as calling the unplug function after a timeout.
277 * A block device may call blk_sync_queue to ensure that any
278 * such activity is cancelled, thus allowing it to release resources
279 * that the callbacks might use. The caller must already have made sure
280 * that its ->make_request_fn will not re-add plugging prior to calling
283 * This function does not cancel any asynchronous activity arising
284 * out of elevator or throttling code. That would require elevaotor_exit()
285 * and blkcg_exit_queue() to be called with queue lock initialized.
288 void blk_sync_queue(struct request_queue *q)
290 del_timer_sync(&q->timeout);
291 cancel_delayed_work_sync(&q->delay_work);
293 EXPORT_SYMBOL(blk_sync_queue);
296 * __blk_run_queue - run a single device queue
297 * @q: The queue to run
300 * See @blk_run_queue. This variant must be called with the queue lock
301 * held and interrupts disabled.
303 void __blk_run_queue(struct request_queue *q)
305 if (unlikely(blk_queue_stopped(q)))
310 EXPORT_SYMBOL(__blk_run_queue);
313 * blk_run_queue_async - run a single device queue in workqueue context
314 * @q: The queue to run
317 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
320 void blk_run_queue_async(struct request_queue *q)
322 if (likely(!blk_queue_stopped(q)))
323 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
325 EXPORT_SYMBOL(blk_run_queue_async);
328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
332 * Invoke request handling on this queue, if it has pending work to do.
333 * May be used to restart queueing when a request has completed.
335 void blk_run_queue(struct request_queue *q)
339 spin_lock_irqsave(q->queue_lock, flags);
341 spin_unlock_irqrestore(q->queue_lock, flags);
343 EXPORT_SYMBOL(blk_run_queue);
345 void blk_put_queue(struct request_queue *q)
347 kobject_put(&q->kobj);
349 EXPORT_SYMBOL(blk_put_queue);
352 * __blk_drain_queue - drain requests from request_queue
354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
360 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
361 __releases(q->queue_lock)
362 __acquires(q->queue_lock)
366 lockdep_assert_held(q->queue_lock);
372 * The caller might be trying to drain @q before its
373 * elevator is initialized.
376 elv_drain_elevator(q);
378 blkcg_drain_queue(q);
381 * This function might be called on a queue which failed
382 * driver init after queue creation or is not yet fully
383 * active yet. Some drivers (e.g. fd and loop) get unhappy
384 * in such cases. Kick queue iff dispatch queue has
385 * something on it and @q has request_fn set.
387 if (!list_empty(&q->queue_head) && q->request_fn)
390 drain |= q->nr_rqs_elvpriv;
393 * Unfortunately, requests are queued at and tracked from
394 * multiple places and there's no single counter which can
395 * be drained. Check all the queues and counters.
398 drain |= !list_empty(&q->queue_head);
399 for (i = 0; i < 2; i++) {
400 drain |= q->nr_rqs[i];
401 drain |= q->in_flight[i];
402 drain |= !list_empty(&q->flush_queue[i]);
409 spin_unlock_irq(q->queue_lock);
413 spin_lock_irq(q->queue_lock);
417 * With queue marked dead, any woken up waiter will fail the
418 * allocation path, so the wakeup chaining is lost and we're
419 * left with hung waiters. We need to wake up those waiters.
422 struct request_list *rl;
424 blk_queue_for_each_rl(rl, q)
425 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
426 wake_up_all(&rl->wait[i]);
431 * blk_queue_bypass_start - enter queue bypass mode
432 * @q: queue of interest
434 * In bypass mode, only the dispatch FIFO queue of @q is used. This
435 * function makes @q enter bypass mode and drains all requests which were
436 * throttled or issued before. On return, it's guaranteed that no request
437 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
438 * inside queue or RCU read lock.
440 void blk_queue_bypass_start(struct request_queue *q)
444 spin_lock_irq(q->queue_lock);
445 drain = !q->bypass_depth++;
446 queue_flag_set(QUEUE_FLAG_BYPASS, q);
447 spin_unlock_irq(q->queue_lock);
450 spin_lock_irq(q->queue_lock);
451 __blk_drain_queue(q, false);
452 spin_unlock_irq(q->queue_lock);
454 /* ensure blk_queue_bypass() is %true inside RCU read lock */
458 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
461 * blk_queue_bypass_end - leave queue bypass mode
462 * @q: queue of interest
464 * Leave bypass mode and restore the normal queueing behavior.
466 void blk_queue_bypass_end(struct request_queue *q)
468 spin_lock_irq(q->queue_lock);
469 if (!--q->bypass_depth)
470 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
471 WARN_ON_ONCE(q->bypass_depth < 0);
472 spin_unlock_irq(q->queue_lock);
474 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
477 * blk_cleanup_queue - shutdown a request queue
478 * @q: request queue to shutdown
480 * Mark @q DYING, drain all pending requests, destroy and put it. All
481 * future requests will be failed immediately with -ENODEV.
483 void blk_cleanup_queue(struct request_queue *q)
485 spinlock_t *lock = q->queue_lock;
487 /* mark @q DYING, no new request or merges will be allowed afterwards */
488 mutex_lock(&q->sysfs_lock);
489 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
493 * A dying queue is permanently in bypass mode till released. Note
494 * that, unlike blk_queue_bypass_start(), we aren't performing
495 * synchronize_rcu() after entering bypass mode to avoid the delay
496 * as some drivers create and destroy a lot of queues while
497 * probing. This is still safe because blk_release_queue() will be
498 * called only after the queue refcnt drops to zero and nothing,
499 * RCU or not, would be traversing the queue by then.
502 queue_flag_set(QUEUE_FLAG_BYPASS, q);
504 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
505 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
506 queue_flag_set(QUEUE_FLAG_DYING, q);
507 spin_unlock_irq(lock);
508 mutex_unlock(&q->sysfs_lock);
510 /* drain all requests queued before DYING marking */
512 __blk_drain_queue(q, true);
513 spin_unlock_irq(lock);
515 /* @q won't process any more request, flush async actions */
516 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
520 if (q->queue_lock != &q->__queue_lock)
521 q->queue_lock = &q->__queue_lock;
522 spin_unlock_irq(lock);
524 /* @q is and will stay empty, shutdown and put */
527 EXPORT_SYMBOL(blk_cleanup_queue);
529 int blk_init_rl(struct request_list *rl, struct request_queue *q,
532 if (unlikely(rl->rq_pool))
536 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
537 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
538 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
539 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
541 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
542 mempool_free_slab, request_cachep,
550 void blk_exit_rl(struct request_list *rl)
553 mempool_destroy(rl->rq_pool);
556 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
558 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
560 EXPORT_SYMBOL(blk_alloc_queue);
562 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
564 struct request_queue *q;
567 q = kmem_cache_alloc_node(blk_requestq_cachep,
568 gfp_mask | __GFP_ZERO, node_id);
572 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
576 q->backing_dev_info.ra_pages =
577 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
578 q->backing_dev_info.state = 0;
579 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
580 q->backing_dev_info.name = "block";
583 err = bdi_init(&q->backing_dev_info);
587 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
588 laptop_mode_timer_fn, (unsigned long) q);
589 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
590 INIT_LIST_HEAD(&q->queue_head);
591 INIT_LIST_HEAD(&q->timeout_list);
592 INIT_LIST_HEAD(&q->icq_list);
593 #ifdef CONFIG_BLK_CGROUP
594 INIT_LIST_HEAD(&q->blkg_list);
596 INIT_LIST_HEAD(&q->flush_queue[0]);
597 INIT_LIST_HEAD(&q->flush_queue[1]);
598 INIT_LIST_HEAD(&q->flush_data_in_flight);
599 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
601 kobject_init(&q->kobj, &blk_queue_ktype);
603 mutex_init(&q->sysfs_lock);
604 spin_lock_init(&q->__queue_lock);
607 * By default initialize queue_lock to internal lock and driver can
608 * override it later if need be.
610 q->queue_lock = &q->__queue_lock;
613 * A queue starts its life with bypass turned on to avoid
614 * unnecessary bypass on/off overhead and nasty surprises during
615 * init. The initial bypass will be finished when the queue is
616 * registered by blk_register_queue().
619 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
621 if (blkcg_init_queue(q))
627 ida_simple_remove(&blk_queue_ida, q->id);
629 kmem_cache_free(blk_requestq_cachep, q);
632 EXPORT_SYMBOL(blk_alloc_queue_node);
635 * blk_init_queue - prepare a request queue for use with a block device
636 * @rfn: The function to be called to process requests that have been
637 * placed on the queue.
638 * @lock: Request queue spin lock
641 * If a block device wishes to use the standard request handling procedures,
642 * which sorts requests and coalesces adjacent requests, then it must
643 * call blk_init_queue(). The function @rfn will be called when there
644 * are requests on the queue that need to be processed. If the device
645 * supports plugging, then @rfn may not be called immediately when requests
646 * are available on the queue, but may be called at some time later instead.
647 * Plugged queues are generally unplugged when a buffer belonging to one
648 * of the requests on the queue is needed, or due to memory pressure.
650 * @rfn is not required, or even expected, to remove all requests off the
651 * queue, but only as many as it can handle at a time. If it does leave
652 * requests on the queue, it is responsible for arranging that the requests
653 * get dealt with eventually.
655 * The queue spin lock must be held while manipulating the requests on the
656 * request queue; this lock will be taken also from interrupt context, so irq
657 * disabling is needed for it.
659 * Function returns a pointer to the initialized request queue, or %NULL if
663 * blk_init_queue() must be paired with a blk_cleanup_queue() call
664 * when the block device is deactivated (such as at module unload).
667 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
669 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
671 EXPORT_SYMBOL(blk_init_queue);
673 struct request_queue *
674 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
676 struct request_queue *uninit_q, *q;
678 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
682 q = blk_init_allocated_queue(uninit_q, rfn, lock);
684 blk_cleanup_queue(uninit_q);
688 EXPORT_SYMBOL(blk_init_queue_node);
690 struct request_queue *
691 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
697 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
701 q->prep_rq_fn = NULL;
702 q->unprep_rq_fn = NULL;
703 q->queue_flags |= QUEUE_FLAG_DEFAULT;
705 /* Override internal queue lock with supplied lock pointer */
707 q->queue_lock = lock;
710 * This also sets hw/phys segments, boundary and size
712 blk_queue_make_request(q, blk_queue_bio);
714 q->sg_reserved_size = INT_MAX;
717 if (elevator_init(q, NULL))
721 EXPORT_SYMBOL(blk_init_allocated_queue);
723 bool blk_get_queue(struct request_queue *q)
725 if (likely(!blk_queue_dying(q))) {
732 EXPORT_SYMBOL(blk_get_queue);
734 static inline void blk_free_request(struct request_list *rl, struct request *rq)
736 if (rq->cmd_flags & REQ_ELVPRIV) {
737 elv_put_request(rl->q, rq);
739 put_io_context(rq->elv.icq->ioc);
742 mempool_free(rq, rl->rq_pool);
746 * ioc_batching returns true if the ioc is a valid batching request and
747 * should be given priority access to a request.
749 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
755 * Make sure the process is able to allocate at least 1 request
756 * even if the batch times out, otherwise we could theoretically
759 return ioc->nr_batch_requests == q->nr_batching ||
760 (ioc->nr_batch_requests > 0
761 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
765 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
766 * will cause the process to be a "batcher" on all queues in the system. This
767 * is the behaviour we want though - once it gets a wakeup it should be given
770 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
772 if (!ioc || ioc_batching(q, ioc))
775 ioc->nr_batch_requests = q->nr_batching;
776 ioc->last_waited = jiffies;
779 static void __freed_request(struct request_list *rl, int sync)
781 struct request_queue *q = rl->q;
784 * bdi isn't aware of blkcg yet. As all async IOs end up root
785 * blkcg anyway, just use root blkcg state.
787 if (rl == &q->root_rl &&
788 rl->count[sync] < queue_congestion_off_threshold(q))
789 blk_clear_queue_congested(q, sync);
791 if (rl->count[sync] + 1 <= q->nr_requests) {
792 if (waitqueue_active(&rl->wait[sync]))
793 wake_up(&rl->wait[sync]);
795 blk_clear_rl_full(rl, sync);
800 * A request has just been released. Account for it, update the full and
801 * congestion status, wake up any waiters. Called under q->queue_lock.
803 static void freed_request(struct request_list *rl, unsigned int flags)
805 struct request_queue *q = rl->q;
806 int sync = rw_is_sync(flags);
810 if (flags & REQ_ELVPRIV)
813 __freed_request(rl, sync);
815 if (unlikely(rl->starved[sync ^ 1]))
816 __freed_request(rl, sync ^ 1);
820 * Determine if elevator data should be initialized when allocating the
821 * request associated with @bio.
823 static bool blk_rq_should_init_elevator(struct bio *bio)
829 * Flush requests do not use the elevator so skip initialization.
830 * This allows a request to share the flush and elevator data.
832 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
839 * rq_ioc - determine io_context for request allocation
840 * @bio: request being allocated is for this bio (can be %NULL)
842 * Determine io_context to use for request allocation for @bio. May return
843 * %NULL if %current->io_context doesn't exist.
845 static struct io_context *rq_ioc(struct bio *bio)
847 #ifdef CONFIG_BLK_CGROUP
848 if (bio && bio->bi_ioc)
851 return current->io_context;
855 * __get_request - get a free request
856 * @rl: request list to allocate from
857 * @rw_flags: RW and SYNC flags
858 * @bio: bio to allocate request for (can be %NULL)
859 * @gfp_mask: allocation mask
861 * Get a free request from @q. This function may fail under memory
862 * pressure or if @q is dead.
864 * Must be callled with @q->queue_lock held and,
865 * Returns %NULL on failure, with @q->queue_lock held.
866 * Returns !%NULL on success, with @q->queue_lock *not held*.
868 static struct request *__get_request(struct request_list *rl, int rw_flags,
869 struct bio *bio, gfp_t gfp_mask)
871 struct request_queue *q = rl->q;
873 struct elevator_type *et = q->elevator->type;
874 struct io_context *ioc = rq_ioc(bio);
875 struct io_cq *icq = NULL;
876 const bool is_sync = rw_is_sync(rw_flags) != 0;
879 if (unlikely(blk_queue_dying(q)))
882 may_queue = elv_may_queue(q, rw_flags);
883 if (may_queue == ELV_MQUEUE_NO)
886 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
887 if (rl->count[is_sync]+1 >= q->nr_requests) {
889 * The queue will fill after this allocation, so set
890 * it as full, and mark this process as "batching".
891 * This process will be allowed to complete a batch of
892 * requests, others will be blocked.
894 if (!blk_rl_full(rl, is_sync)) {
895 ioc_set_batching(q, ioc);
896 blk_set_rl_full(rl, is_sync);
898 if (may_queue != ELV_MQUEUE_MUST
899 && !ioc_batching(q, ioc)) {
901 * The queue is full and the allocating
902 * process is not a "batcher", and not
903 * exempted by the IO scheduler
910 * bdi isn't aware of blkcg yet. As all async IOs end up
911 * root blkcg anyway, just use root blkcg state.
913 if (rl == &q->root_rl)
914 blk_set_queue_congested(q, is_sync);
918 * Only allow batching queuers to allocate up to 50% over the defined
919 * limit of requests, otherwise we could have thousands of requests
920 * allocated with any setting of ->nr_requests
922 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
925 q->nr_rqs[is_sync]++;
926 rl->count[is_sync]++;
927 rl->starved[is_sync] = 0;
930 * Decide whether the new request will be managed by elevator. If
931 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
932 * prevent the current elevator from being destroyed until the new
933 * request is freed. This guarantees icq's won't be destroyed and
934 * makes creating new ones safe.
936 * Also, lookup icq while holding queue_lock. If it doesn't exist,
937 * it will be created after releasing queue_lock.
939 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
940 rw_flags |= REQ_ELVPRIV;
942 if (et->icq_cache && ioc)
943 icq = ioc_lookup_icq(ioc, q);
946 if (blk_queue_io_stat(q))
947 rw_flags |= REQ_IO_STAT;
948 spin_unlock_irq(q->queue_lock);
950 /* allocate and init request */
951 rq = mempool_alloc(rl->rq_pool, gfp_mask);
956 blk_rq_set_rl(rq, rl);
957 rq->cmd_flags = rw_flags | REQ_ALLOCED;
960 if (rw_flags & REQ_ELVPRIV) {
961 if (unlikely(et->icq_cache && !icq)) {
963 icq = ioc_create_icq(ioc, q, gfp_mask);
969 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
972 /* @rq->elv.icq holds io_context until @rq is freed */
974 get_io_context(icq->ioc);
978 * ioc may be NULL here, and ioc_batching will be false. That's
979 * OK, if the queue is under the request limit then requests need
980 * not count toward the nr_batch_requests limit. There will always
981 * be some limit enforced by BLK_BATCH_TIME.
983 if (ioc_batching(q, ioc))
984 ioc->nr_batch_requests--;
986 trace_block_getrq(q, bio, rw_flags & 1);
991 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
992 * and may fail indefinitely under memory pressure and thus
993 * shouldn't stall IO. Treat this request as !elvpriv. This will
994 * disturb iosched and blkcg but weird is bettern than dead.
996 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
997 dev_name(q->backing_dev_info.dev));
999 rq->cmd_flags &= ~REQ_ELVPRIV;
1002 spin_lock_irq(q->queue_lock);
1003 q->nr_rqs_elvpriv--;
1004 spin_unlock_irq(q->queue_lock);
1009 * Allocation failed presumably due to memory. Undo anything we
1010 * might have messed up.
1012 * Allocating task should really be put onto the front of the wait
1013 * queue, but this is pretty rare.
1015 spin_lock_irq(q->queue_lock);
1016 freed_request(rl, rw_flags);
1019 * in the very unlikely event that allocation failed and no
1020 * requests for this direction was pending, mark us starved so that
1021 * freeing of a request in the other direction will notice
1022 * us. another possible fix would be to split the rq mempool into
1026 if (unlikely(rl->count[is_sync] == 0))
1027 rl->starved[is_sync] = 1;
1032 * get_request - get a free request
1033 * @q: request_queue to allocate request from
1034 * @rw_flags: RW and SYNC flags
1035 * @bio: bio to allocate request for (can be %NULL)
1036 * @gfp_mask: allocation mask
1038 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1039 * function keeps retrying under memory pressure and fails iff @q is dead.
1041 * Must be callled with @q->queue_lock held and,
1042 * Returns %NULL on failure, with @q->queue_lock held.
1043 * Returns !%NULL on success, with @q->queue_lock *not held*.
1045 static struct request *get_request(struct request_queue *q, int rw_flags,
1046 struct bio *bio, gfp_t gfp_mask)
1048 const bool is_sync = rw_is_sync(rw_flags) != 0;
1050 struct request_list *rl;
1053 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1055 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1059 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1064 /* wait on @rl and retry */
1065 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1066 TASK_UNINTERRUPTIBLE);
1068 trace_block_sleeprq(q, bio, rw_flags & 1);
1070 spin_unlock_irq(q->queue_lock);
1074 * After sleeping, we become a "batching" process and will be able
1075 * to allocate at least one request, and up to a big batch of them
1076 * for a small period time. See ioc_batching, ioc_set_batching
1078 ioc_set_batching(q, current->io_context);
1080 spin_lock_irq(q->queue_lock);
1081 finish_wait(&rl->wait[is_sync], &wait);
1086 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1090 BUG_ON(rw != READ && rw != WRITE);
1092 /* create ioc upfront */
1093 create_io_context(gfp_mask, q->node);
1095 spin_lock_irq(q->queue_lock);
1096 rq = get_request(q, rw, NULL, gfp_mask);
1098 spin_unlock_irq(q->queue_lock);
1099 /* q->queue_lock is unlocked at this point */
1103 EXPORT_SYMBOL(blk_get_request);
1106 * blk_make_request - given a bio, allocate a corresponding struct request.
1107 * @q: target request queue
1108 * @bio: The bio describing the memory mappings that will be submitted for IO.
1109 * It may be a chained-bio properly constructed by block/bio layer.
1110 * @gfp_mask: gfp flags to be used for memory allocation
1112 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1113 * type commands. Where the struct request needs to be farther initialized by
1114 * the caller. It is passed a &struct bio, which describes the memory info of
1117 * The caller of blk_make_request must make sure that bi_io_vec
1118 * are set to describe the memory buffers. That bio_data_dir() will return
1119 * the needed direction of the request. (And all bio's in the passed bio-chain
1120 * are properly set accordingly)
1122 * If called under none-sleepable conditions, mapped bio buffers must not
1123 * need bouncing, by calling the appropriate masked or flagged allocator,
1124 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1127 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1128 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1129 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1130 * completion of a bio that hasn't been submitted yet, thus resulting in a
1131 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1132 * of bio_alloc(), as that avoids the mempool deadlock.
1133 * If possible a big IO should be split into smaller parts when allocation
1134 * fails. Partial allocation should not be an error, or you risk a live-lock.
1136 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1139 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1142 return ERR_PTR(-ENOMEM);
1145 struct bio *bounce_bio = bio;
1148 blk_queue_bounce(q, &bounce_bio);
1149 ret = blk_rq_append_bio(q, rq, bounce_bio);
1150 if (unlikely(ret)) {
1151 blk_put_request(rq);
1152 return ERR_PTR(ret);
1158 EXPORT_SYMBOL(blk_make_request);
1161 * blk_requeue_request - put a request back on queue
1162 * @q: request queue where request should be inserted
1163 * @rq: request to be inserted
1166 * Drivers often keep queueing requests until the hardware cannot accept
1167 * more, when that condition happens we need to put the request back
1168 * on the queue. Must be called with queue lock held.
1170 void blk_requeue_request(struct request_queue *q, struct request *rq)
1172 blk_delete_timer(rq);
1173 blk_clear_rq_complete(rq);
1174 trace_block_rq_requeue(q, rq);
1176 if (blk_rq_tagged(rq))
1177 blk_queue_end_tag(q, rq);
1179 BUG_ON(blk_queued_rq(rq));
1181 elv_requeue_request(q, rq);
1183 EXPORT_SYMBOL(blk_requeue_request);
1185 static void add_acct_request(struct request_queue *q, struct request *rq,
1188 drive_stat_acct(rq, 1);
1189 __elv_add_request(q, rq, where);
1192 static void part_round_stats_single(int cpu, struct hd_struct *part,
1195 if (now == part->stamp)
1198 if (part_in_flight(part)) {
1199 __part_stat_add(cpu, part, time_in_queue,
1200 part_in_flight(part) * (now - part->stamp));
1201 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1207 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1208 * @cpu: cpu number for stats access
1209 * @part: target partition
1211 * The average IO queue length and utilisation statistics are maintained
1212 * by observing the current state of the queue length and the amount of
1213 * time it has been in this state for.
1215 * Normally, that accounting is done on IO completion, but that can result
1216 * in more than a second's worth of IO being accounted for within any one
1217 * second, leading to >100% utilisation. To deal with that, we call this
1218 * function to do a round-off before returning the results when reading
1219 * /proc/diskstats. This accounts immediately for all queue usage up to
1220 * the current jiffies and restarts the counters again.
1222 void part_round_stats(int cpu, struct hd_struct *part)
1224 unsigned long now = jiffies;
1227 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1228 part_round_stats_single(cpu, part, now);
1230 EXPORT_SYMBOL_GPL(part_round_stats);
1233 * queue lock must be held
1235 void __blk_put_request(struct request_queue *q, struct request *req)
1239 if (unlikely(--req->ref_count))
1242 elv_completed_request(q, req);
1244 /* this is a bio leak */
1245 WARN_ON(req->bio != NULL);
1248 * Request may not have originated from ll_rw_blk. if not,
1249 * it didn't come out of our reserved rq pools
1251 if (req->cmd_flags & REQ_ALLOCED) {
1252 unsigned int flags = req->cmd_flags;
1253 struct request_list *rl = blk_rq_rl(req);
1255 BUG_ON(!list_empty(&req->queuelist));
1256 BUG_ON(!hlist_unhashed(&req->hash));
1258 blk_free_request(rl, req);
1259 freed_request(rl, flags);
1263 EXPORT_SYMBOL_GPL(__blk_put_request);
1265 void blk_put_request(struct request *req)
1267 unsigned long flags;
1268 struct request_queue *q = req->q;
1270 spin_lock_irqsave(q->queue_lock, flags);
1271 __blk_put_request(q, req);
1272 spin_unlock_irqrestore(q->queue_lock, flags);
1274 EXPORT_SYMBOL(blk_put_request);
1277 * blk_add_request_payload - add a payload to a request
1278 * @rq: request to update
1279 * @page: page backing the payload
1280 * @len: length of the payload.
1282 * This allows to later add a payload to an already submitted request by
1283 * a block driver. The driver needs to take care of freeing the payload
1286 * Note that this is a quite horrible hack and nothing but handling of
1287 * discard requests should ever use it.
1289 void blk_add_request_payload(struct request *rq, struct page *page,
1292 struct bio *bio = rq->bio;
1294 bio->bi_io_vec->bv_page = page;
1295 bio->bi_io_vec->bv_offset = 0;
1296 bio->bi_io_vec->bv_len = len;
1300 bio->bi_phys_segments = 1;
1302 rq->__data_len = rq->resid_len = len;
1303 rq->nr_phys_segments = 1;
1304 rq->buffer = bio_data(bio);
1306 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1308 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1311 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1313 if (!ll_back_merge_fn(q, req, bio))
1316 trace_block_bio_backmerge(q, bio);
1318 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1319 blk_rq_set_mixed_merge(req);
1321 req->biotail->bi_next = bio;
1323 req->__data_len += bio->bi_size;
1324 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1326 drive_stat_acct(req, 0);
1330 static bool bio_attempt_front_merge(struct request_queue *q,
1331 struct request *req, struct bio *bio)
1333 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1335 if (!ll_front_merge_fn(q, req, bio))
1338 trace_block_bio_frontmerge(q, bio);
1340 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1341 blk_rq_set_mixed_merge(req);
1343 bio->bi_next = req->bio;
1347 * may not be valid. if the low level driver said
1348 * it didn't need a bounce buffer then it better
1349 * not touch req->buffer either...
1351 req->buffer = bio_data(bio);
1352 req->__sector = bio->bi_sector;
1353 req->__data_len += bio->bi_size;
1354 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1356 drive_stat_acct(req, 0);
1361 * attempt_plug_merge - try to merge with %current's plugged list
1362 * @q: request_queue new bio is being queued at
1363 * @bio: new bio being queued
1364 * @request_count: out parameter for number of traversed plugged requests
1366 * Determine whether @bio being queued on @q can be merged with a request
1367 * on %current's plugged list. Returns %true if merge was successful,
1370 * Plugging coalesces IOs from the same issuer for the same purpose without
1371 * going through @q->queue_lock. As such it's more of an issuing mechanism
1372 * than scheduling, and the request, while may have elvpriv data, is not
1373 * added on the elevator at this point. In addition, we don't have
1374 * reliable access to the elevator outside queue lock. Only check basic
1375 * merging parameters without querying the elevator.
1377 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1378 unsigned int *request_count)
1380 struct blk_plug *plug;
1384 plug = current->plug;
1389 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1395 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1398 el_ret = blk_try_merge(rq, bio);
1399 if (el_ret == ELEVATOR_BACK_MERGE) {
1400 ret = bio_attempt_back_merge(q, rq, bio);
1403 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1404 ret = bio_attempt_front_merge(q, rq, bio);
1413 void init_request_from_bio(struct request *req, struct bio *bio)
1415 req->cmd_type = REQ_TYPE_FS;
1417 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1418 if (bio->bi_rw & REQ_RAHEAD)
1419 req->cmd_flags |= REQ_FAILFAST_MASK;
1422 req->__sector = bio->bi_sector;
1423 req->ioprio = bio_prio(bio);
1424 blk_rq_bio_prep(req->q, req, bio);
1427 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1429 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1430 struct blk_plug *plug;
1431 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1432 struct request *req;
1433 unsigned int request_count = 0;
1436 * low level driver can indicate that it wants pages above a
1437 * certain limit bounced to low memory (ie for highmem, or even
1438 * ISA dma in theory)
1440 blk_queue_bounce(q, &bio);
1442 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1443 spin_lock_irq(q->queue_lock);
1444 where = ELEVATOR_INSERT_FLUSH;
1449 * Check if we can merge with the plugged list before grabbing
1452 if (attempt_plug_merge(q, bio, &request_count))
1455 spin_lock_irq(q->queue_lock);
1457 el_ret = elv_merge(q, &req, bio);
1458 if (el_ret == ELEVATOR_BACK_MERGE) {
1459 if (bio_attempt_back_merge(q, req, bio)) {
1460 elv_bio_merged(q, req, bio);
1461 if (!attempt_back_merge(q, req))
1462 elv_merged_request(q, req, el_ret);
1465 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1466 if (bio_attempt_front_merge(q, req, bio)) {
1467 elv_bio_merged(q, req, bio);
1468 if (!attempt_front_merge(q, req))
1469 elv_merged_request(q, req, el_ret);
1476 * This sync check and mask will be re-done in init_request_from_bio(),
1477 * but we need to set it earlier to expose the sync flag to the
1478 * rq allocator and io schedulers.
1480 rw_flags = bio_data_dir(bio);
1482 rw_flags |= REQ_SYNC;
1485 * Grab a free request. This is might sleep but can not fail.
1486 * Returns with the queue unlocked.
1488 req = get_request(q, rw_flags, bio, GFP_NOIO);
1489 if (unlikely(!req)) {
1490 bio_endio(bio, -ENODEV); /* @q is dead */
1495 * After dropping the lock and possibly sleeping here, our request
1496 * may now be mergeable after it had proven unmergeable (above).
1497 * We don't worry about that case for efficiency. It won't happen
1498 * often, and the elevators are able to handle it.
1500 init_request_from_bio(req, bio);
1502 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1503 req->cpu = raw_smp_processor_id();
1505 plug = current->plug;
1508 * If this is the first request added after a plug, fire
1509 * of a plug trace. If others have been added before, check
1510 * if we have multiple devices in this plug. If so, make a
1511 * note to sort the list before dispatch.
1513 if (list_empty(&plug->list))
1514 trace_block_plug(q);
1516 if (!plug->should_sort) {
1517 struct request *__rq;
1519 __rq = list_entry_rq(plug->list.prev);
1521 plug->should_sort = 1;
1523 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1524 blk_flush_plug_list(plug, false);
1525 trace_block_plug(q);
1528 list_add_tail(&req->queuelist, &plug->list);
1529 drive_stat_acct(req, 1);
1531 spin_lock_irq(q->queue_lock);
1532 add_acct_request(q, req, where);
1535 spin_unlock_irq(q->queue_lock);
1538 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1541 * If bio->bi_dev is a partition, remap the location
1543 static inline void blk_partition_remap(struct bio *bio)
1545 struct block_device *bdev = bio->bi_bdev;
1547 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1548 struct hd_struct *p = bdev->bd_part;
1550 bio->bi_sector += p->start_sect;
1551 bio->bi_bdev = bdev->bd_contains;
1553 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1555 bio->bi_sector - p->start_sect);
1559 static void handle_bad_sector(struct bio *bio)
1561 char b[BDEVNAME_SIZE];
1563 printk(KERN_INFO "attempt to access beyond end of device\n");
1564 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1565 bdevname(bio->bi_bdev, b),
1567 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1568 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1570 set_bit(BIO_EOF, &bio->bi_flags);
1573 #ifdef CONFIG_FAIL_MAKE_REQUEST
1575 static DECLARE_FAULT_ATTR(fail_make_request);
1577 static int __init setup_fail_make_request(char *str)
1579 return setup_fault_attr(&fail_make_request, str);
1581 __setup("fail_make_request=", setup_fail_make_request);
1583 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1585 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1588 static int __init fail_make_request_debugfs(void)
1590 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1591 NULL, &fail_make_request);
1593 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1596 late_initcall(fail_make_request_debugfs);
1598 #else /* CONFIG_FAIL_MAKE_REQUEST */
1600 static inline bool should_fail_request(struct hd_struct *part,
1606 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1609 * Check whether this bio extends beyond the end of the device.
1611 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1618 /* Test device or partition size, when known. */
1619 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1621 sector_t sector = bio->bi_sector;
1623 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1625 * This may well happen - the kernel calls bread()
1626 * without checking the size of the device, e.g., when
1627 * mounting a device.
1629 handle_bad_sector(bio);
1637 static noinline_for_stack bool
1638 generic_make_request_checks(struct bio *bio)
1640 struct request_queue *q;
1641 int nr_sectors = bio_sectors(bio);
1643 char b[BDEVNAME_SIZE];
1644 struct hd_struct *part;
1648 if (bio_check_eod(bio, nr_sectors))
1651 q = bdev_get_queue(bio->bi_bdev);
1654 "generic_make_request: Trying to access "
1655 "nonexistent block-device %s (%Lu)\n",
1656 bdevname(bio->bi_bdev, b),
1657 (long long) bio->bi_sector);
1661 if (likely(bio_is_rw(bio) &&
1662 nr_sectors > queue_max_hw_sectors(q))) {
1663 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1664 bdevname(bio->bi_bdev, b),
1666 queue_max_hw_sectors(q));
1670 part = bio->bi_bdev->bd_part;
1671 if (should_fail_request(part, bio->bi_size) ||
1672 should_fail_request(&part_to_disk(part)->part0,
1677 * If this device has partitions, remap block n
1678 * of partition p to block n+start(p) of the disk.
1680 blk_partition_remap(bio);
1682 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1685 if (bio_check_eod(bio, nr_sectors))
1689 * Filter flush bio's early so that make_request based
1690 * drivers without flush support don't have to worry
1693 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1694 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1701 if ((bio->bi_rw & REQ_DISCARD) &&
1702 (!blk_queue_discard(q) ||
1703 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1708 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1714 * Various block parts want %current->io_context and lazy ioc
1715 * allocation ends up trading a lot of pain for a small amount of
1716 * memory. Just allocate it upfront. This may fail and block
1717 * layer knows how to live with it.
1719 create_io_context(GFP_ATOMIC, q->node);
1721 if (blk_throtl_bio(q, bio))
1722 return false; /* throttled, will be resubmitted later */
1724 trace_block_bio_queue(q, bio);
1728 bio_endio(bio, err);
1733 * generic_make_request - hand a buffer to its device driver for I/O
1734 * @bio: The bio describing the location in memory and on the device.
1736 * generic_make_request() is used to make I/O requests of block
1737 * devices. It is passed a &struct bio, which describes the I/O that needs
1740 * generic_make_request() does not return any status. The
1741 * success/failure status of the request, along with notification of
1742 * completion, is delivered asynchronously through the bio->bi_end_io
1743 * function described (one day) else where.
1745 * The caller of generic_make_request must make sure that bi_io_vec
1746 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1747 * set to describe the device address, and the
1748 * bi_end_io and optionally bi_private are set to describe how
1749 * completion notification should be signaled.
1751 * generic_make_request and the drivers it calls may use bi_next if this
1752 * bio happens to be merged with someone else, and may resubmit the bio to
1753 * a lower device by calling into generic_make_request recursively, which
1754 * means the bio should NOT be touched after the call to ->make_request_fn.
1756 void generic_make_request(struct bio *bio)
1758 struct bio_list bio_list_on_stack;
1760 if (!generic_make_request_checks(bio))
1764 * We only want one ->make_request_fn to be active at a time, else
1765 * stack usage with stacked devices could be a problem. So use
1766 * current->bio_list to keep a list of requests submited by a
1767 * make_request_fn function. current->bio_list is also used as a
1768 * flag to say if generic_make_request is currently active in this
1769 * task or not. If it is NULL, then no make_request is active. If
1770 * it is non-NULL, then a make_request is active, and new requests
1771 * should be added at the tail
1773 if (current->bio_list) {
1774 bio_list_add(current->bio_list, bio);
1778 /* following loop may be a bit non-obvious, and so deserves some
1780 * Before entering the loop, bio->bi_next is NULL (as all callers
1781 * ensure that) so we have a list with a single bio.
1782 * We pretend that we have just taken it off a longer list, so
1783 * we assign bio_list to a pointer to the bio_list_on_stack,
1784 * thus initialising the bio_list of new bios to be
1785 * added. ->make_request() may indeed add some more bios
1786 * through a recursive call to generic_make_request. If it
1787 * did, we find a non-NULL value in bio_list and re-enter the loop
1788 * from the top. In this case we really did just take the bio
1789 * of the top of the list (no pretending) and so remove it from
1790 * bio_list, and call into ->make_request() again.
1792 BUG_ON(bio->bi_next);
1793 bio_list_init(&bio_list_on_stack);
1794 current->bio_list = &bio_list_on_stack;
1796 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1798 q->make_request_fn(q, bio);
1800 bio = bio_list_pop(current->bio_list);
1802 current->bio_list = NULL; /* deactivate */
1804 EXPORT_SYMBOL(generic_make_request);
1807 * submit_bio - submit a bio to the block device layer for I/O
1808 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1809 * @bio: The &struct bio which describes the I/O
1811 * submit_bio() is very similar in purpose to generic_make_request(), and
1812 * uses that function to do most of the work. Both are fairly rough
1813 * interfaces; @bio must be presetup and ready for I/O.
1816 void submit_bio(int rw, struct bio *bio)
1821 * If it's a regular read/write or a barrier with data attached,
1822 * go through the normal accounting stuff before submission.
1824 if (bio_has_data(bio)) {
1827 if (unlikely(rw & REQ_WRITE_SAME))
1828 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1830 count = bio_sectors(bio);
1833 count_vm_events(PGPGOUT, count);
1835 task_io_account_read(bio->bi_size);
1836 count_vm_events(PGPGIN, count);
1839 if (unlikely(block_dump)) {
1840 char b[BDEVNAME_SIZE];
1841 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1842 current->comm, task_pid_nr(current),
1843 (rw & WRITE) ? "WRITE" : "READ",
1844 (unsigned long long)bio->bi_sector,
1845 bdevname(bio->bi_bdev, b),
1850 generic_make_request(bio);
1852 EXPORT_SYMBOL(submit_bio);
1855 * blk_rq_check_limits - Helper function to check a request for the queue limit
1857 * @rq: the request being checked
1860 * @rq may have been made based on weaker limitations of upper-level queues
1861 * in request stacking drivers, and it may violate the limitation of @q.
1862 * Since the block layer and the underlying device driver trust @rq
1863 * after it is inserted to @q, it should be checked against @q before
1864 * the insertion using this generic function.
1866 * This function should also be useful for request stacking drivers
1867 * in some cases below, so export this function.
1868 * Request stacking drivers like request-based dm may change the queue
1869 * limits while requests are in the queue (e.g. dm's table swapping).
1870 * Such request stacking drivers should check those requests agaist
1871 * the new queue limits again when they dispatch those requests,
1872 * although such checkings are also done against the old queue limits
1873 * when submitting requests.
1875 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1877 if (!rq_mergeable(rq))
1880 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1881 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1886 * queue's settings related to segment counting like q->bounce_pfn
1887 * may differ from that of other stacking queues.
1888 * Recalculate it to check the request correctly on this queue's
1891 blk_recalc_rq_segments(rq);
1892 if (rq->nr_phys_segments > queue_max_segments(q)) {
1893 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1899 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1902 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1903 * @q: the queue to submit the request
1904 * @rq: the request being queued
1906 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1908 unsigned long flags;
1909 int where = ELEVATOR_INSERT_BACK;
1911 if (blk_rq_check_limits(q, rq))
1915 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1918 spin_lock_irqsave(q->queue_lock, flags);
1919 if (unlikely(blk_queue_dying(q))) {
1920 spin_unlock_irqrestore(q->queue_lock, flags);
1925 * Submitting request must be dequeued before calling this function
1926 * because it will be linked to another request_queue
1928 BUG_ON(blk_queued_rq(rq));
1930 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1931 where = ELEVATOR_INSERT_FLUSH;
1933 add_acct_request(q, rq, where);
1934 if (where == ELEVATOR_INSERT_FLUSH)
1936 spin_unlock_irqrestore(q->queue_lock, flags);
1940 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1943 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1944 * @rq: request to examine
1947 * A request could be merge of IOs which require different failure
1948 * handling. This function determines the number of bytes which
1949 * can be failed from the beginning of the request without
1950 * crossing into area which need to be retried further.
1953 * The number of bytes to fail.
1956 * queue_lock must be held.
1958 unsigned int blk_rq_err_bytes(const struct request *rq)
1960 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1961 unsigned int bytes = 0;
1964 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1965 return blk_rq_bytes(rq);
1968 * Currently the only 'mixing' which can happen is between
1969 * different fastfail types. We can safely fail portions
1970 * which have all the failfast bits that the first one has -
1971 * the ones which are at least as eager to fail as the first
1974 for (bio = rq->bio; bio; bio = bio->bi_next) {
1975 if ((bio->bi_rw & ff) != ff)
1977 bytes += bio->bi_size;
1980 /* this could lead to infinite loop */
1981 BUG_ON(blk_rq_bytes(rq) && !bytes);
1984 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1986 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1988 if (blk_do_io_stat(req)) {
1989 const int rw = rq_data_dir(req);
1990 struct hd_struct *part;
1993 cpu = part_stat_lock();
1995 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2000 static void blk_account_io_done(struct request *req)
2003 * Account IO completion. flush_rq isn't accounted as a
2004 * normal IO on queueing nor completion. Accounting the
2005 * containing request is enough.
2007 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2008 unsigned long duration = jiffies - req->start_time;
2009 const int rw = rq_data_dir(req);
2010 struct hd_struct *part;
2013 cpu = part_stat_lock();
2016 part_stat_inc(cpu, part, ios[rw]);
2017 part_stat_add(cpu, part, ticks[rw], duration);
2018 part_round_stats(cpu, part);
2019 part_dec_in_flight(part, rw);
2021 hd_struct_put(part);
2027 * blk_peek_request - peek at the top of a request queue
2028 * @q: request queue to peek at
2031 * Return the request at the top of @q. The returned request
2032 * should be started using blk_start_request() before LLD starts
2036 * Pointer to the request at the top of @q if available. Null
2040 * queue_lock must be held.
2042 struct request *blk_peek_request(struct request_queue *q)
2047 while ((rq = __elv_next_request(q)) != NULL) {
2048 if (!(rq->cmd_flags & REQ_STARTED)) {
2050 * This is the first time the device driver
2051 * sees this request (possibly after
2052 * requeueing). Notify IO scheduler.
2054 if (rq->cmd_flags & REQ_SORTED)
2055 elv_activate_rq(q, rq);
2058 * just mark as started even if we don't start
2059 * it, a request that has been delayed should
2060 * not be passed by new incoming requests
2062 rq->cmd_flags |= REQ_STARTED;
2063 trace_block_rq_issue(q, rq);
2066 if (!q->boundary_rq || q->boundary_rq == rq) {
2067 q->end_sector = rq_end_sector(rq);
2068 q->boundary_rq = NULL;
2071 if (rq->cmd_flags & REQ_DONTPREP)
2074 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2076 * make sure space for the drain appears we
2077 * know we can do this because max_hw_segments
2078 * has been adjusted to be one fewer than the
2081 rq->nr_phys_segments++;
2087 ret = q->prep_rq_fn(q, rq);
2088 if (ret == BLKPREP_OK) {
2090 } else if (ret == BLKPREP_DEFER) {
2092 * the request may have been (partially) prepped.
2093 * we need to keep this request in the front to
2094 * avoid resource deadlock. REQ_STARTED will
2095 * prevent other fs requests from passing this one.
2097 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2098 !(rq->cmd_flags & REQ_DONTPREP)) {
2100 * remove the space for the drain we added
2101 * so that we don't add it again
2103 --rq->nr_phys_segments;
2108 } else if (ret == BLKPREP_KILL) {
2109 rq->cmd_flags |= REQ_QUIET;
2111 * Mark this request as started so we don't trigger
2112 * any debug logic in the end I/O path.
2114 blk_start_request(rq);
2115 __blk_end_request_all(rq, -EIO);
2117 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2124 EXPORT_SYMBOL(blk_peek_request);
2126 void blk_dequeue_request(struct request *rq)
2128 struct request_queue *q = rq->q;
2130 BUG_ON(list_empty(&rq->queuelist));
2131 BUG_ON(ELV_ON_HASH(rq));
2133 list_del_init(&rq->queuelist);
2136 * the time frame between a request being removed from the lists
2137 * and to it is freed is accounted as io that is in progress at
2140 if (blk_account_rq(rq)) {
2141 q->in_flight[rq_is_sync(rq)]++;
2142 set_io_start_time_ns(rq);
2147 * blk_start_request - start request processing on the driver
2148 * @req: request to dequeue
2151 * Dequeue @req and start timeout timer on it. This hands off the
2152 * request to the driver.
2154 * Block internal functions which don't want to start timer should
2155 * call blk_dequeue_request().
2158 * queue_lock must be held.
2160 void blk_start_request(struct request *req)
2162 blk_dequeue_request(req);
2165 * We are now handing the request to the hardware, initialize
2166 * resid_len to full count and add the timeout handler.
2168 req->resid_len = blk_rq_bytes(req);
2169 if (unlikely(blk_bidi_rq(req)))
2170 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2174 EXPORT_SYMBOL(blk_start_request);
2177 * blk_fetch_request - fetch a request from a request queue
2178 * @q: request queue to fetch a request from
2181 * Return the request at the top of @q. The request is started on
2182 * return and LLD can start processing it immediately.
2185 * Pointer to the request at the top of @q if available. Null
2189 * queue_lock must be held.
2191 struct request *blk_fetch_request(struct request_queue *q)
2195 rq = blk_peek_request(q);
2197 blk_start_request(rq);
2200 EXPORT_SYMBOL(blk_fetch_request);
2203 * blk_update_request - Special helper function for request stacking drivers
2204 * @req: the request being processed
2205 * @error: %0 for success, < %0 for error
2206 * @nr_bytes: number of bytes to complete @req
2209 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2210 * the request structure even if @req doesn't have leftover.
2211 * If @req has leftover, sets it up for the next range of segments.
2213 * This special helper function is only for request stacking drivers
2214 * (e.g. request-based dm) so that they can handle partial completion.
2215 * Actual device drivers should use blk_end_request instead.
2217 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2218 * %false return from this function.
2221 * %false - this request doesn't have any more data
2222 * %true - this request has more data
2224 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2226 int total_bytes, bio_nbytes, next_idx = 0;
2232 trace_block_rq_complete(req->q, req);
2235 * For fs requests, rq is just carrier of independent bio's
2236 * and each partial completion should be handled separately.
2237 * Reset per-request error on each partial completion.
2239 * TODO: tj: This is too subtle. It would be better to let
2240 * low level drivers do what they see fit.
2242 if (req->cmd_type == REQ_TYPE_FS)
2245 if (error && req->cmd_type == REQ_TYPE_FS &&
2246 !(req->cmd_flags & REQ_QUIET)) {
2251 error_type = "recoverable transport";
2254 error_type = "critical target";
2257 error_type = "critical nexus";
2264 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2265 error_type, req->rq_disk ?
2266 req->rq_disk->disk_name : "?",
2267 (unsigned long long)blk_rq_pos(req));
2271 blk_account_io_completion(req, nr_bytes);
2273 total_bytes = bio_nbytes = 0;
2274 while ((bio = req->bio) != NULL) {
2277 if (nr_bytes >= bio->bi_size) {
2278 req->bio = bio->bi_next;
2279 nbytes = bio->bi_size;
2280 req_bio_endio(req, bio, nbytes, error);
2284 int idx = bio->bi_idx + next_idx;
2286 if (unlikely(idx >= bio->bi_vcnt)) {
2287 blk_dump_rq_flags(req, "__end_that");
2288 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2289 __func__, idx, bio->bi_vcnt);
2293 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2294 BIO_BUG_ON(nbytes > bio->bi_size);
2297 * not a complete bvec done
2299 if (unlikely(nbytes > nr_bytes)) {
2300 bio_nbytes += nr_bytes;
2301 total_bytes += nr_bytes;
2306 * advance to the next vector
2309 bio_nbytes += nbytes;
2312 total_bytes += nbytes;
2318 * end more in this run, or just return 'not-done'
2320 if (unlikely(nr_bytes <= 0))
2330 * Reset counters so that the request stacking driver
2331 * can find how many bytes remain in the request
2334 req->__data_len = 0;
2339 * if the request wasn't completed, update state
2342 req_bio_endio(req, bio, bio_nbytes, error);
2343 bio->bi_idx += next_idx;
2344 bio_iovec(bio)->bv_offset += nr_bytes;
2345 bio_iovec(bio)->bv_len -= nr_bytes;
2348 req->__data_len -= total_bytes;
2349 req->buffer = bio_data(req->bio);
2351 /* update sector only for requests with clear definition of sector */
2352 if (req->cmd_type == REQ_TYPE_FS)
2353 req->__sector += total_bytes >> 9;
2355 /* mixed attributes always follow the first bio */
2356 if (req->cmd_flags & REQ_MIXED_MERGE) {
2357 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2358 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2362 * If total number of sectors is less than the first segment
2363 * size, something has gone terribly wrong.
2365 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2366 blk_dump_rq_flags(req, "request botched");
2367 req->__data_len = blk_rq_cur_bytes(req);
2370 /* recalculate the number of segments */
2371 blk_recalc_rq_segments(req);
2375 EXPORT_SYMBOL_GPL(blk_update_request);
2377 static bool blk_update_bidi_request(struct request *rq, int error,
2378 unsigned int nr_bytes,
2379 unsigned int bidi_bytes)
2381 if (blk_update_request(rq, error, nr_bytes))
2384 /* Bidi request must be completed as a whole */
2385 if (unlikely(blk_bidi_rq(rq)) &&
2386 blk_update_request(rq->next_rq, error, bidi_bytes))
2389 if (blk_queue_add_random(rq->q))
2390 add_disk_randomness(rq->rq_disk);
2396 * blk_unprep_request - unprepare a request
2399 * This function makes a request ready for complete resubmission (or
2400 * completion). It happens only after all error handling is complete,
2401 * so represents the appropriate moment to deallocate any resources
2402 * that were allocated to the request in the prep_rq_fn. The queue
2403 * lock is held when calling this.
2405 void blk_unprep_request(struct request *req)
2407 struct request_queue *q = req->q;
2409 req->cmd_flags &= ~REQ_DONTPREP;
2410 if (q->unprep_rq_fn)
2411 q->unprep_rq_fn(q, req);
2413 EXPORT_SYMBOL_GPL(blk_unprep_request);
2416 * queue lock must be held
2418 static void blk_finish_request(struct request *req, int error)
2420 if (blk_rq_tagged(req))
2421 blk_queue_end_tag(req->q, req);
2423 BUG_ON(blk_queued_rq(req));
2425 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2426 laptop_io_completion(&req->q->backing_dev_info);
2428 blk_delete_timer(req);
2430 if (req->cmd_flags & REQ_DONTPREP)
2431 blk_unprep_request(req);
2434 blk_account_io_done(req);
2437 req->end_io(req, error);
2439 if (blk_bidi_rq(req))
2440 __blk_put_request(req->next_rq->q, req->next_rq);
2442 __blk_put_request(req->q, req);
2447 * blk_end_bidi_request - Complete a bidi request
2448 * @rq: the request to complete
2449 * @error: %0 for success, < %0 for error
2450 * @nr_bytes: number of bytes to complete @rq
2451 * @bidi_bytes: number of bytes to complete @rq->next_rq
2454 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2455 * Drivers that supports bidi can safely call this member for any
2456 * type of request, bidi or uni. In the later case @bidi_bytes is
2460 * %false - we are done with this request
2461 * %true - still buffers pending for this request
2463 static bool blk_end_bidi_request(struct request *rq, int error,
2464 unsigned int nr_bytes, unsigned int bidi_bytes)
2466 struct request_queue *q = rq->q;
2467 unsigned long flags;
2469 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2472 spin_lock_irqsave(q->queue_lock, flags);
2473 blk_finish_request(rq, error);
2474 spin_unlock_irqrestore(q->queue_lock, flags);
2480 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2481 * @rq: the request to complete
2482 * @error: %0 for success, < %0 for error
2483 * @nr_bytes: number of bytes to complete @rq
2484 * @bidi_bytes: number of bytes to complete @rq->next_rq
2487 * Identical to blk_end_bidi_request() except that queue lock is
2488 * assumed to be locked on entry and remains so on return.
2491 * %false - we are done with this request
2492 * %true - still buffers pending for this request
2494 bool __blk_end_bidi_request(struct request *rq, int error,
2495 unsigned int nr_bytes, unsigned int bidi_bytes)
2497 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2500 blk_finish_request(rq, error);
2506 * blk_end_request - Helper function for drivers to complete the request.
2507 * @rq: the request being processed
2508 * @error: %0 for success, < %0 for error
2509 * @nr_bytes: number of bytes to complete
2512 * Ends I/O on a number of bytes attached to @rq.
2513 * If @rq has leftover, sets it up for the next range of segments.
2516 * %false - we are done with this request
2517 * %true - still buffers pending for this request
2519 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2521 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2523 EXPORT_SYMBOL(blk_end_request);
2526 * blk_end_request_all - Helper function for drives to finish the request.
2527 * @rq: the request to finish
2528 * @error: %0 for success, < %0 for error
2531 * Completely finish @rq.
2533 void blk_end_request_all(struct request *rq, int error)
2536 unsigned int bidi_bytes = 0;
2538 if (unlikely(blk_bidi_rq(rq)))
2539 bidi_bytes = blk_rq_bytes(rq->next_rq);
2541 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2544 EXPORT_SYMBOL(blk_end_request_all);
2547 * blk_end_request_cur - Helper function to finish the current request chunk.
2548 * @rq: the request to finish the current chunk for
2549 * @error: %0 for success, < %0 for error
2552 * Complete the current consecutively mapped chunk from @rq.
2555 * %false - we are done with this request
2556 * %true - still buffers pending for this request
2558 bool blk_end_request_cur(struct request *rq, int error)
2560 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2562 EXPORT_SYMBOL(blk_end_request_cur);
2565 * blk_end_request_err - Finish a request till the next failure boundary.
2566 * @rq: the request to finish till the next failure boundary for
2567 * @error: must be negative errno
2570 * Complete @rq till the next failure boundary.
2573 * %false - we are done with this request
2574 * %true - still buffers pending for this request
2576 bool blk_end_request_err(struct request *rq, int error)
2578 WARN_ON(error >= 0);
2579 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2581 EXPORT_SYMBOL_GPL(blk_end_request_err);
2584 * __blk_end_request - Helper function for drivers to complete the request.
2585 * @rq: the request being processed
2586 * @error: %0 for success, < %0 for error
2587 * @nr_bytes: number of bytes to complete
2590 * Must be called with queue lock held unlike blk_end_request().
2593 * %false - we are done with this request
2594 * %true - still buffers pending for this request
2596 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2598 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2600 EXPORT_SYMBOL(__blk_end_request);
2603 * __blk_end_request_all - Helper function for drives to finish the request.
2604 * @rq: the request to finish
2605 * @error: %0 for success, < %0 for error
2608 * Completely finish @rq. Must be called with queue lock held.
2610 void __blk_end_request_all(struct request *rq, int error)
2613 unsigned int bidi_bytes = 0;
2615 if (unlikely(blk_bidi_rq(rq)))
2616 bidi_bytes = blk_rq_bytes(rq->next_rq);
2618 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2621 EXPORT_SYMBOL(__blk_end_request_all);
2624 * __blk_end_request_cur - Helper function to finish the current request chunk.
2625 * @rq: the request to finish the current chunk for
2626 * @error: %0 for success, < %0 for error
2629 * Complete the current consecutively mapped chunk from @rq. Must
2630 * be called with queue lock held.
2633 * %false - we are done with this request
2634 * %true - still buffers pending for this request
2636 bool __blk_end_request_cur(struct request *rq, int error)
2638 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2640 EXPORT_SYMBOL(__blk_end_request_cur);
2643 * __blk_end_request_err - Finish a request till the next failure boundary.
2644 * @rq: the request to finish till the next failure boundary for
2645 * @error: must be negative errno
2648 * Complete @rq till the next failure boundary. Must be called
2649 * with queue lock held.
2652 * %false - we are done with this request
2653 * %true - still buffers pending for this request
2655 bool __blk_end_request_err(struct request *rq, int error)
2657 WARN_ON(error >= 0);
2658 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2660 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2662 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2665 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2666 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2668 if (bio_has_data(bio)) {
2669 rq->nr_phys_segments = bio_phys_segments(q, bio);
2670 rq->buffer = bio_data(bio);
2672 rq->__data_len = bio->bi_size;
2673 rq->bio = rq->biotail = bio;
2676 rq->rq_disk = bio->bi_bdev->bd_disk;
2679 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2681 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2682 * @rq: the request to be flushed
2685 * Flush all pages in @rq.
2687 void rq_flush_dcache_pages(struct request *rq)
2689 struct req_iterator iter;
2690 struct bio_vec *bvec;
2692 rq_for_each_segment(bvec, rq, iter)
2693 flush_dcache_page(bvec->bv_page);
2695 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2699 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2700 * @q : the queue of the device being checked
2703 * Check if underlying low-level drivers of a device are busy.
2704 * If the drivers want to export their busy state, they must set own
2705 * exporting function using blk_queue_lld_busy() first.
2707 * Basically, this function is used only by request stacking drivers
2708 * to stop dispatching requests to underlying devices when underlying
2709 * devices are busy. This behavior helps more I/O merging on the queue
2710 * of the request stacking driver and prevents I/O throughput regression
2711 * on burst I/O load.
2714 * 0 - Not busy (The request stacking driver should dispatch request)
2715 * 1 - Busy (The request stacking driver should stop dispatching request)
2717 int blk_lld_busy(struct request_queue *q)
2720 return q->lld_busy_fn(q);
2724 EXPORT_SYMBOL_GPL(blk_lld_busy);
2727 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2728 * @rq: the clone request to be cleaned up
2731 * Free all bios in @rq for a cloned request.
2733 void blk_rq_unprep_clone(struct request *rq)
2737 while ((bio = rq->bio) != NULL) {
2738 rq->bio = bio->bi_next;
2743 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2746 * Copy attributes of the original request to the clone request.
2747 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2749 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2751 dst->cpu = src->cpu;
2752 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2753 dst->cmd_type = src->cmd_type;
2754 dst->__sector = blk_rq_pos(src);
2755 dst->__data_len = blk_rq_bytes(src);
2756 dst->nr_phys_segments = src->nr_phys_segments;
2757 dst->ioprio = src->ioprio;
2758 dst->extra_len = src->extra_len;
2762 * blk_rq_prep_clone - Helper function to setup clone request
2763 * @rq: the request to be setup
2764 * @rq_src: original request to be cloned
2765 * @bs: bio_set that bios for clone are allocated from
2766 * @gfp_mask: memory allocation mask for bio
2767 * @bio_ctr: setup function to be called for each clone bio.
2768 * Returns %0 for success, non %0 for failure.
2769 * @data: private data to be passed to @bio_ctr
2772 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2773 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2774 * are not copied, and copying such parts is the caller's responsibility.
2775 * Also, pages which the original bios are pointing to are not copied
2776 * and the cloned bios just point same pages.
2777 * So cloned bios must be completed before original bios, which means
2778 * the caller must complete @rq before @rq_src.
2780 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2781 struct bio_set *bs, gfp_t gfp_mask,
2782 int (*bio_ctr)(struct bio *, struct bio *, void *),
2785 struct bio *bio, *bio_src;
2790 blk_rq_init(NULL, rq);
2792 __rq_for_each_bio(bio_src, rq_src) {
2793 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2797 if (bio_ctr && bio_ctr(bio, bio_src, data))
2801 rq->biotail->bi_next = bio;
2804 rq->bio = rq->biotail = bio;
2807 __blk_rq_prep_clone(rq, rq_src);
2814 blk_rq_unprep_clone(rq);
2818 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2820 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2822 return queue_work(kblockd_workqueue, work);
2824 EXPORT_SYMBOL(kblockd_schedule_work);
2826 int kblockd_schedule_delayed_work(struct request_queue *q,
2827 struct delayed_work *dwork, unsigned long delay)
2829 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2831 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2833 #define PLUG_MAGIC 0x91827364
2836 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2837 * @plug: The &struct blk_plug that needs to be initialized
2840 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2841 * pending I/O should the task end up blocking between blk_start_plug() and
2842 * blk_finish_plug(). This is important from a performance perspective, but
2843 * also ensures that we don't deadlock. For instance, if the task is blocking
2844 * for a memory allocation, memory reclaim could end up wanting to free a
2845 * page belonging to that request that is currently residing in our private
2846 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2847 * this kind of deadlock.
2849 void blk_start_plug(struct blk_plug *plug)
2851 struct task_struct *tsk = current;
2853 plug->magic = PLUG_MAGIC;
2854 INIT_LIST_HEAD(&plug->list);
2855 INIT_LIST_HEAD(&plug->cb_list);
2856 plug->should_sort = 0;
2859 * If this is a nested plug, don't actually assign it. It will be
2860 * flushed on its own.
2864 * Store ordering should not be needed here, since a potential
2865 * preempt will imply a full memory barrier
2870 EXPORT_SYMBOL(blk_start_plug);
2872 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2874 struct request *rqa = container_of(a, struct request, queuelist);
2875 struct request *rqb = container_of(b, struct request, queuelist);
2877 return !(rqa->q < rqb->q ||
2878 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2882 * If 'from_schedule' is true, then postpone the dispatch of requests
2883 * until a safe kblockd context. We due this to avoid accidental big
2884 * additional stack usage in driver dispatch, in places where the originally
2885 * plugger did not intend it.
2887 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2889 __releases(q->queue_lock)
2891 trace_block_unplug(q, depth, !from_schedule);
2894 * Don't mess with a dying queue.
2896 if (unlikely(blk_queue_dying(q))) {
2897 spin_unlock(q->queue_lock);
2902 * If we are punting this to kblockd, then we can safely drop
2903 * the queue_lock before waking kblockd (which needs to take
2906 if (from_schedule) {
2907 spin_unlock(q->queue_lock);
2908 blk_run_queue_async(q);
2911 spin_unlock(q->queue_lock);
2916 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2918 LIST_HEAD(callbacks);
2920 while (!list_empty(&plug->cb_list)) {
2921 list_splice_init(&plug->cb_list, &callbacks);
2923 while (!list_empty(&callbacks)) {
2924 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2927 list_del(&cb->list);
2928 cb->callback(cb, from_schedule);
2933 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2936 struct blk_plug *plug = current->plug;
2937 struct blk_plug_cb *cb;
2942 list_for_each_entry(cb, &plug->cb_list, list)
2943 if (cb->callback == unplug && cb->data == data)
2946 /* Not currently on the callback list */
2947 BUG_ON(size < sizeof(*cb));
2948 cb = kzalloc(size, GFP_ATOMIC);
2951 cb->callback = unplug;
2952 list_add(&cb->list, &plug->cb_list);
2956 EXPORT_SYMBOL(blk_check_plugged);
2958 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2960 struct request_queue *q;
2961 unsigned long flags;
2966 BUG_ON(plug->magic != PLUG_MAGIC);
2968 flush_plug_callbacks(plug, from_schedule);
2969 if (list_empty(&plug->list))
2972 list_splice_init(&plug->list, &list);
2974 if (plug->should_sort) {
2975 list_sort(NULL, &list, plug_rq_cmp);
2976 plug->should_sort = 0;
2983 * Save and disable interrupts here, to avoid doing it for every
2984 * queue lock we have to take.
2986 local_irq_save(flags);
2987 while (!list_empty(&list)) {
2988 rq = list_entry_rq(list.next);
2989 list_del_init(&rq->queuelist);
2993 * This drops the queue lock
2996 queue_unplugged(q, depth, from_schedule);
2999 spin_lock(q->queue_lock);
3003 * Short-circuit if @q is dead
3005 if (unlikely(blk_queue_dying(q))) {
3006 __blk_end_request_all(rq, -ENODEV);
3011 * rq is already accounted, so use raw insert
3013 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3014 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3016 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3022 * This drops the queue lock
3025 queue_unplugged(q, depth, from_schedule);
3027 local_irq_restore(flags);
3030 void blk_finish_plug(struct blk_plug *plug)
3032 blk_flush_plug_list(plug, false);
3034 if (plug == current->plug)
3035 current->plug = NULL;
3037 EXPORT_SYMBOL(blk_finish_plug);
3039 int __init blk_dev_init(void)
3041 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3042 sizeof(((struct request *)0)->cmd_flags));
3044 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3045 kblockd_workqueue = alloc_workqueue("kblockd",
3046 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3047 if (!kblockd_workqueue)
3048 panic("Failed to create kblockd\n");
3050 request_cachep = kmem_cache_create("blkdev_requests",
3051 sizeof(struct request), 0, SLAB_PANIC, NULL);
3053 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3054 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);