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_unplug);
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. Queue lock must be held.
224 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
226 if (likely(!blk_queue_dead(q)))
227 queue_delayed_work(kblockd_workqueue, &q->delay_work,
228 msecs_to_jiffies(msecs));
230 EXPORT_SYMBOL(blk_delay_queue);
233 * blk_start_queue - restart a previously stopped queue
234 * @q: The &struct request_queue in question
237 * blk_start_queue() will clear the stop flag on the queue, and call
238 * the request_fn for the queue if it was in a stopped state when
239 * entered. Also see blk_stop_queue(). Queue lock must be held.
241 void blk_start_queue(struct request_queue *q)
243 WARN_ON(!irqs_disabled());
245 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
248 EXPORT_SYMBOL(blk_start_queue);
251 * blk_stop_queue - stop a queue
252 * @q: The &struct request_queue in question
255 * The Linux block layer assumes that a block driver will consume all
256 * entries on the request queue when the request_fn strategy is called.
257 * Often this will not happen, because of hardware limitations (queue
258 * depth settings). If a device driver gets a 'queue full' response,
259 * or if it simply chooses not to queue more I/O at one point, it can
260 * call this function to prevent the request_fn from being called until
261 * the driver has signalled it's ready to go again. This happens by calling
262 * blk_start_queue() to restart queue operations. Queue lock must be held.
264 void blk_stop_queue(struct request_queue *q)
266 cancel_delayed_work(&q->delay_work);
267 queue_flag_set(QUEUE_FLAG_STOPPED, q);
269 EXPORT_SYMBOL(blk_stop_queue);
272 * blk_sync_queue - cancel any pending callbacks on a queue
276 * The block layer may perform asynchronous callback activity
277 * on a queue, such as calling the unplug function after a timeout.
278 * A block device may call blk_sync_queue to ensure that any
279 * such activity is cancelled, thus allowing it to release resources
280 * that the callbacks might use. The caller must already have made sure
281 * that its ->make_request_fn will not re-add plugging prior to calling
284 * This function does not cancel any asynchronous activity arising
285 * out of elevator or throttling code. That would require elevaotor_exit()
286 * and blkcg_exit_queue() to be called with queue lock initialized.
289 void blk_sync_queue(struct request_queue *q)
291 del_timer_sync(&q->timeout);
292 cancel_delayed_work_sync(&q->delay_work);
294 EXPORT_SYMBOL(blk_sync_queue);
297 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
298 * @q: The queue to run
301 * Invoke request handling on a queue if there are any pending requests.
302 * May be used to restart request handling after a request has completed.
303 * This variant runs the queue whether or not the queue has been
304 * stopped. Must be called with the queue lock held and interrupts
305 * disabled. See also @blk_run_queue.
307 inline void __blk_run_queue_uncond(struct request_queue *q)
309 if (unlikely(blk_queue_dead(q)))
313 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
314 * the queue lock internally. As a result multiple threads may be
315 * running such a request function concurrently. Keep track of the
316 * number of active request_fn invocations such that blk_drain_queue()
317 * can wait until all these request_fn calls have finished.
319 q->request_fn_active++;
321 q->request_fn_active--;
325 * __blk_run_queue - run a single device queue
326 * @q: The queue to run
329 * See @blk_run_queue. This variant must be called with the queue lock
330 * held and interrupts disabled.
332 void __blk_run_queue(struct request_queue *q)
334 if (unlikely(blk_queue_stopped(q)))
337 __blk_run_queue_uncond(q);
339 EXPORT_SYMBOL(__blk_run_queue);
342 * blk_run_queue_async - run a single device queue in workqueue context
343 * @q: The queue to run
346 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
347 * of us. The caller must hold the queue lock.
349 void blk_run_queue_async(struct request_queue *q)
351 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
352 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
354 EXPORT_SYMBOL(blk_run_queue_async);
357 * blk_run_queue - run a single device queue
358 * @q: The queue to run
361 * Invoke request handling on this queue, if it has pending work to do.
362 * May be used to restart queueing when a request has completed.
364 void blk_run_queue(struct request_queue *q)
368 spin_lock_irqsave(q->queue_lock, flags);
370 spin_unlock_irqrestore(q->queue_lock, flags);
372 EXPORT_SYMBOL(blk_run_queue);
374 void blk_put_queue(struct request_queue *q)
376 kobject_put(&q->kobj);
378 EXPORT_SYMBOL(blk_put_queue);
381 * __blk_drain_queue - drain requests from request_queue
383 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
385 * Drain requests from @q. If @drain_all is set, all requests are drained.
386 * If not, only ELVPRIV requests are drained. The caller is responsible
387 * for ensuring that no new requests which need to be drained are queued.
389 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
390 __releases(q->queue_lock)
391 __acquires(q->queue_lock)
395 lockdep_assert_held(q->queue_lock);
401 * The caller might be trying to drain @q before its
402 * elevator is initialized.
405 elv_drain_elevator(q);
407 blkcg_drain_queue(q);
410 * This function might be called on a queue which failed
411 * driver init after queue creation or is not yet fully
412 * active yet. Some drivers (e.g. fd and loop) get unhappy
413 * in such cases. Kick queue iff dispatch queue has
414 * something on it and @q has request_fn set.
416 if (!list_empty(&q->queue_head) && q->request_fn)
419 drain |= q->nr_rqs_elvpriv;
420 drain |= q->request_fn_active;
423 * Unfortunately, requests are queued at and tracked from
424 * multiple places and there's no single counter which can
425 * be drained. Check all the queues and counters.
428 drain |= !list_empty(&q->queue_head);
429 for (i = 0; i < 2; i++) {
430 drain |= q->nr_rqs[i];
431 drain |= q->in_flight[i];
432 drain |= !list_empty(&q->flush_queue[i]);
439 spin_unlock_irq(q->queue_lock);
443 spin_lock_irq(q->queue_lock);
447 * With queue marked dead, any woken up waiter will fail the
448 * allocation path, so the wakeup chaining is lost and we're
449 * left with hung waiters. We need to wake up those waiters.
452 struct request_list *rl;
454 blk_queue_for_each_rl(rl, q)
455 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
456 wake_up_all(&rl->wait[i]);
461 * blk_queue_bypass_start - enter queue bypass mode
462 * @q: queue of interest
464 * In bypass mode, only the dispatch FIFO queue of @q is used. This
465 * function makes @q enter bypass mode and drains all requests which were
466 * throttled or issued before. On return, it's guaranteed that no request
467 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
468 * inside queue or RCU read lock.
470 void blk_queue_bypass_start(struct request_queue *q)
474 spin_lock_irq(q->queue_lock);
475 drain = !q->bypass_depth++;
476 queue_flag_set(QUEUE_FLAG_BYPASS, q);
477 spin_unlock_irq(q->queue_lock);
480 spin_lock_irq(q->queue_lock);
481 __blk_drain_queue(q, false);
482 spin_unlock_irq(q->queue_lock);
484 /* ensure blk_queue_bypass() is %true inside RCU read lock */
488 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
491 * blk_queue_bypass_end - leave queue bypass mode
492 * @q: queue of interest
494 * Leave bypass mode and restore the normal queueing behavior.
496 void blk_queue_bypass_end(struct request_queue *q)
498 spin_lock_irq(q->queue_lock);
499 if (!--q->bypass_depth)
500 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
501 WARN_ON_ONCE(q->bypass_depth < 0);
502 spin_unlock_irq(q->queue_lock);
504 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
507 * blk_cleanup_queue - shutdown a request queue
508 * @q: request queue to shutdown
510 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
511 * put it. All future requests will be failed immediately with -ENODEV.
513 void blk_cleanup_queue(struct request_queue *q)
515 spinlock_t *lock = q->queue_lock;
517 /* mark @q DYING, no new request or merges will be allowed afterwards */
518 mutex_lock(&q->sysfs_lock);
519 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
523 * A dying queue is permanently in bypass mode till released. Note
524 * that, unlike blk_queue_bypass_start(), we aren't performing
525 * synchronize_rcu() after entering bypass mode to avoid the delay
526 * as some drivers create and destroy a lot of queues while
527 * probing. This is still safe because blk_release_queue() will be
528 * called only after the queue refcnt drops to zero and nothing,
529 * RCU or not, would be traversing the queue by then.
532 queue_flag_set(QUEUE_FLAG_BYPASS, q);
534 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
535 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
536 queue_flag_set(QUEUE_FLAG_DYING, q);
537 spin_unlock_irq(lock);
538 mutex_unlock(&q->sysfs_lock);
541 * Drain all requests queued before DYING marking. Set DEAD flag to
542 * prevent that q->request_fn() gets invoked after draining finished.
545 __blk_drain_queue(q, true);
546 queue_flag_set(QUEUE_FLAG_DEAD, q);
547 spin_unlock_irq(lock);
549 /* @q won't process any more request, flush async actions */
550 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
554 if (q->queue_lock != &q->__queue_lock)
555 q->queue_lock = &q->__queue_lock;
556 spin_unlock_irq(lock);
558 /* @q is and will stay empty, shutdown and put */
561 EXPORT_SYMBOL(blk_cleanup_queue);
563 int blk_init_rl(struct request_list *rl, struct request_queue *q,
566 if (unlikely(rl->rq_pool))
570 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
571 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
572 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
573 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
575 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
576 mempool_free_slab, request_cachep,
584 void blk_exit_rl(struct request_list *rl)
587 mempool_destroy(rl->rq_pool);
590 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
592 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
594 EXPORT_SYMBOL(blk_alloc_queue);
596 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
598 struct request_queue *q;
601 q = kmem_cache_alloc_node(blk_requestq_cachep,
602 gfp_mask | __GFP_ZERO, node_id);
606 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
610 q->backing_dev_info.ra_pages =
611 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
612 q->backing_dev_info.state = 0;
613 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
614 q->backing_dev_info.name = "block";
617 err = bdi_init(&q->backing_dev_info);
621 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
622 laptop_mode_timer_fn, (unsigned long) q);
623 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
624 INIT_LIST_HEAD(&q->queue_head);
625 INIT_LIST_HEAD(&q->timeout_list);
626 INIT_LIST_HEAD(&q->icq_list);
627 #ifdef CONFIG_BLK_CGROUP
628 INIT_LIST_HEAD(&q->blkg_list);
630 INIT_LIST_HEAD(&q->flush_queue[0]);
631 INIT_LIST_HEAD(&q->flush_queue[1]);
632 INIT_LIST_HEAD(&q->flush_data_in_flight);
633 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
635 kobject_init(&q->kobj, &blk_queue_ktype);
637 mutex_init(&q->sysfs_lock);
638 spin_lock_init(&q->__queue_lock);
641 * By default initialize queue_lock to internal lock and driver can
642 * override it later if need be.
644 q->queue_lock = &q->__queue_lock;
647 * A queue starts its life with bypass turned on to avoid
648 * unnecessary bypass on/off overhead and nasty surprises during
649 * init. The initial bypass will be finished when the queue is
650 * registered by blk_register_queue().
653 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
655 if (blkcg_init_queue(q))
661 ida_simple_remove(&blk_queue_ida, q->id);
663 kmem_cache_free(blk_requestq_cachep, q);
666 EXPORT_SYMBOL(blk_alloc_queue_node);
669 * blk_init_queue - prepare a request queue for use with a block device
670 * @rfn: The function to be called to process requests that have been
671 * placed on the queue.
672 * @lock: Request queue spin lock
675 * If a block device wishes to use the standard request handling procedures,
676 * which sorts requests and coalesces adjacent requests, then it must
677 * call blk_init_queue(). The function @rfn will be called when there
678 * are requests on the queue that need to be processed. If the device
679 * supports plugging, then @rfn may not be called immediately when requests
680 * are available on the queue, but may be called at some time later instead.
681 * Plugged queues are generally unplugged when a buffer belonging to one
682 * of the requests on the queue is needed, or due to memory pressure.
684 * @rfn is not required, or even expected, to remove all requests off the
685 * queue, but only as many as it can handle at a time. If it does leave
686 * requests on the queue, it is responsible for arranging that the requests
687 * get dealt with eventually.
689 * The queue spin lock must be held while manipulating the requests on the
690 * request queue; this lock will be taken also from interrupt context, so irq
691 * disabling is needed for it.
693 * Function returns a pointer to the initialized request queue, or %NULL if
697 * blk_init_queue() must be paired with a blk_cleanup_queue() call
698 * when the block device is deactivated (such as at module unload).
701 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
703 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
705 EXPORT_SYMBOL(blk_init_queue);
707 struct request_queue *
708 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
710 struct request_queue *uninit_q, *q;
712 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
716 q = blk_init_allocated_queue(uninit_q, rfn, lock);
718 blk_cleanup_queue(uninit_q);
722 EXPORT_SYMBOL(blk_init_queue_node);
724 struct request_queue *
725 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
731 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
735 q->prep_rq_fn = NULL;
736 q->unprep_rq_fn = NULL;
737 q->queue_flags |= QUEUE_FLAG_DEFAULT;
739 /* Override internal queue lock with supplied lock pointer */
741 q->queue_lock = lock;
744 * This also sets hw/phys segments, boundary and size
746 blk_queue_make_request(q, blk_queue_bio);
748 q->sg_reserved_size = INT_MAX;
751 if (elevator_init(q, NULL))
755 EXPORT_SYMBOL(blk_init_allocated_queue);
757 bool blk_get_queue(struct request_queue *q)
759 if (likely(!blk_queue_dying(q))) {
766 EXPORT_SYMBOL(blk_get_queue);
768 static inline void blk_free_request(struct request_list *rl, struct request *rq)
770 if (rq->cmd_flags & REQ_ELVPRIV) {
771 elv_put_request(rl->q, rq);
773 put_io_context(rq->elv.icq->ioc);
776 mempool_free(rq, rl->rq_pool);
780 * ioc_batching returns true if the ioc is a valid batching request and
781 * should be given priority access to a request.
783 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
789 * Make sure the process is able to allocate at least 1 request
790 * even if the batch times out, otherwise we could theoretically
793 return ioc->nr_batch_requests == q->nr_batching ||
794 (ioc->nr_batch_requests > 0
795 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
799 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
800 * will cause the process to be a "batcher" on all queues in the system. This
801 * is the behaviour we want though - once it gets a wakeup it should be given
804 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
806 if (!ioc || ioc_batching(q, ioc))
809 ioc->nr_batch_requests = q->nr_batching;
810 ioc->last_waited = jiffies;
813 static void __freed_request(struct request_list *rl, int sync)
815 struct request_queue *q = rl->q;
818 * bdi isn't aware of blkcg yet. As all async IOs end up root
819 * blkcg anyway, just use root blkcg state.
821 if (rl == &q->root_rl &&
822 rl->count[sync] < queue_congestion_off_threshold(q))
823 blk_clear_queue_congested(q, sync);
825 if (rl->count[sync] + 1 <= q->nr_requests) {
826 if (waitqueue_active(&rl->wait[sync]))
827 wake_up(&rl->wait[sync]);
829 blk_clear_rl_full(rl, sync);
834 * A request has just been released. Account for it, update the full and
835 * congestion status, wake up any waiters. Called under q->queue_lock.
837 static void freed_request(struct request_list *rl, unsigned int flags)
839 struct request_queue *q = rl->q;
840 int sync = rw_is_sync(flags);
844 if (flags & REQ_ELVPRIV)
847 __freed_request(rl, sync);
849 if (unlikely(rl->starved[sync ^ 1]))
850 __freed_request(rl, sync ^ 1);
854 * Determine if elevator data should be initialized when allocating the
855 * request associated with @bio.
857 static bool blk_rq_should_init_elevator(struct bio *bio)
863 * Flush requests do not use the elevator so skip initialization.
864 * This allows a request to share the flush and elevator data.
866 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
873 * rq_ioc - determine io_context for request allocation
874 * @bio: request being allocated is for this bio (can be %NULL)
876 * Determine io_context to use for request allocation for @bio. May return
877 * %NULL if %current->io_context doesn't exist.
879 static struct io_context *rq_ioc(struct bio *bio)
881 #ifdef CONFIG_BLK_CGROUP
882 if (bio && bio->bi_ioc)
885 return current->io_context;
889 * __get_request - get a free request
890 * @rl: request list to allocate from
891 * @rw_flags: RW and SYNC flags
892 * @bio: bio to allocate request for (can be %NULL)
893 * @gfp_mask: allocation mask
895 * Get a free request from @q. This function may fail under memory
896 * pressure or if @q is dead.
898 * Must be callled with @q->queue_lock held and,
899 * Returns %NULL on failure, with @q->queue_lock held.
900 * Returns !%NULL on success, with @q->queue_lock *not held*.
902 static struct request *__get_request(struct request_list *rl, int rw_flags,
903 struct bio *bio, gfp_t gfp_mask)
905 struct request_queue *q = rl->q;
907 struct elevator_type *et = q->elevator->type;
908 struct io_context *ioc = rq_ioc(bio);
909 struct io_cq *icq = NULL;
910 const bool is_sync = rw_is_sync(rw_flags) != 0;
913 if (unlikely(blk_queue_dying(q)))
916 may_queue = elv_may_queue(q, rw_flags);
917 if (may_queue == ELV_MQUEUE_NO)
920 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
921 if (rl->count[is_sync]+1 >= q->nr_requests) {
923 * The queue will fill after this allocation, so set
924 * it as full, and mark this process as "batching".
925 * This process will be allowed to complete a batch of
926 * requests, others will be blocked.
928 if (!blk_rl_full(rl, is_sync)) {
929 ioc_set_batching(q, ioc);
930 blk_set_rl_full(rl, is_sync);
932 if (may_queue != ELV_MQUEUE_MUST
933 && !ioc_batching(q, ioc)) {
935 * The queue is full and the allocating
936 * process is not a "batcher", and not
937 * exempted by the IO scheduler
944 * bdi isn't aware of blkcg yet. As all async IOs end up
945 * root blkcg anyway, just use root blkcg state.
947 if (rl == &q->root_rl)
948 blk_set_queue_congested(q, is_sync);
952 * Only allow batching queuers to allocate up to 50% over the defined
953 * limit of requests, otherwise we could have thousands of requests
954 * allocated with any setting of ->nr_requests
956 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
959 q->nr_rqs[is_sync]++;
960 rl->count[is_sync]++;
961 rl->starved[is_sync] = 0;
964 * Decide whether the new request will be managed by elevator. If
965 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
966 * prevent the current elevator from being destroyed until the new
967 * request is freed. This guarantees icq's won't be destroyed and
968 * makes creating new ones safe.
970 * Also, lookup icq while holding queue_lock. If it doesn't exist,
971 * it will be created after releasing queue_lock.
973 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
974 rw_flags |= REQ_ELVPRIV;
976 if (et->icq_cache && ioc)
977 icq = ioc_lookup_icq(ioc, q);
980 if (blk_queue_io_stat(q))
981 rw_flags |= REQ_IO_STAT;
982 spin_unlock_irq(q->queue_lock);
984 /* allocate and init request */
985 rq = mempool_alloc(rl->rq_pool, gfp_mask);
990 blk_rq_set_rl(rq, rl);
991 rq->cmd_flags = rw_flags | REQ_ALLOCED;
994 if (rw_flags & REQ_ELVPRIV) {
995 if (unlikely(et->icq_cache && !icq)) {
997 icq = ioc_create_icq(ioc, q, gfp_mask);
1003 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1006 /* @rq->elv.icq holds io_context until @rq is freed */
1008 get_io_context(icq->ioc);
1012 * ioc may be NULL here, and ioc_batching will be false. That's
1013 * OK, if the queue is under the request limit then requests need
1014 * not count toward the nr_batch_requests limit. There will always
1015 * be some limit enforced by BLK_BATCH_TIME.
1017 if (ioc_batching(q, ioc))
1018 ioc->nr_batch_requests--;
1020 trace_block_getrq(q, bio, rw_flags & 1);
1025 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1026 * and may fail indefinitely under memory pressure and thus
1027 * shouldn't stall IO. Treat this request as !elvpriv. This will
1028 * disturb iosched and blkcg but weird is bettern than dead.
1030 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1031 dev_name(q->backing_dev_info.dev));
1033 rq->cmd_flags &= ~REQ_ELVPRIV;
1036 spin_lock_irq(q->queue_lock);
1037 q->nr_rqs_elvpriv--;
1038 spin_unlock_irq(q->queue_lock);
1043 * Allocation failed presumably due to memory. Undo anything we
1044 * might have messed up.
1046 * Allocating task should really be put onto the front of the wait
1047 * queue, but this is pretty rare.
1049 spin_lock_irq(q->queue_lock);
1050 freed_request(rl, rw_flags);
1053 * in the very unlikely event that allocation failed and no
1054 * requests for this direction was pending, mark us starved so that
1055 * freeing of a request in the other direction will notice
1056 * us. another possible fix would be to split the rq mempool into
1060 if (unlikely(rl->count[is_sync] == 0))
1061 rl->starved[is_sync] = 1;
1066 * get_request - get a free request
1067 * @q: request_queue to allocate request from
1068 * @rw_flags: RW and SYNC flags
1069 * @bio: bio to allocate request for (can be %NULL)
1070 * @gfp_mask: allocation mask
1072 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1073 * function keeps retrying under memory pressure and fails iff @q is dead.
1075 * Must be callled with @q->queue_lock held and,
1076 * Returns %NULL on failure, with @q->queue_lock held.
1077 * Returns !%NULL on success, with @q->queue_lock *not held*.
1079 static struct request *get_request(struct request_queue *q, int rw_flags,
1080 struct bio *bio, gfp_t gfp_mask)
1082 const bool is_sync = rw_is_sync(rw_flags) != 0;
1084 struct request_list *rl;
1087 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1089 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1093 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1098 /* wait on @rl and retry */
1099 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1100 TASK_UNINTERRUPTIBLE);
1102 trace_block_sleeprq(q, bio, rw_flags & 1);
1104 spin_unlock_irq(q->queue_lock);
1108 * After sleeping, we become a "batching" process and will be able
1109 * to allocate at least one request, and up to a big batch of them
1110 * for a small period time. See ioc_batching, ioc_set_batching
1112 ioc_set_batching(q, current->io_context);
1114 spin_lock_irq(q->queue_lock);
1115 finish_wait(&rl->wait[is_sync], &wait);
1120 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1124 BUG_ON(rw != READ && rw != WRITE);
1126 /* create ioc upfront */
1127 create_io_context(gfp_mask, q->node);
1129 spin_lock_irq(q->queue_lock);
1130 rq = get_request(q, rw, NULL, gfp_mask);
1132 spin_unlock_irq(q->queue_lock);
1133 /* q->queue_lock is unlocked at this point */
1137 EXPORT_SYMBOL(blk_get_request);
1140 * blk_make_request - given a bio, allocate a corresponding struct request.
1141 * @q: target request queue
1142 * @bio: The bio describing the memory mappings that will be submitted for IO.
1143 * It may be a chained-bio properly constructed by block/bio layer.
1144 * @gfp_mask: gfp flags to be used for memory allocation
1146 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1147 * type commands. Where the struct request needs to be farther initialized by
1148 * the caller. It is passed a &struct bio, which describes the memory info of
1151 * The caller of blk_make_request must make sure that bi_io_vec
1152 * are set to describe the memory buffers. That bio_data_dir() will return
1153 * the needed direction of the request. (And all bio's in the passed bio-chain
1154 * are properly set accordingly)
1156 * If called under none-sleepable conditions, mapped bio buffers must not
1157 * need bouncing, by calling the appropriate masked or flagged allocator,
1158 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1161 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1162 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1163 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1164 * completion of a bio that hasn't been submitted yet, thus resulting in a
1165 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1166 * of bio_alloc(), as that avoids the mempool deadlock.
1167 * If possible a big IO should be split into smaller parts when allocation
1168 * fails. Partial allocation should not be an error, or you risk a live-lock.
1170 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1173 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1176 return ERR_PTR(-ENOMEM);
1179 struct bio *bounce_bio = bio;
1182 blk_queue_bounce(q, &bounce_bio);
1183 ret = blk_rq_append_bio(q, rq, bounce_bio);
1184 if (unlikely(ret)) {
1185 blk_put_request(rq);
1186 return ERR_PTR(ret);
1192 EXPORT_SYMBOL(blk_make_request);
1195 * blk_requeue_request - put a request back on queue
1196 * @q: request queue where request should be inserted
1197 * @rq: request to be inserted
1200 * Drivers often keep queueing requests until the hardware cannot accept
1201 * more, when that condition happens we need to put the request back
1202 * on the queue. Must be called with queue lock held.
1204 void blk_requeue_request(struct request_queue *q, struct request *rq)
1206 blk_delete_timer(rq);
1207 blk_clear_rq_complete(rq);
1208 trace_block_rq_requeue(q, rq);
1210 if (blk_rq_tagged(rq))
1211 blk_queue_end_tag(q, rq);
1213 BUG_ON(blk_queued_rq(rq));
1215 elv_requeue_request(q, rq);
1217 EXPORT_SYMBOL(blk_requeue_request);
1219 static void add_acct_request(struct request_queue *q, struct request *rq,
1222 drive_stat_acct(rq, 1);
1223 __elv_add_request(q, rq, where);
1226 static void part_round_stats_single(int cpu, struct hd_struct *part,
1229 if (now == part->stamp)
1232 if (part_in_flight(part)) {
1233 __part_stat_add(cpu, part, time_in_queue,
1234 part_in_flight(part) * (now - part->stamp));
1235 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1241 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1242 * @cpu: cpu number for stats access
1243 * @part: target partition
1245 * The average IO queue length and utilisation statistics are maintained
1246 * by observing the current state of the queue length and the amount of
1247 * time it has been in this state for.
1249 * Normally, that accounting is done on IO completion, but that can result
1250 * in more than a second's worth of IO being accounted for within any one
1251 * second, leading to >100% utilisation. To deal with that, we call this
1252 * function to do a round-off before returning the results when reading
1253 * /proc/diskstats. This accounts immediately for all queue usage up to
1254 * the current jiffies and restarts the counters again.
1256 void part_round_stats(int cpu, struct hd_struct *part)
1258 unsigned long now = jiffies;
1261 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1262 part_round_stats_single(cpu, part, now);
1264 EXPORT_SYMBOL_GPL(part_round_stats);
1267 * queue lock must be held
1269 void __blk_put_request(struct request_queue *q, struct request *req)
1273 if (unlikely(--req->ref_count))
1276 elv_completed_request(q, req);
1278 /* this is a bio leak */
1279 WARN_ON(req->bio != NULL);
1282 * Request may not have originated from ll_rw_blk. if not,
1283 * it didn't come out of our reserved rq pools
1285 if (req->cmd_flags & REQ_ALLOCED) {
1286 unsigned int flags = req->cmd_flags;
1287 struct request_list *rl = blk_rq_rl(req);
1289 BUG_ON(!list_empty(&req->queuelist));
1290 BUG_ON(!hlist_unhashed(&req->hash));
1292 blk_free_request(rl, req);
1293 freed_request(rl, flags);
1297 EXPORT_SYMBOL_GPL(__blk_put_request);
1299 void blk_put_request(struct request *req)
1301 unsigned long flags;
1302 struct request_queue *q = req->q;
1304 spin_lock_irqsave(q->queue_lock, flags);
1305 __blk_put_request(q, req);
1306 spin_unlock_irqrestore(q->queue_lock, flags);
1308 EXPORT_SYMBOL(blk_put_request);
1311 * blk_add_request_payload - add a payload to a request
1312 * @rq: request to update
1313 * @page: page backing the payload
1314 * @len: length of the payload.
1316 * This allows to later add a payload to an already submitted request by
1317 * a block driver. The driver needs to take care of freeing the payload
1320 * Note that this is a quite horrible hack and nothing but handling of
1321 * discard requests should ever use it.
1323 void blk_add_request_payload(struct request *rq, struct page *page,
1326 struct bio *bio = rq->bio;
1328 bio->bi_io_vec->bv_page = page;
1329 bio->bi_io_vec->bv_offset = 0;
1330 bio->bi_io_vec->bv_len = len;
1334 bio->bi_phys_segments = 1;
1336 rq->__data_len = rq->resid_len = len;
1337 rq->nr_phys_segments = 1;
1338 rq->buffer = bio_data(bio);
1340 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1342 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1345 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1347 if (!ll_back_merge_fn(q, req, bio))
1350 trace_block_bio_backmerge(q, bio);
1352 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1353 blk_rq_set_mixed_merge(req);
1355 req->biotail->bi_next = bio;
1357 req->__data_len += bio->bi_size;
1358 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1360 drive_stat_acct(req, 0);
1364 static bool bio_attempt_front_merge(struct request_queue *q,
1365 struct request *req, struct bio *bio)
1367 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1369 if (!ll_front_merge_fn(q, req, bio))
1372 trace_block_bio_frontmerge(q, bio);
1374 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1375 blk_rq_set_mixed_merge(req);
1377 bio->bi_next = req->bio;
1381 * may not be valid. if the low level driver said
1382 * it didn't need a bounce buffer then it better
1383 * not touch req->buffer either...
1385 req->buffer = bio_data(bio);
1386 req->__sector = bio->bi_sector;
1387 req->__data_len += bio->bi_size;
1388 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1390 drive_stat_acct(req, 0);
1395 * attempt_plug_merge - try to merge with %current's plugged list
1396 * @q: request_queue new bio is being queued at
1397 * @bio: new bio being queued
1398 * @request_count: out parameter for number of traversed plugged requests
1400 * Determine whether @bio being queued on @q can be merged with a request
1401 * on %current's plugged list. Returns %true if merge was successful,
1404 * Plugging coalesces IOs from the same issuer for the same purpose without
1405 * going through @q->queue_lock. As such it's more of an issuing mechanism
1406 * than scheduling, and the request, while may have elvpriv data, is not
1407 * added on the elevator at this point. In addition, we don't have
1408 * reliable access to the elevator outside queue lock. Only check basic
1409 * merging parameters without querying the elevator.
1411 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1412 unsigned int *request_count)
1414 struct blk_plug *plug;
1418 plug = current->plug;
1423 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1429 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1432 el_ret = blk_try_merge(rq, bio);
1433 if (el_ret == ELEVATOR_BACK_MERGE) {
1434 ret = bio_attempt_back_merge(q, rq, bio);
1437 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1438 ret = bio_attempt_front_merge(q, rq, bio);
1447 void init_request_from_bio(struct request *req, struct bio *bio)
1449 req->cmd_type = REQ_TYPE_FS;
1451 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1452 if (bio->bi_rw & REQ_RAHEAD)
1453 req->cmd_flags |= REQ_FAILFAST_MASK;
1456 req->__sector = bio->bi_sector;
1457 req->ioprio = bio_prio(bio);
1458 blk_rq_bio_prep(req->q, req, bio);
1461 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1463 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1464 struct blk_plug *plug;
1465 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1466 struct request *req;
1467 unsigned int request_count = 0;
1470 * low level driver can indicate that it wants pages above a
1471 * certain limit bounced to low memory (ie for highmem, or even
1472 * ISA dma in theory)
1474 blk_queue_bounce(q, &bio);
1476 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1477 spin_lock_irq(q->queue_lock);
1478 where = ELEVATOR_INSERT_FLUSH;
1483 * Check if we can merge with the plugged list before grabbing
1486 if (attempt_plug_merge(q, bio, &request_count))
1489 spin_lock_irq(q->queue_lock);
1491 el_ret = elv_merge(q, &req, bio);
1492 if (el_ret == ELEVATOR_BACK_MERGE) {
1493 if (bio_attempt_back_merge(q, req, bio)) {
1494 elv_bio_merged(q, req, bio);
1495 if (!attempt_back_merge(q, req))
1496 elv_merged_request(q, req, el_ret);
1499 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1500 if (bio_attempt_front_merge(q, req, bio)) {
1501 elv_bio_merged(q, req, bio);
1502 if (!attempt_front_merge(q, req))
1503 elv_merged_request(q, req, el_ret);
1510 * This sync check and mask will be re-done in init_request_from_bio(),
1511 * but we need to set it earlier to expose the sync flag to the
1512 * rq allocator and io schedulers.
1514 rw_flags = bio_data_dir(bio);
1516 rw_flags |= REQ_SYNC;
1519 * Grab a free request. This is might sleep but can not fail.
1520 * Returns with the queue unlocked.
1522 req = get_request(q, rw_flags, bio, GFP_NOIO);
1523 if (unlikely(!req)) {
1524 bio_endio(bio, -ENODEV); /* @q is dead */
1529 * After dropping the lock and possibly sleeping here, our request
1530 * may now be mergeable after it had proven unmergeable (above).
1531 * We don't worry about that case for efficiency. It won't happen
1532 * often, and the elevators are able to handle it.
1534 init_request_from_bio(req, bio);
1536 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1537 req->cpu = raw_smp_processor_id();
1539 plug = current->plug;
1542 * If this is the first request added after a plug, fire
1543 * of a plug trace. If others have been added before, check
1544 * if we have multiple devices in this plug. If so, make a
1545 * note to sort the list before dispatch.
1547 if (list_empty(&plug->list))
1548 trace_block_plug(q);
1550 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1551 blk_flush_plug_list(plug, false);
1552 trace_block_plug(q);
1555 list_add_tail(&req->queuelist, &plug->list);
1556 drive_stat_acct(req, 1);
1558 spin_lock_irq(q->queue_lock);
1559 add_acct_request(q, req, where);
1562 spin_unlock_irq(q->queue_lock);
1565 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1568 * If bio->bi_dev is a partition, remap the location
1570 static inline void blk_partition_remap(struct bio *bio)
1572 struct block_device *bdev = bio->bi_bdev;
1574 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1575 struct hd_struct *p = bdev->bd_part;
1577 bio->bi_sector += p->start_sect;
1578 bio->bi_bdev = bdev->bd_contains;
1580 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1582 bio->bi_sector - p->start_sect);
1586 static void handle_bad_sector(struct bio *bio)
1588 char b[BDEVNAME_SIZE];
1590 printk(KERN_INFO "attempt to access beyond end of device\n");
1591 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1592 bdevname(bio->bi_bdev, b),
1594 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1595 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1597 set_bit(BIO_EOF, &bio->bi_flags);
1600 #ifdef CONFIG_FAIL_MAKE_REQUEST
1602 static DECLARE_FAULT_ATTR(fail_make_request);
1604 static int __init setup_fail_make_request(char *str)
1606 return setup_fault_attr(&fail_make_request, str);
1608 __setup("fail_make_request=", setup_fail_make_request);
1610 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1612 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1615 static int __init fail_make_request_debugfs(void)
1617 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1618 NULL, &fail_make_request);
1620 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1623 late_initcall(fail_make_request_debugfs);
1625 #else /* CONFIG_FAIL_MAKE_REQUEST */
1627 static inline bool should_fail_request(struct hd_struct *part,
1633 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1636 * Check whether this bio extends beyond the end of the device.
1638 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1645 /* Test device or partition size, when known. */
1646 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1648 sector_t sector = bio->bi_sector;
1650 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1652 * This may well happen - the kernel calls bread()
1653 * without checking the size of the device, e.g., when
1654 * mounting a device.
1656 handle_bad_sector(bio);
1664 static noinline_for_stack bool
1665 generic_make_request_checks(struct bio *bio)
1667 struct request_queue *q;
1668 int nr_sectors = bio_sectors(bio);
1670 char b[BDEVNAME_SIZE];
1671 struct hd_struct *part;
1675 if (bio_check_eod(bio, nr_sectors))
1678 q = bdev_get_queue(bio->bi_bdev);
1681 "generic_make_request: Trying to access "
1682 "nonexistent block-device %s (%Lu)\n",
1683 bdevname(bio->bi_bdev, b),
1684 (long long) bio->bi_sector);
1688 if (likely(bio_is_rw(bio) &&
1689 nr_sectors > queue_max_hw_sectors(q))) {
1690 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1691 bdevname(bio->bi_bdev, b),
1693 queue_max_hw_sectors(q));
1697 part = bio->bi_bdev->bd_part;
1698 if (should_fail_request(part, bio->bi_size) ||
1699 should_fail_request(&part_to_disk(part)->part0,
1704 * If this device has partitions, remap block n
1705 * of partition p to block n+start(p) of the disk.
1707 blk_partition_remap(bio);
1709 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1712 if (bio_check_eod(bio, nr_sectors))
1716 * Filter flush bio's early so that make_request based
1717 * drivers without flush support don't have to worry
1720 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1721 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1728 if ((bio->bi_rw & REQ_DISCARD) &&
1729 (!blk_queue_discard(q) ||
1730 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1735 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1741 * Various block parts want %current->io_context and lazy ioc
1742 * allocation ends up trading a lot of pain for a small amount of
1743 * memory. Just allocate it upfront. This may fail and block
1744 * layer knows how to live with it.
1746 create_io_context(GFP_ATOMIC, q->node);
1748 if (blk_throtl_bio(q, bio))
1749 return false; /* throttled, will be resubmitted later */
1751 trace_block_bio_queue(q, bio);
1755 bio_endio(bio, err);
1760 * generic_make_request - hand a buffer to its device driver for I/O
1761 * @bio: The bio describing the location in memory and on the device.
1763 * generic_make_request() is used to make I/O requests of block
1764 * devices. It is passed a &struct bio, which describes the I/O that needs
1767 * generic_make_request() does not return any status. The
1768 * success/failure status of the request, along with notification of
1769 * completion, is delivered asynchronously through the bio->bi_end_io
1770 * function described (one day) else where.
1772 * The caller of generic_make_request must make sure that bi_io_vec
1773 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1774 * set to describe the device address, and the
1775 * bi_end_io and optionally bi_private are set to describe how
1776 * completion notification should be signaled.
1778 * generic_make_request and the drivers it calls may use bi_next if this
1779 * bio happens to be merged with someone else, and may resubmit the bio to
1780 * a lower device by calling into generic_make_request recursively, which
1781 * means the bio should NOT be touched after the call to ->make_request_fn.
1783 void generic_make_request(struct bio *bio)
1785 struct bio_list bio_list_on_stack;
1787 if (!generic_make_request_checks(bio))
1791 * We only want one ->make_request_fn to be active at a time, else
1792 * stack usage with stacked devices could be a problem. So use
1793 * current->bio_list to keep a list of requests submited by a
1794 * make_request_fn function. current->bio_list is also used as a
1795 * flag to say if generic_make_request is currently active in this
1796 * task or not. If it is NULL, then no make_request is active. If
1797 * it is non-NULL, then a make_request is active, and new requests
1798 * should be added at the tail
1800 if (current->bio_list) {
1801 bio_list_add(current->bio_list, bio);
1805 /* following loop may be a bit non-obvious, and so deserves some
1807 * Before entering the loop, bio->bi_next is NULL (as all callers
1808 * ensure that) so we have a list with a single bio.
1809 * We pretend that we have just taken it off a longer list, so
1810 * we assign bio_list to a pointer to the bio_list_on_stack,
1811 * thus initialising the bio_list of new bios to be
1812 * added. ->make_request() may indeed add some more bios
1813 * through a recursive call to generic_make_request. If it
1814 * did, we find a non-NULL value in bio_list and re-enter the loop
1815 * from the top. In this case we really did just take the bio
1816 * of the top of the list (no pretending) and so remove it from
1817 * bio_list, and call into ->make_request() again.
1819 BUG_ON(bio->bi_next);
1820 bio_list_init(&bio_list_on_stack);
1821 current->bio_list = &bio_list_on_stack;
1823 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1825 q->make_request_fn(q, bio);
1827 bio = bio_list_pop(current->bio_list);
1829 current->bio_list = NULL; /* deactivate */
1831 EXPORT_SYMBOL(generic_make_request);
1834 * submit_bio - submit a bio to the block device layer for I/O
1835 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1836 * @bio: The &struct bio which describes the I/O
1838 * submit_bio() is very similar in purpose to generic_make_request(), and
1839 * uses that function to do most of the work. Both are fairly rough
1840 * interfaces; @bio must be presetup and ready for I/O.
1843 void submit_bio(int rw, struct bio *bio)
1848 * If it's a regular read/write or a barrier with data attached,
1849 * go through the normal accounting stuff before submission.
1851 if (bio_has_data(bio)) {
1854 if (unlikely(rw & REQ_WRITE_SAME))
1855 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1857 count = bio_sectors(bio);
1860 count_vm_events(PGPGOUT, count);
1862 task_io_account_read(bio->bi_size);
1863 count_vm_events(PGPGIN, count);
1866 if (unlikely(block_dump)) {
1867 char b[BDEVNAME_SIZE];
1868 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1869 current->comm, task_pid_nr(current),
1870 (rw & WRITE) ? "WRITE" : "READ",
1871 (unsigned long long)bio->bi_sector,
1872 bdevname(bio->bi_bdev, b),
1877 generic_make_request(bio);
1879 EXPORT_SYMBOL(submit_bio);
1882 * blk_rq_check_limits - Helper function to check a request for the queue limit
1884 * @rq: the request being checked
1887 * @rq may have been made based on weaker limitations of upper-level queues
1888 * in request stacking drivers, and it may violate the limitation of @q.
1889 * Since the block layer and the underlying device driver trust @rq
1890 * after it is inserted to @q, it should be checked against @q before
1891 * the insertion using this generic function.
1893 * This function should also be useful for request stacking drivers
1894 * in some cases below, so export this function.
1895 * Request stacking drivers like request-based dm may change the queue
1896 * limits while requests are in the queue (e.g. dm's table swapping).
1897 * Such request stacking drivers should check those requests agaist
1898 * the new queue limits again when they dispatch those requests,
1899 * although such checkings are also done against the old queue limits
1900 * when submitting requests.
1902 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1904 if (!rq_mergeable(rq))
1907 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1908 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1913 * queue's settings related to segment counting like q->bounce_pfn
1914 * may differ from that of other stacking queues.
1915 * Recalculate it to check the request correctly on this queue's
1918 blk_recalc_rq_segments(rq);
1919 if (rq->nr_phys_segments > queue_max_segments(q)) {
1920 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1926 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1929 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1930 * @q: the queue to submit the request
1931 * @rq: the request being queued
1933 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1935 unsigned long flags;
1936 int where = ELEVATOR_INSERT_BACK;
1938 if (blk_rq_check_limits(q, rq))
1942 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1945 spin_lock_irqsave(q->queue_lock, flags);
1946 if (unlikely(blk_queue_dying(q))) {
1947 spin_unlock_irqrestore(q->queue_lock, flags);
1952 * Submitting request must be dequeued before calling this function
1953 * because it will be linked to another request_queue
1955 BUG_ON(blk_queued_rq(rq));
1957 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1958 where = ELEVATOR_INSERT_FLUSH;
1960 add_acct_request(q, rq, where);
1961 if (where == ELEVATOR_INSERT_FLUSH)
1963 spin_unlock_irqrestore(q->queue_lock, flags);
1967 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1970 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1971 * @rq: request to examine
1974 * A request could be merge of IOs which require different failure
1975 * handling. This function determines the number of bytes which
1976 * can be failed from the beginning of the request without
1977 * crossing into area which need to be retried further.
1980 * The number of bytes to fail.
1983 * queue_lock must be held.
1985 unsigned int blk_rq_err_bytes(const struct request *rq)
1987 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1988 unsigned int bytes = 0;
1991 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1992 return blk_rq_bytes(rq);
1995 * Currently the only 'mixing' which can happen is between
1996 * different fastfail types. We can safely fail portions
1997 * which have all the failfast bits that the first one has -
1998 * the ones which are at least as eager to fail as the first
2001 for (bio = rq->bio; bio; bio = bio->bi_next) {
2002 if ((bio->bi_rw & ff) != ff)
2004 bytes += bio->bi_size;
2007 /* this could lead to infinite loop */
2008 BUG_ON(blk_rq_bytes(rq) && !bytes);
2011 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2013 static void blk_account_io_completion(struct request *req, unsigned int bytes)
2015 if (blk_do_io_stat(req)) {
2016 const int rw = rq_data_dir(req);
2017 struct hd_struct *part;
2020 cpu = part_stat_lock();
2022 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2027 static void blk_account_io_done(struct request *req)
2030 * Account IO completion. flush_rq isn't accounted as a
2031 * normal IO on queueing nor completion. Accounting the
2032 * containing request is enough.
2034 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2035 unsigned long duration = jiffies - req->start_time;
2036 const int rw = rq_data_dir(req);
2037 struct hd_struct *part;
2040 cpu = part_stat_lock();
2043 part_stat_inc(cpu, part, ios[rw]);
2044 part_stat_add(cpu, part, ticks[rw], duration);
2045 part_round_stats(cpu, part);
2046 part_dec_in_flight(part, rw);
2048 hd_struct_put(part);
2054 * blk_peek_request - peek at the top of a request queue
2055 * @q: request queue to peek at
2058 * Return the request at the top of @q. The returned request
2059 * should be started using blk_start_request() before LLD starts
2063 * Pointer to the request at the top of @q if available. Null
2067 * queue_lock must be held.
2069 struct request *blk_peek_request(struct request_queue *q)
2074 while ((rq = __elv_next_request(q)) != NULL) {
2075 if (!(rq->cmd_flags & REQ_STARTED)) {
2077 * This is the first time the device driver
2078 * sees this request (possibly after
2079 * requeueing). Notify IO scheduler.
2081 if (rq->cmd_flags & REQ_SORTED)
2082 elv_activate_rq(q, rq);
2085 * just mark as started even if we don't start
2086 * it, a request that has been delayed should
2087 * not be passed by new incoming requests
2089 rq->cmd_flags |= REQ_STARTED;
2090 trace_block_rq_issue(q, rq);
2093 if (!q->boundary_rq || q->boundary_rq == rq) {
2094 q->end_sector = rq_end_sector(rq);
2095 q->boundary_rq = NULL;
2098 if (rq->cmd_flags & REQ_DONTPREP)
2101 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2103 * make sure space for the drain appears we
2104 * know we can do this because max_hw_segments
2105 * has been adjusted to be one fewer than the
2108 rq->nr_phys_segments++;
2114 ret = q->prep_rq_fn(q, rq);
2115 if (ret == BLKPREP_OK) {
2117 } else if (ret == BLKPREP_DEFER) {
2119 * the request may have been (partially) prepped.
2120 * we need to keep this request in the front to
2121 * avoid resource deadlock. REQ_STARTED will
2122 * prevent other fs requests from passing this one.
2124 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2125 !(rq->cmd_flags & REQ_DONTPREP)) {
2127 * remove the space for the drain we added
2128 * so that we don't add it again
2130 --rq->nr_phys_segments;
2135 } else if (ret == BLKPREP_KILL) {
2136 rq->cmd_flags |= REQ_QUIET;
2138 * Mark this request as started so we don't trigger
2139 * any debug logic in the end I/O path.
2141 blk_start_request(rq);
2142 __blk_end_request_all(rq, -EIO);
2144 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2151 EXPORT_SYMBOL(blk_peek_request);
2153 void blk_dequeue_request(struct request *rq)
2155 struct request_queue *q = rq->q;
2157 BUG_ON(list_empty(&rq->queuelist));
2158 BUG_ON(ELV_ON_HASH(rq));
2160 list_del_init(&rq->queuelist);
2163 * the time frame between a request being removed from the lists
2164 * and to it is freed is accounted as io that is in progress at
2167 if (blk_account_rq(rq)) {
2168 q->in_flight[rq_is_sync(rq)]++;
2169 set_io_start_time_ns(rq);
2174 * blk_start_request - start request processing on the driver
2175 * @req: request to dequeue
2178 * Dequeue @req and start timeout timer on it. This hands off the
2179 * request to the driver.
2181 * Block internal functions which don't want to start timer should
2182 * call blk_dequeue_request().
2185 * queue_lock must be held.
2187 void blk_start_request(struct request *req)
2189 blk_dequeue_request(req);
2192 * We are now handing the request to the hardware, initialize
2193 * resid_len to full count and add the timeout handler.
2195 req->resid_len = blk_rq_bytes(req);
2196 if (unlikely(blk_bidi_rq(req)))
2197 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2201 EXPORT_SYMBOL(blk_start_request);
2204 * blk_fetch_request - fetch a request from a request queue
2205 * @q: request queue to fetch a request from
2208 * Return the request at the top of @q. The request is started on
2209 * return and LLD can start processing it immediately.
2212 * Pointer to the request at the top of @q if available. Null
2216 * queue_lock must be held.
2218 struct request *blk_fetch_request(struct request_queue *q)
2222 rq = blk_peek_request(q);
2224 blk_start_request(rq);
2227 EXPORT_SYMBOL(blk_fetch_request);
2230 * blk_update_request - Special helper function for request stacking drivers
2231 * @req: the request being processed
2232 * @error: %0 for success, < %0 for error
2233 * @nr_bytes: number of bytes to complete @req
2236 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2237 * the request structure even if @req doesn't have leftover.
2238 * If @req has leftover, sets it up for the next range of segments.
2240 * This special helper function is only for request stacking drivers
2241 * (e.g. request-based dm) so that they can handle partial completion.
2242 * Actual device drivers should use blk_end_request instead.
2244 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2245 * %false return from this function.
2248 * %false - this request doesn't have any more data
2249 * %true - this request has more data
2251 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2253 int total_bytes, bio_nbytes, next_idx = 0;
2259 trace_block_rq_complete(req->q, req);
2262 * For fs requests, rq is just carrier of independent bio's
2263 * and each partial completion should be handled separately.
2264 * Reset per-request error on each partial completion.
2266 * TODO: tj: This is too subtle. It would be better to let
2267 * low level drivers do what they see fit.
2269 if (req->cmd_type == REQ_TYPE_FS)
2272 if (error && req->cmd_type == REQ_TYPE_FS &&
2273 !(req->cmd_flags & REQ_QUIET)) {
2278 error_type = "recoverable transport";
2281 error_type = "critical target";
2284 error_type = "critical nexus";
2291 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2292 error_type, req->rq_disk ?
2293 req->rq_disk->disk_name : "?",
2294 (unsigned long long)blk_rq_pos(req));
2298 blk_account_io_completion(req, nr_bytes);
2300 total_bytes = bio_nbytes = 0;
2301 while ((bio = req->bio) != NULL) {
2304 if (nr_bytes >= bio->bi_size) {
2305 req->bio = bio->bi_next;
2306 nbytes = bio->bi_size;
2307 req_bio_endio(req, bio, nbytes, error);
2311 int idx = bio->bi_idx + next_idx;
2313 if (unlikely(idx >= bio->bi_vcnt)) {
2314 blk_dump_rq_flags(req, "__end_that");
2315 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2316 __func__, idx, bio->bi_vcnt);
2320 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2321 BIO_BUG_ON(nbytes > bio->bi_size);
2324 * not a complete bvec done
2326 if (unlikely(nbytes > nr_bytes)) {
2327 bio_nbytes += nr_bytes;
2328 total_bytes += nr_bytes;
2333 * advance to the next vector
2336 bio_nbytes += nbytes;
2339 total_bytes += nbytes;
2345 * end more in this run, or just return 'not-done'
2347 if (unlikely(nr_bytes <= 0))
2357 * Reset counters so that the request stacking driver
2358 * can find how many bytes remain in the request
2361 req->__data_len = 0;
2366 * if the request wasn't completed, update state
2369 req_bio_endio(req, bio, bio_nbytes, error);
2370 bio->bi_idx += next_idx;
2371 bio_iovec(bio)->bv_offset += nr_bytes;
2372 bio_iovec(bio)->bv_len -= nr_bytes;
2375 req->__data_len -= total_bytes;
2376 req->buffer = bio_data(req->bio);
2378 /* update sector only for requests with clear definition of sector */
2379 if (req->cmd_type == REQ_TYPE_FS)
2380 req->__sector += total_bytes >> 9;
2382 /* mixed attributes always follow the first bio */
2383 if (req->cmd_flags & REQ_MIXED_MERGE) {
2384 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2385 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2389 * If total number of sectors is less than the first segment
2390 * size, something has gone terribly wrong.
2392 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2393 blk_dump_rq_flags(req, "request botched");
2394 req->__data_len = blk_rq_cur_bytes(req);
2397 /* recalculate the number of segments */
2398 blk_recalc_rq_segments(req);
2402 EXPORT_SYMBOL_GPL(blk_update_request);
2404 static bool blk_update_bidi_request(struct request *rq, int error,
2405 unsigned int nr_bytes,
2406 unsigned int bidi_bytes)
2408 if (blk_update_request(rq, error, nr_bytes))
2411 /* Bidi request must be completed as a whole */
2412 if (unlikely(blk_bidi_rq(rq)) &&
2413 blk_update_request(rq->next_rq, error, bidi_bytes))
2416 if (blk_queue_add_random(rq->q))
2417 add_disk_randomness(rq->rq_disk);
2423 * blk_unprep_request - unprepare a request
2426 * This function makes a request ready for complete resubmission (or
2427 * completion). It happens only after all error handling is complete,
2428 * so represents the appropriate moment to deallocate any resources
2429 * that were allocated to the request in the prep_rq_fn. The queue
2430 * lock is held when calling this.
2432 void blk_unprep_request(struct request *req)
2434 struct request_queue *q = req->q;
2436 req->cmd_flags &= ~REQ_DONTPREP;
2437 if (q->unprep_rq_fn)
2438 q->unprep_rq_fn(q, req);
2440 EXPORT_SYMBOL_GPL(blk_unprep_request);
2443 * queue lock must be held
2445 static void blk_finish_request(struct request *req, int error)
2447 if (blk_rq_tagged(req))
2448 blk_queue_end_tag(req->q, req);
2450 BUG_ON(blk_queued_rq(req));
2452 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2453 laptop_io_completion(&req->q->backing_dev_info);
2455 blk_delete_timer(req);
2457 if (req->cmd_flags & REQ_DONTPREP)
2458 blk_unprep_request(req);
2461 blk_account_io_done(req);
2464 req->end_io(req, error);
2466 if (blk_bidi_rq(req))
2467 __blk_put_request(req->next_rq->q, req->next_rq);
2469 __blk_put_request(req->q, req);
2474 * blk_end_bidi_request - Complete a bidi request
2475 * @rq: the request to complete
2476 * @error: %0 for success, < %0 for error
2477 * @nr_bytes: number of bytes to complete @rq
2478 * @bidi_bytes: number of bytes to complete @rq->next_rq
2481 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2482 * Drivers that supports bidi can safely call this member for any
2483 * type of request, bidi or uni. In the later case @bidi_bytes is
2487 * %false - we are done with this request
2488 * %true - still buffers pending for this request
2490 static bool blk_end_bidi_request(struct request *rq, int error,
2491 unsigned int nr_bytes, unsigned int bidi_bytes)
2493 struct request_queue *q = rq->q;
2494 unsigned long flags;
2496 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2499 spin_lock_irqsave(q->queue_lock, flags);
2500 blk_finish_request(rq, error);
2501 spin_unlock_irqrestore(q->queue_lock, flags);
2507 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2508 * @rq: the request to complete
2509 * @error: %0 for success, < %0 for error
2510 * @nr_bytes: number of bytes to complete @rq
2511 * @bidi_bytes: number of bytes to complete @rq->next_rq
2514 * Identical to blk_end_bidi_request() except that queue lock is
2515 * assumed to be locked on entry and remains so on return.
2518 * %false - we are done with this request
2519 * %true - still buffers pending for this request
2521 bool __blk_end_bidi_request(struct request *rq, int error,
2522 unsigned int nr_bytes, unsigned int bidi_bytes)
2524 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2527 blk_finish_request(rq, error);
2533 * blk_end_request - Helper function for drivers to complete the request.
2534 * @rq: the request being processed
2535 * @error: %0 for success, < %0 for error
2536 * @nr_bytes: number of bytes to complete
2539 * Ends I/O on a number of bytes attached to @rq.
2540 * If @rq has leftover, sets it up for the next range of segments.
2543 * %false - we are done with this request
2544 * %true - still buffers pending for this request
2546 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2548 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2550 EXPORT_SYMBOL(blk_end_request);
2553 * blk_end_request_all - Helper function for drives to finish the request.
2554 * @rq: the request to finish
2555 * @error: %0 for success, < %0 for error
2558 * Completely finish @rq.
2560 void blk_end_request_all(struct request *rq, int error)
2563 unsigned int bidi_bytes = 0;
2565 if (unlikely(blk_bidi_rq(rq)))
2566 bidi_bytes = blk_rq_bytes(rq->next_rq);
2568 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2571 EXPORT_SYMBOL(blk_end_request_all);
2574 * blk_end_request_cur - Helper function to finish the current request chunk.
2575 * @rq: the request to finish the current chunk for
2576 * @error: %0 for success, < %0 for error
2579 * Complete the current consecutively mapped chunk from @rq.
2582 * %false - we are done with this request
2583 * %true - still buffers pending for this request
2585 bool blk_end_request_cur(struct request *rq, int error)
2587 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2589 EXPORT_SYMBOL(blk_end_request_cur);
2592 * blk_end_request_err - Finish a request till the next failure boundary.
2593 * @rq: the request to finish till the next failure boundary for
2594 * @error: must be negative errno
2597 * Complete @rq till the next failure boundary.
2600 * %false - we are done with this request
2601 * %true - still buffers pending for this request
2603 bool blk_end_request_err(struct request *rq, int error)
2605 WARN_ON(error >= 0);
2606 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2608 EXPORT_SYMBOL_GPL(blk_end_request_err);
2611 * __blk_end_request - Helper function for drivers to complete the request.
2612 * @rq: the request being processed
2613 * @error: %0 for success, < %0 for error
2614 * @nr_bytes: number of bytes to complete
2617 * Must be called with queue lock held unlike blk_end_request().
2620 * %false - we are done with this request
2621 * %true - still buffers pending for this request
2623 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2625 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2627 EXPORT_SYMBOL(__blk_end_request);
2630 * __blk_end_request_all - Helper function for drives to finish the request.
2631 * @rq: the request to finish
2632 * @error: %0 for success, < %0 for error
2635 * Completely finish @rq. Must be called with queue lock held.
2637 void __blk_end_request_all(struct request *rq, int error)
2640 unsigned int bidi_bytes = 0;
2642 if (unlikely(blk_bidi_rq(rq)))
2643 bidi_bytes = blk_rq_bytes(rq->next_rq);
2645 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2648 EXPORT_SYMBOL(__blk_end_request_all);
2651 * __blk_end_request_cur - Helper function to finish the current request chunk.
2652 * @rq: the request to finish the current chunk for
2653 * @error: %0 for success, < %0 for error
2656 * Complete the current consecutively mapped chunk from @rq. Must
2657 * be called with queue lock held.
2660 * %false - we are done with this request
2661 * %true - still buffers pending for this request
2663 bool __blk_end_request_cur(struct request *rq, int error)
2665 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2667 EXPORT_SYMBOL(__blk_end_request_cur);
2670 * __blk_end_request_err - Finish a request till the next failure boundary.
2671 * @rq: the request to finish till the next failure boundary for
2672 * @error: must be negative errno
2675 * Complete @rq till the next failure boundary. Must be called
2676 * with queue lock held.
2679 * %false - we are done with this request
2680 * %true - still buffers pending for this request
2682 bool __blk_end_request_err(struct request *rq, int error)
2684 WARN_ON(error >= 0);
2685 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2687 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2689 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2692 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2693 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2695 if (bio_has_data(bio)) {
2696 rq->nr_phys_segments = bio_phys_segments(q, bio);
2697 rq->buffer = bio_data(bio);
2699 rq->__data_len = bio->bi_size;
2700 rq->bio = rq->biotail = bio;
2703 rq->rq_disk = bio->bi_bdev->bd_disk;
2706 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2708 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2709 * @rq: the request to be flushed
2712 * Flush all pages in @rq.
2714 void rq_flush_dcache_pages(struct request *rq)
2716 struct req_iterator iter;
2717 struct bio_vec *bvec;
2719 rq_for_each_segment(bvec, rq, iter)
2720 flush_dcache_page(bvec->bv_page);
2722 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2726 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2727 * @q : the queue of the device being checked
2730 * Check if underlying low-level drivers of a device are busy.
2731 * If the drivers want to export their busy state, they must set own
2732 * exporting function using blk_queue_lld_busy() first.
2734 * Basically, this function is used only by request stacking drivers
2735 * to stop dispatching requests to underlying devices when underlying
2736 * devices are busy. This behavior helps more I/O merging on the queue
2737 * of the request stacking driver and prevents I/O throughput regression
2738 * on burst I/O load.
2741 * 0 - Not busy (The request stacking driver should dispatch request)
2742 * 1 - Busy (The request stacking driver should stop dispatching request)
2744 int blk_lld_busy(struct request_queue *q)
2747 return q->lld_busy_fn(q);
2751 EXPORT_SYMBOL_GPL(blk_lld_busy);
2754 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2755 * @rq: the clone request to be cleaned up
2758 * Free all bios in @rq for a cloned request.
2760 void blk_rq_unprep_clone(struct request *rq)
2764 while ((bio = rq->bio) != NULL) {
2765 rq->bio = bio->bi_next;
2770 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2773 * Copy attributes of the original request to the clone request.
2774 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2776 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2778 dst->cpu = src->cpu;
2779 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2780 dst->cmd_type = src->cmd_type;
2781 dst->__sector = blk_rq_pos(src);
2782 dst->__data_len = blk_rq_bytes(src);
2783 dst->nr_phys_segments = src->nr_phys_segments;
2784 dst->ioprio = src->ioprio;
2785 dst->extra_len = src->extra_len;
2789 * blk_rq_prep_clone - Helper function to setup clone request
2790 * @rq: the request to be setup
2791 * @rq_src: original request to be cloned
2792 * @bs: bio_set that bios for clone are allocated from
2793 * @gfp_mask: memory allocation mask for bio
2794 * @bio_ctr: setup function to be called for each clone bio.
2795 * Returns %0 for success, non %0 for failure.
2796 * @data: private data to be passed to @bio_ctr
2799 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2800 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2801 * are not copied, and copying such parts is the caller's responsibility.
2802 * Also, pages which the original bios are pointing to are not copied
2803 * and the cloned bios just point same pages.
2804 * So cloned bios must be completed before original bios, which means
2805 * the caller must complete @rq before @rq_src.
2807 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2808 struct bio_set *bs, gfp_t gfp_mask,
2809 int (*bio_ctr)(struct bio *, struct bio *, void *),
2812 struct bio *bio, *bio_src;
2817 blk_rq_init(NULL, rq);
2819 __rq_for_each_bio(bio_src, rq_src) {
2820 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2824 if (bio_ctr && bio_ctr(bio, bio_src, data))
2828 rq->biotail->bi_next = bio;
2831 rq->bio = rq->biotail = bio;
2834 __blk_rq_prep_clone(rq, rq_src);
2841 blk_rq_unprep_clone(rq);
2845 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2847 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2849 return queue_work(kblockd_workqueue, work);
2851 EXPORT_SYMBOL(kblockd_schedule_work);
2853 int kblockd_schedule_delayed_work(struct request_queue *q,
2854 struct delayed_work *dwork, unsigned long delay)
2856 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2858 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2860 #define PLUG_MAGIC 0x91827364
2863 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2864 * @plug: The &struct blk_plug that needs to be initialized
2867 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2868 * pending I/O should the task end up blocking between blk_start_plug() and
2869 * blk_finish_plug(). This is important from a performance perspective, but
2870 * also ensures that we don't deadlock. For instance, if the task is blocking
2871 * for a memory allocation, memory reclaim could end up wanting to free a
2872 * page belonging to that request that is currently residing in our private
2873 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2874 * this kind of deadlock.
2876 void blk_start_plug(struct blk_plug *plug)
2878 struct task_struct *tsk = current;
2880 plug->magic = PLUG_MAGIC;
2881 INIT_LIST_HEAD(&plug->list);
2882 INIT_LIST_HEAD(&plug->cb_list);
2885 * If this is a nested plug, don't actually assign it. It will be
2886 * flushed on its own.
2890 * Store ordering should not be needed here, since a potential
2891 * preempt will imply a full memory barrier
2896 EXPORT_SYMBOL(blk_start_plug);
2898 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2900 struct request *rqa = container_of(a, struct request, queuelist);
2901 struct request *rqb = container_of(b, struct request, queuelist);
2903 return !(rqa->q < rqb->q ||
2904 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2908 * If 'from_schedule' is true, then postpone the dispatch of requests
2909 * until a safe kblockd context. We due this to avoid accidental big
2910 * additional stack usage in driver dispatch, in places where the originally
2911 * plugger did not intend it.
2913 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2915 __releases(q->queue_lock)
2917 trace_block_unplug(q, depth, !from_schedule);
2920 blk_run_queue_async(q);
2923 spin_unlock(q->queue_lock);
2926 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2928 LIST_HEAD(callbacks);
2930 while (!list_empty(&plug->cb_list)) {
2931 list_splice_init(&plug->cb_list, &callbacks);
2933 while (!list_empty(&callbacks)) {
2934 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2937 list_del(&cb->list);
2938 cb->callback(cb, from_schedule);
2943 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2946 struct blk_plug *plug = current->plug;
2947 struct blk_plug_cb *cb;
2952 list_for_each_entry(cb, &plug->cb_list, list)
2953 if (cb->callback == unplug && cb->data == data)
2956 /* Not currently on the callback list */
2957 BUG_ON(size < sizeof(*cb));
2958 cb = kzalloc(size, GFP_ATOMIC);
2961 cb->callback = unplug;
2962 list_add(&cb->list, &plug->cb_list);
2966 EXPORT_SYMBOL(blk_check_plugged);
2968 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2970 struct request_queue *q;
2971 unsigned long flags;
2976 BUG_ON(plug->magic != PLUG_MAGIC);
2978 flush_plug_callbacks(plug, from_schedule);
2979 if (list_empty(&plug->list))
2982 list_splice_init(&plug->list, &list);
2984 list_sort(NULL, &list, plug_rq_cmp);
2990 * Save and disable interrupts here, to avoid doing it for every
2991 * queue lock we have to take.
2993 local_irq_save(flags);
2994 while (!list_empty(&list)) {
2995 rq = list_entry_rq(list.next);
2996 list_del_init(&rq->queuelist);
3000 * This drops the queue lock
3003 queue_unplugged(q, depth, from_schedule);
3006 spin_lock(q->queue_lock);
3010 * Short-circuit if @q is dead
3012 if (unlikely(blk_queue_dying(q))) {
3013 __blk_end_request_all(rq, -ENODEV);
3018 * rq is already accounted, so use raw insert
3020 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3021 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3023 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3029 * This drops the queue lock
3032 queue_unplugged(q, depth, from_schedule);
3034 local_irq_restore(flags);
3037 void blk_finish_plug(struct blk_plug *plug)
3039 blk_flush_plug_list(plug, false);
3041 if (plug == current->plug)
3042 current->plug = NULL;
3044 EXPORT_SYMBOL(blk_finish_plug);
3046 int __init blk_dev_init(void)
3048 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3049 sizeof(((struct request *)0)->cmd_flags));
3051 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3052 kblockd_workqueue = alloc_workqueue("kblockd",
3053 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3054 if (!kblockd_workqueue)
3055 panic("Failed to create kblockd\n");
3057 request_cachep = kmem_cache_create("blkdev_requests",
3058 sizeof(struct request), 0, SLAB_PANIC, NULL);
3060 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3061 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);