2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
32 #include <linux/ratelimit.h>
33 #include <linux/pm_runtime.h>
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/block.h>
39 #include "blk-cgroup.h"
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
46 DEFINE_IDA(blk_queue_ida);
49 * For the allocated request tables
51 static struct kmem_cache *request_cachep;
54 * For queue allocation
56 struct kmem_cache *blk_requestq_cachep;
59 * Controlling structure to kblockd
61 static struct workqueue_struct *kblockd_workqueue;
63 static void drive_stat_acct(struct request *rq, int new_io)
65 struct hd_struct *part;
66 int rw = rq_data_dir(rq);
69 if (!blk_do_io_stat(rq))
72 cpu = part_stat_lock();
76 part_stat_inc(cpu, part, merges[rw]);
78 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
79 if (!hd_struct_try_get(part)) {
81 * The partition is already being removed,
82 * the request will be accounted on the disk only
84 * We take a reference on disk->part0 although that
85 * partition will never be deleted, so we can treat
86 * it as any other partition.
88 part = &rq->rq_disk->part0;
91 part_round_stats(cpu, part);
92 part_inc_in_flight(part, rw);
99 void blk_queue_congestion_threshold(struct request_queue *q)
103 nr = q->nr_requests - (q->nr_requests / 8) + 1;
104 if (nr > q->nr_requests)
106 q->nr_congestion_on = nr;
108 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
111 q->nr_congestion_off = nr;
115 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
118 * Locates the passed device's request queue and returns the address of its
121 * Will return NULL if the request queue cannot be located.
123 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
125 struct backing_dev_info *ret = NULL;
126 struct request_queue *q = bdev_get_queue(bdev);
129 ret = &q->backing_dev_info;
132 EXPORT_SYMBOL(blk_get_backing_dev_info);
134 void blk_rq_init(struct request_queue *q, struct request *rq)
136 memset(rq, 0, sizeof(*rq));
138 INIT_LIST_HEAD(&rq->queuelist);
139 INIT_LIST_HEAD(&rq->timeout_list);
142 rq->__sector = (sector_t) -1;
143 INIT_HLIST_NODE(&rq->hash);
144 RB_CLEAR_NODE(&rq->rb_node);
146 rq->cmd_len = BLK_MAX_CDB;
149 rq->start_time = jiffies;
150 set_start_time_ns(rq);
153 EXPORT_SYMBOL(blk_rq_init);
155 static void req_bio_endio(struct request *rq, struct bio *bio,
156 unsigned int nbytes, int error)
159 clear_bit(BIO_UPTODATE, &bio->bi_flags);
160 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
163 if (unlikely(rq->cmd_flags & REQ_QUIET))
164 set_bit(BIO_QUIET, &bio->bi_flags);
166 bio_advance(bio, nbytes);
168 /* don't actually finish bio if it's part of flush sequence */
169 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
170 bio_endio(bio, error);
173 void blk_dump_rq_flags(struct request *rq, char *msg)
177 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
178 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
181 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
182 (unsigned long long)blk_rq_pos(rq),
183 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
184 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
185 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
187 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
188 printk(KERN_INFO " cdb: ");
189 for (bit = 0; bit < BLK_MAX_CDB; bit++)
190 printk("%02x ", rq->cmd[bit]);
194 EXPORT_SYMBOL(blk_dump_rq_flags);
196 static void blk_delay_work(struct work_struct *work)
198 struct request_queue *q;
200 q = container_of(work, struct request_queue, delay_work.work);
201 spin_lock_irq(q->queue_lock);
203 spin_unlock_irq(q->queue_lock);
207 * blk_delay_queue - restart queueing after defined interval
208 * @q: The &struct request_queue in question
209 * @msecs: Delay in msecs
212 * Sometimes queueing needs to be postponed for a little while, to allow
213 * resources to come back. This function will make sure that queueing is
214 * restarted around the specified time. Queue lock must be held.
216 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
218 if (likely(!blk_queue_dead(q)))
219 queue_delayed_work(kblockd_workqueue, &q->delay_work,
220 msecs_to_jiffies(msecs));
222 EXPORT_SYMBOL(blk_delay_queue);
225 * blk_start_queue - restart a previously stopped queue
226 * @q: The &struct request_queue in question
229 * blk_start_queue() will clear the stop flag on the queue, and call
230 * the request_fn for the queue if it was in a stopped state when
231 * entered. Also see blk_stop_queue(). Queue lock must be held.
233 void blk_start_queue(struct request_queue *q)
235 WARN_ON(!irqs_disabled());
237 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
240 EXPORT_SYMBOL(blk_start_queue);
243 * blk_stop_queue - stop a queue
244 * @q: The &struct request_queue in question
247 * The Linux block layer assumes that a block driver will consume all
248 * entries on the request queue when the request_fn strategy is called.
249 * Often this will not happen, because of hardware limitations (queue
250 * depth settings). If a device driver gets a 'queue full' response,
251 * or if it simply chooses not to queue more I/O at one point, it can
252 * call this function to prevent the request_fn from being called until
253 * the driver has signalled it's ready to go again. This happens by calling
254 * blk_start_queue() to restart queue operations. Queue lock must be held.
256 void blk_stop_queue(struct request_queue *q)
258 cancel_delayed_work(&q->delay_work);
259 queue_flag_set(QUEUE_FLAG_STOPPED, q);
261 EXPORT_SYMBOL(blk_stop_queue);
264 * blk_sync_queue - cancel any pending callbacks on a queue
268 * The block layer may perform asynchronous callback activity
269 * on a queue, such as calling the unplug function after a timeout.
270 * A block device may call blk_sync_queue to ensure that any
271 * such activity is cancelled, thus allowing it to release resources
272 * that the callbacks might use. The caller must already have made sure
273 * that its ->make_request_fn will not re-add plugging prior to calling
276 * This function does not cancel any asynchronous activity arising
277 * out of elevator or throttling code. That would require elevaotor_exit()
278 * and blkcg_exit_queue() to be called with queue lock initialized.
281 void blk_sync_queue(struct request_queue *q)
283 del_timer_sync(&q->timeout);
284 cancel_delayed_work_sync(&q->delay_work);
286 EXPORT_SYMBOL(blk_sync_queue);
289 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
290 * @q: The queue to run
293 * Invoke request handling on a queue if there are any pending requests.
294 * May be used to restart request handling after a request has completed.
295 * This variant runs the queue whether or not the queue has been
296 * stopped. Must be called with the queue lock held and interrupts
297 * disabled. See also @blk_run_queue.
299 inline void __blk_run_queue_uncond(struct request_queue *q)
301 if (unlikely(blk_queue_dead(q)))
305 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
306 * the queue lock internally. As a result multiple threads may be
307 * running such a request function concurrently. Keep track of the
308 * number of active request_fn invocations such that blk_drain_queue()
309 * can wait until all these request_fn calls have finished.
311 q->request_fn_active++;
313 q->request_fn_active--;
317 * __blk_run_queue - run a single device queue
318 * @q: The queue to run
321 * See @blk_run_queue. This variant must be called with the queue lock
322 * held and interrupts disabled.
324 void __blk_run_queue(struct request_queue *q)
326 if (unlikely(blk_queue_stopped(q)))
329 __blk_run_queue_uncond(q);
331 EXPORT_SYMBOL(__blk_run_queue);
334 * blk_run_queue_async - run a single device queue in workqueue context
335 * @q: The queue to run
338 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
339 * of us. The caller must hold the queue lock.
341 void blk_run_queue_async(struct request_queue *q)
343 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
344 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
346 EXPORT_SYMBOL(blk_run_queue_async);
349 * blk_run_queue - run a single device queue
350 * @q: The queue to run
353 * Invoke request handling on this queue, if it has pending work to do.
354 * May be used to restart queueing when a request has completed.
356 void blk_run_queue(struct request_queue *q)
360 spin_lock_irqsave(q->queue_lock, flags);
362 spin_unlock_irqrestore(q->queue_lock, flags);
364 EXPORT_SYMBOL(blk_run_queue);
366 void blk_put_queue(struct request_queue *q)
368 kobject_put(&q->kobj);
370 EXPORT_SYMBOL(blk_put_queue);
373 * __blk_drain_queue - drain requests from request_queue
375 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
377 * Drain requests from @q. If @drain_all is set, all requests are drained.
378 * If not, only ELVPRIV requests are drained. The caller is responsible
379 * for ensuring that no new requests which need to be drained are queued.
381 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
382 __releases(q->queue_lock)
383 __acquires(q->queue_lock)
387 lockdep_assert_held(q->queue_lock);
393 * The caller might be trying to drain @q before its
394 * elevator is initialized.
397 elv_drain_elevator(q);
399 blkcg_drain_queue(q);
402 * This function might be called on a queue which failed
403 * driver init after queue creation or is not yet fully
404 * active yet. Some drivers (e.g. fd and loop) get unhappy
405 * in such cases. Kick queue iff dispatch queue has
406 * something on it and @q has request_fn set.
408 if (!list_empty(&q->queue_head) && q->request_fn)
411 drain |= q->nr_rqs_elvpriv;
412 drain |= q->request_fn_active;
415 * Unfortunately, requests are queued at and tracked from
416 * multiple places and there's no single counter which can
417 * be drained. Check all the queues and counters.
420 drain |= !list_empty(&q->queue_head);
421 for (i = 0; i < 2; i++) {
422 drain |= q->nr_rqs[i];
423 drain |= q->in_flight[i];
424 drain |= !list_empty(&q->flush_queue[i]);
431 spin_unlock_irq(q->queue_lock);
435 spin_lock_irq(q->queue_lock);
439 * With queue marked dead, any woken up waiter will fail the
440 * allocation path, so the wakeup chaining is lost and we're
441 * left with hung waiters. We need to wake up those waiters.
444 struct request_list *rl;
446 blk_queue_for_each_rl(rl, q)
447 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
448 wake_up_all(&rl->wait[i]);
453 * blk_queue_bypass_start - enter queue bypass mode
454 * @q: queue of interest
456 * In bypass mode, only the dispatch FIFO queue of @q is used. This
457 * function makes @q enter bypass mode and drains all requests which were
458 * throttled or issued before. On return, it's guaranteed that no request
459 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
460 * inside queue or RCU read lock.
462 void blk_queue_bypass_start(struct request_queue *q)
466 spin_lock_irq(q->queue_lock);
467 drain = !q->bypass_depth++;
468 queue_flag_set(QUEUE_FLAG_BYPASS, q);
469 spin_unlock_irq(q->queue_lock);
472 spin_lock_irq(q->queue_lock);
473 __blk_drain_queue(q, false);
474 spin_unlock_irq(q->queue_lock);
476 /* ensure blk_queue_bypass() is %true inside RCU read lock */
480 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
483 * blk_queue_bypass_end - leave queue bypass mode
484 * @q: queue of interest
486 * Leave bypass mode and restore the normal queueing behavior.
488 void blk_queue_bypass_end(struct request_queue *q)
490 spin_lock_irq(q->queue_lock);
491 if (!--q->bypass_depth)
492 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
493 WARN_ON_ONCE(q->bypass_depth < 0);
494 spin_unlock_irq(q->queue_lock);
496 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
499 * blk_cleanup_queue - shutdown a request queue
500 * @q: request queue to shutdown
502 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
503 * put it. All future requests will be failed immediately with -ENODEV.
505 void blk_cleanup_queue(struct request_queue *q)
507 spinlock_t *lock = q->queue_lock;
509 /* mark @q DYING, no new request or merges will be allowed afterwards */
510 mutex_lock(&q->sysfs_lock);
511 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
515 * A dying queue is permanently in bypass mode till released. Note
516 * that, unlike blk_queue_bypass_start(), we aren't performing
517 * synchronize_rcu() after entering bypass mode to avoid the delay
518 * as some drivers create and destroy a lot of queues while
519 * probing. This is still safe because blk_release_queue() will be
520 * called only after the queue refcnt drops to zero and nothing,
521 * RCU or not, would be traversing the queue by then.
524 queue_flag_set(QUEUE_FLAG_BYPASS, q);
526 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
527 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
528 queue_flag_set(QUEUE_FLAG_DYING, q);
529 spin_unlock_irq(lock);
530 mutex_unlock(&q->sysfs_lock);
533 * Drain all requests queued before DYING marking. Set DEAD flag to
534 * prevent that q->request_fn() gets invoked after draining finished.
537 __blk_drain_queue(q, true);
538 queue_flag_set(QUEUE_FLAG_DEAD, q);
539 spin_unlock_irq(lock);
541 /* @q won't process any more request, flush async actions */
542 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
546 if (q->queue_lock != &q->__queue_lock)
547 q->queue_lock = &q->__queue_lock;
548 spin_unlock_irq(lock);
550 /* @q is and will stay empty, shutdown and put */
553 EXPORT_SYMBOL(blk_cleanup_queue);
555 int blk_init_rl(struct request_list *rl, struct request_queue *q,
558 if (unlikely(rl->rq_pool))
562 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
563 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
564 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
565 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
567 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
568 mempool_free_slab, request_cachep,
576 void blk_exit_rl(struct request_list *rl)
579 mempool_destroy(rl->rq_pool);
582 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
584 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
586 EXPORT_SYMBOL(blk_alloc_queue);
588 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
590 struct request_queue *q;
593 q = kmem_cache_alloc_node(blk_requestq_cachep,
594 gfp_mask | __GFP_ZERO, node_id);
598 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
602 q->backing_dev_info.ra_pages =
603 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
604 q->backing_dev_info.state = 0;
605 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
606 q->backing_dev_info.name = "block";
609 err = bdi_init(&q->backing_dev_info);
613 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
614 laptop_mode_timer_fn, (unsigned long) q);
615 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
616 INIT_LIST_HEAD(&q->queue_head);
617 INIT_LIST_HEAD(&q->timeout_list);
618 INIT_LIST_HEAD(&q->icq_list);
619 #ifdef CONFIG_BLK_CGROUP
620 INIT_LIST_HEAD(&q->blkg_list);
622 INIT_LIST_HEAD(&q->flush_queue[0]);
623 INIT_LIST_HEAD(&q->flush_queue[1]);
624 INIT_LIST_HEAD(&q->flush_data_in_flight);
625 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
627 kobject_init(&q->kobj, &blk_queue_ktype);
629 mutex_init(&q->sysfs_lock);
630 spin_lock_init(&q->__queue_lock);
633 * By default initialize queue_lock to internal lock and driver can
634 * override it later if need be.
636 q->queue_lock = &q->__queue_lock;
639 * A queue starts its life with bypass turned on to avoid
640 * unnecessary bypass on/off overhead and nasty surprises during
641 * init. The initial bypass will be finished when the queue is
642 * registered by blk_register_queue().
645 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
647 if (blkcg_init_queue(q))
653 bdi_destroy(&q->backing_dev_info);
655 ida_simple_remove(&blk_queue_ida, q->id);
657 kmem_cache_free(blk_requestq_cachep, q);
660 EXPORT_SYMBOL(blk_alloc_queue_node);
663 * blk_init_queue - prepare a request queue for use with a block device
664 * @rfn: The function to be called to process requests that have been
665 * placed on the queue.
666 * @lock: Request queue spin lock
669 * If a block device wishes to use the standard request handling procedures,
670 * which sorts requests and coalesces adjacent requests, then it must
671 * call blk_init_queue(). The function @rfn will be called when there
672 * are requests on the queue that need to be processed. If the device
673 * supports plugging, then @rfn may not be called immediately when requests
674 * are available on the queue, but may be called at some time later instead.
675 * Plugged queues are generally unplugged when a buffer belonging to one
676 * of the requests on the queue is needed, or due to memory pressure.
678 * @rfn is not required, or even expected, to remove all requests off the
679 * queue, but only as many as it can handle at a time. If it does leave
680 * requests on the queue, it is responsible for arranging that the requests
681 * get dealt with eventually.
683 * The queue spin lock must be held while manipulating the requests on the
684 * request queue; this lock will be taken also from interrupt context, so irq
685 * disabling is needed for it.
687 * Function returns a pointer to the initialized request queue, or %NULL if
691 * blk_init_queue() must be paired with a blk_cleanup_queue() call
692 * when the block device is deactivated (such as at module unload).
695 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
697 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
699 EXPORT_SYMBOL(blk_init_queue);
701 struct request_queue *
702 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
704 struct request_queue *uninit_q, *q;
706 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
710 q = blk_init_allocated_queue(uninit_q, rfn, lock);
712 blk_cleanup_queue(uninit_q);
716 EXPORT_SYMBOL(blk_init_queue_node);
718 struct request_queue *
719 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
725 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
729 q->prep_rq_fn = NULL;
730 q->unprep_rq_fn = NULL;
731 q->queue_flags |= QUEUE_FLAG_DEFAULT;
733 /* Override internal queue lock with supplied lock pointer */
735 q->queue_lock = lock;
738 * This also sets hw/phys segments, boundary and size
740 blk_queue_make_request(q, blk_queue_bio);
742 q->sg_reserved_size = INT_MAX;
744 /* Protect q->elevator from elevator_change */
745 mutex_lock(&q->sysfs_lock);
748 if (elevator_init(q, NULL)) {
749 mutex_unlock(&q->sysfs_lock);
753 mutex_unlock(&q->sysfs_lock);
757 EXPORT_SYMBOL(blk_init_allocated_queue);
759 bool blk_get_queue(struct request_queue *q)
761 if (likely(!blk_queue_dying(q))) {
768 EXPORT_SYMBOL(blk_get_queue);
770 static inline void blk_free_request(struct request_list *rl, struct request *rq)
772 if (rq->cmd_flags & REQ_ELVPRIV) {
773 elv_put_request(rl->q, rq);
775 put_io_context(rq->elv.icq->ioc);
778 mempool_free(rq, rl->rq_pool);
782 * ioc_batching returns true if the ioc is a valid batching request and
783 * should be given priority access to a request.
785 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
791 * Make sure the process is able to allocate at least 1 request
792 * even if the batch times out, otherwise we could theoretically
795 return ioc->nr_batch_requests == q->nr_batching ||
796 (ioc->nr_batch_requests > 0
797 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
801 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
802 * will cause the process to be a "batcher" on all queues in the system. This
803 * is the behaviour we want though - once it gets a wakeup it should be given
806 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
808 if (!ioc || ioc_batching(q, ioc))
811 ioc->nr_batch_requests = q->nr_batching;
812 ioc->last_waited = jiffies;
815 static void __freed_request(struct request_list *rl, int sync)
817 struct request_queue *q = rl->q;
820 * bdi isn't aware of blkcg yet. As all async IOs end up root
821 * blkcg anyway, just use root blkcg state.
823 if (rl == &q->root_rl &&
824 rl->count[sync] < queue_congestion_off_threshold(q))
825 blk_clear_queue_congested(q, sync);
827 if (rl->count[sync] + 1 <= q->nr_requests) {
828 if (waitqueue_active(&rl->wait[sync]))
829 wake_up(&rl->wait[sync]);
831 blk_clear_rl_full(rl, sync);
836 * A request has just been released. Account for it, update the full and
837 * congestion status, wake up any waiters. Called under q->queue_lock.
839 static void freed_request(struct request_list *rl, unsigned int flags)
841 struct request_queue *q = rl->q;
842 int sync = rw_is_sync(flags);
846 if (flags & REQ_ELVPRIV)
849 __freed_request(rl, sync);
851 if (unlikely(rl->starved[sync ^ 1]))
852 __freed_request(rl, sync ^ 1);
856 * Determine if elevator data should be initialized when allocating the
857 * request associated with @bio.
859 static bool blk_rq_should_init_elevator(struct bio *bio)
865 * Flush requests do not use the elevator so skip initialization.
866 * This allows a request to share the flush and elevator data.
868 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
875 * rq_ioc - determine io_context for request allocation
876 * @bio: request being allocated is for this bio (can be %NULL)
878 * Determine io_context to use for request allocation for @bio. May return
879 * %NULL if %current->io_context doesn't exist.
881 static struct io_context *rq_ioc(struct bio *bio)
883 #ifdef CONFIG_BLK_CGROUP
884 if (bio && bio->bi_ioc)
887 return current->io_context;
891 * __get_request - get a free request
892 * @rl: request list to allocate from
893 * @rw_flags: RW and SYNC flags
894 * @bio: bio to allocate request for (can be %NULL)
895 * @gfp_mask: allocation mask
897 * Get a free request from @q. This function may fail under memory
898 * pressure or if @q is dead.
900 * Must be callled with @q->queue_lock held and,
901 * Returns %NULL on failure, with @q->queue_lock held.
902 * Returns !%NULL on success, with @q->queue_lock *not held*.
904 static struct request *__get_request(struct request_list *rl, int rw_flags,
905 struct bio *bio, gfp_t gfp_mask)
907 struct request_queue *q = rl->q;
909 struct elevator_type *et = q->elevator->type;
910 struct io_context *ioc = rq_ioc(bio);
911 struct io_cq *icq = NULL;
912 const bool is_sync = rw_is_sync(rw_flags) != 0;
915 if (unlikely(blk_queue_dying(q)))
918 may_queue = elv_may_queue(q, rw_flags);
919 if (may_queue == ELV_MQUEUE_NO)
922 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
923 if (rl->count[is_sync]+1 >= q->nr_requests) {
925 * The queue will fill after this allocation, so set
926 * it as full, and mark this process as "batching".
927 * This process will be allowed to complete a batch of
928 * requests, others will be blocked.
930 if (!blk_rl_full(rl, is_sync)) {
931 ioc_set_batching(q, ioc);
932 blk_set_rl_full(rl, is_sync);
934 if (may_queue != ELV_MQUEUE_MUST
935 && !ioc_batching(q, ioc)) {
937 * The queue is full and the allocating
938 * process is not a "batcher", and not
939 * exempted by the IO scheduler
946 * bdi isn't aware of blkcg yet. As all async IOs end up
947 * root blkcg anyway, just use root blkcg state.
949 if (rl == &q->root_rl)
950 blk_set_queue_congested(q, is_sync);
954 * Only allow batching queuers to allocate up to 50% over the defined
955 * limit of requests, otherwise we could have thousands of requests
956 * allocated with any setting of ->nr_requests
958 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
961 q->nr_rqs[is_sync]++;
962 rl->count[is_sync]++;
963 rl->starved[is_sync] = 0;
966 * Decide whether the new request will be managed by elevator. If
967 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
968 * prevent the current elevator from being destroyed until the new
969 * request is freed. This guarantees icq's won't be destroyed and
970 * makes creating new ones safe.
972 * Also, lookup icq while holding queue_lock. If it doesn't exist,
973 * it will be created after releasing queue_lock.
975 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
976 rw_flags |= REQ_ELVPRIV;
978 if (et->icq_cache && ioc)
979 icq = ioc_lookup_icq(ioc, q);
982 if (blk_queue_io_stat(q))
983 rw_flags |= REQ_IO_STAT;
984 spin_unlock_irq(q->queue_lock);
986 /* allocate and init request */
987 rq = mempool_alloc(rl->rq_pool, gfp_mask);
992 blk_rq_set_rl(rq, rl);
993 rq->cmd_flags = rw_flags | REQ_ALLOCED;
996 if (rw_flags & REQ_ELVPRIV) {
997 if (unlikely(et->icq_cache && !icq)) {
999 icq = ioc_create_icq(ioc, q, gfp_mask);
1005 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1008 /* @rq->elv.icq holds io_context until @rq is freed */
1010 get_io_context(icq->ioc);
1014 * ioc may be NULL here, and ioc_batching will be false. That's
1015 * OK, if the queue is under the request limit then requests need
1016 * not count toward the nr_batch_requests limit. There will always
1017 * be some limit enforced by BLK_BATCH_TIME.
1019 if (ioc_batching(q, ioc))
1020 ioc->nr_batch_requests--;
1022 trace_block_getrq(q, bio, rw_flags & 1);
1027 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1028 * and may fail indefinitely under memory pressure and thus
1029 * shouldn't stall IO. Treat this request as !elvpriv. This will
1030 * disturb iosched and blkcg but weird is bettern than dead.
1032 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1033 dev_name(q->backing_dev_info.dev));
1035 rq->cmd_flags &= ~REQ_ELVPRIV;
1038 spin_lock_irq(q->queue_lock);
1039 q->nr_rqs_elvpriv--;
1040 spin_unlock_irq(q->queue_lock);
1045 * Allocation failed presumably due to memory. Undo anything we
1046 * might have messed up.
1048 * Allocating task should really be put onto the front of the wait
1049 * queue, but this is pretty rare.
1051 spin_lock_irq(q->queue_lock);
1052 freed_request(rl, rw_flags);
1055 * in the very unlikely event that allocation failed and no
1056 * requests for this direction was pending, mark us starved so that
1057 * freeing of a request in the other direction will notice
1058 * us. another possible fix would be to split the rq mempool into
1062 if (unlikely(rl->count[is_sync] == 0))
1063 rl->starved[is_sync] = 1;
1068 * get_request - get a free request
1069 * @q: request_queue to allocate request from
1070 * @rw_flags: RW and SYNC flags
1071 * @bio: bio to allocate request for (can be %NULL)
1072 * @gfp_mask: allocation mask
1074 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1075 * function keeps retrying under memory pressure and fails iff @q is dead.
1077 * Must be callled with @q->queue_lock held and,
1078 * Returns %NULL on failure, with @q->queue_lock held.
1079 * Returns !%NULL on success, with @q->queue_lock *not held*.
1081 static struct request *get_request(struct request_queue *q, int rw_flags,
1082 struct bio *bio, gfp_t gfp_mask)
1084 const bool is_sync = rw_is_sync(rw_flags) != 0;
1086 struct request_list *rl;
1089 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1091 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1095 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1100 /* wait on @rl and retry */
1101 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1102 TASK_UNINTERRUPTIBLE);
1104 trace_block_sleeprq(q, bio, rw_flags & 1);
1106 spin_unlock_irq(q->queue_lock);
1110 * After sleeping, we become a "batching" process and will be able
1111 * to allocate at least one request, and up to a big batch of them
1112 * for a small period time. See ioc_batching, ioc_set_batching
1114 ioc_set_batching(q, current->io_context);
1116 spin_lock_irq(q->queue_lock);
1117 finish_wait(&rl->wait[is_sync], &wait);
1122 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1126 BUG_ON(rw != READ && rw != WRITE);
1128 /* create ioc upfront */
1129 create_io_context(gfp_mask, q->node);
1131 spin_lock_irq(q->queue_lock);
1132 rq = get_request(q, rw, NULL, gfp_mask);
1134 spin_unlock_irq(q->queue_lock);
1135 /* q->queue_lock is unlocked at this point */
1139 EXPORT_SYMBOL(blk_get_request);
1142 * blk_make_request - given a bio, allocate a corresponding struct request.
1143 * @q: target request queue
1144 * @bio: The bio describing the memory mappings that will be submitted for IO.
1145 * It may be a chained-bio properly constructed by block/bio layer.
1146 * @gfp_mask: gfp flags to be used for memory allocation
1148 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1149 * type commands. Where the struct request needs to be farther initialized by
1150 * the caller. It is passed a &struct bio, which describes the memory info of
1153 * The caller of blk_make_request must make sure that bi_io_vec
1154 * are set to describe the memory buffers. That bio_data_dir() will return
1155 * the needed direction of the request. (And all bio's in the passed bio-chain
1156 * are properly set accordingly)
1158 * If called under none-sleepable conditions, mapped bio buffers must not
1159 * need bouncing, by calling the appropriate masked or flagged allocator,
1160 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1163 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1164 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1165 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1166 * completion of a bio that hasn't been submitted yet, thus resulting in a
1167 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1168 * of bio_alloc(), as that avoids the mempool deadlock.
1169 * If possible a big IO should be split into smaller parts when allocation
1170 * fails. Partial allocation should not be an error, or you risk a live-lock.
1172 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1175 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1178 return ERR_PTR(-ENOMEM);
1181 struct bio *bounce_bio = bio;
1184 blk_queue_bounce(q, &bounce_bio);
1185 ret = blk_rq_append_bio(q, rq, bounce_bio);
1186 if (unlikely(ret)) {
1187 blk_put_request(rq);
1188 return ERR_PTR(ret);
1194 EXPORT_SYMBOL(blk_make_request);
1197 * blk_requeue_request - put a request back on queue
1198 * @q: request queue where request should be inserted
1199 * @rq: request to be inserted
1202 * Drivers often keep queueing requests until the hardware cannot accept
1203 * more, when that condition happens we need to put the request back
1204 * on the queue. Must be called with queue lock held.
1206 void blk_requeue_request(struct request_queue *q, struct request *rq)
1208 blk_delete_timer(rq);
1209 blk_clear_rq_complete(rq);
1210 trace_block_rq_requeue(q, rq);
1212 if (blk_rq_tagged(rq))
1213 blk_queue_end_tag(q, rq);
1215 BUG_ON(blk_queued_rq(rq));
1217 elv_requeue_request(q, rq);
1219 EXPORT_SYMBOL(blk_requeue_request);
1221 static void add_acct_request(struct request_queue *q, struct request *rq,
1224 drive_stat_acct(rq, 1);
1225 __elv_add_request(q, rq, where);
1228 static void part_round_stats_single(int cpu, struct hd_struct *part,
1231 if (now == part->stamp)
1234 if (part_in_flight(part)) {
1235 __part_stat_add(cpu, part, time_in_queue,
1236 part_in_flight(part) * (now - part->stamp));
1237 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1243 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1244 * @cpu: cpu number for stats access
1245 * @part: target partition
1247 * The average IO queue length and utilisation statistics are maintained
1248 * by observing the current state of the queue length and the amount of
1249 * time it has been in this state for.
1251 * Normally, that accounting is done on IO completion, but that can result
1252 * in more than a second's worth of IO being accounted for within any one
1253 * second, leading to >100% utilisation. To deal with that, we call this
1254 * function to do a round-off before returning the results when reading
1255 * /proc/diskstats. This accounts immediately for all queue usage up to
1256 * the current jiffies and restarts the counters again.
1258 void part_round_stats(int cpu, struct hd_struct *part)
1260 unsigned long now = jiffies;
1263 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1264 part_round_stats_single(cpu, part, now);
1266 EXPORT_SYMBOL_GPL(part_round_stats);
1268 #ifdef CONFIG_PM_RUNTIME
1269 static void blk_pm_put_request(struct request *rq)
1271 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1272 pm_runtime_mark_last_busy(rq->q->dev);
1275 static inline void blk_pm_put_request(struct request *rq) {}
1279 * queue lock must be held
1281 void __blk_put_request(struct request_queue *q, struct request *req)
1285 if (unlikely(--req->ref_count))
1288 blk_pm_put_request(req);
1290 elv_completed_request(q, req);
1292 /* this is a bio leak */
1293 WARN_ON(req->bio != NULL);
1296 * Request may not have originated from ll_rw_blk. if not,
1297 * it didn't come out of our reserved rq pools
1299 if (req->cmd_flags & REQ_ALLOCED) {
1300 unsigned int flags = req->cmd_flags;
1301 struct request_list *rl = blk_rq_rl(req);
1303 BUG_ON(!list_empty(&req->queuelist));
1304 BUG_ON(!hlist_unhashed(&req->hash));
1306 blk_free_request(rl, req);
1307 freed_request(rl, flags);
1311 EXPORT_SYMBOL_GPL(__blk_put_request);
1313 void blk_put_request(struct request *req)
1315 unsigned long flags;
1316 struct request_queue *q = req->q;
1318 spin_lock_irqsave(q->queue_lock, flags);
1319 __blk_put_request(q, req);
1320 spin_unlock_irqrestore(q->queue_lock, flags);
1322 EXPORT_SYMBOL(blk_put_request);
1325 * blk_add_request_payload - add a payload to a request
1326 * @rq: request to update
1327 * @page: page backing the payload
1328 * @len: length of the payload.
1330 * This allows to later add a payload to an already submitted request by
1331 * a block driver. The driver needs to take care of freeing the payload
1334 * Note that this is a quite horrible hack and nothing but handling of
1335 * discard requests should ever use it.
1337 void blk_add_request_payload(struct request *rq, struct page *page,
1340 struct bio *bio = rq->bio;
1342 bio->bi_io_vec->bv_page = page;
1343 bio->bi_io_vec->bv_offset = 0;
1344 bio->bi_io_vec->bv_len = len;
1348 bio->bi_phys_segments = 1;
1350 rq->__data_len = rq->resid_len = len;
1351 rq->nr_phys_segments = 1;
1352 rq->buffer = bio_data(bio);
1354 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1356 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1359 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1361 if (!ll_back_merge_fn(q, req, bio))
1364 trace_block_bio_backmerge(q, req, bio);
1366 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1367 blk_rq_set_mixed_merge(req);
1369 req->biotail->bi_next = bio;
1371 req->__data_len += bio->bi_size;
1372 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1374 drive_stat_acct(req, 0);
1378 static bool bio_attempt_front_merge(struct request_queue *q,
1379 struct request *req, struct bio *bio)
1381 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1383 if (!ll_front_merge_fn(q, req, bio))
1386 trace_block_bio_frontmerge(q, req, bio);
1388 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1389 blk_rq_set_mixed_merge(req);
1391 bio->bi_next = req->bio;
1395 * may not be valid. if the low level driver said
1396 * it didn't need a bounce buffer then it better
1397 * not touch req->buffer either...
1399 req->buffer = bio_data(bio);
1400 req->__sector = bio->bi_sector;
1401 req->__data_len += bio->bi_size;
1402 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1404 drive_stat_acct(req, 0);
1409 * attempt_plug_merge - try to merge with %current's plugged list
1410 * @q: request_queue new bio is being queued at
1411 * @bio: new bio being queued
1412 * @request_count: out parameter for number of traversed plugged requests
1414 * Determine whether @bio being queued on @q can be merged with a request
1415 * on %current's plugged list. Returns %true if merge was successful,
1418 * Plugging coalesces IOs from the same issuer for the same purpose without
1419 * going through @q->queue_lock. As such it's more of an issuing mechanism
1420 * than scheduling, and the request, while may have elvpriv data, is not
1421 * added on the elevator at this point. In addition, we don't have
1422 * reliable access to the elevator outside queue lock. Only check basic
1423 * merging parameters without querying the elevator.
1425 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1426 unsigned int *request_count)
1428 struct blk_plug *plug;
1432 plug = current->plug;
1437 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1443 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1446 el_ret = blk_try_merge(rq, bio);
1447 if (el_ret == ELEVATOR_BACK_MERGE) {
1448 ret = bio_attempt_back_merge(q, rq, bio);
1451 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1452 ret = bio_attempt_front_merge(q, rq, bio);
1461 void init_request_from_bio(struct request *req, struct bio *bio)
1463 req->cmd_type = REQ_TYPE_FS;
1465 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1466 if (bio->bi_rw & REQ_RAHEAD)
1467 req->cmd_flags |= REQ_FAILFAST_MASK;
1470 req->__sector = bio->bi_sector;
1471 req->ioprio = bio_prio(bio);
1472 blk_rq_bio_prep(req->q, req, bio);
1475 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1477 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1478 struct blk_plug *plug;
1479 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1480 struct request *req;
1481 unsigned int request_count = 0;
1484 * low level driver can indicate that it wants pages above a
1485 * certain limit bounced to low memory (ie for highmem, or even
1486 * ISA dma in theory)
1488 blk_queue_bounce(q, &bio);
1490 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1491 bio_endio(bio, -EIO);
1495 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1496 spin_lock_irq(q->queue_lock);
1497 where = ELEVATOR_INSERT_FLUSH;
1502 * Check if we can merge with the plugged list before grabbing
1505 if (attempt_plug_merge(q, bio, &request_count))
1508 spin_lock_irq(q->queue_lock);
1510 el_ret = elv_merge(q, &req, bio);
1511 if (el_ret == ELEVATOR_BACK_MERGE) {
1512 if (bio_attempt_back_merge(q, req, bio)) {
1513 elv_bio_merged(q, req, bio);
1514 if (!attempt_back_merge(q, req))
1515 elv_merged_request(q, req, el_ret);
1518 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1519 if (bio_attempt_front_merge(q, req, bio)) {
1520 elv_bio_merged(q, req, bio);
1521 if (!attempt_front_merge(q, req))
1522 elv_merged_request(q, req, el_ret);
1529 * This sync check and mask will be re-done in init_request_from_bio(),
1530 * but we need to set it earlier to expose the sync flag to the
1531 * rq allocator and io schedulers.
1533 rw_flags = bio_data_dir(bio);
1535 rw_flags |= REQ_SYNC;
1538 * Grab a free request. This is might sleep but can not fail.
1539 * Returns with the queue unlocked.
1541 req = get_request(q, rw_flags, bio, GFP_NOIO);
1542 if (unlikely(!req)) {
1543 bio_endio(bio, -ENODEV); /* @q is dead */
1548 * After dropping the lock and possibly sleeping here, our request
1549 * may now be mergeable after it had proven unmergeable (above).
1550 * We don't worry about that case for efficiency. It won't happen
1551 * often, and the elevators are able to handle it.
1553 init_request_from_bio(req, bio);
1555 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1556 req->cpu = raw_smp_processor_id();
1558 plug = current->plug;
1561 * If this is the first request added after a plug, fire
1562 * of a plug trace. If others have been added before, check
1563 * if we have multiple devices in this plug. If so, make a
1564 * note to sort the list before dispatch.
1566 if (list_empty(&plug->list))
1567 trace_block_plug(q);
1569 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1570 blk_flush_plug_list(plug, false);
1571 trace_block_plug(q);
1574 list_add_tail(&req->queuelist, &plug->list);
1575 drive_stat_acct(req, 1);
1577 spin_lock_irq(q->queue_lock);
1578 add_acct_request(q, req, where);
1581 spin_unlock_irq(q->queue_lock);
1584 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1587 * If bio->bi_dev is a partition, remap the location
1589 static inline void blk_partition_remap(struct bio *bio)
1591 struct block_device *bdev = bio->bi_bdev;
1593 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1594 struct hd_struct *p = bdev->bd_part;
1596 bio->bi_sector += p->start_sect;
1597 bio->bi_bdev = bdev->bd_contains;
1599 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1601 bio->bi_sector - p->start_sect);
1605 static void handle_bad_sector(struct bio *bio)
1607 char b[BDEVNAME_SIZE];
1609 printk(KERN_INFO "attempt to access beyond end of device\n");
1610 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1611 bdevname(bio->bi_bdev, b),
1613 (unsigned long long)bio_end_sector(bio),
1614 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1616 set_bit(BIO_EOF, &bio->bi_flags);
1619 #ifdef CONFIG_FAIL_MAKE_REQUEST
1621 static DECLARE_FAULT_ATTR(fail_make_request);
1623 static int __init setup_fail_make_request(char *str)
1625 return setup_fault_attr(&fail_make_request, str);
1627 __setup("fail_make_request=", setup_fail_make_request);
1629 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1631 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1634 static int __init fail_make_request_debugfs(void)
1636 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1637 NULL, &fail_make_request);
1639 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1642 late_initcall(fail_make_request_debugfs);
1644 #else /* CONFIG_FAIL_MAKE_REQUEST */
1646 static inline bool should_fail_request(struct hd_struct *part,
1652 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1655 * Check whether this bio extends beyond the end of the device.
1657 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1664 /* Test device or partition size, when known. */
1665 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1667 sector_t sector = bio->bi_sector;
1669 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1671 * This may well happen - the kernel calls bread()
1672 * without checking the size of the device, e.g., when
1673 * mounting a device.
1675 handle_bad_sector(bio);
1683 static noinline_for_stack bool
1684 generic_make_request_checks(struct bio *bio)
1686 struct request_queue *q;
1687 int nr_sectors = bio_sectors(bio);
1689 char b[BDEVNAME_SIZE];
1690 struct hd_struct *part;
1694 if (bio_check_eod(bio, nr_sectors))
1697 q = bdev_get_queue(bio->bi_bdev);
1700 "generic_make_request: Trying to access "
1701 "nonexistent block-device %s (%Lu)\n",
1702 bdevname(bio->bi_bdev, b),
1703 (long long) bio->bi_sector);
1707 if (likely(bio_is_rw(bio) &&
1708 nr_sectors > queue_max_hw_sectors(q))) {
1709 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1710 bdevname(bio->bi_bdev, b),
1712 queue_max_hw_sectors(q));
1716 part = bio->bi_bdev->bd_part;
1717 if (should_fail_request(part, bio->bi_size) ||
1718 should_fail_request(&part_to_disk(part)->part0,
1723 * If this device has partitions, remap block n
1724 * of partition p to block n+start(p) of the disk.
1726 blk_partition_remap(bio);
1728 if (bio_check_eod(bio, nr_sectors))
1732 * Filter flush bio's early so that make_request based
1733 * drivers without flush support don't have to worry
1736 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1737 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1744 if ((bio->bi_rw & REQ_DISCARD) &&
1745 (!blk_queue_discard(q) ||
1746 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1751 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1757 * Various block parts want %current->io_context and lazy ioc
1758 * allocation ends up trading a lot of pain for a small amount of
1759 * memory. Just allocate it upfront. This may fail and block
1760 * layer knows how to live with it.
1762 create_io_context(GFP_ATOMIC, q->node);
1764 if (blk_throtl_bio(q, bio))
1765 return false; /* throttled, will be resubmitted later */
1767 trace_block_bio_queue(q, bio);
1771 bio_endio(bio, err);
1776 * generic_make_request - hand a buffer to its device driver for I/O
1777 * @bio: The bio describing the location in memory and on the device.
1779 * generic_make_request() is used to make I/O requests of block
1780 * devices. It is passed a &struct bio, which describes the I/O that needs
1783 * generic_make_request() does not return any status. The
1784 * success/failure status of the request, along with notification of
1785 * completion, is delivered asynchronously through the bio->bi_end_io
1786 * function described (one day) else where.
1788 * The caller of generic_make_request must make sure that bi_io_vec
1789 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1790 * set to describe the device address, and the
1791 * bi_end_io and optionally bi_private are set to describe how
1792 * completion notification should be signaled.
1794 * generic_make_request and the drivers it calls may use bi_next if this
1795 * bio happens to be merged with someone else, and may resubmit the bio to
1796 * a lower device by calling into generic_make_request recursively, which
1797 * means the bio should NOT be touched after the call to ->make_request_fn.
1799 void generic_make_request(struct bio *bio)
1801 struct bio_list bio_list_on_stack;
1803 if (!generic_make_request_checks(bio))
1807 * We only want one ->make_request_fn to be active at a time, else
1808 * stack usage with stacked devices could be a problem. So use
1809 * current->bio_list to keep a list of requests submited by a
1810 * make_request_fn function. current->bio_list is also used as a
1811 * flag to say if generic_make_request is currently active in this
1812 * task or not. If it is NULL, then no make_request is active. If
1813 * it is non-NULL, then a make_request is active, and new requests
1814 * should be added at the tail
1816 if (current->bio_list) {
1817 bio_list_add(current->bio_list, bio);
1821 /* following loop may be a bit non-obvious, and so deserves some
1823 * Before entering the loop, bio->bi_next is NULL (as all callers
1824 * ensure that) so we have a list with a single bio.
1825 * We pretend that we have just taken it off a longer list, so
1826 * we assign bio_list to a pointer to the bio_list_on_stack,
1827 * thus initialising the bio_list of new bios to be
1828 * added. ->make_request() may indeed add some more bios
1829 * through a recursive call to generic_make_request. If it
1830 * did, we find a non-NULL value in bio_list and re-enter the loop
1831 * from the top. In this case we really did just take the bio
1832 * of the top of the list (no pretending) and so remove it from
1833 * bio_list, and call into ->make_request() again.
1835 BUG_ON(bio->bi_next);
1836 bio_list_init(&bio_list_on_stack);
1837 current->bio_list = &bio_list_on_stack;
1839 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1841 q->make_request_fn(q, bio);
1843 bio = bio_list_pop(current->bio_list);
1845 current->bio_list = NULL; /* deactivate */
1847 EXPORT_SYMBOL(generic_make_request);
1850 * submit_bio - submit a bio to the block device layer for I/O
1851 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1852 * @bio: The &struct bio which describes the I/O
1854 * submit_bio() is very similar in purpose to generic_make_request(), and
1855 * uses that function to do most of the work. Both are fairly rough
1856 * interfaces; @bio must be presetup and ready for I/O.
1859 void submit_bio(int rw, struct bio *bio)
1864 * If it's a regular read/write or a barrier with data attached,
1865 * go through the normal accounting stuff before submission.
1867 if (bio_has_data(bio)) {
1870 if (unlikely(rw & REQ_WRITE_SAME))
1871 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1873 count = bio_sectors(bio);
1876 count_vm_events(PGPGOUT, count);
1878 task_io_account_read(bio->bi_size);
1879 count_vm_events(PGPGIN, count);
1882 if (unlikely(block_dump)) {
1883 char b[BDEVNAME_SIZE];
1884 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1885 current->comm, task_pid_nr(current),
1886 (rw & WRITE) ? "WRITE" : "READ",
1887 (unsigned long long)bio->bi_sector,
1888 bdevname(bio->bi_bdev, b),
1893 generic_make_request(bio);
1895 EXPORT_SYMBOL(submit_bio);
1898 * blk_rq_check_limits - Helper function to check a request for the queue limit
1900 * @rq: the request being checked
1903 * @rq may have been made based on weaker limitations of upper-level queues
1904 * in request stacking drivers, and it may violate the limitation of @q.
1905 * Since the block layer and the underlying device driver trust @rq
1906 * after it is inserted to @q, it should be checked against @q before
1907 * the insertion using this generic function.
1909 * This function should also be useful for request stacking drivers
1910 * in some cases below, so export this function.
1911 * Request stacking drivers like request-based dm may change the queue
1912 * limits while requests are in the queue (e.g. dm's table swapping).
1913 * Such request stacking drivers should check those requests agaist
1914 * the new queue limits again when they dispatch those requests,
1915 * although such checkings are also done against the old queue limits
1916 * when submitting requests.
1918 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1920 if (!rq_mergeable(rq))
1923 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1924 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1929 * queue's settings related to segment counting like q->bounce_pfn
1930 * may differ from that of other stacking queues.
1931 * Recalculate it to check the request correctly on this queue's
1934 blk_recalc_rq_segments(rq);
1935 if (rq->nr_phys_segments > queue_max_segments(q)) {
1936 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1942 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1945 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1946 * @q: the queue to submit the request
1947 * @rq: the request being queued
1949 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1951 unsigned long flags;
1952 int where = ELEVATOR_INSERT_BACK;
1954 if (blk_rq_check_limits(q, rq))
1958 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1961 spin_lock_irqsave(q->queue_lock, flags);
1962 if (unlikely(blk_queue_dying(q))) {
1963 spin_unlock_irqrestore(q->queue_lock, flags);
1968 * Submitting request must be dequeued before calling this function
1969 * because it will be linked to another request_queue
1971 BUG_ON(blk_queued_rq(rq));
1973 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1974 where = ELEVATOR_INSERT_FLUSH;
1976 add_acct_request(q, rq, where);
1977 if (where == ELEVATOR_INSERT_FLUSH)
1979 spin_unlock_irqrestore(q->queue_lock, flags);
1983 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1986 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1987 * @rq: request to examine
1990 * A request could be merge of IOs which require different failure
1991 * handling. This function determines the number of bytes which
1992 * can be failed from the beginning of the request without
1993 * crossing into area which need to be retried further.
1996 * The number of bytes to fail.
1999 * queue_lock must be held.
2001 unsigned int blk_rq_err_bytes(const struct request *rq)
2003 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2004 unsigned int bytes = 0;
2007 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2008 return blk_rq_bytes(rq);
2011 * Currently the only 'mixing' which can happen is between
2012 * different fastfail types. We can safely fail portions
2013 * which have all the failfast bits that the first one has -
2014 * the ones which are at least as eager to fail as the first
2017 for (bio = rq->bio; bio; bio = bio->bi_next) {
2018 if ((bio->bi_rw & ff) != ff)
2020 bytes += bio->bi_size;
2023 /* this could lead to infinite loop */
2024 BUG_ON(blk_rq_bytes(rq) && !bytes);
2027 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2029 static void blk_account_io_completion(struct request *req, unsigned int bytes)
2031 if (blk_do_io_stat(req)) {
2032 const int rw = rq_data_dir(req);
2033 struct hd_struct *part;
2036 cpu = part_stat_lock();
2038 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2043 static void blk_account_io_done(struct request *req)
2046 * Account IO completion. flush_rq isn't accounted as a
2047 * normal IO on queueing nor completion. Accounting the
2048 * containing request is enough.
2050 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2051 unsigned long duration = jiffies - req->start_time;
2052 const int rw = rq_data_dir(req);
2053 struct hd_struct *part;
2056 cpu = part_stat_lock();
2059 part_stat_inc(cpu, part, ios[rw]);
2060 part_stat_add(cpu, part, ticks[rw], duration);
2061 part_round_stats(cpu, part);
2062 part_dec_in_flight(part, rw);
2064 hd_struct_put(part);
2069 #ifdef CONFIG_PM_RUNTIME
2071 * Don't process normal requests when queue is suspended
2072 * or in the process of suspending/resuming
2074 static struct request *blk_pm_peek_request(struct request_queue *q,
2077 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2078 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2084 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2092 * blk_peek_request - peek at the top of a request queue
2093 * @q: request queue to peek at
2096 * Return the request at the top of @q. The returned request
2097 * should be started using blk_start_request() before LLD starts
2101 * Pointer to the request at the top of @q if available. Null
2105 * queue_lock must be held.
2107 struct request *blk_peek_request(struct request_queue *q)
2112 while ((rq = __elv_next_request(q)) != NULL) {
2114 rq = blk_pm_peek_request(q, rq);
2118 if (!(rq->cmd_flags & REQ_STARTED)) {
2120 * This is the first time the device driver
2121 * sees this request (possibly after
2122 * requeueing). Notify IO scheduler.
2124 if (rq->cmd_flags & REQ_SORTED)
2125 elv_activate_rq(q, rq);
2128 * just mark as started even if we don't start
2129 * it, a request that has been delayed should
2130 * not be passed by new incoming requests
2132 rq->cmd_flags |= REQ_STARTED;
2133 trace_block_rq_issue(q, rq);
2136 if (!q->boundary_rq || q->boundary_rq == rq) {
2137 q->end_sector = rq_end_sector(rq);
2138 q->boundary_rq = NULL;
2141 if (rq->cmd_flags & REQ_DONTPREP)
2144 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2146 * make sure space for the drain appears we
2147 * know we can do this because max_hw_segments
2148 * has been adjusted to be one fewer than the
2151 rq->nr_phys_segments++;
2157 ret = q->prep_rq_fn(q, rq);
2158 if (ret == BLKPREP_OK) {
2160 } else if (ret == BLKPREP_DEFER) {
2162 * the request may have been (partially) prepped.
2163 * we need to keep this request in the front to
2164 * avoid resource deadlock. REQ_STARTED will
2165 * prevent other fs requests from passing this one.
2167 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2168 !(rq->cmd_flags & REQ_DONTPREP)) {
2170 * remove the space for the drain we added
2171 * so that we don't add it again
2173 --rq->nr_phys_segments;
2178 } else if (ret == BLKPREP_KILL) {
2179 rq->cmd_flags |= REQ_QUIET;
2181 * Mark this request as started so we don't trigger
2182 * any debug logic in the end I/O path.
2184 blk_start_request(rq);
2185 __blk_end_request_all(rq, -EIO);
2187 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2194 EXPORT_SYMBOL(blk_peek_request);
2196 void blk_dequeue_request(struct request *rq)
2198 struct request_queue *q = rq->q;
2200 BUG_ON(list_empty(&rq->queuelist));
2201 BUG_ON(ELV_ON_HASH(rq));
2203 list_del_init(&rq->queuelist);
2206 * the time frame between a request being removed from the lists
2207 * and to it is freed is accounted as io that is in progress at
2210 if (blk_account_rq(rq)) {
2211 q->in_flight[rq_is_sync(rq)]++;
2212 set_io_start_time_ns(rq);
2217 * blk_start_request - start request processing on the driver
2218 * @req: request to dequeue
2221 * Dequeue @req and start timeout timer on it. This hands off the
2222 * request to the driver.
2224 * Block internal functions which don't want to start timer should
2225 * call blk_dequeue_request().
2228 * queue_lock must be held.
2230 void blk_start_request(struct request *req)
2232 blk_dequeue_request(req);
2235 * We are now handing the request to the hardware, initialize
2236 * resid_len to full count and add the timeout handler.
2238 req->resid_len = blk_rq_bytes(req);
2239 if (unlikely(blk_bidi_rq(req)))
2240 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2242 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2245 EXPORT_SYMBOL(blk_start_request);
2248 * blk_fetch_request - fetch a request from a request queue
2249 * @q: request queue to fetch a request from
2252 * Return the request at the top of @q. The request is started on
2253 * return and LLD can start processing it immediately.
2256 * Pointer to the request at the top of @q if available. Null
2260 * queue_lock must be held.
2262 struct request *blk_fetch_request(struct request_queue *q)
2266 rq = blk_peek_request(q);
2268 blk_start_request(rq);
2271 EXPORT_SYMBOL(blk_fetch_request);
2274 * blk_update_request - Special helper function for request stacking drivers
2275 * @req: the request being processed
2276 * @error: %0 for success, < %0 for error
2277 * @nr_bytes: number of bytes to complete @req
2280 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2281 * the request structure even if @req doesn't have leftover.
2282 * If @req has leftover, sets it up for the next range of segments.
2284 * This special helper function is only for request stacking drivers
2285 * (e.g. request-based dm) so that they can handle partial completion.
2286 * Actual device drivers should use blk_end_request instead.
2288 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2289 * %false return from this function.
2292 * %false - this request doesn't have any more data
2293 * %true - this request has more data
2295 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2302 trace_block_rq_complete(req->q, req, nr_bytes);
2305 * For fs requests, rq is just carrier of independent bio's
2306 * and each partial completion should be handled separately.
2307 * Reset per-request error on each partial completion.
2309 * TODO: tj: This is too subtle. It would be better to let
2310 * low level drivers do what they see fit.
2312 if (req->cmd_type == REQ_TYPE_FS)
2315 if (error && req->cmd_type == REQ_TYPE_FS &&
2316 !(req->cmd_flags & REQ_QUIET)) {
2321 error_type = "recoverable transport";
2324 error_type = "critical target";
2327 error_type = "critical nexus";
2334 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2335 error_type, req->rq_disk ?
2336 req->rq_disk->disk_name : "?",
2337 (unsigned long long)blk_rq_pos(req));
2341 blk_account_io_completion(req, nr_bytes);
2345 struct bio *bio = req->bio;
2346 unsigned bio_bytes = min(bio->bi_size, nr_bytes);
2348 if (bio_bytes == bio->bi_size)
2349 req->bio = bio->bi_next;
2351 req_bio_endio(req, bio, bio_bytes, error);
2353 total_bytes += bio_bytes;
2354 nr_bytes -= bio_bytes;
2365 * Reset counters so that the request stacking driver
2366 * can find how many bytes remain in the request
2369 req->__data_len = 0;
2373 req->__data_len -= total_bytes;
2374 req->buffer = bio_data(req->bio);
2376 /* update sector only for requests with clear definition of sector */
2377 if (req->cmd_type == REQ_TYPE_FS)
2378 req->__sector += total_bytes >> 9;
2380 /* mixed attributes always follow the first bio */
2381 if (req->cmd_flags & REQ_MIXED_MERGE) {
2382 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2383 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2387 * If total number of sectors is less than the first segment
2388 * size, something has gone terribly wrong.
2390 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2391 blk_dump_rq_flags(req, "request botched");
2392 req->__data_len = blk_rq_cur_bytes(req);
2395 /* recalculate the number of segments */
2396 blk_recalc_rq_segments(req);
2400 EXPORT_SYMBOL_GPL(blk_update_request);
2402 static bool blk_update_bidi_request(struct request *rq, int error,
2403 unsigned int nr_bytes,
2404 unsigned int bidi_bytes)
2406 if (blk_update_request(rq, error, nr_bytes))
2409 /* Bidi request must be completed as a whole */
2410 if (unlikely(blk_bidi_rq(rq)) &&
2411 blk_update_request(rq->next_rq, error, bidi_bytes))
2414 if (blk_queue_add_random(rq->q))
2415 add_disk_randomness(rq->rq_disk);
2421 * blk_unprep_request - unprepare a request
2424 * This function makes a request ready for complete resubmission (or
2425 * completion). It happens only after all error handling is complete,
2426 * so represents the appropriate moment to deallocate any resources
2427 * that were allocated to the request in the prep_rq_fn. The queue
2428 * lock is held when calling this.
2430 void blk_unprep_request(struct request *req)
2432 struct request_queue *q = req->q;
2434 req->cmd_flags &= ~REQ_DONTPREP;
2435 if (q->unprep_rq_fn)
2436 q->unprep_rq_fn(q, req);
2438 EXPORT_SYMBOL_GPL(blk_unprep_request);
2441 * queue lock must be held
2443 static void blk_finish_request(struct request *req, int error)
2445 if (blk_rq_tagged(req))
2446 blk_queue_end_tag(req->q, req);
2448 BUG_ON(blk_queued_rq(req));
2450 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2451 laptop_io_completion(&req->q->backing_dev_info);
2453 blk_delete_timer(req);
2455 if (req->cmd_flags & REQ_DONTPREP)
2456 blk_unprep_request(req);
2459 blk_account_io_done(req);
2462 req->end_io(req, error);
2464 if (blk_bidi_rq(req))
2465 __blk_put_request(req->next_rq->q, req->next_rq);
2467 __blk_put_request(req->q, req);
2472 * blk_end_bidi_request - Complete a bidi request
2473 * @rq: the request to complete
2474 * @error: %0 for success, < %0 for error
2475 * @nr_bytes: number of bytes to complete @rq
2476 * @bidi_bytes: number of bytes to complete @rq->next_rq
2479 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2480 * Drivers that supports bidi can safely call this member for any
2481 * type of request, bidi or uni. In the later case @bidi_bytes is
2485 * %false - we are done with this request
2486 * %true - still buffers pending for this request
2488 static bool blk_end_bidi_request(struct request *rq, int error,
2489 unsigned int nr_bytes, unsigned int bidi_bytes)
2491 struct request_queue *q = rq->q;
2492 unsigned long flags;
2494 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2497 spin_lock_irqsave(q->queue_lock, flags);
2498 blk_finish_request(rq, error);
2499 spin_unlock_irqrestore(q->queue_lock, flags);
2505 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2506 * @rq: the request to complete
2507 * @error: %0 for success, < %0 for error
2508 * @nr_bytes: number of bytes to complete @rq
2509 * @bidi_bytes: number of bytes to complete @rq->next_rq
2512 * Identical to blk_end_bidi_request() except that queue lock is
2513 * assumed to be locked on entry and remains so on return.
2516 * %false - we are done with this request
2517 * %true - still buffers pending for this request
2519 bool __blk_end_bidi_request(struct request *rq, int error,
2520 unsigned int nr_bytes, unsigned int bidi_bytes)
2522 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2525 blk_finish_request(rq, error);
2531 * blk_end_request - Helper function for drivers to complete the request.
2532 * @rq: the request being processed
2533 * @error: %0 for success, < %0 for error
2534 * @nr_bytes: number of bytes to complete
2537 * Ends I/O on a number of bytes attached to @rq.
2538 * If @rq has leftover, sets it up for the next range of segments.
2541 * %false - we are done with this request
2542 * %true - still buffers pending for this request
2544 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2546 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2548 EXPORT_SYMBOL(blk_end_request);
2551 * blk_end_request_all - Helper function for drives to finish the request.
2552 * @rq: the request to finish
2553 * @error: %0 for success, < %0 for error
2556 * Completely finish @rq.
2558 void blk_end_request_all(struct request *rq, int error)
2561 unsigned int bidi_bytes = 0;
2563 if (unlikely(blk_bidi_rq(rq)))
2564 bidi_bytes = blk_rq_bytes(rq->next_rq);
2566 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2569 EXPORT_SYMBOL(blk_end_request_all);
2572 * blk_end_request_cur - Helper function to finish the current request chunk.
2573 * @rq: the request to finish the current chunk for
2574 * @error: %0 for success, < %0 for error
2577 * Complete the current consecutively mapped chunk from @rq.
2580 * %false - we are done with this request
2581 * %true - still buffers pending for this request
2583 bool blk_end_request_cur(struct request *rq, int error)
2585 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2587 EXPORT_SYMBOL(blk_end_request_cur);
2590 * blk_end_request_err - Finish a request till the next failure boundary.
2591 * @rq: the request to finish till the next failure boundary for
2592 * @error: must be negative errno
2595 * Complete @rq till the next failure boundary.
2598 * %false - we are done with this request
2599 * %true - still buffers pending for this request
2601 bool blk_end_request_err(struct request *rq, int error)
2603 WARN_ON(error >= 0);
2604 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2606 EXPORT_SYMBOL_GPL(blk_end_request_err);
2609 * __blk_end_request - Helper function for drivers to complete the request.
2610 * @rq: the request being processed
2611 * @error: %0 for success, < %0 for error
2612 * @nr_bytes: number of bytes to complete
2615 * Must be called with queue lock held unlike blk_end_request().
2618 * %false - we are done with this request
2619 * %true - still buffers pending for this request
2621 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2623 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2625 EXPORT_SYMBOL(__blk_end_request);
2628 * __blk_end_request_all - Helper function for drives to finish the request.
2629 * @rq: the request to finish
2630 * @error: %0 for success, < %0 for error
2633 * Completely finish @rq. Must be called with queue lock held.
2635 void __blk_end_request_all(struct request *rq, int error)
2638 unsigned int bidi_bytes = 0;
2640 if (unlikely(blk_bidi_rq(rq)))
2641 bidi_bytes = blk_rq_bytes(rq->next_rq);
2643 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2646 EXPORT_SYMBOL(__blk_end_request_all);
2649 * __blk_end_request_cur - Helper function to finish the current request chunk.
2650 * @rq: the request to finish the current chunk for
2651 * @error: %0 for success, < %0 for error
2654 * Complete the current consecutively mapped chunk from @rq. Must
2655 * be called with queue lock held.
2658 * %false - we are done with this request
2659 * %true - still buffers pending for this request
2661 bool __blk_end_request_cur(struct request *rq, int error)
2663 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2665 EXPORT_SYMBOL(__blk_end_request_cur);
2668 * __blk_end_request_err - Finish a request till the next failure boundary.
2669 * @rq: the request to finish till the next failure boundary for
2670 * @error: must be negative errno
2673 * Complete @rq till the next failure boundary. Must be called
2674 * with queue lock held.
2677 * %false - we are done with this request
2678 * %true - still buffers pending for this request
2680 bool __blk_end_request_err(struct request *rq, int error)
2682 WARN_ON(error >= 0);
2683 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2685 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2687 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2690 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2691 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2693 if (bio_has_data(bio)) {
2694 rq->nr_phys_segments = bio_phys_segments(q, bio);
2695 rq->buffer = bio_data(bio);
2697 rq->__data_len = bio->bi_size;
2698 rq->bio = rq->biotail = bio;
2701 rq->rq_disk = bio->bi_bdev->bd_disk;
2704 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2706 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2707 * @rq: the request to be flushed
2710 * Flush all pages in @rq.
2712 void rq_flush_dcache_pages(struct request *rq)
2714 struct req_iterator iter;
2715 struct bio_vec *bvec;
2717 rq_for_each_segment(bvec, rq, iter)
2718 flush_dcache_page(bvec->bv_page);
2720 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2724 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2725 * @q : the queue of the device being checked
2728 * Check if underlying low-level drivers of a device are busy.
2729 * If the drivers want to export their busy state, they must set own
2730 * exporting function using blk_queue_lld_busy() first.
2732 * Basically, this function is used only by request stacking drivers
2733 * to stop dispatching requests to underlying devices when underlying
2734 * devices are busy. This behavior helps more I/O merging on the queue
2735 * of the request stacking driver and prevents I/O throughput regression
2736 * on burst I/O load.
2739 * 0 - Not busy (The request stacking driver should dispatch request)
2740 * 1 - Busy (The request stacking driver should stop dispatching request)
2742 int blk_lld_busy(struct request_queue *q)
2745 return q->lld_busy_fn(q);
2749 EXPORT_SYMBOL_GPL(blk_lld_busy);
2752 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2753 * @rq: the clone request to be cleaned up
2756 * Free all bios in @rq for a cloned request.
2758 void blk_rq_unprep_clone(struct request *rq)
2762 while ((bio = rq->bio) != NULL) {
2763 rq->bio = bio->bi_next;
2768 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2771 * Copy attributes of the original request to the clone request.
2772 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2774 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2776 dst->cpu = src->cpu;
2777 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2778 dst->cmd_type = src->cmd_type;
2779 dst->__sector = blk_rq_pos(src);
2780 dst->__data_len = blk_rq_bytes(src);
2781 dst->nr_phys_segments = src->nr_phys_segments;
2782 dst->ioprio = src->ioprio;
2783 dst->extra_len = src->extra_len;
2787 * blk_rq_prep_clone - Helper function to setup clone request
2788 * @rq: the request to be setup
2789 * @rq_src: original request to be cloned
2790 * @bs: bio_set that bios for clone are allocated from
2791 * @gfp_mask: memory allocation mask for bio
2792 * @bio_ctr: setup function to be called for each clone bio.
2793 * Returns %0 for success, non %0 for failure.
2794 * @data: private data to be passed to @bio_ctr
2797 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2798 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2799 * are not copied, and copying such parts is the caller's responsibility.
2800 * Also, pages which the original bios are pointing to are not copied
2801 * and the cloned bios just point same pages.
2802 * So cloned bios must be completed before original bios, which means
2803 * the caller must complete @rq before @rq_src.
2805 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2806 struct bio_set *bs, gfp_t gfp_mask,
2807 int (*bio_ctr)(struct bio *, struct bio *, void *),
2810 struct bio *bio, *bio_src;
2815 blk_rq_init(NULL, rq);
2817 __rq_for_each_bio(bio_src, rq_src) {
2818 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2822 if (bio_ctr && bio_ctr(bio, bio_src, data))
2826 rq->biotail->bi_next = bio;
2829 rq->bio = rq->biotail = bio;
2832 __blk_rq_prep_clone(rq, rq_src);
2839 blk_rq_unprep_clone(rq);
2843 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2845 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2847 return queue_work(kblockd_workqueue, work);
2849 EXPORT_SYMBOL(kblockd_schedule_work);
2851 int kblockd_schedule_delayed_work(struct request_queue *q,
2852 struct delayed_work *dwork, unsigned long delay)
2854 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2856 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2858 #define PLUG_MAGIC 0x91827364
2861 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2862 * @plug: The &struct blk_plug that needs to be initialized
2865 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2866 * pending I/O should the task end up blocking between blk_start_plug() and
2867 * blk_finish_plug(). This is important from a performance perspective, but
2868 * also ensures that we don't deadlock. For instance, if the task is blocking
2869 * for a memory allocation, memory reclaim could end up wanting to free a
2870 * page belonging to that request that is currently residing in our private
2871 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2872 * this kind of deadlock.
2874 void blk_start_plug(struct blk_plug *plug)
2876 struct task_struct *tsk = current;
2878 plug->magic = PLUG_MAGIC;
2879 INIT_LIST_HEAD(&plug->list);
2880 INIT_LIST_HEAD(&plug->cb_list);
2883 * If this is a nested plug, don't actually assign it. It will be
2884 * flushed on its own.
2888 * Store ordering should not be needed here, since a potential
2889 * preempt will imply a full memory barrier
2894 EXPORT_SYMBOL(blk_start_plug);
2896 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2898 struct request *rqa = container_of(a, struct request, queuelist);
2899 struct request *rqb = container_of(b, struct request, queuelist);
2901 return !(rqa->q < rqb->q ||
2902 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2906 * If 'from_schedule' is true, then postpone the dispatch of requests
2907 * until a safe kblockd context. We due this to avoid accidental big
2908 * additional stack usage in driver dispatch, in places where the originally
2909 * plugger did not intend it.
2911 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2913 __releases(q->queue_lock)
2915 trace_block_unplug(q, depth, !from_schedule);
2918 blk_run_queue_async(q);
2921 spin_unlock(q->queue_lock);
2924 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2926 LIST_HEAD(callbacks);
2928 while (!list_empty(&plug->cb_list)) {
2929 list_splice_init(&plug->cb_list, &callbacks);
2931 while (!list_empty(&callbacks)) {
2932 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2935 list_del(&cb->list);
2936 cb->callback(cb, from_schedule);
2941 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2944 struct blk_plug *plug = current->plug;
2945 struct blk_plug_cb *cb;
2950 list_for_each_entry(cb, &plug->cb_list, list)
2951 if (cb->callback == unplug && cb->data == data)
2954 /* Not currently on the callback list */
2955 BUG_ON(size < sizeof(*cb));
2956 cb = kzalloc(size, GFP_ATOMIC);
2959 cb->callback = unplug;
2960 list_add(&cb->list, &plug->cb_list);
2964 EXPORT_SYMBOL(blk_check_plugged);
2966 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2968 struct request_queue *q;
2969 unsigned long flags;
2974 BUG_ON(plug->magic != PLUG_MAGIC);
2976 flush_plug_callbacks(plug, from_schedule);
2977 if (list_empty(&plug->list))
2980 list_splice_init(&plug->list, &list);
2982 list_sort(NULL, &list, plug_rq_cmp);
2988 * Save and disable interrupts here, to avoid doing it for every
2989 * queue lock we have to take.
2991 local_irq_save(flags);
2992 while (!list_empty(&list)) {
2993 rq = list_entry_rq(list.next);
2994 list_del_init(&rq->queuelist);
2998 * This drops the queue lock
3001 queue_unplugged(q, depth, from_schedule);
3004 spin_lock(q->queue_lock);
3008 * Short-circuit if @q is dead
3010 if (unlikely(blk_queue_dying(q))) {
3011 __blk_end_request_all(rq, -ENODEV);
3016 * rq is already accounted, so use raw insert
3018 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3019 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3021 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3027 * This drops the queue lock
3030 queue_unplugged(q, depth, from_schedule);
3032 local_irq_restore(flags);
3035 void blk_finish_plug(struct blk_plug *plug)
3037 blk_flush_plug_list(plug, false);
3039 if (plug == current->plug)
3040 current->plug = NULL;
3042 EXPORT_SYMBOL(blk_finish_plug);
3044 #ifdef CONFIG_PM_RUNTIME
3046 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3047 * @q: the queue of the device
3048 * @dev: the device the queue belongs to
3051 * Initialize runtime-PM-related fields for @q and start auto suspend for
3052 * @dev. Drivers that want to take advantage of request-based runtime PM
3053 * should call this function after @dev has been initialized, and its
3054 * request queue @q has been allocated, and runtime PM for it can not happen
3055 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3056 * cases, driver should call this function before any I/O has taken place.
3058 * This function takes care of setting up using auto suspend for the device,
3059 * the autosuspend delay is set to -1 to make runtime suspend impossible
3060 * until an updated value is either set by user or by driver. Drivers do
3061 * not need to touch other autosuspend settings.
3063 * The block layer runtime PM is request based, so only works for drivers
3064 * that use request as their IO unit instead of those directly use bio's.
3066 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3069 q->rpm_status = RPM_ACTIVE;
3070 pm_runtime_set_autosuspend_delay(q->dev, -1);
3071 pm_runtime_use_autosuspend(q->dev);
3073 EXPORT_SYMBOL(blk_pm_runtime_init);
3076 * blk_pre_runtime_suspend - Pre runtime suspend check
3077 * @q: the queue of the device
3080 * This function will check if runtime suspend is allowed for the device
3081 * by examining if there are any requests pending in the queue. If there
3082 * are requests pending, the device can not be runtime suspended; otherwise,
3083 * the queue's status will be updated to SUSPENDING and the driver can
3084 * proceed to suspend the device.
3086 * For the not allowed case, we mark last busy for the device so that
3087 * runtime PM core will try to autosuspend it some time later.
3089 * This function should be called near the start of the device's
3090 * runtime_suspend callback.
3093 * 0 - OK to runtime suspend the device
3094 * -EBUSY - Device should not be runtime suspended
3096 int blk_pre_runtime_suspend(struct request_queue *q)
3100 spin_lock_irq(q->queue_lock);
3101 if (q->nr_pending) {
3103 pm_runtime_mark_last_busy(q->dev);
3105 q->rpm_status = RPM_SUSPENDING;
3107 spin_unlock_irq(q->queue_lock);
3110 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3113 * blk_post_runtime_suspend - Post runtime suspend processing
3114 * @q: the queue of the device
3115 * @err: return value of the device's runtime_suspend function
3118 * Update the queue's runtime status according to the return value of the
3119 * device's runtime suspend function and mark last busy for the device so
3120 * that PM core will try to auto suspend the device at a later time.
3122 * This function should be called near the end of the device's
3123 * runtime_suspend callback.
3125 void blk_post_runtime_suspend(struct request_queue *q, int err)
3127 spin_lock_irq(q->queue_lock);
3129 q->rpm_status = RPM_SUSPENDED;
3131 q->rpm_status = RPM_ACTIVE;
3132 pm_runtime_mark_last_busy(q->dev);
3134 spin_unlock_irq(q->queue_lock);
3136 EXPORT_SYMBOL(blk_post_runtime_suspend);
3139 * blk_pre_runtime_resume - Pre runtime resume processing
3140 * @q: the queue of the device
3143 * Update the queue's runtime status to RESUMING in preparation for the
3144 * runtime resume of the device.
3146 * This function should be called near the start of the device's
3147 * runtime_resume callback.
3149 void blk_pre_runtime_resume(struct request_queue *q)
3151 spin_lock_irq(q->queue_lock);
3152 q->rpm_status = RPM_RESUMING;
3153 spin_unlock_irq(q->queue_lock);
3155 EXPORT_SYMBOL(blk_pre_runtime_resume);
3158 * blk_post_runtime_resume - Post runtime resume processing
3159 * @q: the queue of the device
3160 * @err: return value of the device's runtime_resume function
3163 * Update the queue's runtime status according to the return value of the
3164 * device's runtime_resume function. If it is successfully resumed, process
3165 * the requests that are queued into the device's queue when it is resuming
3166 * and then mark last busy and initiate autosuspend for it.
3168 * This function should be called near the end of the device's
3169 * runtime_resume callback.
3171 void blk_post_runtime_resume(struct request_queue *q, int err)
3173 spin_lock_irq(q->queue_lock);
3175 q->rpm_status = RPM_ACTIVE;
3177 pm_runtime_mark_last_busy(q->dev);
3178 pm_request_autosuspend(q->dev);
3180 q->rpm_status = RPM_SUSPENDED;
3182 spin_unlock_irq(q->queue_lock);
3184 EXPORT_SYMBOL(blk_post_runtime_resume);
3187 int __init blk_dev_init(void)
3189 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3190 sizeof(((struct request *)0)->cmd_flags));
3192 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3193 kblockd_workqueue = alloc_workqueue("kblockd",
3194 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3195 if (!kblockd_workqueue)
3196 panic("Failed to create kblockd\n");
3198 request_cachep = kmem_cache_create("blkdev_requests",
3199 sizeof(struct request), 0, SLAB_PANIC, NULL);
3201 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3202 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);