2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
32 #include <linux/ratelimit.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/block.h>
38 #include "blk-cgroup.h"
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
44 DEFINE_IDA(blk_queue_ida);
47 * For the allocated request tables
49 static struct kmem_cache *request_cachep;
52 * For queue allocation
54 struct kmem_cache *blk_requestq_cachep;
57 * Controlling structure to kblockd
59 static struct workqueue_struct *kblockd_workqueue;
61 static void drive_stat_acct(struct request *rq, int new_io)
63 struct hd_struct *part;
64 int rw = rq_data_dir(rq);
67 if (!blk_do_io_stat(rq))
70 cpu = part_stat_lock();
74 part_stat_inc(cpu, part, merges[rw]);
76 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
77 if (!hd_struct_try_get(part)) {
79 * The partition is already being removed,
80 * the request will be accounted on the disk only
82 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat
84 * it as any other partition.
86 part = &rq->rq_disk->part0;
89 part_round_stats(cpu, part);
90 part_inc_in_flight(part, rw);
97 void blk_queue_congestion_threshold(struct request_queue *q)
101 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests)
104 q->nr_congestion_on = nr;
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
109 q->nr_congestion_off = nr;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
123 struct backing_dev_info *ret = NULL;
124 struct request_queue *q = bdev_get_queue(bdev);
127 ret = &q->backing_dev_info;
130 EXPORT_SYMBOL(blk_get_backing_dev_info);
132 void blk_rq_init(struct request_queue *q, struct request *rq)
134 memset(rq, 0, sizeof(*rq));
136 INIT_LIST_HEAD(&rq->queuelist);
137 INIT_LIST_HEAD(&rq->timeout_list);
140 rq->__sector = (sector_t) -1;
141 INIT_HLIST_NODE(&rq->hash);
142 RB_CLEAR_NODE(&rq->rb_node);
144 rq->cmd_len = BLK_MAX_CDB;
147 rq->start_time = jiffies;
148 set_start_time_ns(rq);
151 EXPORT_SYMBOL(blk_rq_init);
153 static void req_bio_endio(struct request *rq, struct bio *bio,
154 unsigned int nbytes, int error)
157 clear_bit(BIO_UPTODATE, &bio->bi_flags);
158 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
161 if (unlikely(nbytes > bio->bi_size)) {
162 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
163 __func__, nbytes, bio->bi_size);
164 nbytes = bio->bi_size;
167 if (unlikely(rq->cmd_flags & REQ_QUIET))
168 set_bit(BIO_QUIET, &bio->bi_flags);
170 bio->bi_size -= nbytes;
171 bio->bi_sector += (nbytes >> 9);
173 if (bio_integrity(bio))
174 bio_integrity_advance(bio, nbytes);
176 /* don't actually finish bio if it's part of flush sequence */
177 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
178 bio_endio(bio, error);
181 void blk_dump_rq_flags(struct request *rq, char *msg)
185 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
186 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
189 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq),
191 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
192 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
193 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
195 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
196 printk(KERN_INFO " cdb: ");
197 for (bit = 0; bit < BLK_MAX_CDB; bit++)
198 printk("%02x ", rq->cmd[bit]);
202 EXPORT_SYMBOL(blk_dump_rq_flags);
204 static void blk_delay_work(struct work_struct *work)
206 struct request_queue *q;
208 q = container_of(work, struct request_queue, delay_work.work);
209 spin_lock_irq(q->queue_lock);
211 spin_unlock_irq(q->queue_lock);
215 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs
220 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is
222 * restarted around the specified time.
224 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
226 queue_delayed_work(kblockd_workqueue, &q->delay_work,
227 msecs_to_jiffies(msecs));
229 EXPORT_SYMBOL(blk_delay_queue);
232 * blk_start_queue - restart a previously stopped queue
233 * @q: The &struct request_queue in question
236 * blk_start_queue() will clear the stop flag on the queue, and call
237 * the request_fn for the queue if it was in a stopped state when
238 * entered. Also see blk_stop_queue(). Queue lock must be held.
240 void blk_start_queue(struct request_queue *q)
242 WARN_ON(!irqs_disabled());
244 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
247 EXPORT_SYMBOL(blk_start_queue);
250 * blk_stop_queue - stop a queue
251 * @q: The &struct request_queue in question
254 * The Linux block layer assumes that a block driver will consume all
255 * entries on the request queue when the request_fn strategy is called.
256 * Often this will not happen, because of hardware limitations (queue
257 * depth settings). If a device driver gets a 'queue full' response,
258 * or if it simply chooses not to queue more I/O at one point, it can
259 * call this function to prevent the request_fn from being called until
260 * the driver has signalled it's ready to go again. This happens by calling
261 * blk_start_queue() to restart queue operations. Queue lock must be held.
263 void blk_stop_queue(struct request_queue *q)
265 __cancel_delayed_work(&q->delay_work);
266 queue_flag_set(QUEUE_FLAG_STOPPED, q);
268 EXPORT_SYMBOL(blk_stop_queue);
271 * blk_sync_queue - cancel any pending callbacks on a queue
275 * The block layer may perform asynchronous callback activity
276 * on a queue, such as calling the unplug function after a timeout.
277 * A block device may call blk_sync_queue to ensure that any
278 * such activity is cancelled, thus allowing it to release resources
279 * that the callbacks might use. The caller must already have made sure
280 * that its ->make_request_fn will not re-add plugging prior to calling
283 * This function does not cancel any asynchronous activity arising
284 * out of elevator or throttling code. That would require elevaotor_exit()
285 * and blkcg_exit_queue() to be called with queue lock initialized.
288 void blk_sync_queue(struct request_queue *q)
290 del_timer_sync(&q->timeout);
291 cancel_delayed_work_sync(&q->delay_work);
293 EXPORT_SYMBOL(blk_sync_queue);
296 * __blk_run_queue - run a single device queue
297 * @q: The queue to run
300 * See @blk_run_queue. This variant must be called with the queue lock
301 * held and interrupts disabled.
303 void __blk_run_queue(struct request_queue *q)
305 if (unlikely(blk_queue_stopped(q)))
310 EXPORT_SYMBOL(__blk_run_queue);
313 * blk_run_queue_async - run a single device queue in workqueue context
314 * @q: The queue to run
317 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
320 void blk_run_queue_async(struct request_queue *q)
322 if (likely(!blk_queue_stopped(q))) {
323 __cancel_delayed_work(&q->delay_work);
324 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
327 EXPORT_SYMBOL(blk_run_queue_async);
330 * blk_run_queue - run a single device queue
331 * @q: The queue to run
334 * Invoke request handling on this queue, if it has pending work to do.
335 * May be used to restart queueing when a request has completed.
337 void blk_run_queue(struct request_queue *q)
341 spin_lock_irqsave(q->queue_lock, flags);
343 spin_unlock_irqrestore(q->queue_lock, flags);
345 EXPORT_SYMBOL(blk_run_queue);
347 void blk_put_queue(struct request_queue *q)
349 kobject_put(&q->kobj);
351 EXPORT_SYMBOL(blk_put_queue);
354 * blk_drain_queue - drain requests from request_queue
356 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
358 * Drain requests from @q. If @drain_all is set, all requests are drained.
359 * If not, only ELVPRIV requests are drained. The caller is responsible
360 * for ensuring that no new requests which need to be drained are queued.
362 void blk_drain_queue(struct request_queue *q, bool drain_all)
369 spin_lock_irq(q->queue_lock);
372 * The caller might be trying to drain @q before its
373 * elevator is initialized.
376 elv_drain_elevator(q);
378 blkcg_drain_queue(q);
381 * This function might be called on a queue which failed
382 * driver init after queue creation or is not yet fully
383 * active yet. Some drivers (e.g. fd and loop) get unhappy
384 * in such cases. Kick queue iff dispatch queue has
385 * something on it and @q has request_fn set.
387 if (!list_empty(&q->queue_head) && q->request_fn)
390 drain |= q->rq.elvpriv;
393 * Unfortunately, requests are queued at and tracked from
394 * multiple places and there's no single counter which can
395 * be drained. Check all the queues and counters.
398 drain |= !list_empty(&q->queue_head);
399 for (i = 0; i < 2; i++) {
400 drain |= q->rq.count[i];
401 drain |= q->in_flight[i];
402 drain |= !list_empty(&q->flush_queue[i]);
406 spin_unlock_irq(q->queue_lock);
414 * With queue marked dead, any woken up waiter will fail the
415 * allocation path, so the wakeup chaining is lost and we're
416 * left with hung waiters. We need to wake up those waiters.
419 spin_lock_irq(q->queue_lock);
420 for (i = 0; i < ARRAY_SIZE(q->rq.wait); i++)
421 wake_up_all(&q->rq.wait[i]);
422 spin_unlock_irq(q->queue_lock);
427 * blk_queue_bypass_start - enter queue bypass mode
428 * @q: queue of interest
430 * In bypass mode, only the dispatch FIFO queue of @q is used. This
431 * function makes @q enter bypass mode and drains all requests which were
432 * throttled or issued before. On return, it's guaranteed that no request
433 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
434 * inside queue or RCU read lock.
436 void blk_queue_bypass_start(struct request_queue *q)
440 spin_lock_irq(q->queue_lock);
441 drain = !q->bypass_depth++;
442 queue_flag_set(QUEUE_FLAG_BYPASS, q);
443 spin_unlock_irq(q->queue_lock);
446 blk_drain_queue(q, false);
447 /* ensure blk_queue_bypass() is %true inside RCU read lock */
451 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
454 * blk_queue_bypass_end - leave queue bypass mode
455 * @q: queue of interest
457 * Leave bypass mode and restore the normal queueing behavior.
459 void blk_queue_bypass_end(struct request_queue *q)
461 spin_lock_irq(q->queue_lock);
462 if (!--q->bypass_depth)
463 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
464 WARN_ON_ONCE(q->bypass_depth < 0);
465 spin_unlock_irq(q->queue_lock);
467 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
470 * blk_cleanup_queue - shutdown a request queue
471 * @q: request queue to shutdown
473 * Mark @q DEAD, drain all pending requests, destroy and put it. All
474 * future requests will be failed immediately with -ENODEV.
476 void blk_cleanup_queue(struct request_queue *q)
478 spinlock_t *lock = q->queue_lock;
480 /* mark @q DEAD, no new request or merges will be allowed afterwards */
481 mutex_lock(&q->sysfs_lock);
482 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
486 * Dead queue is permanently in bypass mode till released. Note
487 * that, unlike blk_queue_bypass_start(), we aren't performing
488 * synchronize_rcu() after entering bypass mode to avoid the delay
489 * as some drivers create and destroy a lot of queues while
490 * probing. This is still safe because blk_release_queue() will be
491 * called only after the queue refcnt drops to zero and nothing,
492 * RCU or not, would be traversing the queue by then.
495 queue_flag_set(QUEUE_FLAG_BYPASS, q);
497 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
498 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
499 queue_flag_set(QUEUE_FLAG_DEAD, q);
500 spin_unlock_irq(lock);
501 mutex_unlock(&q->sysfs_lock);
503 /* drain all requests queued before DEAD marking */
504 blk_drain_queue(q, true);
506 /* @q won't process any more request, flush async actions */
507 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
511 if (q->queue_lock != &q->__queue_lock)
512 q->queue_lock = &q->__queue_lock;
513 spin_unlock_irq(lock);
515 /* @q is and will stay empty, shutdown and put */
518 EXPORT_SYMBOL(blk_cleanup_queue);
520 static int blk_init_free_list(struct request_queue *q)
522 struct request_list *rl = &q->rq;
524 if (unlikely(rl->rq_pool))
527 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
528 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
530 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
531 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
533 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
534 mempool_free_slab, request_cachep, q->node);
542 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
544 return blk_alloc_queue_node(gfp_mask, -1);
546 EXPORT_SYMBOL(blk_alloc_queue);
548 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
550 struct request_queue *q;
553 q = kmem_cache_alloc_node(blk_requestq_cachep,
554 gfp_mask | __GFP_ZERO, node_id);
558 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
562 q->backing_dev_info.ra_pages =
563 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
564 q->backing_dev_info.state = 0;
565 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
566 q->backing_dev_info.name = "block";
569 err = bdi_init(&q->backing_dev_info);
573 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
574 laptop_mode_timer_fn, (unsigned long) q);
575 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
576 INIT_LIST_HEAD(&q->queue_head);
577 INIT_LIST_HEAD(&q->timeout_list);
578 INIT_LIST_HEAD(&q->icq_list);
579 #ifdef CONFIG_BLK_CGROUP
580 INIT_LIST_HEAD(&q->blkg_list);
582 INIT_LIST_HEAD(&q->flush_queue[0]);
583 INIT_LIST_HEAD(&q->flush_queue[1]);
584 INIT_LIST_HEAD(&q->flush_data_in_flight);
585 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
587 kobject_init(&q->kobj, &blk_queue_ktype);
589 mutex_init(&q->sysfs_lock);
590 spin_lock_init(&q->__queue_lock);
593 * By default initialize queue_lock to internal lock and driver can
594 * override it later if need be.
596 q->queue_lock = &q->__queue_lock;
599 * A queue starts its life with bypass turned on to avoid
600 * unnecessary bypass on/off overhead and nasty surprises during
601 * init. The initial bypass will be finished at the end of
602 * blk_init_allocated_queue().
605 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
607 if (blkcg_init_queue(q))
613 ida_simple_remove(&blk_queue_ida, q->id);
615 kmem_cache_free(blk_requestq_cachep, q);
618 EXPORT_SYMBOL(blk_alloc_queue_node);
621 * blk_init_queue - prepare a request queue for use with a block device
622 * @rfn: The function to be called to process requests that have been
623 * placed on the queue.
624 * @lock: Request queue spin lock
627 * If a block device wishes to use the standard request handling procedures,
628 * which sorts requests and coalesces adjacent requests, then it must
629 * call blk_init_queue(). The function @rfn will be called when there
630 * are requests on the queue that need to be processed. If the device
631 * supports plugging, then @rfn may not be called immediately when requests
632 * are available on the queue, but may be called at some time later instead.
633 * Plugged queues are generally unplugged when a buffer belonging to one
634 * of the requests on the queue is needed, or due to memory pressure.
636 * @rfn is not required, or even expected, to remove all requests off the
637 * queue, but only as many as it can handle at a time. If it does leave
638 * requests on the queue, it is responsible for arranging that the requests
639 * get dealt with eventually.
641 * The queue spin lock must be held while manipulating the requests on the
642 * request queue; this lock will be taken also from interrupt context, so irq
643 * disabling is needed for it.
645 * Function returns a pointer to the initialized request queue, or %NULL if
649 * blk_init_queue() must be paired with a blk_cleanup_queue() call
650 * when the block device is deactivated (such as at module unload).
653 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
655 return blk_init_queue_node(rfn, lock, -1);
657 EXPORT_SYMBOL(blk_init_queue);
659 struct request_queue *
660 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
662 struct request_queue *uninit_q, *q;
664 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
668 q = blk_init_allocated_queue(uninit_q, rfn, lock);
670 blk_cleanup_queue(uninit_q);
674 EXPORT_SYMBOL(blk_init_queue_node);
676 struct request_queue *
677 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
683 if (blk_init_free_list(q))
687 q->prep_rq_fn = NULL;
688 q->unprep_rq_fn = NULL;
689 q->queue_flags = QUEUE_FLAG_DEFAULT;
691 /* Override internal queue lock with supplied lock pointer */
693 q->queue_lock = lock;
696 * This also sets hw/phys segments, boundary and size
698 blk_queue_make_request(q, blk_queue_bio);
700 q->sg_reserved_size = INT_MAX;
703 if (elevator_init(q, NULL))
706 blk_queue_congestion_threshold(q);
708 /* all done, end the initial bypass */
709 blk_queue_bypass_end(q);
712 EXPORT_SYMBOL(blk_init_allocated_queue);
714 bool blk_get_queue(struct request_queue *q)
716 if (likely(!blk_queue_dead(q))) {
723 EXPORT_SYMBOL(blk_get_queue);
725 static inline void blk_free_request(struct request_queue *q, struct request *rq)
727 if (rq->cmd_flags & REQ_ELVPRIV) {
728 elv_put_request(q, rq);
730 put_io_context(rq->elv.icq->ioc);
733 mempool_free(rq, q->rq.rq_pool);
737 * ioc_batching returns true if the ioc is a valid batching request and
738 * should be given priority access to a request.
740 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
746 * Make sure the process is able to allocate at least 1 request
747 * even if the batch times out, otherwise we could theoretically
750 return ioc->nr_batch_requests == q->nr_batching ||
751 (ioc->nr_batch_requests > 0
752 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
756 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
757 * will cause the process to be a "batcher" on all queues in the system. This
758 * is the behaviour we want though - once it gets a wakeup it should be given
761 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
763 if (!ioc || ioc_batching(q, ioc))
766 ioc->nr_batch_requests = q->nr_batching;
767 ioc->last_waited = jiffies;
770 static void __freed_request(struct request_queue *q, int sync)
772 struct request_list *rl = &q->rq;
774 if (rl->count[sync] < queue_congestion_off_threshold(q))
775 blk_clear_queue_congested(q, sync);
777 if (rl->count[sync] + 1 <= q->nr_requests) {
778 if (waitqueue_active(&rl->wait[sync]))
779 wake_up(&rl->wait[sync]);
781 blk_clear_queue_full(q, sync);
786 * A request has just been released. Account for it, update the full and
787 * congestion status, wake up any waiters. Called under q->queue_lock.
789 static void freed_request(struct request_queue *q, unsigned int flags)
791 struct request_list *rl = &q->rq;
792 int sync = rw_is_sync(flags);
795 if (flags & REQ_ELVPRIV)
798 __freed_request(q, sync);
800 if (unlikely(rl->starved[sync ^ 1]))
801 __freed_request(q, sync ^ 1);
805 * Determine if elevator data should be initialized when allocating the
806 * request associated with @bio.
808 static bool blk_rq_should_init_elevator(struct bio *bio)
814 * Flush requests do not use the elevator so skip initialization.
815 * This allows a request to share the flush and elevator data.
817 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
824 * rq_ioc - determine io_context for request allocation
825 * @bio: request being allocated is for this bio (can be %NULL)
827 * Determine io_context to use for request allocation for @bio. May return
828 * %NULL if %current->io_context doesn't exist.
830 static struct io_context *rq_ioc(struct bio *bio)
832 #ifdef CONFIG_BLK_CGROUP
833 if (bio && bio->bi_ioc)
836 return current->io_context;
840 * get_request - get a free request
841 * @q: request_queue to allocate request from
842 * @rw_flags: RW and SYNC flags
843 * @bio: bio to allocate request for (can be %NULL)
844 * @gfp_mask: allocation mask
846 * Get a free request from @q. This function may fail under memory
847 * pressure or if @q is dead.
849 * Must be callled with @q->queue_lock held and,
850 * Returns %NULL on failure, with @q->queue_lock held.
851 * Returns !%NULL on success, with @q->queue_lock *not held*.
853 static struct request *get_request(struct request_queue *q, int rw_flags,
854 struct bio *bio, gfp_t gfp_mask)
857 struct request_list *rl = &q->rq;
858 struct elevator_type *et;
859 struct io_context *ioc;
860 struct io_cq *icq = NULL;
861 const bool is_sync = rw_is_sync(rw_flags) != 0;
862 bool retried = false;
865 et = q->elevator->type;
868 if (unlikely(blk_queue_dead(q)))
871 may_queue = elv_may_queue(q, rw_flags);
872 if (may_queue == ELV_MQUEUE_NO)
875 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
876 if (rl->count[is_sync]+1 >= q->nr_requests) {
878 * We want ioc to record batching state. If it's
879 * not already there, creating a new one requires
880 * dropping queue_lock, which in turn requires
881 * retesting conditions to avoid queue hang.
883 if (!ioc && !retried) {
884 spin_unlock_irq(q->queue_lock);
885 create_io_context(gfp_mask, q->node);
886 spin_lock_irq(q->queue_lock);
892 * The queue will fill after this allocation, so set
893 * it as full, and mark this process as "batching".
894 * This process will be allowed to complete a batch of
895 * requests, others will be blocked.
897 if (!blk_queue_full(q, is_sync)) {
898 ioc_set_batching(q, ioc);
899 blk_set_queue_full(q, is_sync);
901 if (may_queue != ELV_MQUEUE_MUST
902 && !ioc_batching(q, ioc)) {
904 * The queue is full and the allocating
905 * process is not a "batcher", and not
906 * exempted by the IO scheduler
912 blk_set_queue_congested(q, is_sync);
916 * Only allow batching queuers to allocate up to 50% over the defined
917 * limit of requests, otherwise we could have thousands of requests
918 * allocated with any setting of ->nr_requests
920 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
923 rl->count[is_sync]++;
924 rl->starved[is_sync] = 0;
927 * Decide whether the new request will be managed by elevator. If
928 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
929 * prevent the current elevator from being destroyed until the new
930 * request is freed. This guarantees icq's won't be destroyed and
931 * makes creating new ones safe.
933 * Also, lookup icq while holding queue_lock. If it doesn't exist,
934 * it will be created after releasing queue_lock.
936 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
937 rw_flags |= REQ_ELVPRIV;
939 if (et->icq_cache && ioc)
940 icq = ioc_lookup_icq(ioc, q);
943 if (blk_queue_io_stat(q))
944 rw_flags |= REQ_IO_STAT;
945 spin_unlock_irq(q->queue_lock);
947 /* allocate and init request */
948 rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
953 rq->cmd_flags = rw_flags | REQ_ALLOCED;
956 if (rw_flags & REQ_ELVPRIV) {
957 if (unlikely(et->icq_cache && !icq)) {
958 create_io_context(gfp_mask, q->node);
963 icq = ioc_create_icq(ioc, q, gfp_mask);
969 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
972 /* @rq->elv.icq holds io_context until @rq is freed */
974 get_io_context(icq->ioc);
978 * ioc may be NULL here, and ioc_batching will be false. That's
979 * OK, if the queue is under the request limit then requests need
980 * not count toward the nr_batch_requests limit. There will always
981 * be some limit enforced by BLK_BATCH_TIME.
983 if (ioc_batching(q, ioc))
984 ioc->nr_batch_requests--;
986 trace_block_getrq(q, bio, rw_flags & 1);
991 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
992 * and may fail indefinitely under memory pressure and thus
993 * shouldn't stall IO. Treat this request as !elvpriv. This will
994 * disturb iosched and blkcg but weird is bettern than dead.
996 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
997 dev_name(q->backing_dev_info.dev));
999 rq->cmd_flags &= ~REQ_ELVPRIV;
1002 spin_lock_irq(q->queue_lock);
1004 spin_unlock_irq(q->queue_lock);
1009 * Allocation failed presumably due to memory. Undo anything we
1010 * might have messed up.
1012 * Allocating task should really be put onto the front of the wait
1013 * queue, but this is pretty rare.
1015 spin_lock_irq(q->queue_lock);
1016 freed_request(q, rw_flags);
1019 * in the very unlikely event that allocation failed and no
1020 * requests for this direction was pending, mark us starved so that
1021 * freeing of a request in the other direction will notice
1022 * us. another possible fix would be to split the rq mempool into
1026 if (unlikely(rl->count[is_sync] == 0))
1027 rl->starved[is_sync] = 1;
1032 * get_request_wait - get a free request with retry
1033 * @q: request_queue to allocate request from
1034 * @rw_flags: RW and SYNC flags
1035 * @bio: bio to allocate request for (can be %NULL)
1037 * Get a free request from @q. This function keeps retrying under memory
1038 * pressure and fails iff @q is dead.
1040 * Must be callled with @q->queue_lock held and,
1041 * Returns %NULL on failure, with @q->queue_lock held.
1042 * Returns !%NULL on success, with @q->queue_lock *not held*.
1044 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
1047 const bool is_sync = rw_is_sync(rw_flags) != 0;
1050 rq = get_request(q, rw_flags, bio, GFP_NOIO);
1053 struct request_list *rl = &q->rq;
1055 if (unlikely(blk_queue_dead(q)))
1058 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1059 TASK_UNINTERRUPTIBLE);
1061 trace_block_sleeprq(q, bio, rw_flags & 1);
1063 spin_unlock_irq(q->queue_lock);
1067 * After sleeping, we become a "batching" process and
1068 * will be able to allocate at least one request, and
1069 * up to a big batch of them for a small period time.
1070 * See ioc_batching, ioc_set_batching
1072 create_io_context(GFP_NOIO, q->node);
1073 ioc_set_batching(q, current->io_context);
1075 spin_lock_irq(q->queue_lock);
1076 finish_wait(&rl->wait[is_sync], &wait);
1078 rq = get_request(q, rw_flags, bio, GFP_NOIO);
1084 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1088 BUG_ON(rw != READ && rw != WRITE);
1090 spin_lock_irq(q->queue_lock);
1091 if (gfp_mask & __GFP_WAIT)
1092 rq = get_request_wait(q, rw, NULL);
1094 rq = get_request(q, rw, NULL, gfp_mask);
1096 spin_unlock_irq(q->queue_lock);
1097 /* q->queue_lock is unlocked at this point */
1101 EXPORT_SYMBOL(blk_get_request);
1104 * blk_make_request - given a bio, allocate a corresponding struct request.
1105 * @q: target request queue
1106 * @bio: The bio describing the memory mappings that will be submitted for IO.
1107 * It may be a chained-bio properly constructed by block/bio layer.
1108 * @gfp_mask: gfp flags to be used for memory allocation
1110 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1111 * type commands. Where the struct request needs to be farther initialized by
1112 * the caller. It is passed a &struct bio, which describes the memory info of
1115 * The caller of blk_make_request must make sure that bi_io_vec
1116 * are set to describe the memory buffers. That bio_data_dir() will return
1117 * the needed direction of the request. (And all bio's in the passed bio-chain
1118 * are properly set accordingly)
1120 * If called under none-sleepable conditions, mapped bio buffers must not
1121 * need bouncing, by calling the appropriate masked or flagged allocator,
1122 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1125 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1126 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1127 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1128 * completion of a bio that hasn't been submitted yet, thus resulting in a
1129 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1130 * of bio_alloc(), as that avoids the mempool deadlock.
1131 * If possible a big IO should be split into smaller parts when allocation
1132 * fails. Partial allocation should not be an error, or you risk a live-lock.
1134 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1137 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1140 return ERR_PTR(-ENOMEM);
1143 struct bio *bounce_bio = bio;
1146 blk_queue_bounce(q, &bounce_bio);
1147 ret = blk_rq_append_bio(q, rq, bounce_bio);
1148 if (unlikely(ret)) {
1149 blk_put_request(rq);
1150 return ERR_PTR(ret);
1156 EXPORT_SYMBOL(blk_make_request);
1159 * blk_requeue_request - put a request back on queue
1160 * @q: request queue where request should be inserted
1161 * @rq: request to be inserted
1164 * Drivers often keep queueing requests until the hardware cannot accept
1165 * more, when that condition happens we need to put the request back
1166 * on the queue. Must be called with queue lock held.
1168 void blk_requeue_request(struct request_queue *q, struct request *rq)
1170 blk_delete_timer(rq);
1171 blk_clear_rq_complete(rq);
1172 trace_block_rq_requeue(q, rq);
1174 if (blk_rq_tagged(rq))
1175 blk_queue_end_tag(q, rq);
1177 BUG_ON(blk_queued_rq(rq));
1179 elv_requeue_request(q, rq);
1181 EXPORT_SYMBOL(blk_requeue_request);
1183 static void add_acct_request(struct request_queue *q, struct request *rq,
1186 drive_stat_acct(rq, 1);
1187 __elv_add_request(q, rq, where);
1190 static void part_round_stats_single(int cpu, struct hd_struct *part,
1193 if (now == part->stamp)
1196 if (part_in_flight(part)) {
1197 __part_stat_add(cpu, part, time_in_queue,
1198 part_in_flight(part) * (now - part->stamp));
1199 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1205 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1206 * @cpu: cpu number for stats access
1207 * @part: target partition
1209 * The average IO queue length and utilisation statistics are maintained
1210 * by observing the current state of the queue length and the amount of
1211 * time it has been in this state for.
1213 * Normally, that accounting is done on IO completion, but that can result
1214 * in more than a second's worth of IO being accounted for within any one
1215 * second, leading to >100% utilisation. To deal with that, we call this
1216 * function to do a round-off before returning the results when reading
1217 * /proc/diskstats. This accounts immediately for all queue usage up to
1218 * the current jiffies and restarts the counters again.
1220 void part_round_stats(int cpu, struct hd_struct *part)
1222 unsigned long now = jiffies;
1225 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1226 part_round_stats_single(cpu, part, now);
1228 EXPORT_SYMBOL_GPL(part_round_stats);
1231 * queue lock must be held
1233 void __blk_put_request(struct request_queue *q, struct request *req)
1237 if (unlikely(--req->ref_count))
1240 elv_completed_request(q, req);
1242 /* this is a bio leak */
1243 WARN_ON(req->bio != NULL);
1246 * Request may not have originated from ll_rw_blk. if not,
1247 * it didn't come out of our reserved rq pools
1249 if (req->cmd_flags & REQ_ALLOCED) {
1250 unsigned int flags = req->cmd_flags;
1252 BUG_ON(!list_empty(&req->queuelist));
1253 BUG_ON(!hlist_unhashed(&req->hash));
1255 blk_free_request(q, req);
1256 freed_request(q, flags);
1259 EXPORT_SYMBOL_GPL(__blk_put_request);
1261 void blk_put_request(struct request *req)
1263 unsigned long flags;
1264 struct request_queue *q = req->q;
1266 spin_lock_irqsave(q->queue_lock, flags);
1267 __blk_put_request(q, req);
1268 spin_unlock_irqrestore(q->queue_lock, flags);
1270 EXPORT_SYMBOL(blk_put_request);
1273 * blk_add_request_payload - add a payload to a request
1274 * @rq: request to update
1275 * @page: page backing the payload
1276 * @len: length of the payload.
1278 * This allows to later add a payload to an already submitted request by
1279 * a block driver. The driver needs to take care of freeing the payload
1282 * Note that this is a quite horrible hack and nothing but handling of
1283 * discard requests should ever use it.
1285 void blk_add_request_payload(struct request *rq, struct page *page,
1288 struct bio *bio = rq->bio;
1290 bio->bi_io_vec->bv_page = page;
1291 bio->bi_io_vec->bv_offset = 0;
1292 bio->bi_io_vec->bv_len = len;
1296 bio->bi_phys_segments = 1;
1298 rq->__data_len = rq->resid_len = len;
1299 rq->nr_phys_segments = 1;
1300 rq->buffer = bio_data(bio);
1302 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1304 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1307 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1309 if (!ll_back_merge_fn(q, req, bio))
1312 trace_block_bio_backmerge(q, bio);
1314 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1315 blk_rq_set_mixed_merge(req);
1317 req->biotail->bi_next = bio;
1319 req->__data_len += bio->bi_size;
1320 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1322 drive_stat_acct(req, 0);
1326 static bool bio_attempt_front_merge(struct request_queue *q,
1327 struct request *req, struct bio *bio)
1329 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1331 if (!ll_front_merge_fn(q, req, bio))
1334 trace_block_bio_frontmerge(q, bio);
1336 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1337 blk_rq_set_mixed_merge(req);
1339 bio->bi_next = req->bio;
1343 * may not be valid. if the low level driver said
1344 * it didn't need a bounce buffer then it better
1345 * not touch req->buffer either...
1347 req->buffer = bio_data(bio);
1348 req->__sector = bio->bi_sector;
1349 req->__data_len += bio->bi_size;
1350 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1352 drive_stat_acct(req, 0);
1357 * attempt_plug_merge - try to merge with %current's plugged list
1358 * @q: request_queue new bio is being queued at
1359 * @bio: new bio being queued
1360 * @request_count: out parameter for number of traversed plugged requests
1362 * Determine whether @bio being queued on @q can be merged with a request
1363 * on %current's plugged list. Returns %true if merge was successful,
1366 * Plugging coalesces IOs from the same issuer for the same purpose without
1367 * going through @q->queue_lock. As such it's more of an issuing mechanism
1368 * than scheduling, and the request, while may have elvpriv data, is not
1369 * added on the elevator at this point. In addition, we don't have
1370 * reliable access to the elevator outside queue lock. Only check basic
1371 * merging parameters without querying the elevator.
1373 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1374 unsigned int *request_count)
1376 struct blk_plug *plug;
1380 plug = current->plug;
1385 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1391 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1394 el_ret = blk_try_merge(rq, bio);
1395 if (el_ret == ELEVATOR_BACK_MERGE) {
1396 ret = bio_attempt_back_merge(q, rq, bio);
1399 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1400 ret = bio_attempt_front_merge(q, rq, bio);
1409 void init_request_from_bio(struct request *req, struct bio *bio)
1411 req->cmd_type = REQ_TYPE_FS;
1413 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1414 if (bio->bi_rw & REQ_RAHEAD)
1415 req->cmd_flags |= REQ_FAILFAST_MASK;
1418 req->__sector = bio->bi_sector;
1419 req->ioprio = bio_prio(bio);
1420 blk_rq_bio_prep(req->q, req, bio);
1423 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1425 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1426 struct blk_plug *plug;
1427 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1428 struct request *req;
1429 unsigned int request_count = 0;
1432 * low level driver can indicate that it wants pages above a
1433 * certain limit bounced to low memory (ie for highmem, or even
1434 * ISA dma in theory)
1436 blk_queue_bounce(q, &bio);
1438 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1439 spin_lock_irq(q->queue_lock);
1440 where = ELEVATOR_INSERT_FLUSH;
1445 * Check if we can merge with the plugged list before grabbing
1448 if (attempt_plug_merge(q, bio, &request_count))
1451 spin_lock_irq(q->queue_lock);
1453 el_ret = elv_merge(q, &req, bio);
1454 if (el_ret == ELEVATOR_BACK_MERGE) {
1455 if (bio_attempt_back_merge(q, req, bio)) {
1456 elv_bio_merged(q, req, bio);
1457 if (!attempt_back_merge(q, req))
1458 elv_merged_request(q, req, el_ret);
1461 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1462 if (bio_attempt_front_merge(q, req, bio)) {
1463 elv_bio_merged(q, req, bio);
1464 if (!attempt_front_merge(q, req))
1465 elv_merged_request(q, req, el_ret);
1472 * This sync check and mask will be re-done in init_request_from_bio(),
1473 * but we need to set it earlier to expose the sync flag to the
1474 * rq allocator and io schedulers.
1476 rw_flags = bio_data_dir(bio);
1478 rw_flags |= REQ_SYNC;
1481 * Grab a free request. This is might sleep but can not fail.
1482 * Returns with the queue unlocked.
1484 req = get_request_wait(q, rw_flags, bio);
1485 if (unlikely(!req)) {
1486 bio_endio(bio, -ENODEV); /* @q is dead */
1491 * After dropping the lock and possibly sleeping here, our request
1492 * may now be mergeable after it had proven unmergeable (above).
1493 * We don't worry about that case for efficiency. It won't happen
1494 * often, and the elevators are able to handle it.
1496 init_request_from_bio(req, bio);
1498 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1499 req->cpu = raw_smp_processor_id();
1501 plug = current->plug;
1504 * If this is the first request added after a plug, fire
1505 * of a plug trace. If others have been added before, check
1506 * if we have multiple devices in this plug. If so, make a
1507 * note to sort the list before dispatch.
1509 if (list_empty(&plug->list))
1510 trace_block_plug(q);
1512 if (!plug->should_sort) {
1513 struct request *__rq;
1515 __rq = list_entry_rq(plug->list.prev);
1517 plug->should_sort = 1;
1519 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1520 blk_flush_plug_list(plug, false);
1521 trace_block_plug(q);
1524 list_add_tail(&req->queuelist, &plug->list);
1525 drive_stat_acct(req, 1);
1527 spin_lock_irq(q->queue_lock);
1528 add_acct_request(q, req, where);
1531 spin_unlock_irq(q->queue_lock);
1534 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1537 * If bio->bi_dev is a partition, remap the location
1539 static inline void blk_partition_remap(struct bio *bio)
1541 struct block_device *bdev = bio->bi_bdev;
1543 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1544 struct hd_struct *p = bdev->bd_part;
1546 bio->bi_sector += p->start_sect;
1547 bio->bi_bdev = bdev->bd_contains;
1549 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1551 bio->bi_sector - p->start_sect);
1555 static void handle_bad_sector(struct bio *bio)
1557 char b[BDEVNAME_SIZE];
1559 printk(KERN_INFO "attempt to access beyond end of device\n");
1560 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1561 bdevname(bio->bi_bdev, b),
1563 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1564 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1566 set_bit(BIO_EOF, &bio->bi_flags);
1569 #ifdef CONFIG_FAIL_MAKE_REQUEST
1571 static DECLARE_FAULT_ATTR(fail_make_request);
1573 static int __init setup_fail_make_request(char *str)
1575 return setup_fault_attr(&fail_make_request, str);
1577 __setup("fail_make_request=", setup_fail_make_request);
1579 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1581 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1584 static int __init fail_make_request_debugfs(void)
1586 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1587 NULL, &fail_make_request);
1589 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1592 late_initcall(fail_make_request_debugfs);
1594 #else /* CONFIG_FAIL_MAKE_REQUEST */
1596 static inline bool should_fail_request(struct hd_struct *part,
1602 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1605 * Check whether this bio extends beyond the end of the device.
1607 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1614 /* Test device or partition size, when known. */
1615 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1617 sector_t sector = bio->bi_sector;
1619 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1621 * This may well happen - the kernel calls bread()
1622 * without checking the size of the device, e.g., when
1623 * mounting a device.
1625 handle_bad_sector(bio);
1633 static noinline_for_stack bool
1634 generic_make_request_checks(struct bio *bio)
1636 struct request_queue *q;
1637 int nr_sectors = bio_sectors(bio);
1639 char b[BDEVNAME_SIZE];
1640 struct hd_struct *part;
1644 if (bio_check_eod(bio, nr_sectors))
1647 q = bdev_get_queue(bio->bi_bdev);
1650 "generic_make_request: Trying to access "
1651 "nonexistent block-device %s (%Lu)\n",
1652 bdevname(bio->bi_bdev, b),
1653 (long long) bio->bi_sector);
1657 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1658 nr_sectors > queue_max_hw_sectors(q))) {
1659 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1660 bdevname(bio->bi_bdev, b),
1662 queue_max_hw_sectors(q));
1666 part = bio->bi_bdev->bd_part;
1667 if (should_fail_request(part, bio->bi_size) ||
1668 should_fail_request(&part_to_disk(part)->part0,
1673 * If this device has partitions, remap block n
1674 * of partition p to block n+start(p) of the disk.
1676 blk_partition_remap(bio);
1678 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1681 if (bio_check_eod(bio, nr_sectors))
1685 * Filter flush bio's early so that make_request based
1686 * drivers without flush support don't have to worry
1689 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1690 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1697 if ((bio->bi_rw & REQ_DISCARD) &&
1698 (!blk_queue_discard(q) ||
1699 ((bio->bi_rw & REQ_SECURE) &&
1700 !blk_queue_secdiscard(q)))) {
1705 if (blk_throtl_bio(q, bio))
1706 return false; /* throttled, will be resubmitted later */
1708 trace_block_bio_queue(q, bio);
1712 bio_endio(bio, err);
1717 * generic_make_request - hand a buffer to its device driver for I/O
1718 * @bio: The bio describing the location in memory and on the device.
1720 * generic_make_request() is used to make I/O requests of block
1721 * devices. It is passed a &struct bio, which describes the I/O that needs
1724 * generic_make_request() does not return any status. The
1725 * success/failure status of the request, along with notification of
1726 * completion, is delivered asynchronously through the bio->bi_end_io
1727 * function described (one day) else where.
1729 * The caller of generic_make_request must make sure that bi_io_vec
1730 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1731 * set to describe the device address, and the
1732 * bi_end_io and optionally bi_private are set to describe how
1733 * completion notification should be signaled.
1735 * generic_make_request and the drivers it calls may use bi_next if this
1736 * bio happens to be merged with someone else, and may resubmit the bio to
1737 * a lower device by calling into generic_make_request recursively, which
1738 * means the bio should NOT be touched after the call to ->make_request_fn.
1740 void generic_make_request(struct bio *bio)
1742 struct bio_list bio_list_on_stack;
1744 if (!generic_make_request_checks(bio))
1748 * We only want one ->make_request_fn to be active at a time, else
1749 * stack usage with stacked devices could be a problem. So use
1750 * current->bio_list to keep a list of requests submited by a
1751 * make_request_fn function. current->bio_list is also used as a
1752 * flag to say if generic_make_request is currently active in this
1753 * task or not. If it is NULL, then no make_request is active. If
1754 * it is non-NULL, then a make_request is active, and new requests
1755 * should be added at the tail
1757 if (current->bio_list) {
1758 bio_list_add(current->bio_list, bio);
1762 /* following loop may be a bit non-obvious, and so deserves some
1764 * Before entering the loop, bio->bi_next is NULL (as all callers
1765 * ensure that) so we have a list with a single bio.
1766 * We pretend that we have just taken it off a longer list, so
1767 * we assign bio_list to a pointer to the bio_list_on_stack,
1768 * thus initialising the bio_list of new bios to be
1769 * added. ->make_request() may indeed add some more bios
1770 * through a recursive call to generic_make_request. If it
1771 * did, we find a non-NULL value in bio_list and re-enter the loop
1772 * from the top. In this case we really did just take the bio
1773 * of the top of the list (no pretending) and so remove it from
1774 * bio_list, and call into ->make_request() again.
1776 BUG_ON(bio->bi_next);
1777 bio_list_init(&bio_list_on_stack);
1778 current->bio_list = &bio_list_on_stack;
1780 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1782 q->make_request_fn(q, bio);
1784 bio = bio_list_pop(current->bio_list);
1786 current->bio_list = NULL; /* deactivate */
1788 EXPORT_SYMBOL(generic_make_request);
1791 * submit_bio - submit a bio to the block device layer for I/O
1792 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1793 * @bio: The &struct bio which describes the I/O
1795 * submit_bio() is very similar in purpose to generic_make_request(), and
1796 * uses that function to do most of the work. Both are fairly rough
1797 * interfaces; @bio must be presetup and ready for I/O.
1800 void submit_bio(int rw, struct bio *bio)
1802 int count = bio_sectors(bio);
1807 * If it's a regular read/write or a barrier with data attached,
1808 * go through the normal accounting stuff before submission.
1810 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1812 count_vm_events(PGPGOUT, count);
1814 task_io_account_read(bio->bi_size);
1815 count_vm_events(PGPGIN, count);
1818 if (unlikely(block_dump)) {
1819 char b[BDEVNAME_SIZE];
1820 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1821 current->comm, task_pid_nr(current),
1822 (rw & WRITE) ? "WRITE" : "READ",
1823 (unsigned long long)bio->bi_sector,
1824 bdevname(bio->bi_bdev, b),
1829 generic_make_request(bio);
1831 EXPORT_SYMBOL(submit_bio);
1834 * blk_rq_check_limits - Helper function to check a request for the queue limit
1836 * @rq: the request being checked
1839 * @rq may have been made based on weaker limitations of upper-level queues
1840 * in request stacking drivers, and it may violate the limitation of @q.
1841 * Since the block layer and the underlying device driver trust @rq
1842 * after it is inserted to @q, it should be checked against @q before
1843 * the insertion using this generic function.
1845 * This function should also be useful for request stacking drivers
1846 * in some cases below, so export this function.
1847 * Request stacking drivers like request-based dm may change the queue
1848 * limits while requests are in the queue (e.g. dm's table swapping).
1849 * Such request stacking drivers should check those requests agaist
1850 * the new queue limits again when they dispatch those requests,
1851 * although such checkings are also done against the old queue limits
1852 * when submitting requests.
1854 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1856 if (rq->cmd_flags & REQ_DISCARD)
1859 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1860 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1861 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1866 * queue's settings related to segment counting like q->bounce_pfn
1867 * may differ from that of other stacking queues.
1868 * Recalculate it to check the request correctly on this queue's
1871 blk_recalc_rq_segments(rq);
1872 if (rq->nr_phys_segments > queue_max_segments(q)) {
1873 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1879 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1882 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1883 * @q: the queue to submit the request
1884 * @rq: the request being queued
1886 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1888 unsigned long flags;
1889 int where = ELEVATOR_INSERT_BACK;
1891 if (blk_rq_check_limits(q, rq))
1895 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1898 spin_lock_irqsave(q->queue_lock, flags);
1899 if (unlikely(blk_queue_dead(q))) {
1900 spin_unlock_irqrestore(q->queue_lock, flags);
1905 * Submitting request must be dequeued before calling this function
1906 * because it will be linked to another request_queue
1908 BUG_ON(blk_queued_rq(rq));
1910 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1911 where = ELEVATOR_INSERT_FLUSH;
1913 add_acct_request(q, rq, where);
1914 if (where == ELEVATOR_INSERT_FLUSH)
1916 spin_unlock_irqrestore(q->queue_lock, flags);
1920 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1923 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1924 * @rq: request to examine
1927 * A request could be merge of IOs which require different failure
1928 * handling. This function determines the number of bytes which
1929 * can be failed from the beginning of the request without
1930 * crossing into area which need to be retried further.
1933 * The number of bytes to fail.
1936 * queue_lock must be held.
1938 unsigned int blk_rq_err_bytes(const struct request *rq)
1940 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1941 unsigned int bytes = 0;
1944 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1945 return blk_rq_bytes(rq);
1948 * Currently the only 'mixing' which can happen is between
1949 * different fastfail types. We can safely fail portions
1950 * which have all the failfast bits that the first one has -
1951 * the ones which are at least as eager to fail as the first
1954 for (bio = rq->bio; bio; bio = bio->bi_next) {
1955 if ((bio->bi_rw & ff) != ff)
1957 bytes += bio->bi_size;
1960 /* this could lead to infinite loop */
1961 BUG_ON(blk_rq_bytes(rq) && !bytes);
1964 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1966 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1968 if (blk_do_io_stat(req)) {
1969 const int rw = rq_data_dir(req);
1970 struct hd_struct *part;
1973 cpu = part_stat_lock();
1975 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1980 static void blk_account_io_done(struct request *req)
1983 * Account IO completion. flush_rq isn't accounted as a
1984 * normal IO on queueing nor completion. Accounting the
1985 * containing request is enough.
1987 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1988 unsigned long duration = jiffies - req->start_time;
1989 const int rw = rq_data_dir(req);
1990 struct hd_struct *part;
1993 cpu = part_stat_lock();
1996 part_stat_inc(cpu, part, ios[rw]);
1997 part_stat_add(cpu, part, ticks[rw], duration);
1998 part_round_stats(cpu, part);
1999 part_dec_in_flight(part, rw);
2001 hd_struct_put(part);
2007 * blk_peek_request - peek at the top of a request queue
2008 * @q: request queue to peek at
2011 * Return the request at the top of @q. The returned request
2012 * should be started using blk_start_request() before LLD starts
2016 * Pointer to the request at the top of @q if available. Null
2020 * queue_lock must be held.
2022 struct request *blk_peek_request(struct request_queue *q)
2027 while ((rq = __elv_next_request(q)) != NULL) {
2028 if (!(rq->cmd_flags & REQ_STARTED)) {
2030 * This is the first time the device driver
2031 * sees this request (possibly after
2032 * requeueing). Notify IO scheduler.
2034 if (rq->cmd_flags & REQ_SORTED)
2035 elv_activate_rq(q, rq);
2038 * just mark as started even if we don't start
2039 * it, a request that has been delayed should
2040 * not be passed by new incoming requests
2042 rq->cmd_flags |= REQ_STARTED;
2043 trace_block_rq_issue(q, rq);
2046 if (!q->boundary_rq || q->boundary_rq == rq) {
2047 q->end_sector = rq_end_sector(rq);
2048 q->boundary_rq = NULL;
2051 if (rq->cmd_flags & REQ_DONTPREP)
2054 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2056 * make sure space for the drain appears we
2057 * know we can do this because max_hw_segments
2058 * has been adjusted to be one fewer than the
2061 rq->nr_phys_segments++;
2067 ret = q->prep_rq_fn(q, rq);
2068 if (ret == BLKPREP_OK) {
2070 } else if (ret == BLKPREP_DEFER) {
2072 * the request may have been (partially) prepped.
2073 * we need to keep this request in the front to
2074 * avoid resource deadlock. REQ_STARTED will
2075 * prevent other fs requests from passing this one.
2077 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2078 !(rq->cmd_flags & REQ_DONTPREP)) {
2080 * remove the space for the drain we added
2081 * so that we don't add it again
2083 --rq->nr_phys_segments;
2088 } else if (ret == BLKPREP_KILL) {
2089 rq->cmd_flags |= REQ_QUIET;
2091 * Mark this request as started so we don't trigger
2092 * any debug logic in the end I/O path.
2094 blk_start_request(rq);
2095 __blk_end_request_all(rq, -EIO);
2097 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2104 EXPORT_SYMBOL(blk_peek_request);
2106 void blk_dequeue_request(struct request *rq)
2108 struct request_queue *q = rq->q;
2110 BUG_ON(list_empty(&rq->queuelist));
2111 BUG_ON(ELV_ON_HASH(rq));
2113 list_del_init(&rq->queuelist);
2116 * the time frame between a request being removed from the lists
2117 * and to it is freed is accounted as io that is in progress at
2120 if (blk_account_rq(rq)) {
2121 q->in_flight[rq_is_sync(rq)]++;
2122 set_io_start_time_ns(rq);
2127 * blk_start_request - start request processing on the driver
2128 * @req: request to dequeue
2131 * Dequeue @req and start timeout timer on it. This hands off the
2132 * request to the driver.
2134 * Block internal functions which don't want to start timer should
2135 * call blk_dequeue_request().
2138 * queue_lock must be held.
2140 void blk_start_request(struct request *req)
2142 blk_dequeue_request(req);
2145 * We are now handing the request to the hardware, initialize
2146 * resid_len to full count and add the timeout handler.
2148 req->resid_len = blk_rq_bytes(req);
2149 if (unlikely(blk_bidi_rq(req)))
2150 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2154 EXPORT_SYMBOL(blk_start_request);
2157 * blk_fetch_request - fetch a request from a request queue
2158 * @q: request queue to fetch a request from
2161 * Return the request at the top of @q. The request is started on
2162 * return and LLD can start processing it immediately.
2165 * Pointer to the request at the top of @q if available. Null
2169 * queue_lock must be held.
2171 struct request *blk_fetch_request(struct request_queue *q)
2175 rq = blk_peek_request(q);
2177 blk_start_request(rq);
2180 EXPORT_SYMBOL(blk_fetch_request);
2183 * blk_update_request - Special helper function for request stacking drivers
2184 * @req: the request being processed
2185 * @error: %0 for success, < %0 for error
2186 * @nr_bytes: number of bytes to complete @req
2189 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2190 * the request structure even if @req doesn't have leftover.
2191 * If @req has leftover, sets it up for the next range of segments.
2193 * This special helper function is only for request stacking drivers
2194 * (e.g. request-based dm) so that they can handle partial completion.
2195 * Actual device drivers should use blk_end_request instead.
2197 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2198 * %false return from this function.
2201 * %false - this request doesn't have any more data
2202 * %true - this request has more data
2204 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2206 int total_bytes, bio_nbytes, next_idx = 0;
2212 trace_block_rq_complete(req->q, req);
2215 * For fs requests, rq is just carrier of independent bio's
2216 * and each partial completion should be handled separately.
2217 * Reset per-request error on each partial completion.
2219 * TODO: tj: This is too subtle. It would be better to let
2220 * low level drivers do what they see fit.
2222 if (req->cmd_type == REQ_TYPE_FS)
2225 if (error && req->cmd_type == REQ_TYPE_FS &&
2226 !(req->cmd_flags & REQ_QUIET)) {
2231 error_type = "recoverable transport";
2234 error_type = "critical target";
2237 error_type = "critical nexus";
2244 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2245 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2246 (unsigned long long)blk_rq_pos(req));
2249 blk_account_io_completion(req, nr_bytes);
2251 total_bytes = bio_nbytes = 0;
2252 while ((bio = req->bio) != NULL) {
2255 if (nr_bytes >= bio->bi_size) {
2256 req->bio = bio->bi_next;
2257 nbytes = bio->bi_size;
2258 req_bio_endio(req, bio, nbytes, error);
2262 int idx = bio->bi_idx + next_idx;
2264 if (unlikely(idx >= bio->bi_vcnt)) {
2265 blk_dump_rq_flags(req, "__end_that");
2266 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2267 __func__, idx, bio->bi_vcnt);
2271 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2272 BIO_BUG_ON(nbytes > bio->bi_size);
2275 * not a complete bvec done
2277 if (unlikely(nbytes > nr_bytes)) {
2278 bio_nbytes += nr_bytes;
2279 total_bytes += nr_bytes;
2284 * advance to the next vector
2287 bio_nbytes += nbytes;
2290 total_bytes += nbytes;
2296 * end more in this run, or just return 'not-done'
2298 if (unlikely(nr_bytes <= 0))
2308 * Reset counters so that the request stacking driver
2309 * can find how many bytes remain in the request
2312 req->__data_len = 0;
2317 * if the request wasn't completed, update state
2320 req_bio_endio(req, bio, bio_nbytes, error);
2321 bio->bi_idx += next_idx;
2322 bio_iovec(bio)->bv_offset += nr_bytes;
2323 bio_iovec(bio)->bv_len -= nr_bytes;
2326 req->__data_len -= total_bytes;
2327 req->buffer = bio_data(req->bio);
2329 /* update sector only for requests with clear definition of sector */
2330 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2331 req->__sector += total_bytes >> 9;
2333 /* mixed attributes always follow the first bio */
2334 if (req->cmd_flags & REQ_MIXED_MERGE) {
2335 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2336 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2340 * If total number of sectors is less than the first segment
2341 * size, something has gone terribly wrong.
2343 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2344 blk_dump_rq_flags(req, "request botched");
2345 req->__data_len = blk_rq_cur_bytes(req);
2348 /* recalculate the number of segments */
2349 blk_recalc_rq_segments(req);
2353 EXPORT_SYMBOL_GPL(blk_update_request);
2355 static bool blk_update_bidi_request(struct request *rq, int error,
2356 unsigned int nr_bytes,
2357 unsigned int bidi_bytes)
2359 if (blk_update_request(rq, error, nr_bytes))
2362 /* Bidi request must be completed as a whole */
2363 if (unlikely(blk_bidi_rq(rq)) &&
2364 blk_update_request(rq->next_rq, error, bidi_bytes))
2367 if (blk_queue_add_random(rq->q))
2368 add_disk_randomness(rq->rq_disk);
2374 * blk_unprep_request - unprepare a request
2377 * This function makes a request ready for complete resubmission (or
2378 * completion). It happens only after all error handling is complete,
2379 * so represents the appropriate moment to deallocate any resources
2380 * that were allocated to the request in the prep_rq_fn. The queue
2381 * lock is held when calling this.
2383 void blk_unprep_request(struct request *req)
2385 struct request_queue *q = req->q;
2387 req->cmd_flags &= ~REQ_DONTPREP;
2388 if (q->unprep_rq_fn)
2389 q->unprep_rq_fn(q, req);
2391 EXPORT_SYMBOL_GPL(blk_unprep_request);
2394 * queue lock must be held
2396 static void blk_finish_request(struct request *req, int error)
2398 if (blk_rq_tagged(req))
2399 blk_queue_end_tag(req->q, req);
2401 BUG_ON(blk_queued_rq(req));
2403 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2404 laptop_io_completion(&req->q->backing_dev_info);
2406 blk_delete_timer(req);
2408 if (req->cmd_flags & REQ_DONTPREP)
2409 blk_unprep_request(req);
2412 blk_account_io_done(req);
2415 req->end_io(req, error);
2417 if (blk_bidi_rq(req))
2418 __blk_put_request(req->next_rq->q, req->next_rq);
2420 __blk_put_request(req->q, req);
2425 * blk_end_bidi_request - Complete a bidi request
2426 * @rq: the request to complete
2427 * @error: %0 for success, < %0 for error
2428 * @nr_bytes: number of bytes to complete @rq
2429 * @bidi_bytes: number of bytes to complete @rq->next_rq
2432 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2433 * Drivers that supports bidi can safely call this member for any
2434 * type of request, bidi or uni. In the later case @bidi_bytes is
2438 * %false - we are done with this request
2439 * %true - still buffers pending for this request
2441 static bool blk_end_bidi_request(struct request *rq, int error,
2442 unsigned int nr_bytes, unsigned int bidi_bytes)
2444 struct request_queue *q = rq->q;
2445 unsigned long flags;
2447 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2450 spin_lock_irqsave(q->queue_lock, flags);
2451 blk_finish_request(rq, error);
2452 spin_unlock_irqrestore(q->queue_lock, flags);
2458 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2459 * @rq: the request to complete
2460 * @error: %0 for success, < %0 for error
2461 * @nr_bytes: number of bytes to complete @rq
2462 * @bidi_bytes: number of bytes to complete @rq->next_rq
2465 * Identical to blk_end_bidi_request() except that queue lock is
2466 * assumed to be locked on entry and remains so on return.
2469 * %false - we are done with this request
2470 * %true - still buffers pending for this request
2472 bool __blk_end_bidi_request(struct request *rq, int error,
2473 unsigned int nr_bytes, unsigned int bidi_bytes)
2475 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2478 blk_finish_request(rq, error);
2484 * blk_end_request - Helper function for drivers to complete the request.
2485 * @rq: the request being processed
2486 * @error: %0 for success, < %0 for error
2487 * @nr_bytes: number of bytes to complete
2490 * Ends I/O on a number of bytes attached to @rq.
2491 * If @rq has leftover, sets it up for the next range of segments.
2494 * %false - we are done with this request
2495 * %true - still buffers pending for this request
2497 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2499 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2501 EXPORT_SYMBOL(blk_end_request);
2504 * blk_end_request_all - Helper function for drives to finish the request.
2505 * @rq: the request to finish
2506 * @error: %0 for success, < %0 for error
2509 * Completely finish @rq.
2511 void blk_end_request_all(struct request *rq, int error)
2514 unsigned int bidi_bytes = 0;
2516 if (unlikely(blk_bidi_rq(rq)))
2517 bidi_bytes = blk_rq_bytes(rq->next_rq);
2519 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2522 EXPORT_SYMBOL(blk_end_request_all);
2525 * blk_end_request_cur - Helper function to finish the current request chunk.
2526 * @rq: the request to finish the current chunk for
2527 * @error: %0 for success, < %0 for error
2530 * Complete the current consecutively mapped chunk from @rq.
2533 * %false - we are done with this request
2534 * %true - still buffers pending for this request
2536 bool blk_end_request_cur(struct request *rq, int error)
2538 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2540 EXPORT_SYMBOL(blk_end_request_cur);
2543 * blk_end_request_err - Finish a request till the next failure boundary.
2544 * @rq: the request to finish till the next failure boundary for
2545 * @error: must be negative errno
2548 * Complete @rq till the next failure boundary.
2551 * %false - we are done with this request
2552 * %true - still buffers pending for this request
2554 bool blk_end_request_err(struct request *rq, int error)
2556 WARN_ON(error >= 0);
2557 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2559 EXPORT_SYMBOL_GPL(blk_end_request_err);
2562 * __blk_end_request - Helper function for drivers to complete the request.
2563 * @rq: the request being processed
2564 * @error: %0 for success, < %0 for error
2565 * @nr_bytes: number of bytes to complete
2568 * Must be called with queue lock held unlike blk_end_request().
2571 * %false - we are done with this request
2572 * %true - still buffers pending for this request
2574 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2576 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2578 EXPORT_SYMBOL(__blk_end_request);
2581 * __blk_end_request_all - Helper function for drives to finish the request.
2582 * @rq: the request to finish
2583 * @error: %0 for success, < %0 for error
2586 * Completely finish @rq. Must be called with queue lock held.
2588 void __blk_end_request_all(struct request *rq, int error)
2591 unsigned int bidi_bytes = 0;
2593 if (unlikely(blk_bidi_rq(rq)))
2594 bidi_bytes = blk_rq_bytes(rq->next_rq);
2596 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2599 EXPORT_SYMBOL(__blk_end_request_all);
2602 * __blk_end_request_cur - Helper function to finish the current request chunk.
2603 * @rq: the request to finish the current chunk for
2604 * @error: %0 for success, < %0 for error
2607 * Complete the current consecutively mapped chunk from @rq. Must
2608 * be called with queue lock held.
2611 * %false - we are done with this request
2612 * %true - still buffers pending for this request
2614 bool __blk_end_request_cur(struct request *rq, int error)
2616 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2618 EXPORT_SYMBOL(__blk_end_request_cur);
2621 * __blk_end_request_err - Finish a request till the next failure boundary.
2622 * @rq: the request to finish till the next failure boundary for
2623 * @error: must be negative errno
2626 * Complete @rq till the next failure boundary. Must be called
2627 * with queue lock held.
2630 * %false - we are done with this request
2631 * %true - still buffers pending for this request
2633 bool __blk_end_request_err(struct request *rq, int error)
2635 WARN_ON(error >= 0);
2636 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2638 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2640 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2643 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2644 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2646 if (bio_has_data(bio)) {
2647 rq->nr_phys_segments = bio_phys_segments(q, bio);
2648 rq->buffer = bio_data(bio);
2650 rq->__data_len = bio->bi_size;
2651 rq->bio = rq->biotail = bio;
2654 rq->rq_disk = bio->bi_bdev->bd_disk;
2657 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2659 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2660 * @rq: the request to be flushed
2663 * Flush all pages in @rq.
2665 void rq_flush_dcache_pages(struct request *rq)
2667 struct req_iterator iter;
2668 struct bio_vec *bvec;
2670 rq_for_each_segment(bvec, rq, iter)
2671 flush_dcache_page(bvec->bv_page);
2673 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2677 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2678 * @q : the queue of the device being checked
2681 * Check if underlying low-level drivers of a device are busy.
2682 * If the drivers want to export their busy state, they must set own
2683 * exporting function using blk_queue_lld_busy() first.
2685 * Basically, this function is used only by request stacking drivers
2686 * to stop dispatching requests to underlying devices when underlying
2687 * devices are busy. This behavior helps more I/O merging on the queue
2688 * of the request stacking driver and prevents I/O throughput regression
2689 * on burst I/O load.
2692 * 0 - Not busy (The request stacking driver should dispatch request)
2693 * 1 - Busy (The request stacking driver should stop dispatching request)
2695 int blk_lld_busy(struct request_queue *q)
2698 return q->lld_busy_fn(q);
2702 EXPORT_SYMBOL_GPL(blk_lld_busy);
2705 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2706 * @rq: the clone request to be cleaned up
2709 * Free all bios in @rq for a cloned request.
2711 void blk_rq_unprep_clone(struct request *rq)
2715 while ((bio = rq->bio) != NULL) {
2716 rq->bio = bio->bi_next;
2721 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2724 * Copy attributes of the original request to the clone request.
2725 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2727 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2729 dst->cpu = src->cpu;
2730 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2731 dst->cmd_type = src->cmd_type;
2732 dst->__sector = blk_rq_pos(src);
2733 dst->__data_len = blk_rq_bytes(src);
2734 dst->nr_phys_segments = src->nr_phys_segments;
2735 dst->ioprio = src->ioprio;
2736 dst->extra_len = src->extra_len;
2740 * blk_rq_prep_clone - Helper function to setup clone request
2741 * @rq: the request to be setup
2742 * @rq_src: original request to be cloned
2743 * @bs: bio_set that bios for clone are allocated from
2744 * @gfp_mask: memory allocation mask for bio
2745 * @bio_ctr: setup function to be called for each clone bio.
2746 * Returns %0 for success, non %0 for failure.
2747 * @data: private data to be passed to @bio_ctr
2750 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2751 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2752 * are not copied, and copying such parts is the caller's responsibility.
2753 * Also, pages which the original bios are pointing to are not copied
2754 * and the cloned bios just point same pages.
2755 * So cloned bios must be completed before original bios, which means
2756 * the caller must complete @rq before @rq_src.
2758 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2759 struct bio_set *bs, gfp_t gfp_mask,
2760 int (*bio_ctr)(struct bio *, struct bio *, void *),
2763 struct bio *bio, *bio_src;
2768 blk_rq_init(NULL, rq);
2770 __rq_for_each_bio(bio_src, rq_src) {
2771 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2775 __bio_clone(bio, bio_src);
2777 if (bio_integrity(bio_src) &&
2778 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2781 if (bio_ctr && bio_ctr(bio, bio_src, data))
2785 rq->biotail->bi_next = bio;
2788 rq->bio = rq->biotail = bio;
2791 __blk_rq_prep_clone(rq, rq_src);
2798 blk_rq_unprep_clone(rq);
2802 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2804 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2806 return queue_work(kblockd_workqueue, work);
2808 EXPORT_SYMBOL(kblockd_schedule_work);
2810 int kblockd_schedule_delayed_work(struct request_queue *q,
2811 struct delayed_work *dwork, unsigned long delay)
2813 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2815 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2817 #define PLUG_MAGIC 0x91827364
2820 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2821 * @plug: The &struct blk_plug that needs to be initialized
2824 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2825 * pending I/O should the task end up blocking between blk_start_plug() and
2826 * blk_finish_plug(). This is important from a performance perspective, but
2827 * also ensures that we don't deadlock. For instance, if the task is blocking
2828 * for a memory allocation, memory reclaim could end up wanting to free a
2829 * page belonging to that request that is currently residing in our private
2830 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2831 * this kind of deadlock.
2833 void blk_start_plug(struct blk_plug *plug)
2835 struct task_struct *tsk = current;
2837 plug->magic = PLUG_MAGIC;
2838 INIT_LIST_HEAD(&plug->list);
2839 INIT_LIST_HEAD(&plug->cb_list);
2840 plug->should_sort = 0;
2843 * If this is a nested plug, don't actually assign it. It will be
2844 * flushed on its own.
2848 * Store ordering should not be needed here, since a potential
2849 * preempt will imply a full memory barrier
2854 EXPORT_SYMBOL(blk_start_plug);
2856 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2858 struct request *rqa = container_of(a, struct request, queuelist);
2859 struct request *rqb = container_of(b, struct request, queuelist);
2861 return !(rqa->q <= rqb->q);
2865 * If 'from_schedule' is true, then postpone the dispatch of requests
2866 * until a safe kblockd context. We due this to avoid accidental big
2867 * additional stack usage in driver dispatch, in places where the originally
2868 * plugger did not intend it.
2870 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2872 __releases(q->queue_lock)
2874 trace_block_unplug(q, depth, !from_schedule);
2877 * Don't mess with dead queue.
2879 if (unlikely(blk_queue_dead(q))) {
2880 spin_unlock(q->queue_lock);
2885 * If we are punting this to kblockd, then we can safely drop
2886 * the queue_lock before waking kblockd (which needs to take
2889 if (from_schedule) {
2890 spin_unlock(q->queue_lock);
2891 blk_run_queue_async(q);
2894 spin_unlock(q->queue_lock);
2899 static void flush_plug_callbacks(struct blk_plug *plug)
2901 LIST_HEAD(callbacks);
2903 if (list_empty(&plug->cb_list))
2906 list_splice_init(&plug->cb_list, &callbacks);
2908 while (!list_empty(&callbacks)) {
2909 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2912 list_del(&cb->list);
2917 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2919 struct request_queue *q;
2920 unsigned long flags;
2925 BUG_ON(plug->magic != PLUG_MAGIC);
2927 flush_plug_callbacks(plug);
2928 if (list_empty(&plug->list))
2931 list_splice_init(&plug->list, &list);
2933 if (plug->should_sort) {
2934 list_sort(NULL, &list, plug_rq_cmp);
2935 plug->should_sort = 0;
2942 * Save and disable interrupts here, to avoid doing it for every
2943 * queue lock we have to take.
2945 local_irq_save(flags);
2946 while (!list_empty(&list)) {
2947 rq = list_entry_rq(list.next);
2948 list_del_init(&rq->queuelist);
2952 * This drops the queue lock
2955 queue_unplugged(q, depth, from_schedule);
2958 spin_lock(q->queue_lock);
2962 * Short-circuit if @q is dead
2964 if (unlikely(blk_queue_dead(q))) {
2965 __blk_end_request_all(rq, -ENODEV);
2970 * rq is already accounted, so use raw insert
2972 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2973 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2975 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2981 * This drops the queue lock
2984 queue_unplugged(q, depth, from_schedule);
2986 local_irq_restore(flags);
2989 void blk_finish_plug(struct blk_plug *plug)
2991 blk_flush_plug_list(plug, false);
2993 if (plug == current->plug)
2994 current->plug = NULL;
2996 EXPORT_SYMBOL(blk_finish_plug);
2998 int __init blk_dev_init(void)
3000 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3001 sizeof(((struct request *)0)->cmd_flags));
3003 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3004 kblockd_workqueue = alloc_workqueue("kblockd",
3005 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3006 if (!kblockd_workqueue)
3007 panic("Failed to create kblockd\n");
3009 request_cachep = kmem_cache_create("blkdev_requests",
3010 sizeof(struct request), 0, SLAB_PANIC, NULL);
3012 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3013 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);