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);
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
45 DEFINE_IDA(blk_queue_ida);
48 * For the allocated request tables
50 static struct kmem_cache *request_cachep;
53 * For queue allocation
55 struct kmem_cache *blk_requestq_cachep;
58 * Controlling structure to kblockd
60 static struct workqueue_struct *kblockd_workqueue;
62 static void drive_stat_acct(struct request *rq, int new_io)
64 struct hd_struct *part;
65 int rw = rq_data_dir(rq);
68 if (!blk_do_io_stat(rq))
71 cpu = part_stat_lock();
75 part_stat_inc(cpu, part, merges[rw]);
77 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
78 if (!hd_struct_try_get(part)) {
80 * The partition is already being removed,
81 * the request will be accounted on the disk only
83 * We take a reference on disk->part0 although that
84 * partition will never be deleted, so we can treat
85 * it as any other partition.
87 part = &rq->rq_disk->part0;
90 part_round_stats(cpu, part);
91 part_inc_in_flight(part, rw);
98 void blk_queue_congestion_threshold(struct request_queue *q)
102 nr = q->nr_requests - (q->nr_requests / 8) + 1;
103 if (nr > q->nr_requests)
105 q->nr_congestion_on = nr;
107 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
110 q->nr_congestion_off = nr;
114 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
117 * Locates the passed device's request queue and returns the address of its
120 * Will return NULL if the request queue cannot be located.
122 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
124 struct backing_dev_info *ret = NULL;
125 struct request_queue *q = bdev_get_queue(bdev);
128 ret = &q->backing_dev_info;
131 EXPORT_SYMBOL(blk_get_backing_dev_info);
133 void blk_rq_init(struct request_queue *q, struct request *rq)
135 memset(rq, 0, sizeof(*rq));
137 INIT_LIST_HEAD(&rq->queuelist);
138 INIT_LIST_HEAD(&rq->timeout_list);
141 rq->__sector = (sector_t) -1;
142 INIT_HLIST_NODE(&rq->hash);
143 RB_CLEAR_NODE(&rq->rb_node);
145 rq->cmd_len = BLK_MAX_CDB;
148 rq->start_time = jiffies;
149 set_start_time_ns(rq);
152 EXPORT_SYMBOL(blk_rq_init);
154 static void req_bio_endio(struct request *rq, struct bio *bio,
155 unsigned int nbytes, int error)
158 clear_bit(BIO_UPTODATE, &bio->bi_flags);
159 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
162 if (unlikely(nbytes > bio->bi_size)) {
163 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
164 __func__, nbytes, bio->bi_size);
165 nbytes = bio->bi_size;
168 if (unlikely(rq->cmd_flags & REQ_QUIET))
169 set_bit(BIO_QUIET, &bio->bi_flags);
171 bio->bi_size -= nbytes;
172 bio->bi_sector += (nbytes >> 9);
174 if (bio_integrity(bio))
175 bio_integrity_advance(bio, nbytes);
177 /* don't actually finish bio if it's part of flush sequence */
178 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
179 bio_endio(bio, error);
182 void blk_dump_rq_flags(struct request *rq, char *msg)
186 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
187 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
190 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
191 (unsigned long long)blk_rq_pos(rq),
192 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
193 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
194 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
196 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
197 printk(KERN_INFO " cdb: ");
198 for (bit = 0; bit < BLK_MAX_CDB; bit++)
199 printk("%02x ", rq->cmd[bit]);
203 EXPORT_SYMBOL(blk_dump_rq_flags);
205 static void blk_delay_work(struct work_struct *work)
207 struct request_queue *q;
209 q = container_of(work, struct request_queue, delay_work.work);
210 spin_lock_irq(q->queue_lock);
212 spin_unlock_irq(q->queue_lock);
216 * blk_delay_queue - restart queueing after defined interval
217 * @q: The &struct request_queue in question
218 * @msecs: Delay in msecs
221 * Sometimes queueing needs to be postponed for a little while, to allow
222 * resources to come back. This function will make sure that queueing is
223 * restarted around the specified time.
225 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
227 queue_delayed_work(kblockd_workqueue, &q->delay_work,
228 msecs_to_jiffies(msecs));
230 EXPORT_SYMBOL(blk_delay_queue);
233 * blk_start_queue - restart a previously stopped queue
234 * @q: The &struct request_queue in question
237 * blk_start_queue() will clear the stop flag on the queue, and call
238 * the request_fn for the queue if it was in a stopped state when
239 * entered. Also see blk_stop_queue(). Queue lock must be held.
241 void blk_start_queue(struct request_queue *q)
243 WARN_ON(!irqs_disabled());
245 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
248 EXPORT_SYMBOL(blk_start_queue);
251 * blk_stop_queue - stop a queue
252 * @q: The &struct request_queue in question
255 * The Linux block layer assumes that a block driver will consume all
256 * entries on the request queue when the request_fn strategy is called.
257 * Often this will not happen, because of hardware limitations (queue
258 * depth settings). If a device driver gets a 'queue full' response,
259 * or if it simply chooses not to queue more I/O at one point, it can
260 * call this function to prevent the request_fn from being called until
261 * the driver has signalled it's ready to go again. This happens by calling
262 * blk_start_queue() to restart queue operations. Queue lock must be held.
264 void blk_stop_queue(struct request_queue *q)
266 cancel_delayed_work(&q->delay_work);
267 queue_flag_set(QUEUE_FLAG_STOPPED, q);
269 EXPORT_SYMBOL(blk_stop_queue);
272 * blk_sync_queue - cancel any pending callbacks on a queue
276 * The block layer may perform asynchronous callback activity
277 * on a queue, such as calling the unplug function after a timeout.
278 * A block device may call blk_sync_queue to ensure that any
279 * such activity is cancelled, thus allowing it to release resources
280 * that the callbacks might use. The caller must already have made sure
281 * that its ->make_request_fn will not re-add plugging prior to calling
284 * This function does not cancel any asynchronous activity arising
285 * out of elevator or throttling code. That would require elevaotor_exit()
286 * and blkcg_exit_queue() to be called with queue lock initialized.
289 void blk_sync_queue(struct request_queue *q)
291 del_timer_sync(&q->timeout);
292 cancel_delayed_work_sync(&q->delay_work);
294 EXPORT_SYMBOL(blk_sync_queue);
297 * __blk_run_queue - run a single device queue
298 * @q: The queue to run
301 * See @blk_run_queue. This variant must be called with the queue lock
302 * held and interrupts disabled.
304 void __blk_run_queue(struct request_queue *q)
306 if (unlikely(blk_queue_stopped(q)))
311 EXPORT_SYMBOL(__blk_run_queue);
314 * blk_run_queue_async - run a single device queue in workqueue context
315 * @q: The queue to run
318 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
321 void blk_run_queue_async(struct request_queue *q)
323 if (likely(!blk_queue_stopped(q)))
324 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
326 EXPORT_SYMBOL(blk_run_queue_async);
329 * blk_run_queue - run a single device queue
330 * @q: The queue to run
333 * Invoke request handling on this queue, if it has pending work to do.
334 * May be used to restart queueing when a request has completed.
336 void blk_run_queue(struct request_queue *q)
340 spin_lock_irqsave(q->queue_lock, flags);
342 spin_unlock_irqrestore(q->queue_lock, flags);
344 EXPORT_SYMBOL(blk_run_queue);
346 void blk_put_queue(struct request_queue *q)
348 kobject_put(&q->kobj);
350 EXPORT_SYMBOL(blk_put_queue);
353 * blk_drain_queue - drain requests from request_queue
355 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
357 * Drain requests from @q. If @drain_all is set, all requests are drained.
358 * If not, only ELVPRIV requests are drained. The caller is responsible
359 * for ensuring that no new requests which need to be drained are queued.
361 void blk_drain_queue(struct request_queue *q, bool drain_all)
368 spin_lock_irq(q->queue_lock);
371 * The caller might be trying to drain @q before its
372 * elevator is initialized.
375 elv_drain_elevator(q);
377 blkcg_drain_queue(q);
380 * This function might be called on a queue which failed
381 * driver init after queue creation or is not yet fully
382 * active yet. Some drivers (e.g. fd and loop) get unhappy
383 * in such cases. Kick queue iff dispatch queue has
384 * something on it and @q has request_fn set.
386 if (!list_empty(&q->queue_head) && q->request_fn)
389 drain |= q->nr_rqs_elvpriv;
392 * Unfortunately, requests are queued at and tracked from
393 * multiple places and there's no single counter which can
394 * be drained. Check all the queues and counters.
397 drain |= !list_empty(&q->queue_head);
398 for (i = 0; i < 2; i++) {
399 drain |= q->nr_rqs[i];
400 drain |= q->in_flight[i];
401 drain |= !list_empty(&q->flush_queue[i]);
405 spin_unlock_irq(q->queue_lock);
413 * With queue marked dead, any woken up waiter will fail the
414 * allocation path, so the wakeup chaining is lost and we're
415 * left with hung waiters. We need to wake up those waiters.
418 struct request_list *rl;
420 spin_lock_irq(q->queue_lock);
422 blk_queue_for_each_rl(rl, q)
423 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
424 wake_up_all(&rl->wait[i]);
426 spin_unlock_irq(q->queue_lock);
431 * blk_queue_bypass_start - enter queue bypass mode
432 * @q: queue of interest
434 * In bypass mode, only the dispatch FIFO queue of @q is used. This
435 * function makes @q enter bypass mode and drains all requests which were
436 * throttled or issued before. On return, it's guaranteed that no request
437 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
438 * inside queue or RCU read lock.
440 void blk_queue_bypass_start(struct request_queue *q)
444 spin_lock_irq(q->queue_lock);
445 drain = !q->bypass_depth++;
446 queue_flag_set(QUEUE_FLAG_BYPASS, q);
447 spin_unlock_irq(q->queue_lock);
450 blk_drain_queue(q, false);
451 /* ensure blk_queue_bypass() is %true inside RCU read lock */
455 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
458 * blk_queue_bypass_end - leave queue bypass mode
459 * @q: queue of interest
461 * Leave bypass mode and restore the normal queueing behavior.
463 void blk_queue_bypass_end(struct request_queue *q)
465 spin_lock_irq(q->queue_lock);
466 if (!--q->bypass_depth)
467 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
468 WARN_ON_ONCE(q->bypass_depth < 0);
469 spin_unlock_irq(q->queue_lock);
471 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
474 * blk_cleanup_queue - shutdown a request queue
475 * @q: request queue to shutdown
477 * Mark @q DEAD, drain all pending requests, destroy and put it. All
478 * future requests will be failed immediately with -ENODEV.
480 void blk_cleanup_queue(struct request_queue *q)
482 spinlock_t *lock = q->queue_lock;
484 /* mark @q DEAD, no new request or merges will be allowed afterwards */
485 mutex_lock(&q->sysfs_lock);
486 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
490 * Dead queue is permanently in bypass mode till released. Note
491 * that, unlike blk_queue_bypass_start(), we aren't performing
492 * synchronize_rcu() after entering bypass mode to avoid the delay
493 * as some drivers create and destroy a lot of queues while
494 * probing. This is still safe because blk_release_queue() will be
495 * called only after the queue refcnt drops to zero and nothing,
496 * RCU or not, would be traversing the queue by then.
499 queue_flag_set(QUEUE_FLAG_BYPASS, q);
501 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
502 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
503 queue_flag_set(QUEUE_FLAG_DEAD, q);
504 spin_unlock_irq(lock);
505 mutex_unlock(&q->sysfs_lock);
507 /* drain all requests queued before DEAD marking */
508 blk_drain_queue(q, true);
510 /* @q won't process any more request, flush async actions */
511 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
515 if (q->queue_lock != &q->__queue_lock)
516 q->queue_lock = &q->__queue_lock;
517 spin_unlock_irq(lock);
519 /* @q is and will stay empty, shutdown and put */
522 EXPORT_SYMBOL(blk_cleanup_queue);
524 int blk_init_rl(struct request_list *rl, struct request_queue *q,
527 if (unlikely(rl->rq_pool))
531 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
532 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
533 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
534 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
536 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
537 mempool_free_slab, request_cachep,
545 void blk_exit_rl(struct request_list *rl)
548 mempool_destroy(rl->rq_pool);
551 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
553 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
555 EXPORT_SYMBOL(blk_alloc_queue);
557 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
559 struct request_queue *q;
562 q = kmem_cache_alloc_node(blk_requestq_cachep,
563 gfp_mask | __GFP_ZERO, node_id);
567 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
571 q->backing_dev_info.ra_pages =
572 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
573 q->backing_dev_info.state = 0;
574 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
575 q->backing_dev_info.name = "block";
578 err = bdi_init(&q->backing_dev_info);
582 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
583 laptop_mode_timer_fn, (unsigned long) q);
584 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
585 INIT_LIST_HEAD(&q->queue_head);
586 INIT_LIST_HEAD(&q->timeout_list);
587 INIT_LIST_HEAD(&q->icq_list);
588 #ifdef CONFIG_BLK_CGROUP
589 INIT_LIST_HEAD(&q->blkg_list);
591 INIT_LIST_HEAD(&q->flush_queue[0]);
592 INIT_LIST_HEAD(&q->flush_queue[1]);
593 INIT_LIST_HEAD(&q->flush_data_in_flight);
594 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
596 kobject_init(&q->kobj, &blk_queue_ktype);
598 mutex_init(&q->sysfs_lock);
599 spin_lock_init(&q->__queue_lock);
602 * By default initialize queue_lock to internal lock and driver can
603 * override it later if need be.
605 q->queue_lock = &q->__queue_lock;
608 * A queue starts its life with bypass turned on to avoid
609 * unnecessary bypass on/off overhead and nasty surprises during
610 * init. The initial bypass will be finished when the queue is
611 * registered by blk_register_queue().
614 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
616 if (blkcg_init_queue(q))
622 ida_simple_remove(&blk_queue_ida, q->id);
624 kmem_cache_free(blk_requestq_cachep, q);
627 EXPORT_SYMBOL(blk_alloc_queue_node);
630 * blk_init_queue - prepare a request queue for use with a block device
631 * @rfn: The function to be called to process requests that have been
632 * placed on the queue.
633 * @lock: Request queue spin lock
636 * If a block device wishes to use the standard request handling procedures,
637 * which sorts requests and coalesces adjacent requests, then it must
638 * call blk_init_queue(). The function @rfn will be called when there
639 * are requests on the queue that need to be processed. If the device
640 * supports plugging, then @rfn may not be called immediately when requests
641 * are available on the queue, but may be called at some time later instead.
642 * Plugged queues are generally unplugged when a buffer belonging to one
643 * of the requests on the queue is needed, or due to memory pressure.
645 * @rfn is not required, or even expected, to remove all requests off the
646 * queue, but only as many as it can handle at a time. If it does leave
647 * requests on the queue, it is responsible for arranging that the requests
648 * get dealt with eventually.
650 * The queue spin lock must be held while manipulating the requests on the
651 * request queue; this lock will be taken also from interrupt context, so irq
652 * disabling is needed for it.
654 * Function returns a pointer to the initialized request queue, or %NULL if
658 * blk_init_queue() must be paired with a blk_cleanup_queue() call
659 * when the block device is deactivated (such as at module unload).
662 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
664 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
666 EXPORT_SYMBOL(blk_init_queue);
668 struct request_queue *
669 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
671 struct request_queue *uninit_q, *q;
673 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
677 q = blk_init_allocated_queue(uninit_q, rfn, lock);
679 blk_cleanup_queue(uninit_q);
683 EXPORT_SYMBOL(blk_init_queue_node);
685 struct request_queue *
686 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
692 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
696 q->prep_rq_fn = NULL;
697 q->unprep_rq_fn = NULL;
698 q->queue_flags |= QUEUE_FLAG_DEFAULT;
700 /* Override internal queue lock with supplied lock pointer */
702 q->queue_lock = lock;
705 * This also sets hw/phys segments, boundary and size
707 blk_queue_make_request(q, blk_queue_bio);
709 q->sg_reserved_size = INT_MAX;
712 if (elevator_init(q, NULL))
716 EXPORT_SYMBOL(blk_init_allocated_queue);
718 bool blk_get_queue(struct request_queue *q)
720 if (likely(!blk_queue_dead(q))) {
727 EXPORT_SYMBOL(blk_get_queue);
729 static inline void blk_free_request(struct request_list *rl, struct request *rq)
731 if (rq->cmd_flags & REQ_ELVPRIV) {
732 elv_put_request(rl->q, rq);
734 put_io_context(rq->elv.icq->ioc);
737 mempool_free(rq, rl->rq_pool);
741 * ioc_batching returns true if the ioc is a valid batching request and
742 * should be given priority access to a request.
744 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
750 * Make sure the process is able to allocate at least 1 request
751 * even if the batch times out, otherwise we could theoretically
754 return ioc->nr_batch_requests == q->nr_batching ||
755 (ioc->nr_batch_requests > 0
756 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
760 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
761 * will cause the process to be a "batcher" on all queues in the system. This
762 * is the behaviour we want though - once it gets a wakeup it should be given
765 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
767 if (!ioc || ioc_batching(q, ioc))
770 ioc->nr_batch_requests = q->nr_batching;
771 ioc->last_waited = jiffies;
774 static void __freed_request(struct request_list *rl, int sync)
776 struct request_queue *q = rl->q;
779 * bdi isn't aware of blkcg yet. As all async IOs end up root
780 * blkcg anyway, just use root blkcg state.
782 if (rl == &q->root_rl &&
783 rl->count[sync] < queue_congestion_off_threshold(q))
784 blk_clear_queue_congested(q, sync);
786 if (rl->count[sync] + 1 <= q->nr_requests) {
787 if (waitqueue_active(&rl->wait[sync]))
788 wake_up(&rl->wait[sync]);
790 blk_clear_rl_full(rl, sync);
795 * A request has just been released. Account for it, update the full and
796 * congestion status, wake up any waiters. Called under q->queue_lock.
798 static void freed_request(struct request_list *rl, unsigned int flags)
800 struct request_queue *q = rl->q;
801 int sync = rw_is_sync(flags);
805 if (flags & REQ_ELVPRIV)
808 __freed_request(rl, sync);
810 if (unlikely(rl->starved[sync ^ 1]))
811 __freed_request(rl, sync ^ 1);
815 * Determine if elevator data should be initialized when allocating the
816 * request associated with @bio.
818 static bool blk_rq_should_init_elevator(struct bio *bio)
824 * Flush requests do not use the elevator so skip initialization.
825 * This allows a request to share the flush and elevator data.
827 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
834 * rq_ioc - determine io_context for request allocation
835 * @bio: request being allocated is for this bio (can be %NULL)
837 * Determine io_context to use for request allocation for @bio. May return
838 * %NULL if %current->io_context doesn't exist.
840 static struct io_context *rq_ioc(struct bio *bio)
842 #ifdef CONFIG_BLK_CGROUP
843 if (bio && bio->bi_ioc)
846 return current->io_context;
850 * __get_request - get a free request
851 * @rl: request list to allocate from
852 * @rw_flags: RW and SYNC flags
853 * @bio: bio to allocate request for (can be %NULL)
854 * @gfp_mask: allocation mask
856 * Get a free request from @q. This function may fail under memory
857 * pressure or if @q is dead.
859 * Must be callled with @q->queue_lock held and,
860 * Returns %NULL on failure, with @q->queue_lock held.
861 * Returns !%NULL on success, with @q->queue_lock *not held*.
863 static struct request *__get_request(struct request_list *rl, int rw_flags,
864 struct bio *bio, gfp_t gfp_mask)
866 struct request_queue *q = rl->q;
868 struct elevator_type *et = q->elevator->type;
869 struct io_context *ioc = rq_ioc(bio);
870 struct io_cq *icq = NULL;
871 const bool is_sync = rw_is_sync(rw_flags) != 0;
874 if (unlikely(blk_queue_dead(q)))
877 may_queue = elv_may_queue(q, rw_flags);
878 if (may_queue == ELV_MQUEUE_NO)
881 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
882 if (rl->count[is_sync]+1 >= q->nr_requests) {
884 * The queue will fill after this allocation, so set
885 * it as full, and mark this process as "batching".
886 * This process will be allowed to complete a batch of
887 * requests, others will be blocked.
889 if (!blk_rl_full(rl, is_sync)) {
890 ioc_set_batching(q, ioc);
891 blk_set_rl_full(rl, is_sync);
893 if (may_queue != ELV_MQUEUE_MUST
894 && !ioc_batching(q, ioc)) {
896 * The queue is full and the allocating
897 * process is not a "batcher", and not
898 * exempted by the IO scheduler
905 * bdi isn't aware of blkcg yet. As all async IOs end up
906 * root blkcg anyway, just use root blkcg state.
908 if (rl == &q->root_rl)
909 blk_set_queue_congested(q, is_sync);
913 * Only allow batching queuers to allocate up to 50% over the defined
914 * limit of requests, otherwise we could have thousands of requests
915 * allocated with any setting of ->nr_requests
917 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
920 q->nr_rqs[is_sync]++;
921 rl->count[is_sync]++;
922 rl->starved[is_sync] = 0;
925 * Decide whether the new request will be managed by elevator. If
926 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
927 * prevent the current elevator from being destroyed until the new
928 * request is freed. This guarantees icq's won't be destroyed and
929 * makes creating new ones safe.
931 * Also, lookup icq while holding queue_lock. If it doesn't exist,
932 * it will be created after releasing queue_lock.
934 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
935 rw_flags |= REQ_ELVPRIV;
937 if (et->icq_cache && ioc)
938 icq = ioc_lookup_icq(ioc, q);
941 if (blk_queue_io_stat(q))
942 rw_flags |= REQ_IO_STAT;
943 spin_unlock_irq(q->queue_lock);
945 /* allocate and init request */
946 rq = mempool_alloc(rl->rq_pool, gfp_mask);
951 blk_rq_set_rl(rq, rl);
952 rq->cmd_flags = rw_flags | REQ_ALLOCED;
955 if (rw_flags & REQ_ELVPRIV) {
956 if (unlikely(et->icq_cache && !icq)) {
958 icq = ioc_create_icq(ioc, q, gfp_mask);
964 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
967 /* @rq->elv.icq holds io_context until @rq is freed */
969 get_io_context(icq->ioc);
973 * ioc may be NULL here, and ioc_batching will be false. That's
974 * OK, if the queue is under the request limit then requests need
975 * not count toward the nr_batch_requests limit. There will always
976 * be some limit enforced by BLK_BATCH_TIME.
978 if (ioc_batching(q, ioc))
979 ioc->nr_batch_requests--;
981 trace_block_getrq(q, bio, rw_flags & 1);
986 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
987 * and may fail indefinitely under memory pressure and thus
988 * shouldn't stall IO. Treat this request as !elvpriv. This will
989 * disturb iosched and blkcg but weird is bettern than dead.
991 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
992 dev_name(q->backing_dev_info.dev));
994 rq->cmd_flags &= ~REQ_ELVPRIV;
997 spin_lock_irq(q->queue_lock);
999 spin_unlock_irq(q->queue_lock);
1004 * Allocation failed presumably due to memory. Undo anything we
1005 * might have messed up.
1007 * Allocating task should really be put onto the front of the wait
1008 * queue, but this is pretty rare.
1010 spin_lock_irq(q->queue_lock);
1011 freed_request(rl, rw_flags);
1014 * in the very unlikely event that allocation failed and no
1015 * requests for this direction was pending, mark us starved so that
1016 * freeing of a request in the other direction will notice
1017 * us. another possible fix would be to split the rq mempool into
1021 if (unlikely(rl->count[is_sync] == 0))
1022 rl->starved[is_sync] = 1;
1027 * get_request - get a free request
1028 * @q: request_queue to allocate request from
1029 * @rw_flags: RW and SYNC flags
1030 * @bio: bio to allocate request for (can be %NULL)
1031 * @gfp_mask: allocation mask
1033 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1034 * function keeps retrying under memory pressure and fails iff @q is dead.
1036 * Must be callled with @q->queue_lock held and,
1037 * Returns %NULL on failure, with @q->queue_lock held.
1038 * Returns !%NULL on success, with @q->queue_lock *not held*.
1040 static struct request *get_request(struct request_queue *q, int rw_flags,
1041 struct bio *bio, gfp_t gfp_mask)
1043 const bool is_sync = rw_is_sync(rw_flags) != 0;
1045 struct request_list *rl;
1048 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1050 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1054 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dead(q))) {
1059 /* wait on @rl and retry */
1060 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1061 TASK_UNINTERRUPTIBLE);
1063 trace_block_sleeprq(q, bio, rw_flags & 1);
1065 spin_unlock_irq(q->queue_lock);
1069 * After sleeping, we become a "batching" process and will be able
1070 * to allocate at least one request, and up to a big batch of them
1071 * for a small period time. See ioc_batching, ioc_set_batching
1073 ioc_set_batching(q, current->io_context);
1075 spin_lock_irq(q->queue_lock);
1076 finish_wait(&rl->wait[is_sync], &wait);
1081 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1085 BUG_ON(rw != READ && rw != WRITE);
1087 /* create ioc upfront */
1088 create_io_context(gfp_mask, q->node);
1090 spin_lock_irq(q->queue_lock);
1091 rq = get_request(q, rw, NULL, gfp_mask);
1093 spin_unlock_irq(q->queue_lock);
1094 /* q->queue_lock is unlocked at this point */
1098 EXPORT_SYMBOL(blk_get_request);
1101 * blk_make_request - given a bio, allocate a corresponding struct request.
1102 * @q: target request queue
1103 * @bio: The bio describing the memory mappings that will be submitted for IO.
1104 * It may be a chained-bio properly constructed by block/bio layer.
1105 * @gfp_mask: gfp flags to be used for memory allocation
1107 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1108 * type commands. Where the struct request needs to be farther initialized by
1109 * the caller. It is passed a &struct bio, which describes the memory info of
1112 * The caller of blk_make_request must make sure that bi_io_vec
1113 * are set to describe the memory buffers. That bio_data_dir() will return
1114 * the needed direction of the request. (And all bio's in the passed bio-chain
1115 * are properly set accordingly)
1117 * If called under none-sleepable conditions, mapped bio buffers must not
1118 * need bouncing, by calling the appropriate masked or flagged allocator,
1119 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1122 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1123 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1124 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1125 * completion of a bio that hasn't been submitted yet, thus resulting in a
1126 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1127 * of bio_alloc(), as that avoids the mempool deadlock.
1128 * If possible a big IO should be split into smaller parts when allocation
1129 * fails. Partial allocation should not be an error, or you risk a live-lock.
1131 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1134 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1137 return ERR_PTR(-ENOMEM);
1140 struct bio *bounce_bio = bio;
1143 blk_queue_bounce(q, &bounce_bio);
1144 ret = blk_rq_append_bio(q, rq, bounce_bio);
1145 if (unlikely(ret)) {
1146 blk_put_request(rq);
1147 return ERR_PTR(ret);
1153 EXPORT_SYMBOL(blk_make_request);
1156 * blk_requeue_request - put a request back on queue
1157 * @q: request queue where request should be inserted
1158 * @rq: request to be inserted
1161 * Drivers often keep queueing requests until the hardware cannot accept
1162 * more, when that condition happens we need to put the request back
1163 * on the queue. Must be called with queue lock held.
1165 void blk_requeue_request(struct request_queue *q, struct request *rq)
1167 blk_delete_timer(rq);
1168 blk_clear_rq_complete(rq);
1169 trace_block_rq_requeue(q, rq);
1171 if (blk_rq_tagged(rq))
1172 blk_queue_end_tag(q, rq);
1174 BUG_ON(blk_queued_rq(rq));
1176 elv_requeue_request(q, rq);
1178 EXPORT_SYMBOL(blk_requeue_request);
1180 static void add_acct_request(struct request_queue *q, struct request *rq,
1183 drive_stat_acct(rq, 1);
1184 __elv_add_request(q, rq, where);
1187 static void part_round_stats_single(int cpu, struct hd_struct *part,
1190 if (now == part->stamp)
1193 if (part_in_flight(part)) {
1194 __part_stat_add(cpu, part, time_in_queue,
1195 part_in_flight(part) * (now - part->stamp));
1196 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1202 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1203 * @cpu: cpu number for stats access
1204 * @part: target partition
1206 * The average IO queue length and utilisation statistics are maintained
1207 * by observing the current state of the queue length and the amount of
1208 * time it has been in this state for.
1210 * Normally, that accounting is done on IO completion, but that can result
1211 * in more than a second's worth of IO being accounted for within any one
1212 * second, leading to >100% utilisation. To deal with that, we call this
1213 * function to do a round-off before returning the results when reading
1214 * /proc/diskstats. This accounts immediately for all queue usage up to
1215 * the current jiffies and restarts the counters again.
1217 void part_round_stats(int cpu, struct hd_struct *part)
1219 unsigned long now = jiffies;
1222 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1223 part_round_stats_single(cpu, part, now);
1225 EXPORT_SYMBOL_GPL(part_round_stats);
1228 * queue lock must be held
1230 void __blk_put_request(struct request_queue *q, struct request *req)
1234 if (unlikely(--req->ref_count))
1237 elv_completed_request(q, req);
1239 /* this is a bio leak */
1240 WARN_ON(req->bio != NULL);
1243 * Request may not have originated from ll_rw_blk. if not,
1244 * it didn't come out of our reserved rq pools
1246 if (req->cmd_flags & REQ_ALLOCED) {
1247 unsigned int flags = req->cmd_flags;
1248 struct request_list *rl = blk_rq_rl(req);
1250 BUG_ON(!list_empty(&req->queuelist));
1251 BUG_ON(!hlist_unhashed(&req->hash));
1253 blk_free_request(rl, req);
1254 freed_request(rl, flags);
1258 EXPORT_SYMBOL_GPL(__blk_put_request);
1260 void blk_put_request(struct request *req)
1262 unsigned long flags;
1263 struct request_queue *q = req->q;
1265 spin_lock_irqsave(q->queue_lock, flags);
1266 __blk_put_request(q, req);
1267 spin_unlock_irqrestore(q->queue_lock, flags);
1269 EXPORT_SYMBOL(blk_put_request);
1272 * blk_add_request_payload - add a payload to a request
1273 * @rq: request to update
1274 * @page: page backing the payload
1275 * @len: length of the payload.
1277 * This allows to later add a payload to an already submitted request by
1278 * a block driver. The driver needs to take care of freeing the payload
1281 * Note that this is a quite horrible hack and nothing but handling of
1282 * discard requests should ever use it.
1284 void blk_add_request_payload(struct request *rq, struct page *page,
1287 struct bio *bio = rq->bio;
1289 bio->bi_io_vec->bv_page = page;
1290 bio->bi_io_vec->bv_offset = 0;
1291 bio->bi_io_vec->bv_len = len;
1295 bio->bi_phys_segments = 1;
1297 rq->__data_len = rq->resid_len = len;
1298 rq->nr_phys_segments = 1;
1299 rq->buffer = bio_data(bio);
1301 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1303 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1306 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1308 if (!ll_back_merge_fn(q, req, bio))
1311 trace_block_bio_backmerge(q, bio);
1313 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1314 blk_rq_set_mixed_merge(req);
1316 req->biotail->bi_next = bio;
1318 req->__data_len += bio->bi_size;
1319 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1321 drive_stat_acct(req, 0);
1325 static bool bio_attempt_front_merge(struct request_queue *q,
1326 struct request *req, struct bio *bio)
1328 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1330 if (!ll_front_merge_fn(q, req, bio))
1333 trace_block_bio_frontmerge(q, bio);
1335 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1336 blk_rq_set_mixed_merge(req);
1338 bio->bi_next = req->bio;
1342 * may not be valid. if the low level driver said
1343 * it didn't need a bounce buffer then it better
1344 * not touch req->buffer either...
1346 req->buffer = bio_data(bio);
1347 req->__sector = bio->bi_sector;
1348 req->__data_len += bio->bi_size;
1349 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1351 drive_stat_acct(req, 0);
1356 * attempt_plug_merge - try to merge with %current's plugged list
1357 * @q: request_queue new bio is being queued at
1358 * @bio: new bio being queued
1359 * @request_count: out parameter for number of traversed plugged requests
1361 * Determine whether @bio being queued on @q can be merged with a request
1362 * on %current's plugged list. Returns %true if merge was successful,
1365 * Plugging coalesces IOs from the same issuer for the same purpose without
1366 * going through @q->queue_lock. As such it's more of an issuing mechanism
1367 * than scheduling, and the request, while may have elvpriv data, is not
1368 * added on the elevator at this point. In addition, we don't have
1369 * reliable access to the elevator outside queue lock. Only check basic
1370 * merging parameters without querying the elevator.
1372 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1373 unsigned int *request_count)
1375 struct blk_plug *plug;
1379 plug = current->plug;
1384 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1390 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1393 el_ret = blk_try_merge(rq, bio);
1394 if (el_ret == ELEVATOR_BACK_MERGE) {
1395 ret = bio_attempt_back_merge(q, rq, bio);
1398 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1399 ret = bio_attempt_front_merge(q, rq, bio);
1408 void init_request_from_bio(struct request *req, struct bio *bio)
1410 req->cmd_type = REQ_TYPE_FS;
1412 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1413 if (bio->bi_rw & REQ_RAHEAD)
1414 req->cmd_flags |= REQ_FAILFAST_MASK;
1417 req->__sector = bio->bi_sector;
1418 req->ioprio = bio_prio(bio);
1419 blk_rq_bio_prep(req->q, req, bio);
1422 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1424 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1425 struct blk_plug *plug;
1426 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1427 struct request *req;
1428 unsigned int request_count = 0;
1431 * low level driver can indicate that it wants pages above a
1432 * certain limit bounced to low memory (ie for highmem, or even
1433 * ISA dma in theory)
1435 blk_queue_bounce(q, &bio);
1437 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1438 spin_lock_irq(q->queue_lock);
1439 where = ELEVATOR_INSERT_FLUSH;
1444 * Check if we can merge with the plugged list before grabbing
1447 if (attempt_plug_merge(q, bio, &request_count))
1450 spin_lock_irq(q->queue_lock);
1452 el_ret = elv_merge(q, &req, bio);
1453 if (el_ret == ELEVATOR_BACK_MERGE) {
1454 if (bio_attempt_back_merge(q, req, bio)) {
1455 elv_bio_merged(q, req, bio);
1456 if (!attempt_back_merge(q, req))
1457 elv_merged_request(q, req, el_ret);
1460 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1461 if (bio_attempt_front_merge(q, req, bio)) {
1462 elv_bio_merged(q, req, bio);
1463 if (!attempt_front_merge(q, req))
1464 elv_merged_request(q, req, el_ret);
1471 * This sync check and mask will be re-done in init_request_from_bio(),
1472 * but we need to set it earlier to expose the sync flag to the
1473 * rq allocator and io schedulers.
1475 rw_flags = bio_data_dir(bio);
1477 rw_flags |= REQ_SYNC;
1480 * Grab a free request. This is might sleep but can not fail.
1481 * Returns with the queue unlocked.
1483 req = get_request(q, rw_flags, bio, GFP_NOIO);
1484 if (unlikely(!req)) {
1485 bio_endio(bio, -ENODEV); /* @q is dead */
1490 * After dropping the lock and possibly sleeping here, our request
1491 * may now be mergeable after it had proven unmergeable (above).
1492 * We don't worry about that case for efficiency. It won't happen
1493 * often, and the elevators are able to handle it.
1495 init_request_from_bio(req, bio);
1497 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1498 req->cpu = raw_smp_processor_id();
1500 plug = current->plug;
1503 * If this is the first request added after a plug, fire
1504 * of a plug trace. If others have been added before, check
1505 * if we have multiple devices in this plug. If so, make a
1506 * note to sort the list before dispatch.
1508 if (list_empty(&plug->list))
1509 trace_block_plug(q);
1511 if (!plug->should_sort) {
1512 struct request *__rq;
1514 __rq = list_entry_rq(plug->list.prev);
1516 plug->should_sort = 1;
1518 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1519 blk_flush_plug_list(plug, false);
1520 trace_block_plug(q);
1523 list_add_tail(&req->queuelist, &plug->list);
1524 drive_stat_acct(req, 1);
1526 spin_lock_irq(q->queue_lock);
1527 add_acct_request(q, req, where);
1530 spin_unlock_irq(q->queue_lock);
1533 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1536 * If bio->bi_dev is a partition, remap the location
1538 static inline void blk_partition_remap(struct bio *bio)
1540 struct block_device *bdev = bio->bi_bdev;
1542 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1543 struct hd_struct *p = bdev->bd_part;
1545 bio->bi_sector += p->start_sect;
1546 bio->bi_bdev = bdev->bd_contains;
1548 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1550 bio->bi_sector - p->start_sect);
1554 static void handle_bad_sector(struct bio *bio)
1556 char b[BDEVNAME_SIZE];
1558 printk(KERN_INFO "attempt to access beyond end of device\n");
1559 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1560 bdevname(bio->bi_bdev, b),
1562 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1563 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1565 set_bit(BIO_EOF, &bio->bi_flags);
1568 #ifdef CONFIG_FAIL_MAKE_REQUEST
1570 static DECLARE_FAULT_ATTR(fail_make_request);
1572 static int __init setup_fail_make_request(char *str)
1574 return setup_fault_attr(&fail_make_request, str);
1576 __setup("fail_make_request=", setup_fail_make_request);
1578 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1580 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1583 static int __init fail_make_request_debugfs(void)
1585 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1586 NULL, &fail_make_request);
1588 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1591 late_initcall(fail_make_request_debugfs);
1593 #else /* CONFIG_FAIL_MAKE_REQUEST */
1595 static inline bool should_fail_request(struct hd_struct *part,
1601 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1604 * Check whether this bio extends beyond the end of the device.
1606 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1613 /* Test device or partition size, when known. */
1614 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1616 sector_t sector = bio->bi_sector;
1618 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1620 * This may well happen - the kernel calls bread()
1621 * without checking the size of the device, e.g., when
1622 * mounting a device.
1624 handle_bad_sector(bio);
1632 static noinline_for_stack bool
1633 generic_make_request_checks(struct bio *bio)
1635 struct request_queue *q;
1636 int nr_sectors = bio_sectors(bio);
1638 char b[BDEVNAME_SIZE];
1639 struct hd_struct *part;
1643 if (bio_check_eod(bio, nr_sectors))
1646 q = bdev_get_queue(bio->bi_bdev);
1649 "generic_make_request: Trying to access "
1650 "nonexistent block-device %s (%Lu)\n",
1651 bdevname(bio->bi_bdev, b),
1652 (long long) bio->bi_sector);
1656 if (likely(bio_is_rw(bio) &&
1657 nr_sectors > queue_max_hw_sectors(q))) {
1658 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1659 bdevname(bio->bi_bdev, b),
1661 queue_max_hw_sectors(q));
1665 part = bio->bi_bdev->bd_part;
1666 if (should_fail_request(part, bio->bi_size) ||
1667 should_fail_request(&part_to_disk(part)->part0,
1672 * If this device has partitions, remap block n
1673 * of partition p to block n+start(p) of the disk.
1675 blk_partition_remap(bio);
1677 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1680 if (bio_check_eod(bio, nr_sectors))
1684 * Filter flush bio's early so that make_request based
1685 * drivers without flush support don't have to worry
1688 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1689 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1696 if ((bio->bi_rw & REQ_DISCARD) &&
1697 (!blk_queue_discard(q) ||
1698 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1703 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1709 * Various block parts want %current->io_context and lazy ioc
1710 * allocation ends up trading a lot of pain for a small amount of
1711 * memory. Just allocate it upfront. This may fail and block
1712 * layer knows how to live with it.
1714 create_io_context(GFP_ATOMIC, q->node);
1716 if (blk_throtl_bio(q, bio))
1717 return false; /* throttled, will be resubmitted later */
1719 trace_block_bio_queue(q, bio);
1723 bio_endio(bio, err);
1728 * generic_make_request - hand a buffer to its device driver for I/O
1729 * @bio: The bio describing the location in memory and on the device.
1731 * generic_make_request() is used to make I/O requests of block
1732 * devices. It is passed a &struct bio, which describes the I/O that needs
1735 * generic_make_request() does not return any status. The
1736 * success/failure status of the request, along with notification of
1737 * completion, is delivered asynchronously through the bio->bi_end_io
1738 * function described (one day) else where.
1740 * The caller of generic_make_request must make sure that bi_io_vec
1741 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1742 * set to describe the device address, and the
1743 * bi_end_io and optionally bi_private are set to describe how
1744 * completion notification should be signaled.
1746 * generic_make_request and the drivers it calls may use bi_next if this
1747 * bio happens to be merged with someone else, and may resubmit the bio to
1748 * a lower device by calling into generic_make_request recursively, which
1749 * means the bio should NOT be touched after the call to ->make_request_fn.
1751 void generic_make_request(struct bio *bio)
1753 struct bio_list bio_list_on_stack;
1755 if (!generic_make_request_checks(bio))
1759 * We only want one ->make_request_fn to be active at a time, else
1760 * stack usage with stacked devices could be a problem. So use
1761 * current->bio_list to keep a list of requests submited by a
1762 * make_request_fn function. current->bio_list is also used as a
1763 * flag to say if generic_make_request is currently active in this
1764 * task or not. If it is NULL, then no make_request is active. If
1765 * it is non-NULL, then a make_request is active, and new requests
1766 * should be added at the tail
1768 if (current->bio_list) {
1769 bio_list_add(current->bio_list, bio);
1773 /* following loop may be a bit non-obvious, and so deserves some
1775 * Before entering the loop, bio->bi_next is NULL (as all callers
1776 * ensure that) so we have a list with a single bio.
1777 * We pretend that we have just taken it off a longer list, so
1778 * we assign bio_list to a pointer to the bio_list_on_stack,
1779 * thus initialising the bio_list of new bios to be
1780 * added. ->make_request() may indeed add some more bios
1781 * through a recursive call to generic_make_request. If it
1782 * did, we find a non-NULL value in bio_list and re-enter the loop
1783 * from the top. In this case we really did just take the bio
1784 * of the top of the list (no pretending) and so remove it from
1785 * bio_list, and call into ->make_request() again.
1787 BUG_ON(bio->bi_next);
1788 bio_list_init(&bio_list_on_stack);
1789 current->bio_list = &bio_list_on_stack;
1791 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1793 q->make_request_fn(q, bio);
1795 bio = bio_list_pop(current->bio_list);
1797 current->bio_list = NULL; /* deactivate */
1799 EXPORT_SYMBOL(generic_make_request);
1802 * submit_bio - submit a bio to the block device layer for I/O
1803 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1804 * @bio: The &struct bio which describes the I/O
1806 * submit_bio() is very similar in purpose to generic_make_request(), and
1807 * uses that function to do most of the work. Both are fairly rough
1808 * interfaces; @bio must be presetup and ready for I/O.
1811 void submit_bio(int rw, struct bio *bio)
1816 * If it's a regular read/write or a barrier with data attached,
1817 * go through the normal accounting stuff before submission.
1819 if (bio_has_data(bio)) {
1822 if (unlikely(rw & REQ_WRITE_SAME))
1823 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1825 count = bio_sectors(bio);
1828 count_vm_events(PGPGOUT, count);
1830 task_io_account_read(bio->bi_size);
1831 count_vm_events(PGPGIN, count);
1834 if (unlikely(block_dump)) {
1835 char b[BDEVNAME_SIZE];
1836 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1837 current->comm, task_pid_nr(current),
1838 (rw & WRITE) ? "WRITE" : "READ",
1839 (unsigned long long)bio->bi_sector,
1840 bdevname(bio->bi_bdev, b),
1845 generic_make_request(bio);
1847 EXPORT_SYMBOL(submit_bio);
1850 * blk_rq_check_limits - Helper function to check a request for the queue limit
1852 * @rq: the request being checked
1855 * @rq may have been made based on weaker limitations of upper-level queues
1856 * in request stacking drivers, and it may violate the limitation of @q.
1857 * Since the block layer and the underlying device driver trust @rq
1858 * after it is inserted to @q, it should be checked against @q before
1859 * the insertion using this generic function.
1861 * This function should also be useful for request stacking drivers
1862 * in some cases below, so export this function.
1863 * Request stacking drivers like request-based dm may change the queue
1864 * limits while requests are in the queue (e.g. dm's table swapping).
1865 * Such request stacking drivers should check those requests agaist
1866 * the new queue limits again when they dispatch those requests,
1867 * although such checkings are also done against the old queue limits
1868 * when submitting requests.
1870 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1872 if (!rq_mergeable(rq))
1875 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1876 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1881 * queue's settings related to segment counting like q->bounce_pfn
1882 * may differ from that of other stacking queues.
1883 * Recalculate it to check the request correctly on this queue's
1886 blk_recalc_rq_segments(rq);
1887 if (rq->nr_phys_segments > queue_max_segments(q)) {
1888 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1894 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1897 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1898 * @q: the queue to submit the request
1899 * @rq: the request being queued
1901 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1903 unsigned long flags;
1904 int where = ELEVATOR_INSERT_BACK;
1906 if (blk_rq_check_limits(q, rq))
1910 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1913 spin_lock_irqsave(q->queue_lock, flags);
1914 if (unlikely(blk_queue_dead(q))) {
1915 spin_unlock_irqrestore(q->queue_lock, flags);
1920 * Submitting request must be dequeued before calling this function
1921 * because it will be linked to another request_queue
1923 BUG_ON(blk_queued_rq(rq));
1925 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1926 where = ELEVATOR_INSERT_FLUSH;
1928 add_acct_request(q, rq, where);
1929 if (where == ELEVATOR_INSERT_FLUSH)
1931 spin_unlock_irqrestore(q->queue_lock, flags);
1935 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1938 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1939 * @rq: request to examine
1942 * A request could be merge of IOs which require different failure
1943 * handling. This function determines the number of bytes which
1944 * can be failed from the beginning of the request without
1945 * crossing into area which need to be retried further.
1948 * The number of bytes to fail.
1951 * queue_lock must be held.
1953 unsigned int blk_rq_err_bytes(const struct request *rq)
1955 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1956 unsigned int bytes = 0;
1959 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1960 return blk_rq_bytes(rq);
1963 * Currently the only 'mixing' which can happen is between
1964 * different fastfail types. We can safely fail portions
1965 * which have all the failfast bits that the first one has -
1966 * the ones which are at least as eager to fail as the first
1969 for (bio = rq->bio; bio; bio = bio->bi_next) {
1970 if ((bio->bi_rw & ff) != ff)
1972 bytes += bio->bi_size;
1975 /* this could lead to infinite loop */
1976 BUG_ON(blk_rq_bytes(rq) && !bytes);
1979 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1981 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1983 if (blk_do_io_stat(req)) {
1984 const int rw = rq_data_dir(req);
1985 struct hd_struct *part;
1988 cpu = part_stat_lock();
1990 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1995 static void blk_account_io_done(struct request *req)
1998 * Account IO completion. flush_rq isn't accounted as a
1999 * normal IO on queueing nor completion. Accounting the
2000 * containing request is enough.
2002 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2003 unsigned long duration = jiffies - req->start_time;
2004 const int rw = rq_data_dir(req);
2005 struct hd_struct *part;
2008 cpu = part_stat_lock();
2011 part_stat_inc(cpu, part, ios[rw]);
2012 part_stat_add(cpu, part, ticks[rw], duration);
2013 part_round_stats(cpu, part);
2014 part_dec_in_flight(part, rw);
2016 hd_struct_put(part);
2022 * blk_peek_request - peek at the top of a request queue
2023 * @q: request queue to peek at
2026 * Return the request at the top of @q. The returned request
2027 * should be started using blk_start_request() before LLD starts
2031 * Pointer to the request at the top of @q if available. Null
2035 * queue_lock must be held.
2037 struct request *blk_peek_request(struct request_queue *q)
2042 while ((rq = __elv_next_request(q)) != NULL) {
2043 if (!(rq->cmd_flags & REQ_STARTED)) {
2045 * This is the first time the device driver
2046 * sees this request (possibly after
2047 * requeueing). Notify IO scheduler.
2049 if (rq->cmd_flags & REQ_SORTED)
2050 elv_activate_rq(q, rq);
2053 * just mark as started even if we don't start
2054 * it, a request that has been delayed should
2055 * not be passed by new incoming requests
2057 rq->cmd_flags |= REQ_STARTED;
2058 trace_block_rq_issue(q, rq);
2061 if (!q->boundary_rq || q->boundary_rq == rq) {
2062 q->end_sector = rq_end_sector(rq);
2063 q->boundary_rq = NULL;
2066 if (rq->cmd_flags & REQ_DONTPREP)
2069 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2071 * make sure space for the drain appears we
2072 * know we can do this because max_hw_segments
2073 * has been adjusted to be one fewer than the
2076 rq->nr_phys_segments++;
2082 ret = q->prep_rq_fn(q, rq);
2083 if (ret == BLKPREP_OK) {
2085 } else if (ret == BLKPREP_DEFER) {
2087 * the request may have been (partially) prepped.
2088 * we need to keep this request in the front to
2089 * avoid resource deadlock. REQ_STARTED will
2090 * prevent other fs requests from passing this one.
2092 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2093 !(rq->cmd_flags & REQ_DONTPREP)) {
2095 * remove the space for the drain we added
2096 * so that we don't add it again
2098 --rq->nr_phys_segments;
2103 } else if (ret == BLKPREP_KILL) {
2104 rq->cmd_flags |= REQ_QUIET;
2106 * Mark this request as started so we don't trigger
2107 * any debug logic in the end I/O path.
2109 blk_start_request(rq);
2110 __blk_end_request_all(rq, -EIO);
2112 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2119 EXPORT_SYMBOL(blk_peek_request);
2121 void blk_dequeue_request(struct request *rq)
2123 struct request_queue *q = rq->q;
2125 BUG_ON(list_empty(&rq->queuelist));
2126 BUG_ON(ELV_ON_HASH(rq));
2128 list_del_init(&rq->queuelist);
2131 * the time frame between a request being removed from the lists
2132 * and to it is freed is accounted as io that is in progress at
2135 if (blk_account_rq(rq)) {
2136 q->in_flight[rq_is_sync(rq)]++;
2137 set_io_start_time_ns(rq);
2142 * blk_start_request - start request processing on the driver
2143 * @req: request to dequeue
2146 * Dequeue @req and start timeout timer on it. This hands off the
2147 * request to the driver.
2149 * Block internal functions which don't want to start timer should
2150 * call blk_dequeue_request().
2153 * queue_lock must be held.
2155 void blk_start_request(struct request *req)
2157 blk_dequeue_request(req);
2160 * We are now handing the request to the hardware, initialize
2161 * resid_len to full count and add the timeout handler.
2163 req->resid_len = blk_rq_bytes(req);
2164 if (unlikely(blk_bidi_rq(req)))
2165 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2169 EXPORT_SYMBOL(blk_start_request);
2172 * blk_fetch_request - fetch a request from a request queue
2173 * @q: request queue to fetch a request from
2176 * Return the request at the top of @q. The request is started on
2177 * return and LLD can start processing it immediately.
2180 * Pointer to the request at the top of @q if available. Null
2184 * queue_lock must be held.
2186 struct request *blk_fetch_request(struct request_queue *q)
2190 rq = blk_peek_request(q);
2192 blk_start_request(rq);
2195 EXPORT_SYMBOL(blk_fetch_request);
2198 * blk_update_request - Special helper function for request stacking drivers
2199 * @req: the request being processed
2200 * @error: %0 for success, < %0 for error
2201 * @nr_bytes: number of bytes to complete @req
2204 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2205 * the request structure even if @req doesn't have leftover.
2206 * If @req has leftover, sets it up for the next range of segments.
2208 * This special helper function is only for request stacking drivers
2209 * (e.g. request-based dm) so that they can handle partial completion.
2210 * Actual device drivers should use blk_end_request instead.
2212 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2213 * %false return from this function.
2216 * %false - this request doesn't have any more data
2217 * %true - this request has more data
2219 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2221 int total_bytes, bio_nbytes, next_idx = 0;
2227 trace_block_rq_complete(req->q, req);
2230 * For fs requests, rq is just carrier of independent bio's
2231 * and each partial completion should be handled separately.
2232 * Reset per-request error on each partial completion.
2234 * TODO: tj: This is too subtle. It would be better to let
2235 * low level drivers do what they see fit.
2237 if (req->cmd_type == REQ_TYPE_FS)
2240 if (error && req->cmd_type == REQ_TYPE_FS &&
2241 !(req->cmd_flags & REQ_QUIET)) {
2246 error_type = "recoverable transport";
2249 error_type = "critical target";
2252 error_type = "critical nexus";
2259 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2260 error_type, req->rq_disk ?
2261 req->rq_disk->disk_name : "?",
2262 (unsigned long long)blk_rq_pos(req));
2266 blk_account_io_completion(req, nr_bytes);
2268 total_bytes = bio_nbytes = 0;
2269 while ((bio = req->bio) != NULL) {
2272 if (nr_bytes >= bio->bi_size) {
2273 req->bio = bio->bi_next;
2274 nbytes = bio->bi_size;
2275 req_bio_endio(req, bio, nbytes, error);
2279 int idx = bio->bi_idx + next_idx;
2281 if (unlikely(idx >= bio->bi_vcnt)) {
2282 blk_dump_rq_flags(req, "__end_that");
2283 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2284 __func__, idx, bio->bi_vcnt);
2288 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2289 BIO_BUG_ON(nbytes > bio->bi_size);
2292 * not a complete bvec done
2294 if (unlikely(nbytes > nr_bytes)) {
2295 bio_nbytes += nr_bytes;
2296 total_bytes += nr_bytes;
2301 * advance to the next vector
2304 bio_nbytes += nbytes;
2307 total_bytes += nbytes;
2313 * end more in this run, or just return 'not-done'
2315 if (unlikely(nr_bytes <= 0))
2325 * Reset counters so that the request stacking driver
2326 * can find how many bytes remain in the request
2329 req->__data_len = 0;
2334 * if the request wasn't completed, update state
2337 req_bio_endio(req, bio, bio_nbytes, error);
2338 bio->bi_idx += next_idx;
2339 bio_iovec(bio)->bv_offset += nr_bytes;
2340 bio_iovec(bio)->bv_len -= nr_bytes;
2343 req->__data_len -= total_bytes;
2344 req->buffer = bio_data(req->bio);
2346 /* update sector only for requests with clear definition of sector */
2347 if (req->cmd_type == REQ_TYPE_FS)
2348 req->__sector += total_bytes >> 9;
2350 /* mixed attributes always follow the first bio */
2351 if (req->cmd_flags & REQ_MIXED_MERGE) {
2352 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2353 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2357 * If total number of sectors is less than the first segment
2358 * size, something has gone terribly wrong.
2360 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2361 blk_dump_rq_flags(req, "request botched");
2362 req->__data_len = blk_rq_cur_bytes(req);
2365 /* recalculate the number of segments */
2366 blk_recalc_rq_segments(req);
2370 EXPORT_SYMBOL_GPL(blk_update_request);
2372 static bool blk_update_bidi_request(struct request *rq, int error,
2373 unsigned int nr_bytes,
2374 unsigned int bidi_bytes)
2376 if (blk_update_request(rq, error, nr_bytes))
2379 /* Bidi request must be completed as a whole */
2380 if (unlikely(blk_bidi_rq(rq)) &&
2381 blk_update_request(rq->next_rq, error, bidi_bytes))
2384 if (blk_queue_add_random(rq->q))
2385 add_disk_randomness(rq->rq_disk);
2391 * blk_unprep_request - unprepare a request
2394 * This function makes a request ready for complete resubmission (or
2395 * completion). It happens only after all error handling is complete,
2396 * so represents the appropriate moment to deallocate any resources
2397 * that were allocated to the request in the prep_rq_fn. The queue
2398 * lock is held when calling this.
2400 void blk_unprep_request(struct request *req)
2402 struct request_queue *q = req->q;
2404 req->cmd_flags &= ~REQ_DONTPREP;
2405 if (q->unprep_rq_fn)
2406 q->unprep_rq_fn(q, req);
2408 EXPORT_SYMBOL_GPL(blk_unprep_request);
2411 * queue lock must be held
2413 static void blk_finish_request(struct request *req, int error)
2415 if (blk_rq_tagged(req))
2416 blk_queue_end_tag(req->q, req);
2418 BUG_ON(blk_queued_rq(req));
2420 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2421 laptop_io_completion(&req->q->backing_dev_info);
2423 blk_delete_timer(req);
2425 if (req->cmd_flags & REQ_DONTPREP)
2426 blk_unprep_request(req);
2429 blk_account_io_done(req);
2432 req->end_io(req, error);
2434 if (blk_bidi_rq(req))
2435 __blk_put_request(req->next_rq->q, req->next_rq);
2437 __blk_put_request(req->q, req);
2442 * blk_end_bidi_request - Complete a bidi request
2443 * @rq: the request to complete
2444 * @error: %0 for success, < %0 for error
2445 * @nr_bytes: number of bytes to complete @rq
2446 * @bidi_bytes: number of bytes to complete @rq->next_rq
2449 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2450 * Drivers that supports bidi can safely call this member for any
2451 * type of request, bidi or uni. In the later case @bidi_bytes is
2455 * %false - we are done with this request
2456 * %true - still buffers pending for this request
2458 static bool blk_end_bidi_request(struct request *rq, int error,
2459 unsigned int nr_bytes, unsigned int bidi_bytes)
2461 struct request_queue *q = rq->q;
2462 unsigned long flags;
2464 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2467 spin_lock_irqsave(q->queue_lock, flags);
2468 blk_finish_request(rq, error);
2469 spin_unlock_irqrestore(q->queue_lock, flags);
2475 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2476 * @rq: the request to complete
2477 * @error: %0 for success, < %0 for error
2478 * @nr_bytes: number of bytes to complete @rq
2479 * @bidi_bytes: number of bytes to complete @rq->next_rq
2482 * Identical to blk_end_bidi_request() except that queue lock is
2483 * assumed to be locked on entry and remains so on return.
2486 * %false - we are done with this request
2487 * %true - still buffers pending for this request
2489 bool __blk_end_bidi_request(struct request *rq, int error,
2490 unsigned int nr_bytes, unsigned int bidi_bytes)
2492 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2495 blk_finish_request(rq, error);
2501 * blk_end_request - Helper function for drivers to complete the request.
2502 * @rq: the request being processed
2503 * @error: %0 for success, < %0 for error
2504 * @nr_bytes: number of bytes to complete
2507 * Ends I/O on a number of bytes attached to @rq.
2508 * If @rq has leftover, sets it up for the next range of segments.
2511 * %false - we are done with this request
2512 * %true - still buffers pending for this request
2514 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2516 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2518 EXPORT_SYMBOL(blk_end_request);
2521 * blk_end_request_all - Helper function for drives to finish the request.
2522 * @rq: the request to finish
2523 * @error: %0 for success, < %0 for error
2526 * Completely finish @rq.
2528 void blk_end_request_all(struct request *rq, int error)
2531 unsigned int bidi_bytes = 0;
2533 if (unlikely(blk_bidi_rq(rq)))
2534 bidi_bytes = blk_rq_bytes(rq->next_rq);
2536 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2539 EXPORT_SYMBOL(blk_end_request_all);
2542 * blk_end_request_cur - Helper function to finish the current request chunk.
2543 * @rq: the request to finish the current chunk for
2544 * @error: %0 for success, < %0 for error
2547 * Complete the current consecutively mapped chunk from @rq.
2550 * %false - we are done with this request
2551 * %true - still buffers pending for this request
2553 bool blk_end_request_cur(struct request *rq, int error)
2555 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2557 EXPORT_SYMBOL(blk_end_request_cur);
2560 * blk_end_request_err - Finish a request till the next failure boundary.
2561 * @rq: the request to finish till the next failure boundary for
2562 * @error: must be negative errno
2565 * Complete @rq till the next failure boundary.
2568 * %false - we are done with this request
2569 * %true - still buffers pending for this request
2571 bool blk_end_request_err(struct request *rq, int error)
2573 WARN_ON(error >= 0);
2574 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2576 EXPORT_SYMBOL_GPL(blk_end_request_err);
2579 * __blk_end_request - Helper function for drivers to complete the request.
2580 * @rq: the request being processed
2581 * @error: %0 for success, < %0 for error
2582 * @nr_bytes: number of bytes to complete
2585 * Must be called with queue lock held unlike blk_end_request().
2588 * %false - we are done with this request
2589 * %true - still buffers pending for this request
2591 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2593 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2595 EXPORT_SYMBOL(__blk_end_request);
2598 * __blk_end_request_all - Helper function for drives to finish the request.
2599 * @rq: the request to finish
2600 * @error: %0 for success, < %0 for error
2603 * Completely finish @rq. Must be called with queue lock held.
2605 void __blk_end_request_all(struct request *rq, int error)
2608 unsigned int bidi_bytes = 0;
2610 if (unlikely(blk_bidi_rq(rq)))
2611 bidi_bytes = blk_rq_bytes(rq->next_rq);
2613 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2616 EXPORT_SYMBOL(__blk_end_request_all);
2619 * __blk_end_request_cur - Helper function to finish the current request chunk.
2620 * @rq: the request to finish the current chunk for
2621 * @error: %0 for success, < %0 for error
2624 * Complete the current consecutively mapped chunk from @rq. Must
2625 * be called with queue lock held.
2628 * %false - we are done with this request
2629 * %true - still buffers pending for this request
2631 bool __blk_end_request_cur(struct request *rq, int error)
2633 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2635 EXPORT_SYMBOL(__blk_end_request_cur);
2638 * __blk_end_request_err - Finish a request till the next failure boundary.
2639 * @rq: the request to finish till the next failure boundary for
2640 * @error: must be negative errno
2643 * Complete @rq till the next failure boundary. Must be called
2644 * with queue lock held.
2647 * %false - we are done with this request
2648 * %true - still buffers pending for this request
2650 bool __blk_end_request_err(struct request *rq, int error)
2652 WARN_ON(error >= 0);
2653 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2655 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2657 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2660 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2661 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2663 if (bio_has_data(bio)) {
2664 rq->nr_phys_segments = bio_phys_segments(q, bio);
2665 rq->buffer = bio_data(bio);
2667 rq->__data_len = bio->bi_size;
2668 rq->bio = rq->biotail = bio;
2671 rq->rq_disk = bio->bi_bdev->bd_disk;
2674 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2676 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2677 * @rq: the request to be flushed
2680 * Flush all pages in @rq.
2682 void rq_flush_dcache_pages(struct request *rq)
2684 struct req_iterator iter;
2685 struct bio_vec *bvec;
2687 rq_for_each_segment(bvec, rq, iter)
2688 flush_dcache_page(bvec->bv_page);
2690 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2694 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2695 * @q : the queue of the device being checked
2698 * Check if underlying low-level drivers of a device are busy.
2699 * If the drivers want to export their busy state, they must set own
2700 * exporting function using blk_queue_lld_busy() first.
2702 * Basically, this function is used only by request stacking drivers
2703 * to stop dispatching requests to underlying devices when underlying
2704 * devices are busy. This behavior helps more I/O merging on the queue
2705 * of the request stacking driver and prevents I/O throughput regression
2706 * on burst I/O load.
2709 * 0 - Not busy (The request stacking driver should dispatch request)
2710 * 1 - Busy (The request stacking driver should stop dispatching request)
2712 int blk_lld_busy(struct request_queue *q)
2715 return q->lld_busy_fn(q);
2719 EXPORT_SYMBOL_GPL(blk_lld_busy);
2722 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2723 * @rq: the clone request to be cleaned up
2726 * Free all bios in @rq for a cloned request.
2728 void blk_rq_unprep_clone(struct request *rq)
2732 while ((bio = rq->bio) != NULL) {
2733 rq->bio = bio->bi_next;
2738 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2741 * Copy attributes of the original request to the clone request.
2742 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2744 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2746 dst->cpu = src->cpu;
2747 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2748 dst->cmd_type = src->cmd_type;
2749 dst->__sector = blk_rq_pos(src);
2750 dst->__data_len = blk_rq_bytes(src);
2751 dst->nr_phys_segments = src->nr_phys_segments;
2752 dst->ioprio = src->ioprio;
2753 dst->extra_len = src->extra_len;
2757 * blk_rq_prep_clone - Helper function to setup clone request
2758 * @rq: the request to be setup
2759 * @rq_src: original request to be cloned
2760 * @bs: bio_set that bios for clone are allocated from
2761 * @gfp_mask: memory allocation mask for bio
2762 * @bio_ctr: setup function to be called for each clone bio.
2763 * Returns %0 for success, non %0 for failure.
2764 * @data: private data to be passed to @bio_ctr
2767 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2768 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2769 * are not copied, and copying such parts is the caller's responsibility.
2770 * Also, pages which the original bios are pointing to are not copied
2771 * and the cloned bios just point same pages.
2772 * So cloned bios must be completed before original bios, which means
2773 * the caller must complete @rq before @rq_src.
2775 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2776 struct bio_set *bs, gfp_t gfp_mask,
2777 int (*bio_ctr)(struct bio *, struct bio *, void *),
2780 struct bio *bio, *bio_src;
2785 blk_rq_init(NULL, rq);
2787 __rq_for_each_bio(bio_src, rq_src) {
2788 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2792 if (bio_ctr && bio_ctr(bio, bio_src, data))
2796 rq->biotail->bi_next = bio;
2799 rq->bio = rq->biotail = bio;
2802 __blk_rq_prep_clone(rq, rq_src);
2809 blk_rq_unprep_clone(rq);
2813 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2815 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2817 return queue_work(kblockd_workqueue, work);
2819 EXPORT_SYMBOL(kblockd_schedule_work);
2821 int kblockd_schedule_delayed_work(struct request_queue *q,
2822 struct delayed_work *dwork, unsigned long delay)
2824 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2826 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2828 #define PLUG_MAGIC 0x91827364
2831 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2832 * @plug: The &struct blk_plug that needs to be initialized
2835 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2836 * pending I/O should the task end up blocking between blk_start_plug() and
2837 * blk_finish_plug(). This is important from a performance perspective, but
2838 * also ensures that we don't deadlock. For instance, if the task is blocking
2839 * for a memory allocation, memory reclaim could end up wanting to free a
2840 * page belonging to that request that is currently residing in our private
2841 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2842 * this kind of deadlock.
2844 void blk_start_plug(struct blk_plug *plug)
2846 struct task_struct *tsk = current;
2848 plug->magic = PLUG_MAGIC;
2849 INIT_LIST_HEAD(&plug->list);
2850 INIT_LIST_HEAD(&plug->cb_list);
2851 plug->should_sort = 0;
2854 * If this is a nested plug, don't actually assign it. It will be
2855 * flushed on its own.
2859 * Store ordering should not be needed here, since a potential
2860 * preempt will imply a full memory barrier
2865 EXPORT_SYMBOL(blk_start_plug);
2867 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2869 struct request *rqa = container_of(a, struct request, queuelist);
2870 struct request *rqb = container_of(b, struct request, queuelist);
2872 return !(rqa->q < rqb->q ||
2873 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2877 * If 'from_schedule' is true, then postpone the dispatch of requests
2878 * until a safe kblockd context. We due this to avoid accidental big
2879 * additional stack usage in driver dispatch, in places where the originally
2880 * plugger did not intend it.
2882 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2884 __releases(q->queue_lock)
2886 trace_block_unplug(q, depth, !from_schedule);
2889 * Don't mess with dead queue.
2891 if (unlikely(blk_queue_dead(q))) {
2892 spin_unlock(q->queue_lock);
2897 * If we are punting this to kblockd, then we can safely drop
2898 * the queue_lock before waking kblockd (which needs to take
2901 if (from_schedule) {
2902 spin_unlock(q->queue_lock);
2903 blk_run_queue_async(q);
2906 spin_unlock(q->queue_lock);
2911 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2913 LIST_HEAD(callbacks);
2915 while (!list_empty(&plug->cb_list)) {
2916 list_splice_init(&plug->cb_list, &callbacks);
2918 while (!list_empty(&callbacks)) {
2919 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2922 list_del(&cb->list);
2923 cb->callback(cb, from_schedule);
2928 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2931 struct blk_plug *plug = current->plug;
2932 struct blk_plug_cb *cb;
2937 list_for_each_entry(cb, &plug->cb_list, list)
2938 if (cb->callback == unplug && cb->data == data)
2941 /* Not currently on the callback list */
2942 BUG_ON(size < sizeof(*cb));
2943 cb = kzalloc(size, GFP_ATOMIC);
2946 cb->callback = unplug;
2947 list_add(&cb->list, &plug->cb_list);
2951 EXPORT_SYMBOL(blk_check_plugged);
2953 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2955 struct request_queue *q;
2956 unsigned long flags;
2961 BUG_ON(plug->magic != PLUG_MAGIC);
2963 flush_plug_callbacks(plug, from_schedule);
2964 if (list_empty(&plug->list))
2967 list_splice_init(&plug->list, &list);
2969 if (plug->should_sort) {
2970 list_sort(NULL, &list, plug_rq_cmp);
2971 plug->should_sort = 0;
2978 * Save and disable interrupts here, to avoid doing it for every
2979 * queue lock we have to take.
2981 local_irq_save(flags);
2982 while (!list_empty(&list)) {
2983 rq = list_entry_rq(list.next);
2984 list_del_init(&rq->queuelist);
2988 * This drops the queue lock
2991 queue_unplugged(q, depth, from_schedule);
2994 spin_lock(q->queue_lock);
2998 * Short-circuit if @q is dead
3000 if (unlikely(blk_queue_dead(q))) {
3001 __blk_end_request_all(rq, -ENODEV);
3006 * rq is already accounted, so use raw insert
3008 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3009 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3011 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3017 * This drops the queue lock
3020 queue_unplugged(q, depth, from_schedule);
3022 local_irq_restore(flags);
3025 void blk_finish_plug(struct blk_plug *plug)
3027 blk_flush_plug_list(plug, false);
3029 if (plug == current->plug)
3030 current->plug = NULL;
3032 EXPORT_SYMBOL(blk_finish_plug);
3034 int __init blk_dev_init(void)
3036 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3037 sizeof(((struct request *)0)->cmd_flags));
3039 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3040 kblockd_workqueue = alloc_workqueue("kblockd",
3041 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3042 if (!kblockd_workqueue)
3043 panic("Failed to create kblockd\n");
3045 request_cachep = kmem_cache_create("blkdev_requests",
3046 sizeof(struct request), 0, SLAB_PANIC, NULL);
3048 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3049 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);