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/blk-mq.h>
20 #include <linux/highmem.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/block.h>
43 #include "blk-mq-sched.h"
46 #ifdef CONFIG_DEBUG_FS
47 struct dentry
*blk_debugfs_root
;
50 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
51 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
56 DEFINE_IDA(blk_queue_ida
);
59 * For the allocated request tables
61 struct kmem_cache
*request_cachep
;
64 * For queue allocation
66 struct kmem_cache
*blk_requestq_cachep
;
69 * Controlling structure to kblockd
71 static struct workqueue_struct
*kblockd_workqueue
;
73 static void blk_clear_congested(struct request_list
*rl
, int sync
)
75 #ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
79 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
80 * flip its congestion state for events on other blkcgs.
82 if (rl
== &rl
->q
->root_rl
)
83 clear_wb_congested(rl
->q
->backing_dev_info
->wb
.congested
, sync
);
87 static void blk_set_congested(struct request_list
*rl
, int sync
)
89 #ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
92 /* see blk_clear_congested() */
93 if (rl
== &rl
->q
->root_rl
)
94 set_wb_congested(rl
->q
->backing_dev_info
->wb
.congested
, sync
);
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
;
113 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
115 memset(rq
, 0, sizeof(*rq
));
117 INIT_LIST_HEAD(&rq
->queuelist
);
118 INIT_LIST_HEAD(&rq
->timeout_list
);
121 rq
->__sector
= (sector_t
) -1;
122 INIT_HLIST_NODE(&rq
->hash
);
123 RB_CLEAR_NODE(&rq
->rb_node
);
125 rq
->internal_tag
= -1;
126 rq
->start_time
= jiffies
;
127 set_start_time_ns(rq
);
130 EXPORT_SYMBOL(blk_rq_init
);
132 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
133 unsigned int nbytes
, int error
)
136 bio
->bi_error
= error
;
138 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
139 bio_set_flag(bio
, BIO_QUIET
);
141 bio_advance(bio
, nbytes
);
143 /* don't actually finish bio if it's part of flush sequence */
144 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
148 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
150 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
151 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
152 (unsigned long long) rq
->cmd_flags
);
154 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
155 (unsigned long long)blk_rq_pos(rq
),
156 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
157 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
158 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
160 EXPORT_SYMBOL(blk_dump_rq_flags
);
162 static void blk_delay_work(struct work_struct
*work
)
164 struct request_queue
*q
;
166 q
= container_of(work
, struct request_queue
, delay_work
.work
);
167 spin_lock_irq(q
->queue_lock
);
169 spin_unlock_irq(q
->queue_lock
);
173 * blk_delay_queue - restart queueing after defined interval
174 * @q: The &struct request_queue in question
175 * @msecs: Delay in msecs
178 * Sometimes queueing needs to be postponed for a little while, to allow
179 * resources to come back. This function will make sure that queueing is
180 * restarted around the specified time. Queue lock must be held.
182 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
184 if (likely(!blk_queue_dead(q
)))
185 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
186 msecs_to_jiffies(msecs
));
188 EXPORT_SYMBOL(blk_delay_queue
);
191 * blk_start_queue_async - asynchronously restart a previously stopped queue
192 * @q: The &struct request_queue in question
195 * blk_start_queue_async() will clear the stop flag on the queue, and
196 * ensure that the request_fn for the queue is run from an async
199 void blk_start_queue_async(struct request_queue
*q
)
201 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
202 blk_run_queue_async(q
);
204 EXPORT_SYMBOL(blk_start_queue_async
);
207 * blk_start_queue - restart a previously stopped queue
208 * @q: The &struct request_queue in question
211 * blk_start_queue() will clear the stop flag on the queue, and call
212 * the request_fn for the queue if it was in a stopped state when
213 * entered. Also see blk_stop_queue(). Queue lock must be held.
215 void blk_start_queue(struct request_queue
*q
)
217 WARN_ON(!irqs_disabled());
219 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
222 EXPORT_SYMBOL(blk_start_queue
);
225 * blk_stop_queue - stop a queue
226 * @q: The &struct request_queue in question
229 * The Linux block layer assumes that a block driver will consume all
230 * entries on the request queue when the request_fn strategy is called.
231 * Often this will not happen, because of hardware limitations (queue
232 * depth settings). If a device driver gets a 'queue full' response,
233 * or if it simply chooses not to queue more I/O at one point, it can
234 * call this function to prevent the request_fn from being called until
235 * the driver has signalled it's ready to go again. This happens by calling
236 * blk_start_queue() to restart queue operations. Queue lock must be held.
238 void blk_stop_queue(struct request_queue
*q
)
240 cancel_delayed_work(&q
->delay_work
);
241 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
243 EXPORT_SYMBOL(blk_stop_queue
);
246 * blk_sync_queue - cancel any pending callbacks on a queue
250 * The block layer may perform asynchronous callback activity
251 * on a queue, such as calling the unplug function after a timeout.
252 * A block device may call blk_sync_queue to ensure that any
253 * such activity is cancelled, thus allowing it to release resources
254 * that the callbacks might use. The caller must already have made sure
255 * that its ->make_request_fn will not re-add plugging prior to calling
258 * This function does not cancel any asynchronous activity arising
259 * out of elevator or throttling code. That would require elevator_exit()
260 * and blkcg_exit_queue() to be called with queue lock initialized.
263 void blk_sync_queue(struct request_queue
*q
)
265 del_timer_sync(&q
->timeout
);
268 struct blk_mq_hw_ctx
*hctx
;
271 queue_for_each_hw_ctx(q
, hctx
, i
) {
272 cancel_work_sync(&hctx
->run_work
);
273 cancel_delayed_work_sync(&hctx
->delay_work
);
276 cancel_delayed_work_sync(&q
->delay_work
);
279 EXPORT_SYMBOL(blk_sync_queue
);
282 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
283 * @q: The queue to run
286 * Invoke request handling on a queue if there are any pending requests.
287 * May be used to restart request handling after a request has completed.
288 * This variant runs the queue whether or not the queue has been
289 * stopped. Must be called with the queue lock held and interrupts
290 * disabled. See also @blk_run_queue.
292 inline void __blk_run_queue_uncond(struct request_queue
*q
)
294 if (unlikely(blk_queue_dead(q
)))
298 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
299 * the queue lock internally. As a result multiple threads may be
300 * running such a request function concurrently. Keep track of the
301 * number of active request_fn invocations such that blk_drain_queue()
302 * can wait until all these request_fn calls have finished.
304 q
->request_fn_active
++;
306 q
->request_fn_active
--;
308 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
311 * __blk_run_queue - run a single device queue
312 * @q: The queue to run
315 * See @blk_run_queue. This variant must be called with the queue lock
316 * held and interrupts disabled.
318 void __blk_run_queue(struct request_queue
*q
)
320 if (unlikely(blk_queue_stopped(q
)))
323 __blk_run_queue_uncond(q
);
325 EXPORT_SYMBOL(__blk_run_queue
);
328 * blk_run_queue_async - run a single device queue in workqueue context
329 * @q: The queue to run
332 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
333 * of us. The caller must hold the queue lock.
335 void blk_run_queue_async(struct request_queue
*q
)
337 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
338 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
340 EXPORT_SYMBOL(blk_run_queue_async
);
343 * blk_run_queue - run a single device queue
344 * @q: The queue to run
347 * Invoke request handling on this queue, if it has pending work to do.
348 * May be used to restart queueing when a request has completed.
350 void blk_run_queue(struct request_queue
*q
)
354 spin_lock_irqsave(q
->queue_lock
, flags
);
356 spin_unlock_irqrestore(q
->queue_lock
, flags
);
358 EXPORT_SYMBOL(blk_run_queue
);
360 void blk_put_queue(struct request_queue
*q
)
362 kobject_put(&q
->kobj
);
364 EXPORT_SYMBOL(blk_put_queue
);
367 * __blk_drain_queue - drain requests from request_queue
369 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
371 * Drain requests from @q. If @drain_all is set, all requests are drained.
372 * If not, only ELVPRIV requests are drained. The caller is responsible
373 * for ensuring that no new requests which need to be drained are queued.
375 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
376 __releases(q
->queue_lock
)
377 __acquires(q
->queue_lock
)
381 lockdep_assert_held(q
->queue_lock
);
387 * The caller might be trying to drain @q before its
388 * elevator is initialized.
391 elv_drain_elevator(q
);
393 blkcg_drain_queue(q
);
396 * This function might be called on a queue which failed
397 * driver init after queue creation or is not yet fully
398 * active yet. Some drivers (e.g. fd and loop) get unhappy
399 * in such cases. Kick queue iff dispatch queue has
400 * something on it and @q has request_fn set.
402 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
405 drain
|= q
->nr_rqs_elvpriv
;
406 drain
|= q
->request_fn_active
;
409 * Unfortunately, requests are queued at and tracked from
410 * multiple places and there's no single counter which can
411 * be drained. Check all the queues and counters.
414 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
415 drain
|= !list_empty(&q
->queue_head
);
416 for (i
= 0; i
< 2; i
++) {
417 drain
|= q
->nr_rqs
[i
];
418 drain
|= q
->in_flight
[i
];
420 drain
|= !list_empty(&fq
->flush_queue
[i
]);
427 spin_unlock_irq(q
->queue_lock
);
431 spin_lock_irq(q
->queue_lock
);
435 * With queue marked dead, any woken up waiter will fail the
436 * allocation path, so the wakeup chaining is lost and we're
437 * left with hung waiters. We need to wake up those waiters.
440 struct request_list
*rl
;
442 blk_queue_for_each_rl(rl
, q
)
443 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
444 wake_up_all(&rl
->wait
[i
]);
449 * blk_queue_bypass_start - enter queue bypass mode
450 * @q: queue of interest
452 * In bypass mode, only the dispatch FIFO queue of @q is used. This
453 * function makes @q enter bypass mode and drains all requests which were
454 * throttled or issued before. On return, it's guaranteed that no request
455 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
456 * inside queue or RCU read lock.
458 void blk_queue_bypass_start(struct request_queue
*q
)
460 spin_lock_irq(q
->queue_lock
);
462 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
463 spin_unlock_irq(q
->queue_lock
);
466 * Queues start drained. Skip actual draining till init is
467 * complete. This avoids lenghty delays during queue init which
468 * can happen many times during boot.
470 if (blk_queue_init_done(q
)) {
471 spin_lock_irq(q
->queue_lock
);
472 __blk_drain_queue(q
, false);
473 spin_unlock_irq(q
->queue_lock
);
475 /* ensure blk_queue_bypass() is %true inside RCU read lock */
479 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
482 * blk_queue_bypass_end - leave queue bypass mode
483 * @q: queue of interest
485 * Leave bypass mode and restore the normal queueing behavior.
487 void blk_queue_bypass_end(struct request_queue
*q
)
489 spin_lock_irq(q
->queue_lock
);
490 if (!--q
->bypass_depth
)
491 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
492 WARN_ON_ONCE(q
->bypass_depth
< 0);
493 spin_unlock_irq(q
->queue_lock
);
495 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
497 void blk_set_queue_dying(struct request_queue
*q
)
499 spin_lock_irq(q
->queue_lock
);
500 queue_flag_set(QUEUE_FLAG_DYING
, q
);
501 spin_unlock_irq(q
->queue_lock
);
504 blk_mq_wake_waiters(q
);
506 struct request_list
*rl
;
508 spin_lock_irq(q
->queue_lock
);
509 blk_queue_for_each_rl(rl
, q
) {
511 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
512 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
515 spin_unlock_irq(q
->queue_lock
);
518 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
521 * blk_cleanup_queue - shutdown a request queue
522 * @q: request queue to shutdown
524 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
525 * put it. All future requests will be failed immediately with -ENODEV.
527 void blk_cleanup_queue(struct request_queue
*q
)
529 spinlock_t
*lock
= q
->queue_lock
;
531 /* mark @q DYING, no new request or merges will be allowed afterwards */
532 mutex_lock(&q
->sysfs_lock
);
533 blk_set_queue_dying(q
);
537 * A dying queue is permanently in bypass mode till released. Note
538 * that, unlike blk_queue_bypass_start(), we aren't performing
539 * synchronize_rcu() after entering bypass mode to avoid the delay
540 * as some drivers create and destroy a lot of queues while
541 * probing. This is still safe because blk_release_queue() will be
542 * called only after the queue refcnt drops to zero and nothing,
543 * RCU or not, would be traversing the queue by then.
546 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
548 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
549 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
550 queue_flag_set(QUEUE_FLAG_DYING
, q
);
551 spin_unlock_irq(lock
);
552 mutex_unlock(&q
->sysfs_lock
);
555 * Drain all requests queued before DYING marking. Set DEAD flag to
556 * prevent that q->request_fn() gets invoked after draining finished.
561 __blk_drain_queue(q
, true);
562 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
563 spin_unlock_irq(lock
);
565 /* for synchronous bio-based driver finish in-flight integrity i/o */
566 blk_flush_integrity();
568 /* @q won't process any more request, flush async actions */
569 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
573 blk_mq_free_queue(q
);
574 percpu_ref_exit(&q
->q_usage_counter
);
577 if (q
->queue_lock
!= &q
->__queue_lock
)
578 q
->queue_lock
= &q
->__queue_lock
;
579 spin_unlock_irq(lock
);
581 /* @q is and will stay empty, shutdown and put */
584 EXPORT_SYMBOL(blk_cleanup_queue
);
586 /* Allocate memory local to the request queue */
587 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
589 struct request_queue
*q
= data
;
591 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
594 static void free_request_simple(void *element
, void *data
)
596 kmem_cache_free(request_cachep
, element
);
599 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
601 struct request_queue
*q
= data
;
604 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
606 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
613 static void free_request_size(void *element
, void *data
)
615 struct request_queue
*q
= data
;
618 q
->exit_rq_fn(q
, element
);
622 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
625 if (unlikely(rl
->rq_pool
))
629 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
630 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
631 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
632 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
635 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
636 alloc_request_size
, free_request_size
,
637 q
, gfp_mask
, q
->node
);
639 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
640 alloc_request_simple
, free_request_simple
,
641 q
, gfp_mask
, q
->node
);
649 void blk_exit_rl(struct request_list
*rl
)
652 mempool_destroy(rl
->rq_pool
);
655 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
657 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
659 EXPORT_SYMBOL(blk_alloc_queue
);
661 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
666 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
672 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
673 !atomic_read(&q
->mq_freeze_depth
) ||
675 if (blk_queue_dying(q
))
682 void blk_queue_exit(struct request_queue
*q
)
684 percpu_ref_put(&q
->q_usage_counter
);
687 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
689 struct request_queue
*q
=
690 container_of(ref
, struct request_queue
, q_usage_counter
);
692 wake_up_all(&q
->mq_freeze_wq
);
695 static void blk_rq_timed_out_timer(unsigned long data
)
697 struct request_queue
*q
= (struct request_queue
*)data
;
699 kblockd_schedule_work(&q
->timeout_work
);
702 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
704 struct request_queue
*q
;
706 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
707 gfp_mask
| __GFP_ZERO
, node_id
);
711 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
715 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
719 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
720 if (!q
->backing_dev_info
)
723 q
->backing_dev_info
->ra_pages
=
724 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
725 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
726 q
->backing_dev_info
->name
= "block";
729 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
730 laptop_mode_timer_fn
, (unsigned long) q
);
731 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
732 INIT_LIST_HEAD(&q
->queue_head
);
733 INIT_LIST_HEAD(&q
->timeout_list
);
734 INIT_LIST_HEAD(&q
->icq_list
);
735 #ifdef CONFIG_BLK_CGROUP
736 INIT_LIST_HEAD(&q
->blkg_list
);
738 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
740 kobject_init(&q
->kobj
, &blk_queue_ktype
);
742 mutex_init(&q
->sysfs_lock
);
743 spin_lock_init(&q
->__queue_lock
);
746 * By default initialize queue_lock to internal lock and driver can
747 * override it later if need be.
749 q
->queue_lock
= &q
->__queue_lock
;
752 * A queue starts its life with bypass turned on to avoid
753 * unnecessary bypass on/off overhead and nasty surprises during
754 * init. The initial bypass will be finished when the queue is
755 * registered by blk_register_queue().
758 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
760 init_waitqueue_head(&q
->mq_freeze_wq
);
763 * Init percpu_ref in atomic mode so that it's faster to shutdown.
764 * See blk_register_queue() for details.
766 if (percpu_ref_init(&q
->q_usage_counter
,
767 blk_queue_usage_counter_release
,
768 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
771 if (blkcg_init_queue(q
))
777 percpu_ref_exit(&q
->q_usage_counter
);
779 bdi_put(q
->backing_dev_info
);
781 bioset_free(q
->bio_split
);
783 ida_simple_remove(&blk_queue_ida
, q
->id
);
785 kmem_cache_free(blk_requestq_cachep
, q
);
788 EXPORT_SYMBOL(blk_alloc_queue_node
);
791 * blk_init_queue - prepare a request queue for use with a block device
792 * @rfn: The function to be called to process requests that have been
793 * placed on the queue.
794 * @lock: Request queue spin lock
797 * If a block device wishes to use the standard request handling procedures,
798 * which sorts requests and coalesces adjacent requests, then it must
799 * call blk_init_queue(). The function @rfn will be called when there
800 * are requests on the queue that need to be processed. If the device
801 * supports plugging, then @rfn may not be called immediately when requests
802 * are available on the queue, but may be called at some time later instead.
803 * Plugged queues are generally unplugged when a buffer belonging to one
804 * of the requests on the queue is needed, or due to memory pressure.
806 * @rfn is not required, or even expected, to remove all requests off the
807 * queue, but only as many as it can handle at a time. If it does leave
808 * requests on the queue, it is responsible for arranging that the requests
809 * get dealt with eventually.
811 * The queue spin lock must be held while manipulating the requests on the
812 * request queue; this lock will be taken also from interrupt context, so irq
813 * disabling is needed for it.
815 * Function returns a pointer to the initialized request queue, or %NULL if
819 * blk_init_queue() must be paired with a blk_cleanup_queue() call
820 * when the block device is deactivated (such as at module unload).
823 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
825 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
827 EXPORT_SYMBOL(blk_init_queue
);
829 struct request_queue
*
830 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
832 struct request_queue
*q
;
834 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
840 q
->queue_lock
= lock
;
841 if (blk_init_allocated_queue(q
) < 0) {
842 blk_cleanup_queue(q
);
848 EXPORT_SYMBOL(blk_init_queue_node
);
850 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
853 int blk_init_allocated_queue(struct request_queue
*q
)
855 q
->stats
= blk_alloc_queue_stats();
859 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
863 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
864 goto out_free_flush_queue
;
866 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
867 goto out_exit_flush_rq
;
869 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
870 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
873 * This also sets hw/phys segments, boundary and size
875 blk_queue_make_request(q
, blk_queue_bio
);
877 q
->sg_reserved_size
= INT_MAX
;
879 /* Protect q->elevator from elevator_change */
880 mutex_lock(&q
->sysfs_lock
);
883 if (elevator_init(q
, NULL
)) {
884 mutex_unlock(&q
->sysfs_lock
);
885 goto out_exit_flush_rq
;
888 mutex_unlock(&q
->sysfs_lock
);
893 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
894 out_free_flush_queue
:
895 blk_free_flush_queue(q
->fq
);
898 EXPORT_SYMBOL(blk_init_allocated_queue
);
900 bool blk_get_queue(struct request_queue
*q
)
902 if (likely(!blk_queue_dying(q
))) {
909 EXPORT_SYMBOL(blk_get_queue
);
911 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
913 if (rq
->rq_flags
& RQF_ELVPRIV
) {
914 elv_put_request(rl
->q
, rq
);
916 put_io_context(rq
->elv
.icq
->ioc
);
919 mempool_free(rq
, rl
->rq_pool
);
923 * ioc_batching returns true if the ioc is a valid batching request and
924 * should be given priority access to a request.
926 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
932 * Make sure the process is able to allocate at least 1 request
933 * even if the batch times out, otherwise we could theoretically
936 return ioc
->nr_batch_requests
== q
->nr_batching
||
937 (ioc
->nr_batch_requests
> 0
938 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
942 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
943 * will cause the process to be a "batcher" on all queues in the system. This
944 * is the behaviour we want though - once it gets a wakeup it should be given
947 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
949 if (!ioc
|| ioc_batching(q
, ioc
))
952 ioc
->nr_batch_requests
= q
->nr_batching
;
953 ioc
->last_waited
= jiffies
;
956 static void __freed_request(struct request_list
*rl
, int sync
)
958 struct request_queue
*q
= rl
->q
;
960 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
961 blk_clear_congested(rl
, sync
);
963 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
964 if (waitqueue_active(&rl
->wait
[sync
]))
965 wake_up(&rl
->wait
[sync
]);
967 blk_clear_rl_full(rl
, sync
);
972 * A request has just been released. Account for it, update the full and
973 * congestion status, wake up any waiters. Called under q->queue_lock.
975 static void freed_request(struct request_list
*rl
, bool sync
,
976 req_flags_t rq_flags
)
978 struct request_queue
*q
= rl
->q
;
982 if (rq_flags
& RQF_ELVPRIV
)
985 __freed_request(rl
, sync
);
987 if (unlikely(rl
->starved
[sync
^ 1]))
988 __freed_request(rl
, sync
^ 1);
991 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
993 struct request_list
*rl
;
994 int on_thresh
, off_thresh
;
996 spin_lock_irq(q
->queue_lock
);
998 blk_queue_congestion_threshold(q
);
999 on_thresh
= queue_congestion_on_threshold(q
);
1000 off_thresh
= queue_congestion_off_threshold(q
);
1002 blk_queue_for_each_rl(rl
, q
) {
1003 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1004 blk_set_congested(rl
, BLK_RW_SYNC
);
1005 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1006 blk_clear_congested(rl
, BLK_RW_SYNC
);
1008 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1009 blk_set_congested(rl
, BLK_RW_ASYNC
);
1010 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1011 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1013 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1014 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1016 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1017 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1020 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1021 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1023 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1024 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1028 spin_unlock_irq(q
->queue_lock
);
1033 * __get_request - get a free request
1034 * @rl: request list to allocate from
1035 * @op: operation and flags
1036 * @bio: bio to allocate request for (can be %NULL)
1037 * @gfp_mask: allocation mask
1039 * Get a free request from @q. This function may fail under memory
1040 * pressure or if @q is dead.
1042 * Must be called with @q->queue_lock held and,
1043 * Returns ERR_PTR on failure, with @q->queue_lock held.
1044 * Returns request pointer on success, with @q->queue_lock *not held*.
1046 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1047 struct bio
*bio
, gfp_t gfp_mask
)
1049 struct request_queue
*q
= rl
->q
;
1051 struct elevator_type
*et
= q
->elevator
->type
;
1052 struct io_context
*ioc
= rq_ioc(bio
);
1053 struct io_cq
*icq
= NULL
;
1054 const bool is_sync
= op_is_sync(op
);
1056 req_flags_t rq_flags
= RQF_ALLOCED
;
1058 if (unlikely(blk_queue_dying(q
)))
1059 return ERR_PTR(-ENODEV
);
1061 may_queue
= elv_may_queue(q
, op
);
1062 if (may_queue
== ELV_MQUEUE_NO
)
1065 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1066 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1068 * The queue will fill after this allocation, so set
1069 * it as full, and mark this process as "batching".
1070 * This process will be allowed to complete a batch of
1071 * requests, others will be blocked.
1073 if (!blk_rl_full(rl
, is_sync
)) {
1074 ioc_set_batching(q
, ioc
);
1075 blk_set_rl_full(rl
, is_sync
);
1077 if (may_queue
!= ELV_MQUEUE_MUST
1078 && !ioc_batching(q
, ioc
)) {
1080 * The queue is full and the allocating
1081 * process is not a "batcher", and not
1082 * exempted by the IO scheduler
1084 return ERR_PTR(-ENOMEM
);
1088 blk_set_congested(rl
, is_sync
);
1092 * Only allow batching queuers to allocate up to 50% over the defined
1093 * limit of requests, otherwise we could have thousands of requests
1094 * allocated with any setting of ->nr_requests
1096 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1097 return ERR_PTR(-ENOMEM
);
1099 q
->nr_rqs
[is_sync
]++;
1100 rl
->count
[is_sync
]++;
1101 rl
->starved
[is_sync
] = 0;
1104 * Decide whether the new request will be managed by elevator. If
1105 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1106 * prevent the current elevator from being destroyed until the new
1107 * request is freed. This guarantees icq's won't be destroyed and
1108 * makes creating new ones safe.
1110 * Flush requests do not use the elevator so skip initialization.
1111 * This allows a request to share the flush and elevator data.
1113 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1114 * it will be created after releasing queue_lock.
1116 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1117 rq_flags
|= RQF_ELVPRIV
;
1118 q
->nr_rqs_elvpriv
++;
1119 if (et
->icq_cache
&& ioc
)
1120 icq
= ioc_lookup_icq(ioc
, q
);
1123 if (blk_queue_io_stat(q
))
1124 rq_flags
|= RQF_IO_STAT
;
1125 spin_unlock_irq(q
->queue_lock
);
1127 /* allocate and init request */
1128 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1133 blk_rq_set_rl(rq
, rl
);
1134 blk_rq_set_prio(rq
, ioc
);
1136 rq
->rq_flags
= rq_flags
;
1139 if (rq_flags
& RQF_ELVPRIV
) {
1140 if (unlikely(et
->icq_cache
&& !icq
)) {
1142 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1148 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1151 /* @rq->elv.icq holds io_context until @rq is freed */
1153 get_io_context(icq
->ioc
);
1157 * ioc may be NULL here, and ioc_batching will be false. That's
1158 * OK, if the queue is under the request limit then requests need
1159 * not count toward the nr_batch_requests limit. There will always
1160 * be some limit enforced by BLK_BATCH_TIME.
1162 if (ioc_batching(q
, ioc
))
1163 ioc
->nr_batch_requests
--;
1165 trace_block_getrq(q
, bio
, op
);
1170 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1171 * and may fail indefinitely under memory pressure and thus
1172 * shouldn't stall IO. Treat this request as !elvpriv. This will
1173 * disturb iosched and blkcg but weird is bettern than dead.
1175 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1176 __func__
, dev_name(q
->backing_dev_info
->dev
));
1178 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1181 spin_lock_irq(q
->queue_lock
);
1182 q
->nr_rqs_elvpriv
--;
1183 spin_unlock_irq(q
->queue_lock
);
1188 * Allocation failed presumably due to memory. Undo anything we
1189 * might have messed up.
1191 * Allocating task should really be put onto the front of the wait
1192 * queue, but this is pretty rare.
1194 spin_lock_irq(q
->queue_lock
);
1195 freed_request(rl
, is_sync
, rq_flags
);
1198 * in the very unlikely event that allocation failed and no
1199 * requests for this direction was pending, mark us starved so that
1200 * freeing of a request in the other direction will notice
1201 * us. another possible fix would be to split the rq mempool into
1205 if (unlikely(rl
->count
[is_sync
] == 0))
1206 rl
->starved
[is_sync
] = 1;
1207 return ERR_PTR(-ENOMEM
);
1211 * get_request - get a free request
1212 * @q: request_queue to allocate request from
1213 * @op: operation and flags
1214 * @bio: bio to allocate request for (can be %NULL)
1215 * @gfp_mask: allocation mask
1217 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1218 * this function keeps retrying under memory pressure and fails iff @q is dead.
1220 * Must be called with @q->queue_lock held and,
1221 * Returns ERR_PTR on failure, with @q->queue_lock held.
1222 * Returns request pointer on success, with @q->queue_lock *not held*.
1224 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1225 struct bio
*bio
, gfp_t gfp_mask
)
1227 const bool is_sync
= op_is_sync(op
);
1229 struct request_list
*rl
;
1232 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1234 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1238 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1243 /* wait on @rl and retry */
1244 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1245 TASK_UNINTERRUPTIBLE
);
1247 trace_block_sleeprq(q
, bio
, op
);
1249 spin_unlock_irq(q
->queue_lock
);
1253 * After sleeping, we become a "batching" process and will be able
1254 * to allocate at least one request, and up to a big batch of them
1255 * for a small period time. See ioc_batching, ioc_set_batching
1257 ioc_set_batching(q
, current
->io_context
);
1259 spin_lock_irq(q
->queue_lock
);
1260 finish_wait(&rl
->wait
[is_sync
], &wait
);
1265 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1270 /* create ioc upfront */
1271 create_io_context(gfp_mask
, q
->node
);
1273 spin_lock_irq(q
->queue_lock
);
1274 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1276 spin_unlock_irq(q
->queue_lock
);
1280 /* q->queue_lock is unlocked at this point */
1282 rq
->__sector
= (sector_t
) -1;
1283 rq
->bio
= rq
->biotail
= NULL
;
1287 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1290 return blk_mq_alloc_request(q
, rw
,
1291 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1292 0 : BLK_MQ_REQ_NOWAIT
);
1294 return blk_old_get_request(q
, rw
, gfp_mask
);
1296 EXPORT_SYMBOL(blk_get_request
);
1299 * blk_requeue_request - put a request back on queue
1300 * @q: request queue where request should be inserted
1301 * @rq: request to be inserted
1304 * Drivers often keep queueing requests until the hardware cannot accept
1305 * more, when that condition happens we need to put the request back
1306 * on the queue. Must be called with queue lock held.
1308 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1310 blk_delete_timer(rq
);
1311 blk_clear_rq_complete(rq
);
1312 trace_block_rq_requeue(q
, rq
);
1313 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1315 if (rq
->rq_flags
& RQF_QUEUED
)
1316 blk_queue_end_tag(q
, rq
);
1318 BUG_ON(blk_queued_rq(rq
));
1320 elv_requeue_request(q
, rq
);
1322 EXPORT_SYMBOL(blk_requeue_request
);
1324 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1327 blk_account_io_start(rq
, true);
1328 __elv_add_request(q
, rq
, where
);
1331 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1336 if (now
== part
->stamp
)
1339 inflight
= part_in_flight(part
);
1341 __part_stat_add(cpu
, part
, time_in_queue
,
1342 inflight
* (now
- part
->stamp
));
1343 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1349 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1350 * @cpu: cpu number for stats access
1351 * @part: target partition
1353 * The average IO queue length and utilisation statistics are maintained
1354 * by observing the current state of the queue length and the amount of
1355 * time it has been in this state for.
1357 * Normally, that accounting is done on IO completion, but that can result
1358 * in more than a second's worth of IO being accounted for within any one
1359 * second, leading to >100% utilisation. To deal with that, we call this
1360 * function to do a round-off before returning the results when reading
1361 * /proc/diskstats. This accounts immediately for all queue usage up to
1362 * the current jiffies and restarts the counters again.
1364 void part_round_stats(int cpu
, struct hd_struct
*part
)
1366 unsigned long now
= jiffies
;
1369 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1370 part_round_stats_single(cpu
, part
, now
);
1372 EXPORT_SYMBOL_GPL(part_round_stats
);
1375 static void blk_pm_put_request(struct request
*rq
)
1377 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1378 pm_runtime_mark_last_busy(rq
->q
->dev
);
1381 static inline void blk_pm_put_request(struct request
*rq
) {}
1385 * queue lock must be held
1387 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1389 req_flags_t rq_flags
= req
->rq_flags
;
1395 blk_mq_free_request(req
);
1399 blk_pm_put_request(req
);
1401 elv_completed_request(q
, req
);
1403 /* this is a bio leak */
1404 WARN_ON(req
->bio
!= NULL
);
1406 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1409 * Request may not have originated from ll_rw_blk. if not,
1410 * it didn't come out of our reserved rq pools
1412 if (rq_flags
& RQF_ALLOCED
) {
1413 struct request_list
*rl
= blk_rq_rl(req
);
1414 bool sync
= op_is_sync(req
->cmd_flags
);
1416 BUG_ON(!list_empty(&req
->queuelist
));
1417 BUG_ON(ELV_ON_HASH(req
));
1419 blk_free_request(rl
, req
);
1420 freed_request(rl
, sync
, rq_flags
);
1424 EXPORT_SYMBOL_GPL(__blk_put_request
);
1426 void blk_put_request(struct request
*req
)
1428 struct request_queue
*q
= req
->q
;
1431 blk_mq_free_request(req
);
1433 unsigned long flags
;
1435 spin_lock_irqsave(q
->queue_lock
, flags
);
1436 __blk_put_request(q
, req
);
1437 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1440 EXPORT_SYMBOL(blk_put_request
);
1442 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1445 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1447 if (!ll_back_merge_fn(q
, req
, bio
))
1450 trace_block_bio_backmerge(q
, req
, bio
);
1452 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1453 blk_rq_set_mixed_merge(req
);
1455 req
->biotail
->bi_next
= bio
;
1457 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1458 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1460 blk_account_io_start(req
, false);
1464 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1467 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1469 if (!ll_front_merge_fn(q
, req
, bio
))
1472 trace_block_bio_frontmerge(q
, req
, bio
);
1474 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1475 blk_rq_set_mixed_merge(req
);
1477 bio
->bi_next
= req
->bio
;
1480 req
->__sector
= bio
->bi_iter
.bi_sector
;
1481 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1482 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1484 blk_account_io_start(req
, false);
1488 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1491 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1493 if (segments
>= queue_max_discard_segments(q
))
1495 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1496 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1499 req
->biotail
->bi_next
= bio
;
1501 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1502 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1503 req
->nr_phys_segments
= segments
+ 1;
1505 blk_account_io_start(req
, false);
1508 req_set_nomerge(q
, req
);
1513 * blk_attempt_plug_merge - try to merge with %current's plugged list
1514 * @q: request_queue new bio is being queued at
1515 * @bio: new bio being queued
1516 * @request_count: out parameter for number of traversed plugged requests
1517 * @same_queue_rq: pointer to &struct request that gets filled in when
1518 * another request associated with @q is found on the plug list
1519 * (optional, may be %NULL)
1521 * Determine whether @bio being queued on @q can be merged with a request
1522 * on %current's plugged list. Returns %true if merge was successful,
1525 * Plugging coalesces IOs from the same issuer for the same purpose without
1526 * going through @q->queue_lock. As such it's more of an issuing mechanism
1527 * than scheduling, and the request, while may have elvpriv data, is not
1528 * added on the elevator at this point. In addition, we don't have
1529 * reliable access to the elevator outside queue lock. Only check basic
1530 * merging parameters without querying the elevator.
1532 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1534 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1535 unsigned int *request_count
,
1536 struct request
**same_queue_rq
)
1538 struct blk_plug
*plug
;
1540 struct list_head
*plug_list
;
1542 plug
= current
->plug
;
1548 plug_list
= &plug
->mq_list
;
1550 plug_list
= &plug
->list
;
1552 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1553 bool merged
= false;
1558 * Only blk-mq multiple hardware queues case checks the
1559 * rq in the same queue, there should be only one such
1563 *same_queue_rq
= rq
;
1566 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1569 switch (blk_try_merge(rq
, bio
)) {
1570 case ELEVATOR_BACK_MERGE
:
1571 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1573 case ELEVATOR_FRONT_MERGE
:
1574 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1576 case ELEVATOR_DISCARD_MERGE
:
1577 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1590 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1592 struct blk_plug
*plug
;
1594 struct list_head
*plug_list
;
1595 unsigned int ret
= 0;
1597 plug
= current
->plug
;
1602 plug_list
= &plug
->mq_list
;
1604 plug_list
= &plug
->list
;
1606 list_for_each_entry(rq
, plug_list
, queuelist
) {
1614 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1616 if (bio
->bi_opf
& REQ_RAHEAD
)
1617 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1620 req
->__sector
= bio
->bi_iter
.bi_sector
;
1621 if (ioprio_valid(bio_prio(bio
)))
1622 req
->ioprio
= bio_prio(bio
);
1623 blk_rq_bio_prep(req
->q
, req
, bio
);
1626 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1628 struct blk_plug
*plug
;
1629 int where
= ELEVATOR_INSERT_SORT
;
1630 struct request
*req
, *free
;
1631 unsigned int request_count
= 0;
1632 unsigned int wb_acct
;
1635 * low level driver can indicate that it wants pages above a
1636 * certain limit bounced to low memory (ie for highmem, or even
1637 * ISA dma in theory)
1639 blk_queue_bounce(q
, &bio
);
1641 blk_queue_split(q
, &bio
, q
->bio_split
);
1643 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1644 bio
->bi_error
= -EIO
;
1646 return BLK_QC_T_NONE
;
1649 if (op_is_flush(bio
->bi_opf
)) {
1650 spin_lock_irq(q
->queue_lock
);
1651 where
= ELEVATOR_INSERT_FLUSH
;
1656 * Check if we can merge with the plugged list before grabbing
1659 if (!blk_queue_nomerges(q
)) {
1660 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1661 return BLK_QC_T_NONE
;
1663 request_count
= blk_plug_queued_count(q
);
1665 spin_lock_irq(q
->queue_lock
);
1667 switch (elv_merge(q
, &req
, bio
)) {
1668 case ELEVATOR_BACK_MERGE
:
1669 if (!bio_attempt_back_merge(q
, req
, bio
))
1671 elv_bio_merged(q
, req
, bio
);
1672 free
= attempt_back_merge(q
, req
);
1674 __blk_put_request(q
, free
);
1676 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1678 case ELEVATOR_FRONT_MERGE
:
1679 if (!bio_attempt_front_merge(q
, req
, bio
))
1681 elv_bio_merged(q
, req
, bio
);
1682 free
= attempt_front_merge(q
, req
);
1684 __blk_put_request(q
, free
);
1686 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1693 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1696 * Grab a free request. This is might sleep but can not fail.
1697 * Returns with the queue unlocked.
1699 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1701 __wbt_done(q
->rq_wb
, wb_acct
);
1702 bio
->bi_error
= PTR_ERR(req
);
1707 wbt_track(&req
->issue_stat
, wb_acct
);
1710 * After dropping the lock and possibly sleeping here, our request
1711 * may now be mergeable after it had proven unmergeable (above).
1712 * We don't worry about that case for efficiency. It won't happen
1713 * often, and the elevators are able to handle it.
1715 init_request_from_bio(req
, bio
);
1717 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1718 req
->cpu
= raw_smp_processor_id();
1720 plug
= current
->plug
;
1723 * If this is the first request added after a plug, fire
1726 * @request_count may become stale because of schedule
1727 * out, so check plug list again.
1729 if (!request_count
|| list_empty(&plug
->list
))
1730 trace_block_plug(q
);
1732 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1733 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1734 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1735 blk_flush_plug_list(plug
, false);
1736 trace_block_plug(q
);
1739 list_add_tail(&req
->queuelist
, &plug
->list
);
1740 blk_account_io_start(req
, true);
1742 spin_lock_irq(q
->queue_lock
);
1743 add_acct_request(q
, req
, where
);
1746 spin_unlock_irq(q
->queue_lock
);
1749 return BLK_QC_T_NONE
;
1753 * If bio->bi_dev is a partition, remap the location
1755 static inline void blk_partition_remap(struct bio
*bio
)
1757 struct block_device
*bdev
= bio
->bi_bdev
;
1760 * Zone reset does not include bi_size so bio_sectors() is always 0.
1761 * Include a test for the reset op code and perform the remap if needed.
1763 if (bdev
!= bdev
->bd_contains
&&
1764 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1765 struct hd_struct
*p
= bdev
->bd_part
;
1767 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1768 bio
->bi_bdev
= bdev
->bd_contains
;
1770 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1772 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1776 static void handle_bad_sector(struct bio
*bio
)
1778 char b
[BDEVNAME_SIZE
];
1780 printk(KERN_INFO
"attempt to access beyond end of device\n");
1781 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1782 bdevname(bio
->bi_bdev
, b
),
1784 (unsigned long long)bio_end_sector(bio
),
1785 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1788 #ifdef CONFIG_FAIL_MAKE_REQUEST
1790 static DECLARE_FAULT_ATTR(fail_make_request
);
1792 static int __init
setup_fail_make_request(char *str
)
1794 return setup_fault_attr(&fail_make_request
, str
);
1796 __setup("fail_make_request=", setup_fail_make_request
);
1798 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1800 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1803 static int __init
fail_make_request_debugfs(void)
1805 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1806 NULL
, &fail_make_request
);
1808 return PTR_ERR_OR_ZERO(dir
);
1811 late_initcall(fail_make_request_debugfs
);
1813 #else /* CONFIG_FAIL_MAKE_REQUEST */
1815 static inline bool should_fail_request(struct hd_struct
*part
,
1821 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1824 * Check whether this bio extends beyond the end of the device.
1826 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1833 /* Test device or partition size, when known. */
1834 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1836 sector_t sector
= bio
->bi_iter
.bi_sector
;
1838 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1840 * This may well happen - the kernel calls bread()
1841 * without checking the size of the device, e.g., when
1842 * mounting a device.
1844 handle_bad_sector(bio
);
1852 static noinline_for_stack
bool
1853 generic_make_request_checks(struct bio
*bio
)
1855 struct request_queue
*q
;
1856 int nr_sectors
= bio_sectors(bio
);
1858 char b
[BDEVNAME_SIZE
];
1859 struct hd_struct
*part
;
1863 if (bio_check_eod(bio
, nr_sectors
))
1866 q
= bdev_get_queue(bio
->bi_bdev
);
1869 "generic_make_request: Trying to access "
1870 "nonexistent block-device %s (%Lu)\n",
1871 bdevname(bio
->bi_bdev
, b
),
1872 (long long) bio
->bi_iter
.bi_sector
);
1876 part
= bio
->bi_bdev
->bd_part
;
1877 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1878 should_fail_request(&part_to_disk(part
)->part0
,
1879 bio
->bi_iter
.bi_size
))
1883 * If this device has partitions, remap block n
1884 * of partition p to block n+start(p) of the disk.
1886 blk_partition_remap(bio
);
1888 if (bio_check_eod(bio
, nr_sectors
))
1892 * Filter flush bio's early so that make_request based
1893 * drivers without flush support don't have to worry
1896 if (op_is_flush(bio
->bi_opf
) &&
1897 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1898 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1905 switch (bio_op(bio
)) {
1906 case REQ_OP_DISCARD
:
1907 if (!blk_queue_discard(q
))
1910 case REQ_OP_SECURE_ERASE
:
1911 if (!blk_queue_secure_erase(q
))
1914 case REQ_OP_WRITE_SAME
:
1915 if (!bdev_write_same(bio
->bi_bdev
))
1918 case REQ_OP_ZONE_REPORT
:
1919 case REQ_OP_ZONE_RESET
:
1920 if (!bdev_is_zoned(bio
->bi_bdev
))
1923 case REQ_OP_WRITE_ZEROES
:
1924 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1932 * Various block parts want %current->io_context and lazy ioc
1933 * allocation ends up trading a lot of pain for a small amount of
1934 * memory. Just allocate it upfront. This may fail and block
1935 * layer knows how to live with it.
1937 create_io_context(GFP_ATOMIC
, q
->node
);
1939 if (!blkcg_bio_issue_check(q
, bio
))
1942 trace_block_bio_queue(q
, bio
);
1948 bio
->bi_error
= err
;
1954 * generic_make_request - hand a buffer to its device driver for I/O
1955 * @bio: The bio describing the location in memory and on the device.
1957 * generic_make_request() is used to make I/O requests of block
1958 * devices. It is passed a &struct bio, which describes the I/O that needs
1961 * generic_make_request() does not return any status. The
1962 * success/failure status of the request, along with notification of
1963 * completion, is delivered asynchronously through the bio->bi_end_io
1964 * function described (one day) else where.
1966 * The caller of generic_make_request must make sure that bi_io_vec
1967 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1968 * set to describe the device address, and the
1969 * bi_end_io and optionally bi_private are set to describe how
1970 * completion notification should be signaled.
1972 * generic_make_request and the drivers it calls may use bi_next if this
1973 * bio happens to be merged with someone else, and may resubmit the bio to
1974 * a lower device by calling into generic_make_request recursively, which
1975 * means the bio should NOT be touched after the call to ->make_request_fn.
1977 blk_qc_t
generic_make_request(struct bio
*bio
)
1980 * bio_list_on_stack[0] contains bios submitted by the current
1982 * bio_list_on_stack[1] contains bios that were submitted before
1983 * the current make_request_fn, but that haven't been processed
1986 struct bio_list bio_list_on_stack
[2];
1987 blk_qc_t ret
= BLK_QC_T_NONE
;
1989 if (!generic_make_request_checks(bio
))
1993 * We only want one ->make_request_fn to be active at a time, else
1994 * stack usage with stacked devices could be a problem. So use
1995 * current->bio_list to keep a list of requests submited by a
1996 * make_request_fn function. current->bio_list is also used as a
1997 * flag to say if generic_make_request is currently active in this
1998 * task or not. If it is NULL, then no make_request is active. If
1999 * it is non-NULL, then a make_request is active, and new requests
2000 * should be added at the tail
2002 if (current
->bio_list
) {
2003 bio_list_add(¤t
->bio_list
[0], bio
);
2007 /* following loop may be a bit non-obvious, and so deserves some
2009 * Before entering the loop, bio->bi_next is NULL (as all callers
2010 * ensure that) so we have a list with a single bio.
2011 * We pretend that we have just taken it off a longer list, so
2012 * we assign bio_list to a pointer to the bio_list_on_stack,
2013 * thus initialising the bio_list of new bios to be
2014 * added. ->make_request() may indeed add some more bios
2015 * through a recursive call to generic_make_request. If it
2016 * did, we find a non-NULL value in bio_list and re-enter the loop
2017 * from the top. In this case we really did just take the bio
2018 * of the top of the list (no pretending) and so remove it from
2019 * bio_list, and call into ->make_request() again.
2021 BUG_ON(bio
->bi_next
);
2022 bio_list_init(&bio_list_on_stack
[0]);
2023 current
->bio_list
= bio_list_on_stack
;
2025 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2027 if (likely(blk_queue_enter(q
, false) == 0)) {
2028 struct bio_list lower
, same
;
2030 /* Create a fresh bio_list for all subordinate requests */
2031 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2032 bio_list_init(&bio_list_on_stack
[0]);
2033 ret
= q
->make_request_fn(q
, bio
);
2037 /* sort new bios into those for a lower level
2038 * and those for the same level
2040 bio_list_init(&lower
);
2041 bio_list_init(&same
);
2042 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2043 if (q
== bdev_get_queue(bio
->bi_bdev
))
2044 bio_list_add(&same
, bio
);
2046 bio_list_add(&lower
, bio
);
2047 /* now assemble so we handle the lowest level first */
2048 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2049 bio_list_merge(&bio_list_on_stack
[0], &same
);
2050 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2054 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2056 current
->bio_list
= NULL
; /* deactivate */
2061 EXPORT_SYMBOL(generic_make_request
);
2064 * submit_bio - submit a bio to the block device layer for I/O
2065 * @bio: The &struct bio which describes the I/O
2067 * submit_bio() is very similar in purpose to generic_make_request(), and
2068 * uses that function to do most of the work. Both are fairly rough
2069 * interfaces; @bio must be presetup and ready for I/O.
2072 blk_qc_t
submit_bio(struct bio
*bio
)
2075 * If it's a regular read/write or a barrier with data attached,
2076 * go through the normal accounting stuff before submission.
2078 if (bio_has_data(bio
)) {
2081 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2082 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2084 count
= bio_sectors(bio
);
2086 if (op_is_write(bio_op(bio
))) {
2087 count_vm_events(PGPGOUT
, count
);
2089 task_io_account_read(bio
->bi_iter
.bi_size
);
2090 count_vm_events(PGPGIN
, count
);
2093 if (unlikely(block_dump
)) {
2094 char b
[BDEVNAME_SIZE
];
2095 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2096 current
->comm
, task_pid_nr(current
),
2097 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2098 (unsigned long long)bio
->bi_iter
.bi_sector
,
2099 bdevname(bio
->bi_bdev
, b
),
2104 return generic_make_request(bio
);
2106 EXPORT_SYMBOL(submit_bio
);
2109 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2110 * for new the queue limits
2112 * @rq: the request being checked
2115 * @rq may have been made based on weaker limitations of upper-level queues
2116 * in request stacking drivers, and it may violate the limitation of @q.
2117 * Since the block layer and the underlying device driver trust @rq
2118 * after it is inserted to @q, it should be checked against @q before
2119 * the insertion using this generic function.
2121 * Request stacking drivers like request-based dm may change the queue
2122 * limits when retrying requests on other queues. Those requests need
2123 * to be checked against the new queue limits again during dispatch.
2125 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2128 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2129 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2134 * queue's settings related to segment counting like q->bounce_pfn
2135 * may differ from that of other stacking queues.
2136 * Recalculate it to check the request correctly on this queue's
2139 blk_recalc_rq_segments(rq
);
2140 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2141 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2149 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2150 * @q: the queue to submit the request
2151 * @rq: the request being queued
2153 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2155 unsigned long flags
;
2156 int where
= ELEVATOR_INSERT_BACK
;
2158 if (blk_cloned_rq_check_limits(q
, rq
))
2162 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2166 if (blk_queue_io_stat(q
))
2167 blk_account_io_start(rq
, true);
2168 blk_mq_sched_insert_request(rq
, false, true, false, false);
2172 spin_lock_irqsave(q
->queue_lock
, flags
);
2173 if (unlikely(blk_queue_dying(q
))) {
2174 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2179 * Submitting request must be dequeued before calling this function
2180 * because it will be linked to another request_queue
2182 BUG_ON(blk_queued_rq(rq
));
2184 if (op_is_flush(rq
->cmd_flags
))
2185 where
= ELEVATOR_INSERT_FLUSH
;
2187 add_acct_request(q
, rq
, where
);
2188 if (where
== ELEVATOR_INSERT_FLUSH
)
2190 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2194 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2197 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2198 * @rq: request to examine
2201 * A request could be merge of IOs which require different failure
2202 * handling. This function determines the number of bytes which
2203 * can be failed from the beginning of the request without
2204 * crossing into area which need to be retried further.
2207 * The number of bytes to fail.
2210 * queue_lock must be held.
2212 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2214 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2215 unsigned int bytes
= 0;
2218 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2219 return blk_rq_bytes(rq
);
2222 * Currently the only 'mixing' which can happen is between
2223 * different fastfail types. We can safely fail portions
2224 * which have all the failfast bits that the first one has -
2225 * the ones which are at least as eager to fail as the first
2228 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2229 if ((bio
->bi_opf
& ff
) != ff
)
2231 bytes
+= bio
->bi_iter
.bi_size
;
2234 /* this could lead to infinite loop */
2235 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2238 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2240 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2242 if (blk_do_io_stat(req
)) {
2243 const int rw
= rq_data_dir(req
);
2244 struct hd_struct
*part
;
2247 cpu
= part_stat_lock();
2249 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2254 void blk_account_io_done(struct request
*req
)
2257 * Account IO completion. flush_rq isn't accounted as a
2258 * normal IO on queueing nor completion. Accounting the
2259 * containing request is enough.
2261 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2262 unsigned long duration
= jiffies
- req
->start_time
;
2263 const int rw
= rq_data_dir(req
);
2264 struct hd_struct
*part
;
2267 cpu
= part_stat_lock();
2270 part_stat_inc(cpu
, part
, ios
[rw
]);
2271 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2272 part_round_stats(cpu
, part
);
2273 part_dec_in_flight(part
, rw
);
2275 hd_struct_put(part
);
2282 * Don't process normal requests when queue is suspended
2283 * or in the process of suspending/resuming
2285 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2288 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2289 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2295 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2302 void blk_account_io_start(struct request
*rq
, bool new_io
)
2304 struct hd_struct
*part
;
2305 int rw
= rq_data_dir(rq
);
2308 if (!blk_do_io_stat(rq
))
2311 cpu
= part_stat_lock();
2315 part_stat_inc(cpu
, part
, merges
[rw
]);
2317 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2318 if (!hd_struct_try_get(part
)) {
2320 * The partition is already being removed,
2321 * the request will be accounted on the disk only
2323 * We take a reference on disk->part0 although that
2324 * partition will never be deleted, so we can treat
2325 * it as any other partition.
2327 part
= &rq
->rq_disk
->part0
;
2328 hd_struct_get(part
);
2330 part_round_stats(cpu
, part
);
2331 part_inc_in_flight(part
, rw
);
2339 * blk_peek_request - peek at the top of a request queue
2340 * @q: request queue to peek at
2343 * Return the request at the top of @q. The returned request
2344 * should be started using blk_start_request() before LLD starts
2348 * Pointer to the request at the top of @q if available. Null
2352 * queue_lock must be held.
2354 struct request
*blk_peek_request(struct request_queue
*q
)
2359 while ((rq
= __elv_next_request(q
)) != NULL
) {
2361 rq
= blk_pm_peek_request(q
, rq
);
2365 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2367 * This is the first time the device driver
2368 * sees this request (possibly after
2369 * requeueing). Notify IO scheduler.
2371 if (rq
->rq_flags
& RQF_SORTED
)
2372 elv_activate_rq(q
, rq
);
2375 * just mark as started even if we don't start
2376 * it, a request that has been delayed should
2377 * not be passed by new incoming requests
2379 rq
->rq_flags
|= RQF_STARTED
;
2380 trace_block_rq_issue(q
, rq
);
2383 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2384 q
->end_sector
= rq_end_sector(rq
);
2385 q
->boundary_rq
= NULL
;
2388 if (rq
->rq_flags
& RQF_DONTPREP
)
2391 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2393 * make sure space for the drain appears we
2394 * know we can do this because max_hw_segments
2395 * has been adjusted to be one fewer than the
2398 rq
->nr_phys_segments
++;
2404 ret
= q
->prep_rq_fn(q
, rq
);
2405 if (ret
== BLKPREP_OK
) {
2407 } else if (ret
== BLKPREP_DEFER
) {
2409 * the request may have been (partially) prepped.
2410 * we need to keep this request in the front to
2411 * avoid resource deadlock. RQF_STARTED will
2412 * prevent other fs requests from passing this one.
2414 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2415 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2417 * remove the space for the drain we added
2418 * so that we don't add it again
2420 --rq
->nr_phys_segments
;
2425 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2426 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2428 rq
->rq_flags
|= RQF_QUIET
;
2430 * Mark this request as started so we don't trigger
2431 * any debug logic in the end I/O path.
2433 blk_start_request(rq
);
2434 __blk_end_request_all(rq
, err
);
2436 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2443 EXPORT_SYMBOL(blk_peek_request
);
2445 void blk_dequeue_request(struct request
*rq
)
2447 struct request_queue
*q
= rq
->q
;
2449 BUG_ON(list_empty(&rq
->queuelist
));
2450 BUG_ON(ELV_ON_HASH(rq
));
2452 list_del_init(&rq
->queuelist
);
2455 * the time frame between a request being removed from the lists
2456 * and to it is freed is accounted as io that is in progress at
2459 if (blk_account_rq(rq
)) {
2460 q
->in_flight
[rq_is_sync(rq
)]++;
2461 set_io_start_time_ns(rq
);
2466 * blk_start_request - start request processing on the driver
2467 * @req: request to dequeue
2470 * Dequeue @req and start timeout timer on it. This hands off the
2471 * request to the driver.
2473 * Block internal functions which don't want to start timer should
2474 * call blk_dequeue_request().
2477 * queue_lock must be held.
2479 void blk_start_request(struct request
*req
)
2481 blk_dequeue_request(req
);
2483 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2484 blk_stat_set_issue_time(&req
->issue_stat
);
2485 req
->rq_flags
|= RQF_STATS
;
2486 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2489 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2492 EXPORT_SYMBOL(blk_start_request
);
2495 * blk_fetch_request - fetch a request from a request queue
2496 * @q: request queue to fetch a request from
2499 * Return the request at the top of @q. The request is started on
2500 * return and LLD can start processing it immediately.
2503 * Pointer to the request at the top of @q if available. Null
2507 * queue_lock must be held.
2509 struct request
*blk_fetch_request(struct request_queue
*q
)
2513 rq
= blk_peek_request(q
);
2515 blk_start_request(rq
);
2518 EXPORT_SYMBOL(blk_fetch_request
);
2521 * blk_update_request - Special helper function for request stacking drivers
2522 * @req: the request being processed
2523 * @error: %0 for success, < %0 for error
2524 * @nr_bytes: number of bytes to complete @req
2527 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2528 * the request structure even if @req doesn't have leftover.
2529 * If @req has leftover, sets it up for the next range of segments.
2531 * This special helper function is only for request stacking drivers
2532 * (e.g. request-based dm) so that they can handle partial completion.
2533 * Actual device drivers should use blk_end_request instead.
2535 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2536 * %false return from this function.
2539 * %false - this request doesn't have any more data
2540 * %true - this request has more data
2542 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2546 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2552 * For fs requests, rq is just carrier of independent bio's
2553 * and each partial completion should be handled separately.
2554 * Reset per-request error on each partial completion.
2556 * TODO: tj: This is too subtle. It would be better to let
2557 * low level drivers do what they see fit.
2559 if (!blk_rq_is_passthrough(req
))
2562 if (error
&& !blk_rq_is_passthrough(req
) &&
2563 !(req
->rq_flags
& RQF_QUIET
)) {
2568 error_type
= "recoverable transport";
2571 error_type
= "critical target";
2574 error_type
= "critical nexus";
2577 error_type
= "timeout";
2580 error_type
= "critical space allocation";
2583 error_type
= "critical medium";
2590 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2591 __func__
, error_type
, req
->rq_disk
?
2592 req
->rq_disk
->disk_name
: "?",
2593 (unsigned long long)blk_rq_pos(req
));
2597 blk_account_io_completion(req
, nr_bytes
);
2601 struct bio
*bio
= req
->bio
;
2602 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2604 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2605 req
->bio
= bio
->bi_next
;
2607 req_bio_endio(req
, bio
, bio_bytes
, error
);
2609 total_bytes
+= bio_bytes
;
2610 nr_bytes
-= bio_bytes
;
2621 * Reset counters so that the request stacking driver
2622 * can find how many bytes remain in the request
2625 req
->__data_len
= 0;
2629 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2631 req
->__data_len
-= total_bytes
;
2633 /* update sector only for requests with clear definition of sector */
2634 if (!blk_rq_is_passthrough(req
))
2635 req
->__sector
+= total_bytes
>> 9;
2637 /* mixed attributes always follow the first bio */
2638 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2639 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2640 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2644 * If total number of sectors is less than the first segment
2645 * size, something has gone terribly wrong.
2647 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2648 blk_dump_rq_flags(req
, "request botched");
2649 req
->__data_len
= blk_rq_cur_bytes(req
);
2652 /* recalculate the number of segments */
2653 blk_recalc_rq_segments(req
);
2657 EXPORT_SYMBOL_GPL(blk_update_request
);
2659 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2660 unsigned int nr_bytes
,
2661 unsigned int bidi_bytes
)
2663 if (blk_update_request(rq
, error
, nr_bytes
))
2666 /* Bidi request must be completed as a whole */
2667 if (unlikely(blk_bidi_rq(rq
)) &&
2668 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2671 if (blk_queue_add_random(rq
->q
))
2672 add_disk_randomness(rq
->rq_disk
);
2678 * blk_unprep_request - unprepare a request
2681 * This function makes a request ready for complete resubmission (or
2682 * completion). It happens only after all error handling is complete,
2683 * so represents the appropriate moment to deallocate any resources
2684 * that were allocated to the request in the prep_rq_fn. The queue
2685 * lock is held when calling this.
2687 void blk_unprep_request(struct request
*req
)
2689 struct request_queue
*q
= req
->q
;
2691 req
->rq_flags
&= ~RQF_DONTPREP
;
2692 if (q
->unprep_rq_fn
)
2693 q
->unprep_rq_fn(q
, req
);
2695 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2698 * queue lock must be held
2700 void blk_finish_request(struct request
*req
, int error
)
2702 struct request_queue
*q
= req
->q
;
2704 if (req
->rq_flags
& RQF_STATS
)
2707 if (req
->rq_flags
& RQF_QUEUED
)
2708 blk_queue_end_tag(q
, req
);
2710 BUG_ON(blk_queued_rq(req
));
2712 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2713 laptop_io_completion(req
->q
->backing_dev_info
);
2715 blk_delete_timer(req
);
2717 if (req
->rq_flags
& RQF_DONTPREP
)
2718 blk_unprep_request(req
);
2720 blk_account_io_done(req
);
2723 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2724 req
->end_io(req
, error
);
2726 if (blk_bidi_rq(req
))
2727 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2729 __blk_put_request(q
, req
);
2732 EXPORT_SYMBOL(blk_finish_request
);
2735 * blk_end_bidi_request - Complete a bidi request
2736 * @rq: the request to complete
2737 * @error: %0 for success, < %0 for error
2738 * @nr_bytes: number of bytes to complete @rq
2739 * @bidi_bytes: number of bytes to complete @rq->next_rq
2742 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2743 * Drivers that supports bidi can safely call this member for any
2744 * type of request, bidi or uni. In the later case @bidi_bytes is
2748 * %false - we are done with this request
2749 * %true - still buffers pending for this request
2751 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2752 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2754 struct request_queue
*q
= rq
->q
;
2755 unsigned long flags
;
2757 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2760 spin_lock_irqsave(q
->queue_lock
, flags
);
2761 blk_finish_request(rq
, error
);
2762 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2768 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2769 * @rq: the request to complete
2770 * @error: %0 for success, < %0 for error
2771 * @nr_bytes: number of bytes to complete @rq
2772 * @bidi_bytes: number of bytes to complete @rq->next_rq
2775 * Identical to blk_end_bidi_request() except that queue lock is
2776 * assumed to be locked on entry and remains so on return.
2779 * %false - we are done with this request
2780 * %true - still buffers pending for this request
2782 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2783 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2785 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2788 blk_finish_request(rq
, error
);
2794 * blk_end_request - Helper function for drivers to complete the request.
2795 * @rq: the request being processed
2796 * @error: %0 for success, < %0 for error
2797 * @nr_bytes: number of bytes to complete
2800 * Ends I/O on a number of bytes attached to @rq.
2801 * If @rq has leftover, sets it up for the next range of segments.
2804 * %false - we are done with this request
2805 * %true - still buffers pending for this request
2807 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2809 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2811 EXPORT_SYMBOL(blk_end_request
);
2814 * blk_end_request_all - Helper function for drives to finish the request.
2815 * @rq: the request to finish
2816 * @error: %0 for success, < %0 for error
2819 * Completely finish @rq.
2821 void blk_end_request_all(struct request
*rq
, int error
)
2824 unsigned int bidi_bytes
= 0;
2826 if (unlikely(blk_bidi_rq(rq
)))
2827 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2829 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2832 EXPORT_SYMBOL(blk_end_request_all
);
2835 * blk_end_request_cur - Helper function to finish the current request chunk.
2836 * @rq: the request to finish the current chunk for
2837 * @error: %0 for success, < %0 for error
2840 * Complete the current consecutively mapped chunk from @rq.
2843 * %false - we are done with this request
2844 * %true - still buffers pending for this request
2846 bool blk_end_request_cur(struct request
*rq
, int error
)
2848 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2850 EXPORT_SYMBOL(blk_end_request_cur
);
2853 * blk_end_request_err - Finish a request till the next failure boundary.
2854 * @rq: the request to finish till the next failure boundary for
2855 * @error: must be negative errno
2858 * Complete @rq till the next failure boundary.
2861 * %false - we are done with this request
2862 * %true - still buffers pending for this request
2864 bool blk_end_request_err(struct request
*rq
, int error
)
2866 WARN_ON(error
>= 0);
2867 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2869 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2872 * __blk_end_request - Helper function for drivers to complete the request.
2873 * @rq: the request being processed
2874 * @error: %0 for success, < %0 for error
2875 * @nr_bytes: number of bytes to complete
2878 * Must be called with queue lock held unlike blk_end_request().
2881 * %false - we are done with this request
2882 * %true - still buffers pending for this request
2884 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2886 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2888 EXPORT_SYMBOL(__blk_end_request
);
2891 * __blk_end_request_all - Helper function for drives to finish the request.
2892 * @rq: the request to finish
2893 * @error: %0 for success, < %0 for error
2896 * Completely finish @rq. Must be called with queue lock held.
2898 void __blk_end_request_all(struct request
*rq
, int error
)
2901 unsigned int bidi_bytes
= 0;
2903 if (unlikely(blk_bidi_rq(rq
)))
2904 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2906 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2909 EXPORT_SYMBOL(__blk_end_request_all
);
2912 * __blk_end_request_cur - Helper function to finish the current request chunk.
2913 * @rq: the request to finish the current chunk for
2914 * @error: %0 for success, < %0 for error
2917 * Complete the current consecutively mapped chunk from @rq. Must
2918 * be called with queue lock held.
2921 * %false - we are done with this request
2922 * %true - still buffers pending for this request
2924 bool __blk_end_request_cur(struct request
*rq
, int error
)
2926 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2928 EXPORT_SYMBOL(__blk_end_request_cur
);
2931 * __blk_end_request_err - Finish a request till the next failure boundary.
2932 * @rq: the request to finish till the next failure boundary for
2933 * @error: must be negative errno
2936 * Complete @rq till the next failure boundary. Must be called
2937 * with queue lock held.
2940 * %false - we are done with this request
2941 * %true - still buffers pending for this request
2943 bool __blk_end_request_err(struct request
*rq
, int error
)
2945 WARN_ON(error
>= 0);
2946 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2948 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2950 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2953 if (bio_has_data(bio
))
2954 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2956 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2957 rq
->bio
= rq
->biotail
= bio
;
2960 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2963 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2965 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2966 * @rq: the request to be flushed
2969 * Flush all pages in @rq.
2971 void rq_flush_dcache_pages(struct request
*rq
)
2973 struct req_iterator iter
;
2974 struct bio_vec bvec
;
2976 rq_for_each_segment(bvec
, rq
, iter
)
2977 flush_dcache_page(bvec
.bv_page
);
2979 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2983 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2984 * @q : the queue of the device being checked
2987 * Check if underlying low-level drivers of a device are busy.
2988 * If the drivers want to export their busy state, they must set own
2989 * exporting function using blk_queue_lld_busy() first.
2991 * Basically, this function is used only by request stacking drivers
2992 * to stop dispatching requests to underlying devices when underlying
2993 * devices are busy. This behavior helps more I/O merging on the queue
2994 * of the request stacking driver and prevents I/O throughput regression
2995 * on burst I/O load.
2998 * 0 - Not busy (The request stacking driver should dispatch request)
2999 * 1 - Busy (The request stacking driver should stop dispatching request)
3001 int blk_lld_busy(struct request_queue
*q
)
3004 return q
->lld_busy_fn(q
);
3008 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3011 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3012 * @rq: the clone request to be cleaned up
3015 * Free all bios in @rq for a cloned request.
3017 void blk_rq_unprep_clone(struct request
*rq
)
3021 while ((bio
= rq
->bio
) != NULL
) {
3022 rq
->bio
= bio
->bi_next
;
3027 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3030 * Copy attributes of the original request to the clone request.
3031 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3033 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3035 dst
->cpu
= src
->cpu
;
3036 dst
->__sector
= blk_rq_pos(src
);
3037 dst
->__data_len
= blk_rq_bytes(src
);
3038 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3039 dst
->ioprio
= src
->ioprio
;
3040 dst
->extra_len
= src
->extra_len
;
3044 * blk_rq_prep_clone - Helper function to setup clone request
3045 * @rq: the request to be setup
3046 * @rq_src: original request to be cloned
3047 * @bs: bio_set that bios for clone are allocated from
3048 * @gfp_mask: memory allocation mask for bio
3049 * @bio_ctr: setup function to be called for each clone bio.
3050 * Returns %0 for success, non %0 for failure.
3051 * @data: private data to be passed to @bio_ctr
3054 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3055 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3056 * are not copied, and copying such parts is the caller's responsibility.
3057 * Also, pages which the original bios are pointing to are not copied
3058 * and the cloned bios just point same pages.
3059 * So cloned bios must be completed before original bios, which means
3060 * the caller must complete @rq before @rq_src.
3062 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3063 struct bio_set
*bs
, gfp_t gfp_mask
,
3064 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3067 struct bio
*bio
, *bio_src
;
3072 __rq_for_each_bio(bio_src
, rq_src
) {
3073 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3077 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3081 rq
->biotail
->bi_next
= bio
;
3084 rq
->bio
= rq
->biotail
= bio
;
3087 __blk_rq_prep_clone(rq
, rq_src
);
3094 blk_rq_unprep_clone(rq
);
3098 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3100 int kblockd_schedule_work(struct work_struct
*work
)
3102 return queue_work(kblockd_workqueue
, work
);
3104 EXPORT_SYMBOL(kblockd_schedule_work
);
3106 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3108 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3110 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3112 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3113 unsigned long delay
)
3115 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3117 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3119 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3120 unsigned long delay
)
3122 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3124 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3127 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3128 * @plug: The &struct blk_plug that needs to be initialized
3131 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3132 * pending I/O should the task end up blocking between blk_start_plug() and
3133 * blk_finish_plug(). This is important from a performance perspective, but
3134 * also ensures that we don't deadlock. For instance, if the task is blocking
3135 * for a memory allocation, memory reclaim could end up wanting to free a
3136 * page belonging to that request that is currently residing in our private
3137 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3138 * this kind of deadlock.
3140 void blk_start_plug(struct blk_plug
*plug
)
3142 struct task_struct
*tsk
= current
;
3145 * If this is a nested plug, don't actually assign it.
3150 INIT_LIST_HEAD(&plug
->list
);
3151 INIT_LIST_HEAD(&plug
->mq_list
);
3152 INIT_LIST_HEAD(&plug
->cb_list
);
3154 * Store ordering should not be needed here, since a potential
3155 * preempt will imply a full memory barrier
3159 EXPORT_SYMBOL(blk_start_plug
);
3161 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3163 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3164 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3166 return !(rqa
->q
< rqb
->q
||
3167 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3171 * If 'from_schedule' is true, then postpone the dispatch of requests
3172 * until a safe kblockd context. We due this to avoid accidental big
3173 * additional stack usage in driver dispatch, in places where the originally
3174 * plugger did not intend it.
3176 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3178 __releases(q
->queue_lock
)
3180 trace_block_unplug(q
, depth
, !from_schedule
);
3183 blk_run_queue_async(q
);
3186 spin_unlock(q
->queue_lock
);
3189 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3191 LIST_HEAD(callbacks
);
3193 while (!list_empty(&plug
->cb_list
)) {
3194 list_splice_init(&plug
->cb_list
, &callbacks
);
3196 while (!list_empty(&callbacks
)) {
3197 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3200 list_del(&cb
->list
);
3201 cb
->callback(cb
, from_schedule
);
3206 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3209 struct blk_plug
*plug
= current
->plug
;
3210 struct blk_plug_cb
*cb
;
3215 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3216 if (cb
->callback
== unplug
&& cb
->data
== data
)
3219 /* Not currently on the callback list */
3220 BUG_ON(size
< sizeof(*cb
));
3221 cb
= kzalloc(size
, GFP_ATOMIC
);
3224 cb
->callback
= unplug
;
3225 list_add(&cb
->list
, &plug
->cb_list
);
3229 EXPORT_SYMBOL(blk_check_plugged
);
3231 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3233 struct request_queue
*q
;
3234 unsigned long flags
;
3239 flush_plug_callbacks(plug
, from_schedule
);
3241 if (!list_empty(&plug
->mq_list
))
3242 blk_mq_flush_plug_list(plug
, from_schedule
);
3244 if (list_empty(&plug
->list
))
3247 list_splice_init(&plug
->list
, &list
);
3249 list_sort(NULL
, &list
, plug_rq_cmp
);
3255 * Save and disable interrupts here, to avoid doing it for every
3256 * queue lock we have to take.
3258 local_irq_save(flags
);
3259 while (!list_empty(&list
)) {
3260 rq
= list_entry_rq(list
.next
);
3261 list_del_init(&rq
->queuelist
);
3265 * This drops the queue lock
3268 queue_unplugged(q
, depth
, from_schedule
);
3271 spin_lock(q
->queue_lock
);
3275 * Short-circuit if @q is dead
3277 if (unlikely(blk_queue_dying(q
))) {
3278 __blk_end_request_all(rq
, -ENODEV
);
3283 * rq is already accounted, so use raw insert
3285 if (op_is_flush(rq
->cmd_flags
))
3286 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3288 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3294 * This drops the queue lock
3297 queue_unplugged(q
, depth
, from_schedule
);
3299 local_irq_restore(flags
);
3302 void blk_finish_plug(struct blk_plug
*plug
)
3304 if (plug
!= current
->plug
)
3306 blk_flush_plug_list(plug
, false);
3308 current
->plug
= NULL
;
3310 EXPORT_SYMBOL(blk_finish_plug
);
3314 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3315 * @q: the queue of the device
3316 * @dev: the device the queue belongs to
3319 * Initialize runtime-PM-related fields for @q and start auto suspend for
3320 * @dev. Drivers that want to take advantage of request-based runtime PM
3321 * should call this function after @dev has been initialized, and its
3322 * request queue @q has been allocated, and runtime PM for it can not happen
3323 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3324 * cases, driver should call this function before any I/O has taken place.
3326 * This function takes care of setting up using auto suspend for the device,
3327 * the autosuspend delay is set to -1 to make runtime suspend impossible
3328 * until an updated value is either set by user or by driver. Drivers do
3329 * not need to touch other autosuspend settings.
3331 * The block layer runtime PM is request based, so only works for drivers
3332 * that use request as their IO unit instead of those directly use bio's.
3334 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3337 q
->rpm_status
= RPM_ACTIVE
;
3338 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3339 pm_runtime_use_autosuspend(q
->dev
);
3341 EXPORT_SYMBOL(blk_pm_runtime_init
);
3344 * blk_pre_runtime_suspend - Pre runtime suspend check
3345 * @q: the queue of the device
3348 * This function will check if runtime suspend is allowed for the device
3349 * by examining if there are any requests pending in the queue. If there
3350 * are requests pending, the device can not be runtime suspended; otherwise,
3351 * the queue's status will be updated to SUSPENDING and the driver can
3352 * proceed to suspend the device.
3354 * For the not allowed case, we mark last busy for the device so that
3355 * runtime PM core will try to autosuspend it some time later.
3357 * This function should be called near the start of the device's
3358 * runtime_suspend callback.
3361 * 0 - OK to runtime suspend the device
3362 * -EBUSY - Device should not be runtime suspended
3364 int blk_pre_runtime_suspend(struct request_queue
*q
)
3371 spin_lock_irq(q
->queue_lock
);
3372 if (q
->nr_pending
) {
3374 pm_runtime_mark_last_busy(q
->dev
);
3376 q
->rpm_status
= RPM_SUSPENDING
;
3378 spin_unlock_irq(q
->queue_lock
);
3381 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3384 * blk_post_runtime_suspend - Post runtime suspend processing
3385 * @q: the queue of the device
3386 * @err: return value of the device's runtime_suspend function
3389 * Update the queue's runtime status according to the return value of the
3390 * device's runtime suspend function and mark last busy for the device so
3391 * that PM core will try to auto suspend the device at a later time.
3393 * This function should be called near the end of the device's
3394 * runtime_suspend callback.
3396 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3401 spin_lock_irq(q
->queue_lock
);
3403 q
->rpm_status
= RPM_SUSPENDED
;
3405 q
->rpm_status
= RPM_ACTIVE
;
3406 pm_runtime_mark_last_busy(q
->dev
);
3408 spin_unlock_irq(q
->queue_lock
);
3410 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3413 * blk_pre_runtime_resume - Pre runtime resume processing
3414 * @q: the queue of the device
3417 * Update the queue's runtime status to RESUMING in preparation for the
3418 * runtime resume of the device.
3420 * This function should be called near the start of the device's
3421 * runtime_resume callback.
3423 void blk_pre_runtime_resume(struct request_queue
*q
)
3428 spin_lock_irq(q
->queue_lock
);
3429 q
->rpm_status
= RPM_RESUMING
;
3430 spin_unlock_irq(q
->queue_lock
);
3432 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3435 * blk_post_runtime_resume - Post runtime resume processing
3436 * @q: the queue of the device
3437 * @err: return value of the device's runtime_resume function
3440 * Update the queue's runtime status according to the return value of the
3441 * device's runtime_resume function. If it is successfully resumed, process
3442 * the requests that are queued into the device's queue when it is resuming
3443 * and then mark last busy and initiate autosuspend for it.
3445 * This function should be called near the end of the device's
3446 * runtime_resume callback.
3448 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3453 spin_lock_irq(q
->queue_lock
);
3455 q
->rpm_status
= RPM_ACTIVE
;
3457 pm_runtime_mark_last_busy(q
->dev
);
3458 pm_request_autosuspend(q
->dev
);
3460 q
->rpm_status
= RPM_SUSPENDED
;
3462 spin_unlock_irq(q
->queue_lock
);
3464 EXPORT_SYMBOL(blk_post_runtime_resume
);
3467 * blk_set_runtime_active - Force runtime status of the queue to be active
3468 * @q: the queue of the device
3470 * If the device is left runtime suspended during system suspend the resume
3471 * hook typically resumes the device and corrects runtime status
3472 * accordingly. However, that does not affect the queue runtime PM status
3473 * which is still "suspended". This prevents processing requests from the
3476 * This function can be used in driver's resume hook to correct queue
3477 * runtime PM status and re-enable peeking requests from the queue. It
3478 * should be called before first request is added to the queue.
3480 void blk_set_runtime_active(struct request_queue
*q
)
3482 spin_lock_irq(q
->queue_lock
);
3483 q
->rpm_status
= RPM_ACTIVE
;
3484 pm_runtime_mark_last_busy(q
->dev
);
3485 pm_request_autosuspend(q
->dev
);
3486 spin_unlock_irq(q
->queue_lock
);
3488 EXPORT_SYMBOL(blk_set_runtime_active
);
3491 int __init
blk_dev_init(void)
3493 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3494 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3495 FIELD_SIZEOF(struct request
, cmd_flags
));
3496 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3497 FIELD_SIZEOF(struct bio
, bi_opf
));
3499 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3500 kblockd_workqueue
= alloc_workqueue("kblockd",
3501 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3502 if (!kblockd_workqueue
)
3503 panic("Failed to create kblockd\n");
3505 request_cachep
= kmem_cache_create("blkdev_requests",
3506 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3508 blk_requestq_cachep
= kmem_cache_create("request_queue",
3509 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3511 #ifdef CONFIG_DEBUG_FS
3512 blk_debugfs_root
= debugfs_create_dir("block", NULL
);