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
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
859 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
860 goto out_free_flush_queue
;
862 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
863 goto out_exit_flush_rq
;
865 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
866 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
869 * This also sets hw/phys segments, boundary and size
871 blk_queue_make_request(q
, blk_queue_bio
);
873 q
->sg_reserved_size
= INT_MAX
;
875 /* Protect q->elevator from elevator_change */
876 mutex_lock(&q
->sysfs_lock
);
879 if (elevator_init(q
, NULL
)) {
880 mutex_unlock(&q
->sysfs_lock
);
881 goto out_exit_flush_rq
;
884 mutex_unlock(&q
->sysfs_lock
);
889 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
890 out_free_flush_queue
:
891 blk_free_flush_queue(q
->fq
);
894 EXPORT_SYMBOL(blk_init_allocated_queue
);
896 bool blk_get_queue(struct request_queue
*q
)
898 if (likely(!blk_queue_dying(q
))) {
905 EXPORT_SYMBOL(blk_get_queue
);
907 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
909 if (rq
->rq_flags
& RQF_ELVPRIV
) {
910 elv_put_request(rl
->q
, rq
);
912 put_io_context(rq
->elv
.icq
->ioc
);
915 mempool_free(rq
, rl
->rq_pool
);
919 * ioc_batching returns true if the ioc is a valid batching request and
920 * should be given priority access to a request.
922 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
928 * Make sure the process is able to allocate at least 1 request
929 * even if the batch times out, otherwise we could theoretically
932 return ioc
->nr_batch_requests
== q
->nr_batching
||
933 (ioc
->nr_batch_requests
> 0
934 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
938 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
939 * will cause the process to be a "batcher" on all queues in the system. This
940 * is the behaviour we want though - once it gets a wakeup it should be given
943 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
945 if (!ioc
|| ioc_batching(q
, ioc
))
948 ioc
->nr_batch_requests
= q
->nr_batching
;
949 ioc
->last_waited
= jiffies
;
952 static void __freed_request(struct request_list
*rl
, int sync
)
954 struct request_queue
*q
= rl
->q
;
956 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
957 blk_clear_congested(rl
, sync
);
959 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
960 if (waitqueue_active(&rl
->wait
[sync
]))
961 wake_up(&rl
->wait
[sync
]);
963 blk_clear_rl_full(rl
, sync
);
968 * A request has just been released. Account for it, update the full and
969 * congestion status, wake up any waiters. Called under q->queue_lock.
971 static void freed_request(struct request_list
*rl
, bool sync
,
972 req_flags_t rq_flags
)
974 struct request_queue
*q
= rl
->q
;
978 if (rq_flags
& RQF_ELVPRIV
)
981 __freed_request(rl
, sync
);
983 if (unlikely(rl
->starved
[sync
^ 1]))
984 __freed_request(rl
, sync
^ 1);
987 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
989 struct request_list
*rl
;
990 int on_thresh
, off_thresh
;
992 spin_lock_irq(q
->queue_lock
);
994 blk_queue_congestion_threshold(q
);
995 on_thresh
= queue_congestion_on_threshold(q
);
996 off_thresh
= queue_congestion_off_threshold(q
);
998 blk_queue_for_each_rl(rl
, q
) {
999 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1000 blk_set_congested(rl
, BLK_RW_SYNC
);
1001 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1002 blk_clear_congested(rl
, BLK_RW_SYNC
);
1004 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1005 blk_set_congested(rl
, BLK_RW_ASYNC
);
1006 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1007 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1009 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1010 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1012 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1013 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1016 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1017 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1019 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1020 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1024 spin_unlock_irq(q
->queue_lock
);
1029 * __get_request - get a free request
1030 * @rl: request list to allocate from
1031 * @op: operation and flags
1032 * @bio: bio to allocate request for (can be %NULL)
1033 * @gfp_mask: allocation mask
1035 * Get a free request from @q. This function may fail under memory
1036 * pressure or if @q is dead.
1038 * Must be called with @q->queue_lock held and,
1039 * Returns ERR_PTR on failure, with @q->queue_lock held.
1040 * Returns request pointer on success, with @q->queue_lock *not held*.
1042 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1043 struct bio
*bio
, gfp_t gfp_mask
)
1045 struct request_queue
*q
= rl
->q
;
1047 struct elevator_type
*et
= q
->elevator
->type
;
1048 struct io_context
*ioc
= rq_ioc(bio
);
1049 struct io_cq
*icq
= NULL
;
1050 const bool is_sync
= op_is_sync(op
);
1052 req_flags_t rq_flags
= RQF_ALLOCED
;
1054 if (unlikely(blk_queue_dying(q
)))
1055 return ERR_PTR(-ENODEV
);
1057 may_queue
= elv_may_queue(q
, op
);
1058 if (may_queue
== ELV_MQUEUE_NO
)
1061 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1062 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1064 * The queue will fill after this allocation, so set
1065 * it as full, and mark this process as "batching".
1066 * This process will be allowed to complete a batch of
1067 * requests, others will be blocked.
1069 if (!blk_rl_full(rl
, is_sync
)) {
1070 ioc_set_batching(q
, ioc
);
1071 blk_set_rl_full(rl
, is_sync
);
1073 if (may_queue
!= ELV_MQUEUE_MUST
1074 && !ioc_batching(q
, ioc
)) {
1076 * The queue is full and the allocating
1077 * process is not a "batcher", and not
1078 * exempted by the IO scheduler
1080 return ERR_PTR(-ENOMEM
);
1084 blk_set_congested(rl
, is_sync
);
1088 * Only allow batching queuers to allocate up to 50% over the defined
1089 * limit of requests, otherwise we could have thousands of requests
1090 * allocated with any setting of ->nr_requests
1092 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1093 return ERR_PTR(-ENOMEM
);
1095 q
->nr_rqs
[is_sync
]++;
1096 rl
->count
[is_sync
]++;
1097 rl
->starved
[is_sync
] = 0;
1100 * Decide whether the new request will be managed by elevator. If
1101 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1102 * prevent the current elevator from being destroyed until the new
1103 * request is freed. This guarantees icq's won't be destroyed and
1104 * makes creating new ones safe.
1106 * Flush requests do not use the elevator so skip initialization.
1107 * This allows a request to share the flush and elevator data.
1109 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1110 * it will be created after releasing queue_lock.
1112 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1113 rq_flags
|= RQF_ELVPRIV
;
1114 q
->nr_rqs_elvpriv
++;
1115 if (et
->icq_cache
&& ioc
)
1116 icq
= ioc_lookup_icq(ioc
, q
);
1119 if (blk_queue_io_stat(q
))
1120 rq_flags
|= RQF_IO_STAT
;
1121 spin_unlock_irq(q
->queue_lock
);
1123 /* allocate and init request */
1124 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1129 blk_rq_set_rl(rq
, rl
);
1130 blk_rq_set_prio(rq
, ioc
);
1132 rq
->rq_flags
= rq_flags
;
1135 if (rq_flags
& RQF_ELVPRIV
) {
1136 if (unlikely(et
->icq_cache
&& !icq
)) {
1138 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1144 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1147 /* @rq->elv.icq holds io_context until @rq is freed */
1149 get_io_context(icq
->ioc
);
1153 * ioc may be NULL here, and ioc_batching will be false. That's
1154 * OK, if the queue is under the request limit then requests need
1155 * not count toward the nr_batch_requests limit. There will always
1156 * be some limit enforced by BLK_BATCH_TIME.
1158 if (ioc_batching(q
, ioc
))
1159 ioc
->nr_batch_requests
--;
1161 trace_block_getrq(q
, bio
, op
);
1166 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1167 * and may fail indefinitely under memory pressure and thus
1168 * shouldn't stall IO. Treat this request as !elvpriv. This will
1169 * disturb iosched and blkcg but weird is bettern than dead.
1171 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1172 __func__
, dev_name(q
->backing_dev_info
->dev
));
1174 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1177 spin_lock_irq(q
->queue_lock
);
1178 q
->nr_rqs_elvpriv
--;
1179 spin_unlock_irq(q
->queue_lock
);
1184 * Allocation failed presumably due to memory. Undo anything we
1185 * might have messed up.
1187 * Allocating task should really be put onto the front of the wait
1188 * queue, but this is pretty rare.
1190 spin_lock_irq(q
->queue_lock
);
1191 freed_request(rl
, is_sync
, rq_flags
);
1194 * in the very unlikely event that allocation failed and no
1195 * requests for this direction was pending, mark us starved so that
1196 * freeing of a request in the other direction will notice
1197 * us. another possible fix would be to split the rq mempool into
1201 if (unlikely(rl
->count
[is_sync
] == 0))
1202 rl
->starved
[is_sync
] = 1;
1203 return ERR_PTR(-ENOMEM
);
1207 * get_request - get a free request
1208 * @q: request_queue to allocate request from
1209 * @op: operation and flags
1210 * @bio: bio to allocate request for (can be %NULL)
1211 * @gfp_mask: allocation mask
1213 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1214 * this function keeps retrying under memory pressure and fails iff @q is dead.
1216 * Must be called with @q->queue_lock held and,
1217 * Returns ERR_PTR on failure, with @q->queue_lock held.
1218 * Returns request pointer on success, with @q->queue_lock *not held*.
1220 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1221 struct bio
*bio
, gfp_t gfp_mask
)
1223 const bool is_sync
= op_is_sync(op
);
1225 struct request_list
*rl
;
1228 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1230 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1234 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1239 /* wait on @rl and retry */
1240 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1241 TASK_UNINTERRUPTIBLE
);
1243 trace_block_sleeprq(q
, bio
, op
);
1245 spin_unlock_irq(q
->queue_lock
);
1249 * After sleeping, we become a "batching" process and will be able
1250 * to allocate at least one request, and up to a big batch of them
1251 * for a small period time. See ioc_batching, ioc_set_batching
1253 ioc_set_batching(q
, current
->io_context
);
1255 spin_lock_irq(q
->queue_lock
);
1256 finish_wait(&rl
->wait
[is_sync
], &wait
);
1261 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1266 /* create ioc upfront */
1267 create_io_context(gfp_mask
, q
->node
);
1269 spin_lock_irq(q
->queue_lock
);
1270 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1272 spin_unlock_irq(q
->queue_lock
);
1276 /* q->queue_lock is unlocked at this point */
1278 rq
->__sector
= (sector_t
) -1;
1279 rq
->bio
= rq
->biotail
= NULL
;
1283 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1286 return blk_mq_alloc_request(q
, rw
,
1287 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1288 0 : BLK_MQ_REQ_NOWAIT
);
1290 return blk_old_get_request(q
, rw
, gfp_mask
);
1292 EXPORT_SYMBOL(blk_get_request
);
1295 * blk_requeue_request - put a request back on queue
1296 * @q: request queue where request should be inserted
1297 * @rq: request to be inserted
1300 * Drivers often keep queueing requests until the hardware cannot accept
1301 * more, when that condition happens we need to put the request back
1302 * on the queue. Must be called with queue lock held.
1304 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1306 blk_delete_timer(rq
);
1307 blk_clear_rq_complete(rq
);
1308 trace_block_rq_requeue(q
, rq
);
1309 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1311 if (rq
->rq_flags
& RQF_QUEUED
)
1312 blk_queue_end_tag(q
, rq
);
1314 BUG_ON(blk_queued_rq(rq
));
1316 elv_requeue_request(q
, rq
);
1318 EXPORT_SYMBOL(blk_requeue_request
);
1320 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1323 blk_account_io_start(rq
, true);
1324 __elv_add_request(q
, rq
, where
);
1327 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1332 if (now
== part
->stamp
)
1335 inflight
= part_in_flight(part
);
1337 __part_stat_add(cpu
, part
, time_in_queue
,
1338 inflight
* (now
- part
->stamp
));
1339 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1345 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1346 * @cpu: cpu number for stats access
1347 * @part: target partition
1349 * The average IO queue length and utilisation statistics are maintained
1350 * by observing the current state of the queue length and the amount of
1351 * time it has been in this state for.
1353 * Normally, that accounting is done on IO completion, but that can result
1354 * in more than a second's worth of IO being accounted for within any one
1355 * second, leading to >100% utilisation. To deal with that, we call this
1356 * function to do a round-off before returning the results when reading
1357 * /proc/diskstats. This accounts immediately for all queue usage up to
1358 * the current jiffies and restarts the counters again.
1360 void part_round_stats(int cpu
, struct hd_struct
*part
)
1362 unsigned long now
= jiffies
;
1365 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1366 part_round_stats_single(cpu
, part
, now
);
1368 EXPORT_SYMBOL_GPL(part_round_stats
);
1371 static void blk_pm_put_request(struct request
*rq
)
1373 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1374 pm_runtime_mark_last_busy(rq
->q
->dev
);
1377 static inline void blk_pm_put_request(struct request
*rq
) {}
1381 * queue lock must be held
1383 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1385 req_flags_t rq_flags
= req
->rq_flags
;
1391 blk_mq_free_request(req
);
1395 blk_pm_put_request(req
);
1397 elv_completed_request(q
, req
);
1399 /* this is a bio leak */
1400 WARN_ON(req
->bio
!= NULL
);
1402 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1405 * Request may not have originated from ll_rw_blk. if not,
1406 * it didn't come out of our reserved rq pools
1408 if (rq_flags
& RQF_ALLOCED
) {
1409 struct request_list
*rl
= blk_rq_rl(req
);
1410 bool sync
= op_is_sync(req
->cmd_flags
);
1412 BUG_ON(!list_empty(&req
->queuelist
));
1413 BUG_ON(ELV_ON_HASH(req
));
1415 blk_free_request(rl
, req
);
1416 freed_request(rl
, sync
, rq_flags
);
1420 EXPORT_SYMBOL_GPL(__blk_put_request
);
1422 void blk_put_request(struct request
*req
)
1424 struct request_queue
*q
= req
->q
;
1427 blk_mq_free_request(req
);
1429 unsigned long flags
;
1431 spin_lock_irqsave(q
->queue_lock
, flags
);
1432 __blk_put_request(q
, req
);
1433 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1436 EXPORT_SYMBOL(blk_put_request
);
1438 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1441 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1443 if (!ll_back_merge_fn(q
, req
, bio
))
1446 trace_block_bio_backmerge(q
, req
, bio
);
1448 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1449 blk_rq_set_mixed_merge(req
);
1451 req
->biotail
->bi_next
= bio
;
1453 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1454 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1456 blk_account_io_start(req
, false);
1460 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1463 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1465 if (!ll_front_merge_fn(q
, req
, bio
))
1468 trace_block_bio_frontmerge(q
, req
, bio
);
1470 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1471 blk_rq_set_mixed_merge(req
);
1473 bio
->bi_next
= req
->bio
;
1476 req
->__sector
= bio
->bi_iter
.bi_sector
;
1477 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1478 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1480 blk_account_io_start(req
, false);
1484 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1487 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1489 if (segments
>= queue_max_discard_segments(q
))
1491 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1492 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1495 req
->biotail
->bi_next
= bio
;
1497 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1498 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1499 req
->nr_phys_segments
= segments
+ 1;
1501 blk_account_io_start(req
, false);
1504 req_set_nomerge(q
, req
);
1509 * blk_attempt_plug_merge - try to merge with %current's plugged list
1510 * @q: request_queue new bio is being queued at
1511 * @bio: new bio being queued
1512 * @request_count: out parameter for number of traversed plugged requests
1513 * @same_queue_rq: pointer to &struct request that gets filled in when
1514 * another request associated with @q is found on the plug list
1515 * (optional, may be %NULL)
1517 * Determine whether @bio being queued on @q can be merged with a request
1518 * on %current's plugged list. Returns %true if merge was successful,
1521 * Plugging coalesces IOs from the same issuer for the same purpose without
1522 * going through @q->queue_lock. As such it's more of an issuing mechanism
1523 * than scheduling, and the request, while may have elvpriv data, is not
1524 * added on the elevator at this point. In addition, we don't have
1525 * reliable access to the elevator outside queue lock. Only check basic
1526 * merging parameters without querying the elevator.
1528 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1530 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1531 unsigned int *request_count
,
1532 struct request
**same_queue_rq
)
1534 struct blk_plug
*plug
;
1536 struct list_head
*plug_list
;
1538 plug
= current
->plug
;
1544 plug_list
= &plug
->mq_list
;
1546 plug_list
= &plug
->list
;
1548 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1549 bool merged
= false;
1554 * Only blk-mq multiple hardware queues case checks the
1555 * rq in the same queue, there should be only one such
1559 *same_queue_rq
= rq
;
1562 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1565 switch (blk_try_merge(rq
, bio
)) {
1566 case ELEVATOR_BACK_MERGE
:
1567 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1569 case ELEVATOR_FRONT_MERGE
:
1570 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1572 case ELEVATOR_DISCARD_MERGE
:
1573 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1586 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1588 struct blk_plug
*plug
;
1590 struct list_head
*plug_list
;
1591 unsigned int ret
= 0;
1593 plug
= current
->plug
;
1598 plug_list
= &plug
->mq_list
;
1600 plug_list
= &plug
->list
;
1602 list_for_each_entry(rq
, plug_list
, queuelist
) {
1610 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1612 if (bio
->bi_opf
& REQ_RAHEAD
)
1613 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1616 req
->__sector
= bio
->bi_iter
.bi_sector
;
1617 if (ioprio_valid(bio_prio(bio
)))
1618 req
->ioprio
= bio_prio(bio
);
1619 blk_rq_bio_prep(req
->q
, req
, bio
);
1622 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1624 struct blk_plug
*plug
;
1625 int where
= ELEVATOR_INSERT_SORT
;
1626 struct request
*req
, *free
;
1627 unsigned int request_count
= 0;
1628 unsigned int wb_acct
;
1631 * low level driver can indicate that it wants pages above a
1632 * certain limit bounced to low memory (ie for highmem, or even
1633 * ISA dma in theory)
1635 blk_queue_bounce(q
, &bio
);
1637 blk_queue_split(q
, &bio
, q
->bio_split
);
1639 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1640 bio
->bi_error
= -EIO
;
1642 return BLK_QC_T_NONE
;
1645 if (op_is_flush(bio
->bi_opf
)) {
1646 spin_lock_irq(q
->queue_lock
);
1647 where
= ELEVATOR_INSERT_FLUSH
;
1652 * Check if we can merge with the plugged list before grabbing
1655 if (!blk_queue_nomerges(q
)) {
1656 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1657 return BLK_QC_T_NONE
;
1659 request_count
= blk_plug_queued_count(q
);
1661 spin_lock_irq(q
->queue_lock
);
1663 switch (elv_merge(q
, &req
, bio
)) {
1664 case ELEVATOR_BACK_MERGE
:
1665 if (!bio_attempt_back_merge(q
, req
, bio
))
1667 elv_bio_merged(q
, req
, bio
);
1668 free
= attempt_back_merge(q
, req
);
1670 __blk_put_request(q
, free
);
1672 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1674 case ELEVATOR_FRONT_MERGE
:
1675 if (!bio_attempt_front_merge(q
, req
, bio
))
1677 elv_bio_merged(q
, req
, bio
);
1678 free
= attempt_front_merge(q
, req
);
1680 __blk_put_request(q
, free
);
1682 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1689 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1692 * Grab a free request. This is might sleep but can not fail.
1693 * Returns with the queue unlocked.
1695 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1697 __wbt_done(q
->rq_wb
, wb_acct
);
1698 bio
->bi_error
= PTR_ERR(req
);
1703 wbt_track(&req
->issue_stat
, wb_acct
);
1706 * After dropping the lock and possibly sleeping here, our request
1707 * may now be mergeable after it had proven unmergeable (above).
1708 * We don't worry about that case for efficiency. It won't happen
1709 * often, and the elevators are able to handle it.
1711 init_request_from_bio(req
, bio
);
1713 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1714 req
->cpu
= raw_smp_processor_id();
1716 plug
= current
->plug
;
1719 * If this is the first request added after a plug, fire
1722 * @request_count may become stale because of schedule
1723 * out, so check plug list again.
1725 if (!request_count
|| list_empty(&plug
->list
))
1726 trace_block_plug(q
);
1728 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1729 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1730 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1731 blk_flush_plug_list(plug
, false);
1732 trace_block_plug(q
);
1735 list_add_tail(&req
->queuelist
, &plug
->list
);
1736 blk_account_io_start(req
, true);
1738 spin_lock_irq(q
->queue_lock
);
1739 add_acct_request(q
, req
, where
);
1742 spin_unlock_irq(q
->queue_lock
);
1745 return BLK_QC_T_NONE
;
1749 * If bio->bi_dev is a partition, remap the location
1751 static inline void blk_partition_remap(struct bio
*bio
)
1753 struct block_device
*bdev
= bio
->bi_bdev
;
1756 * Zone reset does not include bi_size so bio_sectors() is always 0.
1757 * Include a test for the reset op code and perform the remap if needed.
1759 if (bdev
!= bdev
->bd_contains
&&
1760 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1761 struct hd_struct
*p
= bdev
->bd_part
;
1763 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1764 bio
->bi_bdev
= bdev
->bd_contains
;
1766 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1768 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1772 static void handle_bad_sector(struct bio
*bio
)
1774 char b
[BDEVNAME_SIZE
];
1776 printk(KERN_INFO
"attempt to access beyond end of device\n");
1777 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1778 bdevname(bio
->bi_bdev
, b
),
1780 (unsigned long long)bio_end_sector(bio
),
1781 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1784 #ifdef CONFIG_FAIL_MAKE_REQUEST
1786 static DECLARE_FAULT_ATTR(fail_make_request
);
1788 static int __init
setup_fail_make_request(char *str
)
1790 return setup_fault_attr(&fail_make_request
, str
);
1792 __setup("fail_make_request=", setup_fail_make_request
);
1794 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1796 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1799 static int __init
fail_make_request_debugfs(void)
1801 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1802 NULL
, &fail_make_request
);
1804 return PTR_ERR_OR_ZERO(dir
);
1807 late_initcall(fail_make_request_debugfs
);
1809 #else /* CONFIG_FAIL_MAKE_REQUEST */
1811 static inline bool should_fail_request(struct hd_struct
*part
,
1817 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1820 * Check whether this bio extends beyond the end of the device.
1822 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1829 /* Test device or partition size, when known. */
1830 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1832 sector_t sector
= bio
->bi_iter
.bi_sector
;
1834 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1836 * This may well happen - the kernel calls bread()
1837 * without checking the size of the device, e.g., when
1838 * mounting a device.
1840 handle_bad_sector(bio
);
1848 static noinline_for_stack
bool
1849 generic_make_request_checks(struct bio
*bio
)
1851 struct request_queue
*q
;
1852 int nr_sectors
= bio_sectors(bio
);
1854 char b
[BDEVNAME_SIZE
];
1855 struct hd_struct
*part
;
1859 if (bio_check_eod(bio
, nr_sectors
))
1862 q
= bdev_get_queue(bio
->bi_bdev
);
1865 "generic_make_request: Trying to access "
1866 "nonexistent block-device %s (%Lu)\n",
1867 bdevname(bio
->bi_bdev
, b
),
1868 (long long) bio
->bi_iter
.bi_sector
);
1872 part
= bio
->bi_bdev
->bd_part
;
1873 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1874 should_fail_request(&part_to_disk(part
)->part0
,
1875 bio
->bi_iter
.bi_size
))
1879 * If this device has partitions, remap block n
1880 * of partition p to block n+start(p) of the disk.
1882 blk_partition_remap(bio
);
1884 if (bio_check_eod(bio
, nr_sectors
))
1888 * Filter flush bio's early so that make_request based
1889 * drivers without flush support don't have to worry
1892 if (op_is_flush(bio
->bi_opf
) &&
1893 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1894 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1901 switch (bio_op(bio
)) {
1902 case REQ_OP_DISCARD
:
1903 if (!blk_queue_discard(q
))
1906 case REQ_OP_SECURE_ERASE
:
1907 if (!blk_queue_secure_erase(q
))
1910 case REQ_OP_WRITE_SAME
:
1911 if (!bdev_write_same(bio
->bi_bdev
))
1914 case REQ_OP_ZONE_REPORT
:
1915 case REQ_OP_ZONE_RESET
:
1916 if (!bdev_is_zoned(bio
->bi_bdev
))
1919 case REQ_OP_WRITE_ZEROES
:
1920 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1928 * Various block parts want %current->io_context and lazy ioc
1929 * allocation ends up trading a lot of pain for a small amount of
1930 * memory. Just allocate it upfront. This may fail and block
1931 * layer knows how to live with it.
1933 create_io_context(GFP_ATOMIC
, q
->node
);
1935 if (!blkcg_bio_issue_check(q
, bio
))
1938 trace_block_bio_queue(q
, bio
);
1944 bio
->bi_error
= err
;
1950 * generic_make_request - hand a buffer to its device driver for I/O
1951 * @bio: The bio describing the location in memory and on the device.
1953 * generic_make_request() is used to make I/O requests of block
1954 * devices. It is passed a &struct bio, which describes the I/O that needs
1957 * generic_make_request() does not return any status. The
1958 * success/failure status of the request, along with notification of
1959 * completion, is delivered asynchronously through the bio->bi_end_io
1960 * function described (one day) else where.
1962 * The caller of generic_make_request must make sure that bi_io_vec
1963 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1964 * set to describe the device address, and the
1965 * bi_end_io and optionally bi_private are set to describe how
1966 * completion notification should be signaled.
1968 * generic_make_request and the drivers it calls may use bi_next if this
1969 * bio happens to be merged with someone else, and may resubmit the bio to
1970 * a lower device by calling into generic_make_request recursively, which
1971 * means the bio should NOT be touched after the call to ->make_request_fn.
1973 blk_qc_t
generic_make_request(struct bio
*bio
)
1976 * bio_list_on_stack[0] contains bios submitted by the current
1978 * bio_list_on_stack[1] contains bios that were submitted before
1979 * the current make_request_fn, but that haven't been processed
1982 struct bio_list bio_list_on_stack
[2];
1983 blk_qc_t ret
= BLK_QC_T_NONE
;
1985 if (!generic_make_request_checks(bio
))
1989 * We only want one ->make_request_fn to be active at a time, else
1990 * stack usage with stacked devices could be a problem. So use
1991 * current->bio_list to keep a list of requests submited by a
1992 * make_request_fn function. current->bio_list is also used as a
1993 * flag to say if generic_make_request is currently active in this
1994 * task or not. If it is NULL, then no make_request is active. If
1995 * it is non-NULL, then a make_request is active, and new requests
1996 * should be added at the tail
1998 if (current
->bio_list
) {
1999 bio_list_add(¤t
->bio_list
[0], bio
);
2003 /* following loop may be a bit non-obvious, and so deserves some
2005 * Before entering the loop, bio->bi_next is NULL (as all callers
2006 * ensure that) so we have a list with a single bio.
2007 * We pretend that we have just taken it off a longer list, so
2008 * we assign bio_list to a pointer to the bio_list_on_stack,
2009 * thus initialising the bio_list of new bios to be
2010 * added. ->make_request() may indeed add some more bios
2011 * through a recursive call to generic_make_request. If it
2012 * did, we find a non-NULL value in bio_list and re-enter the loop
2013 * from the top. In this case we really did just take the bio
2014 * of the top of the list (no pretending) and so remove it from
2015 * bio_list, and call into ->make_request() again.
2017 BUG_ON(bio
->bi_next
);
2018 bio_list_init(&bio_list_on_stack
[0]);
2019 current
->bio_list
= bio_list_on_stack
;
2021 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2023 if (likely(blk_queue_enter(q
, false) == 0)) {
2024 struct bio_list lower
, same
;
2026 /* Create a fresh bio_list for all subordinate requests */
2027 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2028 bio_list_init(&bio_list_on_stack
[0]);
2029 ret
= q
->make_request_fn(q
, bio
);
2033 /* sort new bios into those for a lower level
2034 * and those for the same level
2036 bio_list_init(&lower
);
2037 bio_list_init(&same
);
2038 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2039 if (q
== bdev_get_queue(bio
->bi_bdev
))
2040 bio_list_add(&same
, bio
);
2042 bio_list_add(&lower
, bio
);
2043 /* now assemble so we handle the lowest level first */
2044 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2045 bio_list_merge(&bio_list_on_stack
[0], &same
);
2046 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2050 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2052 current
->bio_list
= NULL
; /* deactivate */
2057 EXPORT_SYMBOL(generic_make_request
);
2060 * submit_bio - submit a bio to the block device layer for I/O
2061 * @bio: The &struct bio which describes the I/O
2063 * submit_bio() is very similar in purpose to generic_make_request(), and
2064 * uses that function to do most of the work. Both are fairly rough
2065 * interfaces; @bio must be presetup and ready for I/O.
2068 blk_qc_t
submit_bio(struct bio
*bio
)
2071 * If it's a regular read/write or a barrier with data attached,
2072 * go through the normal accounting stuff before submission.
2074 if (bio_has_data(bio
)) {
2077 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2078 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2080 count
= bio_sectors(bio
);
2082 if (op_is_write(bio_op(bio
))) {
2083 count_vm_events(PGPGOUT
, count
);
2085 task_io_account_read(bio
->bi_iter
.bi_size
);
2086 count_vm_events(PGPGIN
, count
);
2089 if (unlikely(block_dump
)) {
2090 char b
[BDEVNAME_SIZE
];
2091 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2092 current
->comm
, task_pid_nr(current
),
2093 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2094 (unsigned long long)bio
->bi_iter
.bi_sector
,
2095 bdevname(bio
->bi_bdev
, b
),
2100 return generic_make_request(bio
);
2102 EXPORT_SYMBOL(submit_bio
);
2105 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2106 * for new the queue limits
2108 * @rq: the request being checked
2111 * @rq may have been made based on weaker limitations of upper-level queues
2112 * in request stacking drivers, and it may violate the limitation of @q.
2113 * Since the block layer and the underlying device driver trust @rq
2114 * after it is inserted to @q, it should be checked against @q before
2115 * the insertion using this generic function.
2117 * Request stacking drivers like request-based dm may change the queue
2118 * limits when retrying requests on other queues. Those requests need
2119 * to be checked against the new queue limits again during dispatch.
2121 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2124 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2125 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2130 * queue's settings related to segment counting like q->bounce_pfn
2131 * may differ from that of other stacking queues.
2132 * Recalculate it to check the request correctly on this queue's
2135 blk_recalc_rq_segments(rq
);
2136 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2137 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2145 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2146 * @q: the queue to submit the request
2147 * @rq: the request being queued
2149 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2151 unsigned long flags
;
2152 int where
= ELEVATOR_INSERT_BACK
;
2154 if (blk_cloned_rq_check_limits(q
, rq
))
2158 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2162 if (blk_queue_io_stat(q
))
2163 blk_account_io_start(rq
, true);
2164 blk_mq_sched_insert_request(rq
, false, true, false, false);
2168 spin_lock_irqsave(q
->queue_lock
, flags
);
2169 if (unlikely(blk_queue_dying(q
))) {
2170 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2175 * Submitting request must be dequeued before calling this function
2176 * because it will be linked to another request_queue
2178 BUG_ON(blk_queued_rq(rq
));
2180 if (op_is_flush(rq
->cmd_flags
))
2181 where
= ELEVATOR_INSERT_FLUSH
;
2183 add_acct_request(q
, rq
, where
);
2184 if (where
== ELEVATOR_INSERT_FLUSH
)
2186 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2190 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2193 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2194 * @rq: request to examine
2197 * A request could be merge of IOs which require different failure
2198 * handling. This function determines the number of bytes which
2199 * can be failed from the beginning of the request without
2200 * crossing into area which need to be retried further.
2203 * The number of bytes to fail.
2206 * queue_lock must be held.
2208 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2210 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2211 unsigned int bytes
= 0;
2214 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2215 return blk_rq_bytes(rq
);
2218 * Currently the only 'mixing' which can happen is between
2219 * different fastfail types. We can safely fail portions
2220 * which have all the failfast bits that the first one has -
2221 * the ones which are at least as eager to fail as the first
2224 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2225 if ((bio
->bi_opf
& ff
) != ff
)
2227 bytes
+= bio
->bi_iter
.bi_size
;
2230 /* this could lead to infinite loop */
2231 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2234 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2236 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2238 if (blk_do_io_stat(req
)) {
2239 const int rw
= rq_data_dir(req
);
2240 struct hd_struct
*part
;
2243 cpu
= part_stat_lock();
2245 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2250 void blk_account_io_done(struct request
*req
)
2253 * Account IO completion. flush_rq isn't accounted as a
2254 * normal IO on queueing nor completion. Accounting the
2255 * containing request is enough.
2257 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2258 unsigned long duration
= jiffies
- req
->start_time
;
2259 const int rw
= rq_data_dir(req
);
2260 struct hd_struct
*part
;
2263 cpu
= part_stat_lock();
2266 part_stat_inc(cpu
, part
, ios
[rw
]);
2267 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2268 part_round_stats(cpu
, part
);
2269 part_dec_in_flight(part
, rw
);
2271 hd_struct_put(part
);
2278 * Don't process normal requests when queue is suspended
2279 * or in the process of suspending/resuming
2281 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2284 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2285 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2291 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2298 void blk_account_io_start(struct request
*rq
, bool new_io
)
2300 struct hd_struct
*part
;
2301 int rw
= rq_data_dir(rq
);
2304 if (!blk_do_io_stat(rq
))
2307 cpu
= part_stat_lock();
2311 part_stat_inc(cpu
, part
, merges
[rw
]);
2313 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2314 if (!hd_struct_try_get(part
)) {
2316 * The partition is already being removed,
2317 * the request will be accounted on the disk only
2319 * We take a reference on disk->part0 although that
2320 * partition will never be deleted, so we can treat
2321 * it as any other partition.
2323 part
= &rq
->rq_disk
->part0
;
2324 hd_struct_get(part
);
2326 part_round_stats(cpu
, part
);
2327 part_inc_in_flight(part
, rw
);
2335 * blk_peek_request - peek at the top of a request queue
2336 * @q: request queue to peek at
2339 * Return the request at the top of @q. The returned request
2340 * should be started using blk_start_request() before LLD starts
2344 * Pointer to the request at the top of @q if available. Null
2348 * queue_lock must be held.
2350 struct request
*blk_peek_request(struct request_queue
*q
)
2355 while ((rq
= __elv_next_request(q
)) != NULL
) {
2357 rq
= blk_pm_peek_request(q
, rq
);
2361 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2363 * This is the first time the device driver
2364 * sees this request (possibly after
2365 * requeueing). Notify IO scheduler.
2367 if (rq
->rq_flags
& RQF_SORTED
)
2368 elv_activate_rq(q
, rq
);
2371 * just mark as started even if we don't start
2372 * it, a request that has been delayed should
2373 * not be passed by new incoming requests
2375 rq
->rq_flags
|= RQF_STARTED
;
2376 trace_block_rq_issue(q
, rq
);
2379 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2380 q
->end_sector
= rq_end_sector(rq
);
2381 q
->boundary_rq
= NULL
;
2384 if (rq
->rq_flags
& RQF_DONTPREP
)
2387 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2389 * make sure space for the drain appears we
2390 * know we can do this because max_hw_segments
2391 * has been adjusted to be one fewer than the
2394 rq
->nr_phys_segments
++;
2400 ret
= q
->prep_rq_fn(q
, rq
);
2401 if (ret
== BLKPREP_OK
) {
2403 } else if (ret
== BLKPREP_DEFER
) {
2405 * the request may have been (partially) prepped.
2406 * we need to keep this request in the front to
2407 * avoid resource deadlock. RQF_STARTED will
2408 * prevent other fs requests from passing this one.
2410 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2411 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2413 * remove the space for the drain we added
2414 * so that we don't add it again
2416 --rq
->nr_phys_segments
;
2421 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2422 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2424 rq
->rq_flags
|= RQF_QUIET
;
2426 * Mark this request as started so we don't trigger
2427 * any debug logic in the end I/O path.
2429 blk_start_request(rq
);
2430 __blk_end_request_all(rq
, err
);
2432 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2439 EXPORT_SYMBOL(blk_peek_request
);
2441 void blk_dequeue_request(struct request
*rq
)
2443 struct request_queue
*q
= rq
->q
;
2445 BUG_ON(list_empty(&rq
->queuelist
));
2446 BUG_ON(ELV_ON_HASH(rq
));
2448 list_del_init(&rq
->queuelist
);
2451 * the time frame between a request being removed from the lists
2452 * and to it is freed is accounted as io that is in progress at
2455 if (blk_account_rq(rq
)) {
2456 q
->in_flight
[rq_is_sync(rq
)]++;
2457 set_io_start_time_ns(rq
);
2462 * blk_start_request - start request processing on the driver
2463 * @req: request to dequeue
2466 * Dequeue @req and start timeout timer on it. This hands off the
2467 * request to the driver.
2469 * Block internal functions which don't want to start timer should
2470 * call blk_dequeue_request().
2473 * queue_lock must be held.
2475 void blk_start_request(struct request
*req
)
2477 blk_dequeue_request(req
);
2479 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2480 blk_stat_set_issue_time(&req
->issue_stat
);
2481 req
->rq_flags
|= RQF_STATS
;
2482 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2485 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2488 EXPORT_SYMBOL(blk_start_request
);
2491 * blk_fetch_request - fetch a request from a request queue
2492 * @q: request queue to fetch a request from
2495 * Return the request at the top of @q. The request is started on
2496 * return and LLD can start processing it immediately.
2499 * Pointer to the request at the top of @q if available. Null
2503 * queue_lock must be held.
2505 struct request
*blk_fetch_request(struct request_queue
*q
)
2509 rq
= blk_peek_request(q
);
2511 blk_start_request(rq
);
2514 EXPORT_SYMBOL(blk_fetch_request
);
2517 * blk_update_request - Special helper function for request stacking drivers
2518 * @req: the request being processed
2519 * @error: %0 for success, < %0 for error
2520 * @nr_bytes: number of bytes to complete @req
2523 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2524 * the request structure even if @req doesn't have leftover.
2525 * If @req has leftover, sets it up for the next range of segments.
2527 * This special helper function is only for request stacking drivers
2528 * (e.g. request-based dm) so that they can handle partial completion.
2529 * Actual device drivers should use blk_end_request instead.
2531 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2532 * %false return from this function.
2535 * %false - this request doesn't have any more data
2536 * %true - this request has more data
2538 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2542 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2548 * For fs requests, rq is just carrier of independent bio's
2549 * and each partial completion should be handled separately.
2550 * Reset per-request error on each partial completion.
2552 * TODO: tj: This is too subtle. It would be better to let
2553 * low level drivers do what they see fit.
2555 if (!blk_rq_is_passthrough(req
))
2558 if (error
&& !blk_rq_is_passthrough(req
) &&
2559 !(req
->rq_flags
& RQF_QUIET
)) {
2564 error_type
= "recoverable transport";
2567 error_type
= "critical target";
2570 error_type
= "critical nexus";
2573 error_type
= "timeout";
2576 error_type
= "critical space allocation";
2579 error_type
= "critical medium";
2586 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2587 __func__
, error_type
, req
->rq_disk
?
2588 req
->rq_disk
->disk_name
: "?",
2589 (unsigned long long)blk_rq_pos(req
));
2593 blk_account_io_completion(req
, nr_bytes
);
2597 struct bio
*bio
= req
->bio
;
2598 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2600 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2601 req
->bio
= bio
->bi_next
;
2603 req_bio_endio(req
, bio
, bio_bytes
, error
);
2605 total_bytes
+= bio_bytes
;
2606 nr_bytes
-= bio_bytes
;
2617 * Reset counters so that the request stacking driver
2618 * can find how many bytes remain in the request
2621 req
->__data_len
= 0;
2625 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2627 req
->__data_len
-= total_bytes
;
2629 /* update sector only for requests with clear definition of sector */
2630 if (!blk_rq_is_passthrough(req
))
2631 req
->__sector
+= total_bytes
>> 9;
2633 /* mixed attributes always follow the first bio */
2634 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2635 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2636 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2640 * If total number of sectors is less than the first segment
2641 * size, something has gone terribly wrong.
2643 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2644 blk_dump_rq_flags(req
, "request botched");
2645 req
->__data_len
= blk_rq_cur_bytes(req
);
2648 /* recalculate the number of segments */
2649 blk_recalc_rq_segments(req
);
2653 EXPORT_SYMBOL_GPL(blk_update_request
);
2655 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2656 unsigned int nr_bytes
,
2657 unsigned int bidi_bytes
)
2659 if (blk_update_request(rq
, error
, nr_bytes
))
2662 /* Bidi request must be completed as a whole */
2663 if (unlikely(blk_bidi_rq(rq
)) &&
2664 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2667 if (blk_queue_add_random(rq
->q
))
2668 add_disk_randomness(rq
->rq_disk
);
2674 * blk_unprep_request - unprepare a request
2677 * This function makes a request ready for complete resubmission (or
2678 * completion). It happens only after all error handling is complete,
2679 * so represents the appropriate moment to deallocate any resources
2680 * that were allocated to the request in the prep_rq_fn. The queue
2681 * lock is held when calling this.
2683 void blk_unprep_request(struct request
*req
)
2685 struct request_queue
*q
= req
->q
;
2687 req
->rq_flags
&= ~RQF_DONTPREP
;
2688 if (q
->unprep_rq_fn
)
2689 q
->unprep_rq_fn(q
, req
);
2691 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2694 * queue lock must be held
2696 void blk_finish_request(struct request
*req
, int error
)
2698 struct request_queue
*q
= req
->q
;
2700 if (req
->rq_flags
& RQF_STATS
)
2701 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2703 if (req
->rq_flags
& RQF_QUEUED
)
2704 blk_queue_end_tag(q
, req
);
2706 BUG_ON(blk_queued_rq(req
));
2708 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2709 laptop_io_completion(req
->q
->backing_dev_info
);
2711 blk_delete_timer(req
);
2713 if (req
->rq_flags
& RQF_DONTPREP
)
2714 blk_unprep_request(req
);
2716 blk_account_io_done(req
);
2719 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2720 req
->end_io(req
, error
);
2722 if (blk_bidi_rq(req
))
2723 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2725 __blk_put_request(q
, req
);
2728 EXPORT_SYMBOL(blk_finish_request
);
2731 * blk_end_bidi_request - Complete a bidi request
2732 * @rq: the request to complete
2733 * @error: %0 for success, < %0 for error
2734 * @nr_bytes: number of bytes to complete @rq
2735 * @bidi_bytes: number of bytes to complete @rq->next_rq
2738 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2739 * Drivers that supports bidi can safely call this member for any
2740 * type of request, bidi or uni. In the later case @bidi_bytes is
2744 * %false - we are done with this request
2745 * %true - still buffers pending for this request
2747 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2748 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2750 struct request_queue
*q
= rq
->q
;
2751 unsigned long flags
;
2753 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2756 spin_lock_irqsave(q
->queue_lock
, flags
);
2757 blk_finish_request(rq
, error
);
2758 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2764 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2765 * @rq: the request to complete
2766 * @error: %0 for success, < %0 for error
2767 * @nr_bytes: number of bytes to complete @rq
2768 * @bidi_bytes: number of bytes to complete @rq->next_rq
2771 * Identical to blk_end_bidi_request() except that queue lock is
2772 * assumed to be locked on entry and remains so on return.
2775 * %false - we are done with this request
2776 * %true - still buffers pending for this request
2778 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2779 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2781 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2784 blk_finish_request(rq
, error
);
2790 * blk_end_request - Helper function for drivers to complete the request.
2791 * @rq: the request being processed
2792 * @error: %0 for success, < %0 for error
2793 * @nr_bytes: number of bytes to complete
2796 * Ends I/O on a number of bytes attached to @rq.
2797 * If @rq has leftover, sets it up for the next range of segments.
2800 * %false - we are done with this request
2801 * %true - still buffers pending for this request
2803 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2805 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2807 EXPORT_SYMBOL(blk_end_request
);
2810 * blk_end_request_all - Helper function for drives to finish the request.
2811 * @rq: the request to finish
2812 * @error: %0 for success, < %0 for error
2815 * Completely finish @rq.
2817 void blk_end_request_all(struct request
*rq
, int error
)
2820 unsigned int bidi_bytes
= 0;
2822 if (unlikely(blk_bidi_rq(rq
)))
2823 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2825 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2828 EXPORT_SYMBOL(blk_end_request_all
);
2831 * blk_end_request_cur - Helper function to finish the current request chunk.
2832 * @rq: the request to finish the current chunk for
2833 * @error: %0 for success, < %0 for error
2836 * Complete the current consecutively mapped chunk from @rq.
2839 * %false - we are done with this request
2840 * %true - still buffers pending for this request
2842 bool blk_end_request_cur(struct request
*rq
, int error
)
2844 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2846 EXPORT_SYMBOL(blk_end_request_cur
);
2849 * blk_end_request_err - Finish a request till the next failure boundary.
2850 * @rq: the request to finish till the next failure boundary for
2851 * @error: must be negative errno
2854 * Complete @rq till the next failure boundary.
2857 * %false - we are done with this request
2858 * %true - still buffers pending for this request
2860 bool blk_end_request_err(struct request
*rq
, int error
)
2862 WARN_ON(error
>= 0);
2863 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2865 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2868 * __blk_end_request - Helper function for drivers to complete the request.
2869 * @rq: the request being processed
2870 * @error: %0 for success, < %0 for error
2871 * @nr_bytes: number of bytes to complete
2874 * Must be called with queue lock held unlike blk_end_request().
2877 * %false - we are done with this request
2878 * %true - still buffers pending for this request
2880 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2882 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2884 EXPORT_SYMBOL(__blk_end_request
);
2887 * __blk_end_request_all - Helper function for drives to finish the request.
2888 * @rq: the request to finish
2889 * @error: %0 for success, < %0 for error
2892 * Completely finish @rq. Must be called with queue lock held.
2894 void __blk_end_request_all(struct request
*rq
, int error
)
2897 unsigned int bidi_bytes
= 0;
2899 if (unlikely(blk_bidi_rq(rq
)))
2900 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2902 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2905 EXPORT_SYMBOL(__blk_end_request_all
);
2908 * __blk_end_request_cur - Helper function to finish the current request chunk.
2909 * @rq: the request to finish the current chunk for
2910 * @error: %0 for success, < %0 for error
2913 * Complete the current consecutively mapped chunk from @rq. Must
2914 * be called with queue lock held.
2917 * %false - we are done with this request
2918 * %true - still buffers pending for this request
2920 bool __blk_end_request_cur(struct request
*rq
, int error
)
2922 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2924 EXPORT_SYMBOL(__blk_end_request_cur
);
2927 * __blk_end_request_err - Finish a request till the next failure boundary.
2928 * @rq: the request to finish till the next failure boundary for
2929 * @error: must be negative errno
2932 * Complete @rq till the next failure boundary. Must be called
2933 * with queue lock held.
2936 * %false - we are done with this request
2937 * %true - still buffers pending for this request
2939 bool __blk_end_request_err(struct request
*rq
, int error
)
2941 WARN_ON(error
>= 0);
2942 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2944 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2946 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2949 if (bio_has_data(bio
))
2950 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2952 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2953 rq
->bio
= rq
->biotail
= bio
;
2956 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2959 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2961 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2962 * @rq: the request to be flushed
2965 * Flush all pages in @rq.
2967 void rq_flush_dcache_pages(struct request
*rq
)
2969 struct req_iterator iter
;
2970 struct bio_vec bvec
;
2972 rq_for_each_segment(bvec
, rq
, iter
)
2973 flush_dcache_page(bvec
.bv_page
);
2975 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2979 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2980 * @q : the queue of the device being checked
2983 * Check if underlying low-level drivers of a device are busy.
2984 * If the drivers want to export their busy state, they must set own
2985 * exporting function using blk_queue_lld_busy() first.
2987 * Basically, this function is used only by request stacking drivers
2988 * to stop dispatching requests to underlying devices when underlying
2989 * devices are busy. This behavior helps more I/O merging on the queue
2990 * of the request stacking driver and prevents I/O throughput regression
2991 * on burst I/O load.
2994 * 0 - Not busy (The request stacking driver should dispatch request)
2995 * 1 - Busy (The request stacking driver should stop dispatching request)
2997 int blk_lld_busy(struct request_queue
*q
)
3000 return q
->lld_busy_fn(q
);
3004 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3007 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3008 * @rq: the clone request to be cleaned up
3011 * Free all bios in @rq for a cloned request.
3013 void blk_rq_unprep_clone(struct request
*rq
)
3017 while ((bio
= rq
->bio
) != NULL
) {
3018 rq
->bio
= bio
->bi_next
;
3023 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3026 * Copy attributes of the original request to the clone request.
3027 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3029 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3031 dst
->cpu
= src
->cpu
;
3032 dst
->__sector
= blk_rq_pos(src
);
3033 dst
->__data_len
= blk_rq_bytes(src
);
3034 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3035 dst
->ioprio
= src
->ioprio
;
3036 dst
->extra_len
= src
->extra_len
;
3040 * blk_rq_prep_clone - Helper function to setup clone request
3041 * @rq: the request to be setup
3042 * @rq_src: original request to be cloned
3043 * @bs: bio_set that bios for clone are allocated from
3044 * @gfp_mask: memory allocation mask for bio
3045 * @bio_ctr: setup function to be called for each clone bio.
3046 * Returns %0 for success, non %0 for failure.
3047 * @data: private data to be passed to @bio_ctr
3050 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3051 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3052 * are not copied, and copying such parts is the caller's responsibility.
3053 * Also, pages which the original bios are pointing to are not copied
3054 * and the cloned bios just point same pages.
3055 * So cloned bios must be completed before original bios, which means
3056 * the caller must complete @rq before @rq_src.
3058 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3059 struct bio_set
*bs
, gfp_t gfp_mask
,
3060 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3063 struct bio
*bio
, *bio_src
;
3068 __rq_for_each_bio(bio_src
, rq_src
) {
3069 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3073 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3077 rq
->biotail
->bi_next
= bio
;
3080 rq
->bio
= rq
->biotail
= bio
;
3083 __blk_rq_prep_clone(rq
, rq_src
);
3090 blk_rq_unprep_clone(rq
);
3094 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3096 int kblockd_schedule_work(struct work_struct
*work
)
3098 return queue_work(kblockd_workqueue
, work
);
3100 EXPORT_SYMBOL(kblockd_schedule_work
);
3102 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3104 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3106 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3108 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3109 unsigned long delay
)
3111 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3113 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3115 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3116 unsigned long delay
)
3118 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3120 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3123 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3124 * @plug: The &struct blk_plug that needs to be initialized
3127 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3128 * pending I/O should the task end up blocking between blk_start_plug() and
3129 * blk_finish_plug(). This is important from a performance perspective, but
3130 * also ensures that we don't deadlock. For instance, if the task is blocking
3131 * for a memory allocation, memory reclaim could end up wanting to free a
3132 * page belonging to that request that is currently residing in our private
3133 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3134 * this kind of deadlock.
3136 void blk_start_plug(struct blk_plug
*plug
)
3138 struct task_struct
*tsk
= current
;
3141 * If this is a nested plug, don't actually assign it.
3146 INIT_LIST_HEAD(&plug
->list
);
3147 INIT_LIST_HEAD(&plug
->mq_list
);
3148 INIT_LIST_HEAD(&plug
->cb_list
);
3150 * Store ordering should not be needed here, since a potential
3151 * preempt will imply a full memory barrier
3155 EXPORT_SYMBOL(blk_start_plug
);
3157 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3159 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3160 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3162 return !(rqa
->q
< rqb
->q
||
3163 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3167 * If 'from_schedule' is true, then postpone the dispatch of requests
3168 * until a safe kblockd context. We due this to avoid accidental big
3169 * additional stack usage in driver dispatch, in places where the originally
3170 * plugger did not intend it.
3172 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3174 __releases(q
->queue_lock
)
3176 trace_block_unplug(q
, depth
, !from_schedule
);
3179 blk_run_queue_async(q
);
3182 spin_unlock(q
->queue_lock
);
3185 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3187 LIST_HEAD(callbacks
);
3189 while (!list_empty(&plug
->cb_list
)) {
3190 list_splice_init(&plug
->cb_list
, &callbacks
);
3192 while (!list_empty(&callbacks
)) {
3193 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3196 list_del(&cb
->list
);
3197 cb
->callback(cb
, from_schedule
);
3202 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3205 struct blk_plug
*plug
= current
->plug
;
3206 struct blk_plug_cb
*cb
;
3211 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3212 if (cb
->callback
== unplug
&& cb
->data
== data
)
3215 /* Not currently on the callback list */
3216 BUG_ON(size
< sizeof(*cb
));
3217 cb
= kzalloc(size
, GFP_ATOMIC
);
3220 cb
->callback
= unplug
;
3221 list_add(&cb
->list
, &plug
->cb_list
);
3225 EXPORT_SYMBOL(blk_check_plugged
);
3227 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3229 struct request_queue
*q
;
3230 unsigned long flags
;
3235 flush_plug_callbacks(plug
, from_schedule
);
3237 if (!list_empty(&plug
->mq_list
))
3238 blk_mq_flush_plug_list(plug
, from_schedule
);
3240 if (list_empty(&plug
->list
))
3243 list_splice_init(&plug
->list
, &list
);
3245 list_sort(NULL
, &list
, plug_rq_cmp
);
3251 * Save and disable interrupts here, to avoid doing it for every
3252 * queue lock we have to take.
3254 local_irq_save(flags
);
3255 while (!list_empty(&list
)) {
3256 rq
= list_entry_rq(list
.next
);
3257 list_del_init(&rq
->queuelist
);
3261 * This drops the queue lock
3264 queue_unplugged(q
, depth
, from_schedule
);
3267 spin_lock(q
->queue_lock
);
3271 * Short-circuit if @q is dead
3273 if (unlikely(blk_queue_dying(q
))) {
3274 __blk_end_request_all(rq
, -ENODEV
);
3279 * rq is already accounted, so use raw insert
3281 if (op_is_flush(rq
->cmd_flags
))
3282 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3284 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3290 * This drops the queue lock
3293 queue_unplugged(q
, depth
, from_schedule
);
3295 local_irq_restore(flags
);
3298 void blk_finish_plug(struct blk_plug
*plug
)
3300 if (plug
!= current
->plug
)
3302 blk_flush_plug_list(plug
, false);
3304 current
->plug
= NULL
;
3306 EXPORT_SYMBOL(blk_finish_plug
);
3310 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3311 * @q: the queue of the device
3312 * @dev: the device the queue belongs to
3315 * Initialize runtime-PM-related fields for @q and start auto suspend for
3316 * @dev. Drivers that want to take advantage of request-based runtime PM
3317 * should call this function after @dev has been initialized, and its
3318 * request queue @q has been allocated, and runtime PM for it can not happen
3319 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3320 * cases, driver should call this function before any I/O has taken place.
3322 * This function takes care of setting up using auto suspend for the device,
3323 * the autosuspend delay is set to -1 to make runtime suspend impossible
3324 * until an updated value is either set by user or by driver. Drivers do
3325 * not need to touch other autosuspend settings.
3327 * The block layer runtime PM is request based, so only works for drivers
3328 * that use request as their IO unit instead of those directly use bio's.
3330 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3333 q
->rpm_status
= RPM_ACTIVE
;
3334 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3335 pm_runtime_use_autosuspend(q
->dev
);
3337 EXPORT_SYMBOL(blk_pm_runtime_init
);
3340 * blk_pre_runtime_suspend - Pre runtime suspend check
3341 * @q: the queue of the device
3344 * This function will check if runtime suspend is allowed for the device
3345 * by examining if there are any requests pending in the queue. If there
3346 * are requests pending, the device can not be runtime suspended; otherwise,
3347 * the queue's status will be updated to SUSPENDING and the driver can
3348 * proceed to suspend the device.
3350 * For the not allowed case, we mark last busy for the device so that
3351 * runtime PM core will try to autosuspend it some time later.
3353 * This function should be called near the start of the device's
3354 * runtime_suspend callback.
3357 * 0 - OK to runtime suspend the device
3358 * -EBUSY - Device should not be runtime suspended
3360 int blk_pre_runtime_suspend(struct request_queue
*q
)
3367 spin_lock_irq(q
->queue_lock
);
3368 if (q
->nr_pending
) {
3370 pm_runtime_mark_last_busy(q
->dev
);
3372 q
->rpm_status
= RPM_SUSPENDING
;
3374 spin_unlock_irq(q
->queue_lock
);
3377 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3380 * blk_post_runtime_suspend - Post runtime suspend processing
3381 * @q: the queue of the device
3382 * @err: return value of the device's runtime_suspend function
3385 * Update the queue's runtime status according to the return value of the
3386 * device's runtime suspend function and mark last busy for the device so
3387 * that PM core will try to auto suspend the device at a later time.
3389 * This function should be called near the end of the device's
3390 * runtime_suspend callback.
3392 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3397 spin_lock_irq(q
->queue_lock
);
3399 q
->rpm_status
= RPM_SUSPENDED
;
3401 q
->rpm_status
= RPM_ACTIVE
;
3402 pm_runtime_mark_last_busy(q
->dev
);
3404 spin_unlock_irq(q
->queue_lock
);
3406 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3409 * blk_pre_runtime_resume - Pre runtime resume processing
3410 * @q: the queue of the device
3413 * Update the queue's runtime status to RESUMING in preparation for the
3414 * runtime resume of the device.
3416 * This function should be called near the start of the device's
3417 * runtime_resume callback.
3419 void blk_pre_runtime_resume(struct request_queue
*q
)
3424 spin_lock_irq(q
->queue_lock
);
3425 q
->rpm_status
= RPM_RESUMING
;
3426 spin_unlock_irq(q
->queue_lock
);
3428 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3431 * blk_post_runtime_resume - Post runtime resume processing
3432 * @q: the queue of the device
3433 * @err: return value of the device's runtime_resume function
3436 * Update the queue's runtime status according to the return value of the
3437 * device's runtime_resume function. If it is successfully resumed, process
3438 * the requests that are queued into the device's queue when it is resuming
3439 * and then mark last busy and initiate autosuspend for it.
3441 * This function should be called near the end of the device's
3442 * runtime_resume callback.
3444 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3449 spin_lock_irq(q
->queue_lock
);
3451 q
->rpm_status
= RPM_ACTIVE
;
3453 pm_runtime_mark_last_busy(q
->dev
);
3454 pm_request_autosuspend(q
->dev
);
3456 q
->rpm_status
= RPM_SUSPENDED
;
3458 spin_unlock_irq(q
->queue_lock
);
3460 EXPORT_SYMBOL(blk_post_runtime_resume
);
3463 * blk_set_runtime_active - Force runtime status of the queue to be active
3464 * @q: the queue of the device
3466 * If the device is left runtime suspended during system suspend the resume
3467 * hook typically resumes the device and corrects runtime status
3468 * accordingly. However, that does not affect the queue runtime PM status
3469 * which is still "suspended". This prevents processing requests from the
3472 * This function can be used in driver's resume hook to correct queue
3473 * runtime PM status and re-enable peeking requests from the queue. It
3474 * should be called before first request is added to the queue.
3476 void blk_set_runtime_active(struct request_queue
*q
)
3478 spin_lock_irq(q
->queue_lock
);
3479 q
->rpm_status
= RPM_ACTIVE
;
3480 pm_runtime_mark_last_busy(q
->dev
);
3481 pm_request_autosuspend(q
->dev
);
3482 spin_unlock_irq(q
->queue_lock
);
3484 EXPORT_SYMBOL(blk_set_runtime_active
);
3487 int __init
blk_dev_init(void)
3489 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3490 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3491 FIELD_SIZEOF(struct request
, cmd_flags
));
3492 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3493 FIELD_SIZEOF(struct bio
, bi_opf
));
3495 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3496 kblockd_workqueue
= alloc_workqueue("kblockd",
3497 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3498 if (!kblockd_workqueue
)
3499 panic("Failed to create kblockd\n");
3501 request_cachep
= kmem_cache_create("blkdev_requests",
3502 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3504 blk_requestq_cachep
= kmem_cache_create("request_queue",
3505 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3507 #ifdef CONFIG_DEBUG_FS
3508 blk_debugfs_root
= debugfs_create_dir("block", NULL
);