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>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/block.h>
42 #include "blk-mq-sched.h"
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
48 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
49 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
51 DEFINE_IDA(blk_queue_ida
);
54 * For the allocated request tables
56 struct kmem_cache
*request_cachep
;
59 * For queue allocation
61 struct kmem_cache
*blk_requestq_cachep
;
64 * Controlling structure to kblockd
66 static struct workqueue_struct
*kblockd_workqueue
;
68 static void blk_clear_congested(struct request_list
*rl
, int sync
)
70 #ifdef CONFIG_CGROUP_WRITEBACK
71 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
74 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
75 * flip its congestion state for events on other blkcgs.
77 if (rl
== &rl
->q
->root_rl
)
78 clear_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
82 static void blk_set_congested(struct request_list
*rl
, int sync
)
84 #ifdef CONFIG_CGROUP_WRITEBACK
85 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
87 /* see blk_clear_congested() */
88 if (rl
== &rl
->q
->root_rl
)
89 set_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
93 void blk_queue_congestion_threshold(struct request_queue
*q
)
97 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
98 if (nr
> q
->nr_requests
)
100 q
->nr_congestion_on
= nr
;
102 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
105 q
->nr_congestion_off
= nr
;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
113 * backing_dev_info. This function can only be called if @bdev is opened
114 * and the return value is never NULL.
116 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
118 struct request_queue
*q
= bdev_get_queue(bdev
);
120 return &q
->backing_dev_info
;
122 EXPORT_SYMBOL(blk_get_backing_dev_info
);
124 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
126 memset(rq
, 0, sizeof(*rq
));
128 INIT_LIST_HEAD(&rq
->queuelist
);
129 INIT_LIST_HEAD(&rq
->timeout_list
);
132 rq
->__sector
= (sector_t
) -1;
133 INIT_HLIST_NODE(&rq
->hash
);
134 RB_CLEAR_NODE(&rq
->rb_node
);
136 rq
->cmd_len
= BLK_MAX_CDB
;
138 rq
->internal_tag
= -1;
139 rq
->start_time
= jiffies
;
140 set_start_time_ns(rq
);
143 EXPORT_SYMBOL(blk_rq_init
);
145 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
146 unsigned int nbytes
, int error
)
149 bio
->bi_error
= error
;
151 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
152 bio_set_flag(bio
, BIO_QUIET
);
154 bio_advance(bio
, nbytes
);
156 /* don't actually finish bio if it's part of flush sequence */
157 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
161 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
165 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
166 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
167 (unsigned long long) rq
->cmd_flags
);
169 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
170 (unsigned long long)blk_rq_pos(rq
),
171 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
172 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
173 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
175 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
176 printk(KERN_INFO
" cdb: ");
177 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
178 printk("%02x ", rq
->cmd
[bit
]);
182 EXPORT_SYMBOL(blk_dump_rq_flags
);
184 static void blk_delay_work(struct work_struct
*work
)
186 struct request_queue
*q
;
188 q
= container_of(work
, struct request_queue
, delay_work
.work
);
189 spin_lock_irq(q
->queue_lock
);
191 spin_unlock_irq(q
->queue_lock
);
195 * blk_delay_queue - restart queueing after defined interval
196 * @q: The &struct request_queue in question
197 * @msecs: Delay in msecs
200 * Sometimes queueing needs to be postponed for a little while, to allow
201 * resources to come back. This function will make sure that queueing is
202 * restarted around the specified time. Queue lock must be held.
204 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
206 if (likely(!blk_queue_dead(q
)))
207 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
208 msecs_to_jiffies(msecs
));
210 EXPORT_SYMBOL(blk_delay_queue
);
213 * blk_start_queue_async - asynchronously restart a previously stopped queue
214 * @q: The &struct request_queue in question
217 * blk_start_queue_async() will clear the stop flag on the queue, and
218 * ensure that the request_fn for the queue is run from an async
221 void blk_start_queue_async(struct request_queue
*q
)
223 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
224 blk_run_queue_async(q
);
226 EXPORT_SYMBOL(blk_start_queue_async
);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue
*q
)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
244 EXPORT_SYMBOL(blk_start_queue
);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue
*q
)
262 cancel_delayed_work(&q
->delay_work
);
263 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
265 EXPORT_SYMBOL(blk_stop_queue
);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevator_exit()
282 * and blkcg_exit_queue() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue
*q
)
287 del_timer_sync(&q
->timeout
);
290 struct blk_mq_hw_ctx
*hctx
;
293 queue_for_each_hw_ctx(q
, hctx
, i
) {
294 cancel_work_sync(&hctx
->run_work
);
295 cancel_delayed_work_sync(&hctx
->delay_work
);
298 cancel_delayed_work_sync(&q
->delay_work
);
301 EXPORT_SYMBOL(blk_sync_queue
);
304 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
305 * @q: The queue to run
308 * Invoke request handling on a queue if there are any pending requests.
309 * May be used to restart request handling after a request has completed.
310 * This variant runs the queue whether or not the queue has been
311 * stopped. Must be called with the queue lock held and interrupts
312 * disabled. See also @blk_run_queue.
314 inline void __blk_run_queue_uncond(struct request_queue
*q
)
316 if (unlikely(blk_queue_dead(q
)))
320 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
321 * the queue lock internally. As a result multiple threads may be
322 * running such a request function concurrently. Keep track of the
323 * number of active request_fn invocations such that blk_drain_queue()
324 * can wait until all these request_fn calls have finished.
326 q
->request_fn_active
++;
328 q
->request_fn_active
--;
330 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
333 * __blk_run_queue - run a single device queue
334 * @q: The queue to run
337 * See @blk_run_queue. This variant must be called with the queue lock
338 * held and interrupts disabled.
340 void __blk_run_queue(struct request_queue
*q
)
342 if (unlikely(blk_queue_stopped(q
)))
345 __blk_run_queue_uncond(q
);
347 EXPORT_SYMBOL(__blk_run_queue
);
350 * blk_run_queue_async - run a single device queue in workqueue context
351 * @q: The queue to run
354 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
355 * of us. The caller must hold the queue lock.
357 void blk_run_queue_async(struct request_queue
*q
)
359 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
360 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
362 EXPORT_SYMBOL(blk_run_queue_async
);
365 * blk_run_queue - run a single device queue
366 * @q: The queue to run
369 * Invoke request handling on this queue, if it has pending work to do.
370 * May be used to restart queueing when a request has completed.
372 void blk_run_queue(struct request_queue
*q
)
376 spin_lock_irqsave(q
->queue_lock
, flags
);
378 spin_unlock_irqrestore(q
->queue_lock
, flags
);
380 EXPORT_SYMBOL(blk_run_queue
);
382 void blk_put_queue(struct request_queue
*q
)
384 kobject_put(&q
->kobj
);
386 EXPORT_SYMBOL(blk_put_queue
);
389 * __blk_drain_queue - drain requests from request_queue
391 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
393 * Drain requests from @q. If @drain_all is set, all requests are drained.
394 * If not, only ELVPRIV requests are drained. The caller is responsible
395 * for ensuring that no new requests which need to be drained are queued.
397 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
398 __releases(q
->queue_lock
)
399 __acquires(q
->queue_lock
)
403 lockdep_assert_held(q
->queue_lock
);
409 * The caller might be trying to drain @q before its
410 * elevator is initialized.
413 elv_drain_elevator(q
);
415 blkcg_drain_queue(q
);
418 * This function might be called on a queue which failed
419 * driver init after queue creation or is not yet fully
420 * active yet. Some drivers (e.g. fd and loop) get unhappy
421 * in such cases. Kick queue iff dispatch queue has
422 * something on it and @q has request_fn set.
424 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
427 drain
|= q
->nr_rqs_elvpriv
;
428 drain
|= q
->request_fn_active
;
431 * Unfortunately, requests are queued at and tracked from
432 * multiple places and there's no single counter which can
433 * be drained. Check all the queues and counters.
436 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
437 drain
|= !list_empty(&q
->queue_head
);
438 for (i
= 0; i
< 2; i
++) {
439 drain
|= q
->nr_rqs
[i
];
440 drain
|= q
->in_flight
[i
];
442 drain
|= !list_empty(&fq
->flush_queue
[i
]);
449 spin_unlock_irq(q
->queue_lock
);
453 spin_lock_irq(q
->queue_lock
);
457 * With queue marked dead, any woken up waiter will fail the
458 * allocation path, so the wakeup chaining is lost and we're
459 * left with hung waiters. We need to wake up those waiters.
462 struct request_list
*rl
;
464 blk_queue_for_each_rl(rl
, q
)
465 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
466 wake_up_all(&rl
->wait
[i
]);
471 * blk_queue_bypass_start - enter queue bypass mode
472 * @q: queue of interest
474 * In bypass mode, only the dispatch FIFO queue of @q is used. This
475 * function makes @q enter bypass mode and drains all requests which were
476 * throttled or issued before. On return, it's guaranteed that no request
477 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
478 * inside queue or RCU read lock.
480 void blk_queue_bypass_start(struct request_queue
*q
)
482 spin_lock_irq(q
->queue_lock
);
484 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
485 spin_unlock_irq(q
->queue_lock
);
488 * Queues start drained. Skip actual draining till init is
489 * complete. This avoids lenghty delays during queue init which
490 * can happen many times during boot.
492 if (blk_queue_init_done(q
)) {
493 spin_lock_irq(q
->queue_lock
);
494 __blk_drain_queue(q
, false);
495 spin_unlock_irq(q
->queue_lock
);
497 /* ensure blk_queue_bypass() is %true inside RCU read lock */
501 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
504 * blk_queue_bypass_end - leave queue bypass mode
505 * @q: queue of interest
507 * Leave bypass mode and restore the normal queueing behavior.
509 void blk_queue_bypass_end(struct request_queue
*q
)
511 spin_lock_irq(q
->queue_lock
);
512 if (!--q
->bypass_depth
)
513 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
514 WARN_ON_ONCE(q
->bypass_depth
< 0);
515 spin_unlock_irq(q
->queue_lock
);
517 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
519 void blk_set_queue_dying(struct request_queue
*q
)
521 spin_lock_irq(q
->queue_lock
);
522 queue_flag_set(QUEUE_FLAG_DYING
, q
);
523 spin_unlock_irq(q
->queue_lock
);
526 blk_mq_wake_waiters(q
);
528 struct request_list
*rl
;
530 blk_queue_for_each_rl(rl
, q
) {
532 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
533 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
538 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
541 * blk_cleanup_queue - shutdown a request queue
542 * @q: request queue to shutdown
544 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
545 * put it. All future requests will be failed immediately with -ENODEV.
547 void blk_cleanup_queue(struct request_queue
*q
)
549 spinlock_t
*lock
= q
->queue_lock
;
551 /* mark @q DYING, no new request or merges will be allowed afterwards */
552 mutex_lock(&q
->sysfs_lock
);
553 blk_set_queue_dying(q
);
557 * A dying queue is permanently in bypass mode till released. Note
558 * that, unlike blk_queue_bypass_start(), we aren't performing
559 * synchronize_rcu() after entering bypass mode to avoid the delay
560 * as some drivers create and destroy a lot of queues while
561 * probing. This is still safe because blk_release_queue() will be
562 * called only after the queue refcnt drops to zero and nothing,
563 * RCU or not, would be traversing the queue by then.
566 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
568 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
569 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
570 queue_flag_set(QUEUE_FLAG_DYING
, q
);
571 spin_unlock_irq(lock
);
572 mutex_unlock(&q
->sysfs_lock
);
575 * Drain all requests queued before DYING marking. Set DEAD flag to
576 * prevent that q->request_fn() gets invoked after draining finished.
581 __blk_drain_queue(q
, true);
582 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
583 spin_unlock_irq(lock
);
585 /* for synchronous bio-based driver finish in-flight integrity i/o */
586 blk_flush_integrity();
588 /* @q won't process any more request, flush async actions */
589 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
593 blk_mq_free_queue(q
);
594 percpu_ref_exit(&q
->q_usage_counter
);
597 if (q
->queue_lock
!= &q
->__queue_lock
)
598 q
->queue_lock
= &q
->__queue_lock
;
599 spin_unlock_irq(lock
);
601 bdi_unregister(&q
->backing_dev_info
);
603 /* @q is and will stay empty, shutdown and put */
606 EXPORT_SYMBOL(blk_cleanup_queue
);
608 /* Allocate memory local to the request queue */
609 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
611 int nid
= (int)(long)data
;
612 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
615 static void free_request_struct(void *element
, void *unused
)
617 kmem_cache_free(request_cachep
, element
);
620 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
623 if (unlikely(rl
->rq_pool
))
627 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
628 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
629 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
630 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
632 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
634 (void *)(long)q
->node
, gfp_mask
,
642 void blk_exit_rl(struct request_list
*rl
)
645 mempool_destroy(rl
->rq_pool
);
648 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
650 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
652 EXPORT_SYMBOL(blk_alloc_queue
);
654 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
659 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
665 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
666 !atomic_read(&q
->mq_freeze_depth
) ||
668 if (blk_queue_dying(q
))
675 void blk_queue_exit(struct request_queue
*q
)
677 percpu_ref_put(&q
->q_usage_counter
);
680 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
682 struct request_queue
*q
=
683 container_of(ref
, struct request_queue
, q_usage_counter
);
685 wake_up_all(&q
->mq_freeze_wq
);
688 static void blk_rq_timed_out_timer(unsigned long data
)
690 struct request_queue
*q
= (struct request_queue
*)data
;
692 kblockd_schedule_work(&q
->timeout_work
);
695 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
697 struct request_queue
*q
;
700 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
701 gfp_mask
| __GFP_ZERO
, node_id
);
705 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
709 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
713 q
->backing_dev_info
.ra_pages
=
714 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
715 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
716 q
->backing_dev_info
.name
= "block";
719 err
= bdi_init(&q
->backing_dev_info
);
723 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
724 laptop_mode_timer_fn
, (unsigned long) q
);
725 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
726 INIT_LIST_HEAD(&q
->queue_head
);
727 INIT_LIST_HEAD(&q
->timeout_list
);
728 INIT_LIST_HEAD(&q
->icq_list
);
729 #ifdef CONFIG_BLK_CGROUP
730 INIT_LIST_HEAD(&q
->blkg_list
);
732 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
734 kobject_init(&q
->kobj
, &blk_queue_ktype
);
736 mutex_init(&q
->sysfs_lock
);
737 spin_lock_init(&q
->__queue_lock
);
740 * By default initialize queue_lock to internal lock and driver can
741 * override it later if need be.
743 q
->queue_lock
= &q
->__queue_lock
;
746 * A queue starts its life with bypass turned on to avoid
747 * unnecessary bypass on/off overhead and nasty surprises during
748 * init. The initial bypass will be finished when the queue is
749 * registered by blk_register_queue().
752 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
754 init_waitqueue_head(&q
->mq_freeze_wq
);
757 * Init percpu_ref in atomic mode so that it's faster to shutdown.
758 * See blk_register_queue() for details.
760 if (percpu_ref_init(&q
->q_usage_counter
,
761 blk_queue_usage_counter_release
,
762 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
765 if (blkcg_init_queue(q
))
771 percpu_ref_exit(&q
->q_usage_counter
);
773 bdi_destroy(&q
->backing_dev_info
);
775 bioset_free(q
->bio_split
);
777 ida_simple_remove(&blk_queue_ida
, q
->id
);
779 kmem_cache_free(blk_requestq_cachep
, q
);
782 EXPORT_SYMBOL(blk_alloc_queue_node
);
785 * blk_init_queue - prepare a request queue for use with a block device
786 * @rfn: The function to be called to process requests that have been
787 * placed on the queue.
788 * @lock: Request queue spin lock
791 * If a block device wishes to use the standard request handling procedures,
792 * which sorts requests and coalesces adjacent requests, then it must
793 * call blk_init_queue(). The function @rfn will be called when there
794 * are requests on the queue that need to be processed. If the device
795 * supports plugging, then @rfn may not be called immediately when requests
796 * are available on the queue, but may be called at some time later instead.
797 * Plugged queues are generally unplugged when a buffer belonging to one
798 * of the requests on the queue is needed, or due to memory pressure.
800 * @rfn is not required, or even expected, to remove all requests off the
801 * queue, but only as many as it can handle at a time. If it does leave
802 * requests on the queue, it is responsible for arranging that the requests
803 * get dealt with eventually.
805 * The queue spin lock must be held while manipulating the requests on the
806 * request queue; this lock will be taken also from interrupt context, so irq
807 * disabling is needed for it.
809 * Function returns a pointer to the initialized request queue, or %NULL if
813 * blk_init_queue() must be paired with a blk_cleanup_queue() call
814 * when the block device is deactivated (such as at module unload).
817 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
819 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
821 EXPORT_SYMBOL(blk_init_queue
);
823 struct request_queue
*
824 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
826 struct request_queue
*uninit_q
, *q
;
828 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
832 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
834 blk_cleanup_queue(uninit_q
);
838 EXPORT_SYMBOL(blk_init_queue_node
);
840 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
842 struct request_queue
*
843 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
849 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
853 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
856 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
858 q
->prep_rq_fn
= NULL
;
859 q
->unprep_rq_fn
= NULL
;
860 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
862 /* Override internal queue lock with supplied lock pointer */
864 q
->queue_lock
= lock
;
867 * This also sets hw/phys segments, boundary and size
869 blk_queue_make_request(q
, blk_queue_bio
);
871 q
->sg_reserved_size
= INT_MAX
;
873 /* Protect q->elevator from elevator_change */
874 mutex_lock(&q
->sysfs_lock
);
877 if (elevator_init(q
, NULL
)) {
878 mutex_unlock(&q
->sysfs_lock
);
882 mutex_unlock(&q
->sysfs_lock
);
887 blk_free_flush_queue(q
->fq
);
891 EXPORT_SYMBOL(blk_init_allocated_queue
);
893 bool blk_get_queue(struct request_queue
*q
)
895 if (likely(!blk_queue_dying(q
))) {
902 EXPORT_SYMBOL(blk_get_queue
);
904 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
906 if (rq
->rq_flags
& RQF_ELVPRIV
) {
907 elv_put_request(rl
->q
, rq
);
909 put_io_context(rq
->elv
.icq
->ioc
);
912 mempool_free(rq
, rl
->rq_pool
);
916 * ioc_batching returns true if the ioc is a valid batching request and
917 * should be given priority access to a request.
919 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
925 * Make sure the process is able to allocate at least 1 request
926 * even if the batch times out, otherwise we could theoretically
929 return ioc
->nr_batch_requests
== q
->nr_batching
||
930 (ioc
->nr_batch_requests
> 0
931 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
935 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
936 * will cause the process to be a "batcher" on all queues in the system. This
937 * is the behaviour we want though - once it gets a wakeup it should be given
940 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
942 if (!ioc
|| ioc_batching(q
, ioc
))
945 ioc
->nr_batch_requests
= q
->nr_batching
;
946 ioc
->last_waited
= jiffies
;
949 static void __freed_request(struct request_list
*rl
, int sync
)
951 struct request_queue
*q
= rl
->q
;
953 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
954 blk_clear_congested(rl
, sync
);
956 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
957 if (waitqueue_active(&rl
->wait
[sync
]))
958 wake_up(&rl
->wait
[sync
]);
960 blk_clear_rl_full(rl
, sync
);
965 * A request has just been released. Account for it, update the full and
966 * congestion status, wake up any waiters. Called under q->queue_lock.
968 static void freed_request(struct request_list
*rl
, bool sync
,
969 req_flags_t rq_flags
)
971 struct request_queue
*q
= rl
->q
;
975 if (rq_flags
& RQF_ELVPRIV
)
978 __freed_request(rl
, sync
);
980 if (unlikely(rl
->starved
[sync
^ 1]))
981 __freed_request(rl
, sync
^ 1);
984 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
986 struct request_list
*rl
;
987 int on_thresh
, off_thresh
;
989 spin_lock_irq(q
->queue_lock
);
991 blk_queue_congestion_threshold(q
);
992 on_thresh
= queue_congestion_on_threshold(q
);
993 off_thresh
= queue_congestion_off_threshold(q
);
995 blk_queue_for_each_rl(rl
, q
) {
996 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
997 blk_set_congested(rl
, BLK_RW_SYNC
);
998 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
999 blk_clear_congested(rl
, BLK_RW_SYNC
);
1001 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1002 blk_set_congested(rl
, BLK_RW_ASYNC
);
1003 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1004 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1006 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1007 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1009 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1010 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1013 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1014 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1016 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1017 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1021 spin_unlock_irq(q
->queue_lock
);
1026 * Determine if elevator data should be initialized when allocating the
1027 * request associated with @bio.
1029 static bool blk_rq_should_init_elevator(struct bio
*bio
)
1035 * Flush requests do not use the elevator so skip initialization.
1036 * This allows a request to share the flush and elevator data.
1038 if (op_is_flush(bio
->bi_opf
))
1045 * __get_request - get a free request
1046 * @rl: request list to allocate from
1047 * @op: operation and flags
1048 * @bio: bio to allocate request for (can be %NULL)
1049 * @gfp_mask: allocation mask
1051 * Get a free request from @q. This function may fail under memory
1052 * pressure or if @q is dead.
1054 * Must be called with @q->queue_lock held and,
1055 * Returns ERR_PTR on failure, with @q->queue_lock held.
1056 * Returns request pointer on success, with @q->queue_lock *not held*.
1058 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1059 struct bio
*bio
, gfp_t gfp_mask
)
1061 struct request_queue
*q
= rl
->q
;
1063 struct elevator_type
*et
= q
->elevator
->type
;
1064 struct io_context
*ioc
= rq_ioc(bio
);
1065 struct io_cq
*icq
= NULL
;
1066 const bool is_sync
= op_is_sync(op
);
1068 req_flags_t rq_flags
= RQF_ALLOCED
;
1070 if (unlikely(blk_queue_dying(q
)))
1071 return ERR_PTR(-ENODEV
);
1073 may_queue
= elv_may_queue(q
, op
);
1074 if (may_queue
== ELV_MQUEUE_NO
)
1077 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1078 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1080 * The queue will fill after this allocation, so set
1081 * it as full, and mark this process as "batching".
1082 * This process will be allowed to complete a batch of
1083 * requests, others will be blocked.
1085 if (!blk_rl_full(rl
, is_sync
)) {
1086 ioc_set_batching(q
, ioc
);
1087 blk_set_rl_full(rl
, is_sync
);
1089 if (may_queue
!= ELV_MQUEUE_MUST
1090 && !ioc_batching(q
, ioc
)) {
1092 * The queue is full and the allocating
1093 * process is not a "batcher", and not
1094 * exempted by the IO scheduler
1096 return ERR_PTR(-ENOMEM
);
1100 blk_set_congested(rl
, is_sync
);
1104 * Only allow batching queuers to allocate up to 50% over the defined
1105 * limit of requests, otherwise we could have thousands of requests
1106 * allocated with any setting of ->nr_requests
1108 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1109 return ERR_PTR(-ENOMEM
);
1111 q
->nr_rqs
[is_sync
]++;
1112 rl
->count
[is_sync
]++;
1113 rl
->starved
[is_sync
] = 0;
1116 * Decide whether the new request will be managed by elevator. If
1117 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1118 * prevent the current elevator from being destroyed until the new
1119 * request is freed. This guarantees icq's won't be destroyed and
1120 * makes creating new ones safe.
1122 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1123 * it will be created after releasing queue_lock.
1125 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1126 rq_flags
|= RQF_ELVPRIV
;
1127 q
->nr_rqs_elvpriv
++;
1128 if (et
->icq_cache
&& ioc
)
1129 icq
= ioc_lookup_icq(ioc
, q
);
1132 if (blk_queue_io_stat(q
))
1133 rq_flags
|= RQF_IO_STAT
;
1134 spin_unlock_irq(q
->queue_lock
);
1136 /* allocate and init request */
1137 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1142 blk_rq_set_rl(rq
, rl
);
1143 blk_rq_set_prio(rq
, ioc
);
1145 rq
->rq_flags
= rq_flags
;
1148 if (rq_flags
& RQF_ELVPRIV
) {
1149 if (unlikely(et
->icq_cache
&& !icq
)) {
1151 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1157 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1160 /* @rq->elv.icq holds io_context until @rq is freed */
1162 get_io_context(icq
->ioc
);
1166 * ioc may be NULL here, and ioc_batching will be false. That's
1167 * OK, if the queue is under the request limit then requests need
1168 * not count toward the nr_batch_requests limit. There will always
1169 * be some limit enforced by BLK_BATCH_TIME.
1171 if (ioc_batching(q
, ioc
))
1172 ioc
->nr_batch_requests
--;
1174 trace_block_getrq(q
, bio
, op
);
1179 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1180 * and may fail indefinitely under memory pressure and thus
1181 * shouldn't stall IO. Treat this request as !elvpriv. This will
1182 * disturb iosched and blkcg but weird is bettern than dead.
1184 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1185 __func__
, dev_name(q
->backing_dev_info
.dev
));
1187 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1190 spin_lock_irq(q
->queue_lock
);
1191 q
->nr_rqs_elvpriv
--;
1192 spin_unlock_irq(q
->queue_lock
);
1197 * Allocation failed presumably due to memory. Undo anything we
1198 * might have messed up.
1200 * Allocating task should really be put onto the front of the wait
1201 * queue, but this is pretty rare.
1203 spin_lock_irq(q
->queue_lock
);
1204 freed_request(rl
, is_sync
, rq_flags
);
1207 * in the very unlikely event that allocation failed and no
1208 * requests for this direction was pending, mark us starved so that
1209 * freeing of a request in the other direction will notice
1210 * us. another possible fix would be to split the rq mempool into
1214 if (unlikely(rl
->count
[is_sync
] == 0))
1215 rl
->starved
[is_sync
] = 1;
1216 return ERR_PTR(-ENOMEM
);
1220 * get_request - get a free request
1221 * @q: request_queue to allocate request from
1222 * @op: operation and flags
1223 * @bio: bio to allocate request for (can be %NULL)
1224 * @gfp_mask: allocation mask
1226 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1227 * this function keeps retrying under memory pressure and fails iff @q is dead.
1229 * Must be called with @q->queue_lock held and,
1230 * Returns ERR_PTR on failure, with @q->queue_lock held.
1231 * Returns request pointer on success, with @q->queue_lock *not held*.
1233 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1234 struct bio
*bio
, gfp_t gfp_mask
)
1236 const bool is_sync
= op_is_sync(op
);
1238 struct request_list
*rl
;
1241 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1243 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1247 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1252 /* wait on @rl and retry */
1253 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1254 TASK_UNINTERRUPTIBLE
);
1256 trace_block_sleeprq(q
, bio
, op
);
1258 spin_unlock_irq(q
->queue_lock
);
1262 * After sleeping, we become a "batching" process and will be able
1263 * to allocate at least one request, and up to a big batch of them
1264 * for a small period time. See ioc_batching, ioc_set_batching
1266 ioc_set_batching(q
, current
->io_context
);
1268 spin_lock_irq(q
->queue_lock
);
1269 finish_wait(&rl
->wait
[is_sync
], &wait
);
1274 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1279 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1281 /* create ioc upfront */
1282 create_io_context(gfp_mask
, q
->node
);
1284 spin_lock_irq(q
->queue_lock
);
1285 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1287 spin_unlock_irq(q
->queue_lock
);
1291 /* q->queue_lock is unlocked at this point */
1293 rq
->__sector
= (sector_t
) -1;
1294 rq
->bio
= rq
->biotail
= NULL
;
1298 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1301 return blk_mq_alloc_request(q
, rw
,
1302 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1303 0 : BLK_MQ_REQ_NOWAIT
);
1305 return blk_old_get_request(q
, rw
, gfp_mask
);
1307 EXPORT_SYMBOL(blk_get_request
);
1310 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1311 * @rq: request to be initialized
1314 void blk_rq_set_block_pc(struct request
*rq
)
1316 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1317 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1319 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1322 * blk_requeue_request - put a request back on queue
1323 * @q: request queue where request should be inserted
1324 * @rq: request to be inserted
1327 * Drivers often keep queueing requests until the hardware cannot accept
1328 * more, when that condition happens we need to put the request back
1329 * on the queue. Must be called with queue lock held.
1331 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1333 blk_delete_timer(rq
);
1334 blk_clear_rq_complete(rq
);
1335 trace_block_rq_requeue(q
, rq
);
1336 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1338 if (rq
->rq_flags
& RQF_QUEUED
)
1339 blk_queue_end_tag(q
, rq
);
1341 BUG_ON(blk_queued_rq(rq
));
1343 elv_requeue_request(q
, rq
);
1345 EXPORT_SYMBOL(blk_requeue_request
);
1347 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1350 blk_account_io_start(rq
, true);
1351 __elv_add_request(q
, rq
, where
);
1354 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1359 if (now
== part
->stamp
)
1362 inflight
= part_in_flight(part
);
1364 __part_stat_add(cpu
, part
, time_in_queue
,
1365 inflight
* (now
- part
->stamp
));
1366 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1372 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1373 * @cpu: cpu number for stats access
1374 * @part: target partition
1376 * The average IO queue length and utilisation statistics are maintained
1377 * by observing the current state of the queue length and the amount of
1378 * time it has been in this state for.
1380 * Normally, that accounting is done on IO completion, but that can result
1381 * in more than a second's worth of IO being accounted for within any one
1382 * second, leading to >100% utilisation. To deal with that, we call this
1383 * function to do a round-off before returning the results when reading
1384 * /proc/diskstats. This accounts immediately for all queue usage up to
1385 * the current jiffies and restarts the counters again.
1387 void part_round_stats(int cpu
, struct hd_struct
*part
)
1389 unsigned long now
= jiffies
;
1392 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1393 part_round_stats_single(cpu
, part
, now
);
1395 EXPORT_SYMBOL_GPL(part_round_stats
);
1398 static void blk_pm_put_request(struct request
*rq
)
1400 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1401 pm_runtime_mark_last_busy(rq
->q
->dev
);
1404 static inline void blk_pm_put_request(struct request
*rq
) {}
1408 * queue lock must be held
1410 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1412 req_flags_t rq_flags
= req
->rq_flags
;
1418 blk_mq_free_request(req
);
1422 blk_pm_put_request(req
);
1424 elv_completed_request(q
, req
);
1426 /* this is a bio leak */
1427 WARN_ON(req
->bio
!= NULL
);
1429 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1432 * Request may not have originated from ll_rw_blk. if not,
1433 * it didn't come out of our reserved rq pools
1435 if (rq_flags
& RQF_ALLOCED
) {
1436 struct request_list
*rl
= blk_rq_rl(req
);
1437 bool sync
= op_is_sync(req
->cmd_flags
);
1439 BUG_ON(!list_empty(&req
->queuelist
));
1440 BUG_ON(ELV_ON_HASH(req
));
1442 blk_free_request(rl
, req
);
1443 freed_request(rl
, sync
, rq_flags
);
1447 EXPORT_SYMBOL_GPL(__blk_put_request
);
1449 void blk_put_request(struct request
*req
)
1451 struct request_queue
*q
= req
->q
;
1454 blk_mq_free_request(req
);
1456 unsigned long flags
;
1458 spin_lock_irqsave(q
->queue_lock
, flags
);
1459 __blk_put_request(q
, req
);
1460 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1463 EXPORT_SYMBOL(blk_put_request
);
1465 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1468 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1470 if (!ll_back_merge_fn(q
, req
, bio
))
1473 trace_block_bio_backmerge(q
, req
, bio
);
1475 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1476 blk_rq_set_mixed_merge(req
);
1478 req
->biotail
->bi_next
= bio
;
1480 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1481 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1483 blk_account_io_start(req
, false);
1487 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1490 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1492 if (!ll_front_merge_fn(q
, req
, bio
))
1495 trace_block_bio_frontmerge(q
, req
, bio
);
1497 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1498 blk_rq_set_mixed_merge(req
);
1500 bio
->bi_next
= req
->bio
;
1503 req
->__sector
= bio
->bi_iter
.bi_sector
;
1504 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1505 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1507 blk_account_io_start(req
, false);
1512 * blk_attempt_plug_merge - try to merge with %current's plugged list
1513 * @q: request_queue new bio is being queued at
1514 * @bio: new bio being queued
1515 * @request_count: out parameter for number of traversed plugged requests
1516 * @same_queue_rq: pointer to &struct request that gets filled in when
1517 * another request associated with @q is found on the plug list
1518 * (optional, may be %NULL)
1520 * Determine whether @bio being queued on @q can be merged with a request
1521 * on %current's plugged list. Returns %true if merge was successful,
1524 * Plugging coalesces IOs from the same issuer for the same purpose without
1525 * going through @q->queue_lock. As such it's more of an issuing mechanism
1526 * than scheduling, and the request, while may have elvpriv data, is not
1527 * added on the elevator at this point. In addition, we don't have
1528 * reliable access to the elevator outside queue lock. Only check basic
1529 * merging parameters without querying the elevator.
1531 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1533 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1534 unsigned int *request_count
,
1535 struct request
**same_queue_rq
)
1537 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
) {
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 el_ret
= blk_try_merge(rq
, bio
);
1570 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1571 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1574 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1575 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1584 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1586 struct blk_plug
*plug
;
1588 struct list_head
*plug_list
;
1589 unsigned int ret
= 0;
1591 plug
= current
->plug
;
1596 plug_list
= &plug
->mq_list
;
1598 plug_list
= &plug
->list
;
1600 list_for_each_entry(rq
, plug_list
, queuelist
) {
1608 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1610 req
->cmd_type
= REQ_TYPE_FS
;
1611 if (bio
->bi_opf
& REQ_RAHEAD
)
1612 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1615 req
->__sector
= bio
->bi_iter
.bi_sector
;
1616 if (ioprio_valid(bio_prio(bio
)))
1617 req
->ioprio
= bio_prio(bio
);
1618 blk_rq_bio_prep(req
->q
, req
, bio
);
1621 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1623 struct blk_plug
*plug
;
1624 int el_ret
, where
= ELEVATOR_INSERT_SORT
;
1625 struct request
*req
;
1626 unsigned int request_count
= 0;
1627 unsigned int wb_acct
;
1630 * low level driver can indicate that it wants pages above a
1631 * certain limit bounced to low memory (ie for highmem, or even
1632 * ISA dma in theory)
1634 blk_queue_bounce(q
, &bio
);
1636 blk_queue_split(q
, &bio
, q
->bio_split
);
1638 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1639 bio
->bi_error
= -EIO
;
1641 return BLK_QC_T_NONE
;
1644 if (op_is_flush(bio
->bi_opf
)) {
1645 spin_lock_irq(q
->queue_lock
);
1646 where
= ELEVATOR_INSERT_FLUSH
;
1651 * Check if we can merge with the plugged list before grabbing
1654 if (!blk_queue_nomerges(q
)) {
1655 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1656 return BLK_QC_T_NONE
;
1658 request_count
= blk_plug_queued_count(q
);
1660 spin_lock_irq(q
->queue_lock
);
1662 el_ret
= elv_merge(q
, &req
, bio
);
1663 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1664 if (bio_attempt_back_merge(q
, req
, bio
)) {
1665 elv_bio_merged(q
, req
, bio
);
1666 if (!attempt_back_merge(q
, req
))
1667 elv_merged_request(q
, req
, el_ret
);
1670 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1671 if (bio_attempt_front_merge(q
, req
, bio
)) {
1672 elv_bio_merged(q
, req
, bio
);
1673 if (!attempt_front_merge(q
, req
))
1674 elv_merged_request(q
, req
, el_ret
);
1680 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1683 * Grab a free request. This is might sleep but can not fail.
1684 * Returns with the queue unlocked.
1686 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1688 __wbt_done(q
->rq_wb
, wb_acct
);
1689 bio
->bi_error
= PTR_ERR(req
);
1694 wbt_track(&req
->issue_stat
, wb_acct
);
1697 * After dropping the lock and possibly sleeping here, our request
1698 * may now be mergeable after it had proven unmergeable (above).
1699 * We don't worry about that case for efficiency. It won't happen
1700 * often, and the elevators are able to handle it.
1702 init_request_from_bio(req
, bio
);
1704 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1705 req
->cpu
= raw_smp_processor_id();
1707 plug
= current
->plug
;
1710 * If this is the first request added after a plug, fire
1713 * @request_count may become stale because of schedule
1714 * out, so check plug list again.
1716 if (!request_count
|| list_empty(&plug
->list
))
1717 trace_block_plug(q
);
1719 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1720 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1721 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1722 blk_flush_plug_list(plug
, false);
1723 trace_block_plug(q
);
1726 list_add_tail(&req
->queuelist
, &plug
->list
);
1727 blk_account_io_start(req
, true);
1729 spin_lock_irq(q
->queue_lock
);
1730 add_acct_request(q
, req
, where
);
1733 spin_unlock_irq(q
->queue_lock
);
1736 return BLK_QC_T_NONE
;
1740 * If bio->bi_dev is a partition, remap the location
1742 static inline void blk_partition_remap(struct bio
*bio
)
1744 struct block_device
*bdev
= bio
->bi_bdev
;
1747 * Zone reset does not include bi_size so bio_sectors() is always 0.
1748 * Include a test for the reset op code and perform the remap if needed.
1750 if (bdev
!= bdev
->bd_contains
&&
1751 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1752 struct hd_struct
*p
= bdev
->bd_part
;
1754 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1755 bio
->bi_bdev
= bdev
->bd_contains
;
1757 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1759 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1763 static void handle_bad_sector(struct bio
*bio
)
1765 char b
[BDEVNAME_SIZE
];
1767 printk(KERN_INFO
"attempt to access beyond end of device\n");
1768 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1769 bdevname(bio
->bi_bdev
, b
),
1771 (unsigned long long)bio_end_sector(bio
),
1772 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1775 #ifdef CONFIG_FAIL_MAKE_REQUEST
1777 static DECLARE_FAULT_ATTR(fail_make_request
);
1779 static int __init
setup_fail_make_request(char *str
)
1781 return setup_fault_attr(&fail_make_request
, str
);
1783 __setup("fail_make_request=", setup_fail_make_request
);
1785 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1787 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1790 static int __init
fail_make_request_debugfs(void)
1792 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1793 NULL
, &fail_make_request
);
1795 return PTR_ERR_OR_ZERO(dir
);
1798 late_initcall(fail_make_request_debugfs
);
1800 #else /* CONFIG_FAIL_MAKE_REQUEST */
1802 static inline bool should_fail_request(struct hd_struct
*part
,
1808 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1811 * Check whether this bio extends beyond the end of the device.
1813 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1820 /* Test device or partition size, when known. */
1821 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1823 sector_t sector
= bio
->bi_iter
.bi_sector
;
1825 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1827 * This may well happen - the kernel calls bread()
1828 * without checking the size of the device, e.g., when
1829 * mounting a device.
1831 handle_bad_sector(bio
);
1839 static noinline_for_stack
bool
1840 generic_make_request_checks(struct bio
*bio
)
1842 struct request_queue
*q
;
1843 int nr_sectors
= bio_sectors(bio
);
1845 char b
[BDEVNAME_SIZE
];
1846 struct hd_struct
*part
;
1850 if (bio_check_eod(bio
, nr_sectors
))
1853 q
= bdev_get_queue(bio
->bi_bdev
);
1856 "generic_make_request: Trying to access "
1857 "nonexistent block-device %s (%Lu)\n",
1858 bdevname(bio
->bi_bdev
, b
),
1859 (long long) bio
->bi_iter
.bi_sector
);
1863 part
= bio
->bi_bdev
->bd_part
;
1864 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1865 should_fail_request(&part_to_disk(part
)->part0
,
1866 bio
->bi_iter
.bi_size
))
1870 * If this device has partitions, remap block n
1871 * of partition p to block n+start(p) of the disk.
1873 blk_partition_remap(bio
);
1875 if (bio_check_eod(bio
, nr_sectors
))
1879 * Filter flush bio's early so that make_request based
1880 * drivers without flush support don't have to worry
1883 if ((bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
)) &&
1884 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1885 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1892 switch (bio_op(bio
)) {
1893 case REQ_OP_DISCARD
:
1894 if (!blk_queue_discard(q
))
1897 case REQ_OP_SECURE_ERASE
:
1898 if (!blk_queue_secure_erase(q
))
1901 case REQ_OP_WRITE_SAME
:
1902 if (!bdev_write_same(bio
->bi_bdev
))
1905 case REQ_OP_ZONE_REPORT
:
1906 case REQ_OP_ZONE_RESET
:
1907 if (!bdev_is_zoned(bio
->bi_bdev
))
1910 case REQ_OP_WRITE_ZEROES
:
1911 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1919 * Various block parts want %current->io_context and lazy ioc
1920 * allocation ends up trading a lot of pain for a small amount of
1921 * memory. Just allocate it upfront. This may fail and block
1922 * layer knows how to live with it.
1924 create_io_context(GFP_ATOMIC
, q
->node
);
1926 if (!blkcg_bio_issue_check(q
, bio
))
1929 trace_block_bio_queue(q
, bio
);
1935 bio
->bi_error
= err
;
1941 * generic_make_request - hand a buffer to its device driver for I/O
1942 * @bio: The bio describing the location in memory and on the device.
1944 * generic_make_request() is used to make I/O requests of block
1945 * devices. It is passed a &struct bio, which describes the I/O that needs
1948 * generic_make_request() does not return any status. The
1949 * success/failure status of the request, along with notification of
1950 * completion, is delivered asynchronously through the bio->bi_end_io
1951 * function described (one day) else where.
1953 * The caller of generic_make_request must make sure that bi_io_vec
1954 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1955 * set to describe the device address, and the
1956 * bi_end_io and optionally bi_private are set to describe how
1957 * completion notification should be signaled.
1959 * generic_make_request and the drivers it calls may use bi_next if this
1960 * bio happens to be merged with someone else, and may resubmit the bio to
1961 * a lower device by calling into generic_make_request recursively, which
1962 * means the bio should NOT be touched after the call to ->make_request_fn.
1964 blk_qc_t
generic_make_request(struct bio
*bio
)
1966 struct bio_list bio_list_on_stack
;
1967 blk_qc_t ret
= BLK_QC_T_NONE
;
1969 if (!generic_make_request_checks(bio
))
1973 * We only want one ->make_request_fn to be active at a time, else
1974 * stack usage with stacked devices could be a problem. So use
1975 * current->bio_list to keep a list of requests submited by a
1976 * make_request_fn function. current->bio_list is also used as a
1977 * flag to say if generic_make_request is currently active in this
1978 * task or not. If it is NULL, then no make_request is active. If
1979 * it is non-NULL, then a make_request is active, and new requests
1980 * should be added at the tail
1982 if (current
->bio_list
) {
1983 bio_list_add(current
->bio_list
, bio
);
1987 /* following loop may be a bit non-obvious, and so deserves some
1989 * Before entering the loop, bio->bi_next is NULL (as all callers
1990 * ensure that) so we have a list with a single bio.
1991 * We pretend that we have just taken it off a longer list, so
1992 * we assign bio_list to a pointer to the bio_list_on_stack,
1993 * thus initialising the bio_list of new bios to be
1994 * added. ->make_request() may indeed add some more bios
1995 * through a recursive call to generic_make_request. If it
1996 * did, we find a non-NULL value in bio_list and re-enter the loop
1997 * from the top. In this case we really did just take the bio
1998 * of the top of the list (no pretending) and so remove it from
1999 * bio_list, and call into ->make_request() again.
2001 BUG_ON(bio
->bi_next
);
2002 bio_list_init(&bio_list_on_stack
);
2003 current
->bio_list
= &bio_list_on_stack
;
2005 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2007 if (likely(blk_queue_enter(q
, false) == 0)) {
2008 ret
= q
->make_request_fn(q
, bio
);
2012 bio
= bio_list_pop(current
->bio_list
);
2014 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2020 current
->bio_list
= NULL
; /* deactivate */
2025 EXPORT_SYMBOL(generic_make_request
);
2028 * submit_bio - submit a bio to the block device layer for I/O
2029 * @bio: The &struct bio which describes the I/O
2031 * submit_bio() is very similar in purpose to generic_make_request(), and
2032 * uses that function to do most of the work. Both are fairly rough
2033 * interfaces; @bio must be presetup and ready for I/O.
2036 blk_qc_t
submit_bio(struct bio
*bio
)
2039 * If it's a regular read/write or a barrier with data attached,
2040 * go through the normal accounting stuff before submission.
2042 if (bio_has_data(bio
)) {
2045 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2046 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2048 count
= bio_sectors(bio
);
2050 if (op_is_write(bio_op(bio
))) {
2051 count_vm_events(PGPGOUT
, count
);
2053 task_io_account_read(bio
->bi_iter
.bi_size
);
2054 count_vm_events(PGPGIN
, count
);
2057 if (unlikely(block_dump
)) {
2058 char b
[BDEVNAME_SIZE
];
2059 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2060 current
->comm
, task_pid_nr(current
),
2061 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2062 (unsigned long long)bio
->bi_iter
.bi_sector
,
2063 bdevname(bio
->bi_bdev
, b
),
2068 return generic_make_request(bio
);
2070 EXPORT_SYMBOL(submit_bio
);
2073 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2074 * for new the queue limits
2076 * @rq: the request being checked
2079 * @rq may have been made based on weaker limitations of upper-level queues
2080 * in request stacking drivers, and it may violate the limitation of @q.
2081 * Since the block layer and the underlying device driver trust @rq
2082 * after it is inserted to @q, it should be checked against @q before
2083 * the insertion using this generic function.
2085 * Request stacking drivers like request-based dm may change the queue
2086 * limits when retrying requests on other queues. Those requests need
2087 * to be checked against the new queue limits again during dispatch.
2089 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2092 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2093 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2098 * queue's settings related to segment counting like q->bounce_pfn
2099 * may differ from that of other stacking queues.
2100 * Recalculate it to check the request correctly on this queue's
2103 blk_recalc_rq_segments(rq
);
2104 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2105 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2113 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2114 * @q: the queue to submit the request
2115 * @rq: the request being queued
2117 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2119 unsigned long flags
;
2120 int where
= ELEVATOR_INSERT_BACK
;
2122 if (blk_cloned_rq_check_limits(q
, rq
))
2126 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2130 if (blk_queue_io_stat(q
))
2131 blk_account_io_start(rq
, true);
2132 blk_mq_sched_insert_request(rq
, false, true, false);
2136 spin_lock_irqsave(q
->queue_lock
, flags
);
2137 if (unlikely(blk_queue_dying(q
))) {
2138 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2143 * Submitting request must be dequeued before calling this function
2144 * because it will be linked to another request_queue
2146 BUG_ON(blk_queued_rq(rq
));
2148 if (op_is_flush(rq
->cmd_flags
))
2149 where
= ELEVATOR_INSERT_FLUSH
;
2151 add_acct_request(q
, rq
, where
);
2152 if (where
== ELEVATOR_INSERT_FLUSH
)
2154 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2158 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2161 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2162 * @rq: request to examine
2165 * A request could be merge of IOs which require different failure
2166 * handling. This function determines the number of bytes which
2167 * can be failed from the beginning of the request without
2168 * crossing into area which need to be retried further.
2171 * The number of bytes to fail.
2174 * queue_lock must be held.
2176 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2178 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2179 unsigned int bytes
= 0;
2182 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2183 return blk_rq_bytes(rq
);
2186 * Currently the only 'mixing' which can happen is between
2187 * different fastfail types. We can safely fail portions
2188 * which have all the failfast bits that the first one has -
2189 * the ones which are at least as eager to fail as the first
2192 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2193 if ((bio
->bi_opf
& ff
) != ff
)
2195 bytes
+= bio
->bi_iter
.bi_size
;
2198 /* this could lead to infinite loop */
2199 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2202 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2204 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2206 if (blk_do_io_stat(req
)) {
2207 const int rw
= rq_data_dir(req
);
2208 struct hd_struct
*part
;
2211 cpu
= part_stat_lock();
2213 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2218 void blk_account_io_done(struct request
*req
)
2221 * Account IO completion. flush_rq isn't accounted as a
2222 * normal IO on queueing nor completion. Accounting the
2223 * containing request is enough.
2225 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2226 unsigned long duration
= jiffies
- req
->start_time
;
2227 const int rw
= rq_data_dir(req
);
2228 struct hd_struct
*part
;
2231 cpu
= part_stat_lock();
2234 part_stat_inc(cpu
, part
, ios
[rw
]);
2235 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2236 part_round_stats(cpu
, part
);
2237 part_dec_in_flight(part
, rw
);
2239 hd_struct_put(part
);
2246 * Don't process normal requests when queue is suspended
2247 * or in the process of suspending/resuming
2249 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2252 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2253 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2259 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2266 void blk_account_io_start(struct request
*rq
, bool new_io
)
2268 struct hd_struct
*part
;
2269 int rw
= rq_data_dir(rq
);
2272 if (!blk_do_io_stat(rq
))
2275 cpu
= part_stat_lock();
2279 part_stat_inc(cpu
, part
, merges
[rw
]);
2281 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2282 if (!hd_struct_try_get(part
)) {
2284 * The partition is already being removed,
2285 * the request will be accounted on the disk only
2287 * We take a reference on disk->part0 although that
2288 * partition will never be deleted, so we can treat
2289 * it as any other partition.
2291 part
= &rq
->rq_disk
->part0
;
2292 hd_struct_get(part
);
2294 part_round_stats(cpu
, part
);
2295 part_inc_in_flight(part
, rw
);
2303 * blk_peek_request - peek at the top of a request queue
2304 * @q: request queue to peek at
2307 * Return the request at the top of @q. The returned request
2308 * should be started using blk_start_request() before LLD starts
2312 * Pointer to the request at the top of @q if available. Null
2316 * queue_lock must be held.
2318 struct request
*blk_peek_request(struct request_queue
*q
)
2323 while ((rq
= __elv_next_request(q
)) != NULL
) {
2325 rq
= blk_pm_peek_request(q
, rq
);
2329 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2331 * This is the first time the device driver
2332 * sees this request (possibly after
2333 * requeueing). Notify IO scheduler.
2335 if (rq
->rq_flags
& RQF_SORTED
)
2336 elv_activate_rq(q
, rq
);
2339 * just mark as started even if we don't start
2340 * it, a request that has been delayed should
2341 * not be passed by new incoming requests
2343 rq
->rq_flags
|= RQF_STARTED
;
2344 trace_block_rq_issue(q
, rq
);
2347 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2348 q
->end_sector
= rq_end_sector(rq
);
2349 q
->boundary_rq
= NULL
;
2352 if (rq
->rq_flags
& RQF_DONTPREP
)
2355 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2357 * make sure space for the drain appears we
2358 * know we can do this because max_hw_segments
2359 * has been adjusted to be one fewer than the
2362 rq
->nr_phys_segments
++;
2368 ret
= q
->prep_rq_fn(q
, rq
);
2369 if (ret
== BLKPREP_OK
) {
2371 } else if (ret
== BLKPREP_DEFER
) {
2373 * the request may have been (partially) prepped.
2374 * we need to keep this request in the front to
2375 * avoid resource deadlock. RQF_STARTED will
2376 * prevent other fs requests from passing this one.
2378 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2379 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2381 * remove the space for the drain we added
2382 * so that we don't add it again
2384 --rq
->nr_phys_segments
;
2389 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2390 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2392 rq
->rq_flags
|= RQF_QUIET
;
2394 * Mark this request as started so we don't trigger
2395 * any debug logic in the end I/O path.
2397 blk_start_request(rq
);
2398 __blk_end_request_all(rq
, err
);
2400 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2407 EXPORT_SYMBOL(blk_peek_request
);
2409 void blk_dequeue_request(struct request
*rq
)
2411 struct request_queue
*q
= rq
->q
;
2413 BUG_ON(list_empty(&rq
->queuelist
));
2414 BUG_ON(ELV_ON_HASH(rq
));
2416 list_del_init(&rq
->queuelist
);
2419 * the time frame between a request being removed from the lists
2420 * and to it is freed is accounted as io that is in progress at
2423 if (blk_account_rq(rq
)) {
2424 q
->in_flight
[rq_is_sync(rq
)]++;
2425 set_io_start_time_ns(rq
);
2430 * blk_start_request - start request processing on the driver
2431 * @req: request to dequeue
2434 * Dequeue @req and start timeout timer on it. This hands off the
2435 * request to the driver.
2437 * Block internal functions which don't want to start timer should
2438 * call blk_dequeue_request().
2441 * queue_lock must be held.
2443 void blk_start_request(struct request
*req
)
2445 blk_dequeue_request(req
);
2447 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2448 blk_stat_set_issue_time(&req
->issue_stat
);
2449 req
->rq_flags
|= RQF_STATS
;
2450 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2454 * We are now handing the request to the hardware, initialize
2455 * resid_len to full count and add the timeout handler.
2457 req
->resid_len
= blk_rq_bytes(req
);
2458 if (unlikely(blk_bidi_rq(req
)))
2459 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2461 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2464 EXPORT_SYMBOL(blk_start_request
);
2467 * blk_fetch_request - fetch a request from a request queue
2468 * @q: request queue to fetch a request from
2471 * Return the request at the top of @q. The request is started on
2472 * return and LLD can start processing it immediately.
2475 * Pointer to the request at the top of @q if available. Null
2479 * queue_lock must be held.
2481 struct request
*blk_fetch_request(struct request_queue
*q
)
2485 rq
= blk_peek_request(q
);
2487 blk_start_request(rq
);
2490 EXPORT_SYMBOL(blk_fetch_request
);
2493 * blk_update_request - Special helper function for request stacking drivers
2494 * @req: the request being processed
2495 * @error: %0 for success, < %0 for error
2496 * @nr_bytes: number of bytes to complete @req
2499 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2500 * the request structure even if @req doesn't have leftover.
2501 * If @req has leftover, sets it up for the next range of segments.
2503 * This special helper function is only for request stacking drivers
2504 * (e.g. request-based dm) so that they can handle partial completion.
2505 * Actual device drivers should use blk_end_request instead.
2507 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2508 * %false return from this function.
2511 * %false - this request doesn't have any more data
2512 * %true - this request has more data
2514 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2518 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2524 * For fs requests, rq is just carrier of independent bio's
2525 * and each partial completion should be handled separately.
2526 * Reset per-request error on each partial completion.
2528 * TODO: tj: This is too subtle. It would be better to let
2529 * low level drivers do what they see fit.
2531 if (req
->cmd_type
== REQ_TYPE_FS
)
2534 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2535 !(req
->rq_flags
& RQF_QUIET
)) {
2540 error_type
= "recoverable transport";
2543 error_type
= "critical target";
2546 error_type
= "critical nexus";
2549 error_type
= "timeout";
2552 error_type
= "critical space allocation";
2555 error_type
= "critical medium";
2562 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2563 __func__
, error_type
, req
->rq_disk
?
2564 req
->rq_disk
->disk_name
: "?",
2565 (unsigned long long)blk_rq_pos(req
));
2569 blk_account_io_completion(req
, nr_bytes
);
2573 struct bio
*bio
= req
->bio
;
2574 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2576 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2577 req
->bio
= bio
->bi_next
;
2579 req_bio_endio(req
, bio
, bio_bytes
, error
);
2581 total_bytes
+= bio_bytes
;
2582 nr_bytes
-= bio_bytes
;
2593 * Reset counters so that the request stacking driver
2594 * can find how many bytes remain in the request
2597 req
->__data_len
= 0;
2601 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2603 req
->__data_len
-= total_bytes
;
2605 /* update sector only for requests with clear definition of sector */
2606 if (req
->cmd_type
== REQ_TYPE_FS
)
2607 req
->__sector
+= total_bytes
>> 9;
2609 /* mixed attributes always follow the first bio */
2610 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2611 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2612 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2616 * If total number of sectors is less than the first segment
2617 * size, something has gone terribly wrong.
2619 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2620 blk_dump_rq_flags(req
, "request botched");
2621 req
->__data_len
= blk_rq_cur_bytes(req
);
2624 /* recalculate the number of segments */
2625 blk_recalc_rq_segments(req
);
2629 EXPORT_SYMBOL_GPL(blk_update_request
);
2631 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2632 unsigned int nr_bytes
,
2633 unsigned int bidi_bytes
)
2635 if (blk_update_request(rq
, error
, nr_bytes
))
2638 /* Bidi request must be completed as a whole */
2639 if (unlikely(blk_bidi_rq(rq
)) &&
2640 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2643 if (blk_queue_add_random(rq
->q
))
2644 add_disk_randomness(rq
->rq_disk
);
2650 * blk_unprep_request - unprepare a request
2653 * This function makes a request ready for complete resubmission (or
2654 * completion). It happens only after all error handling is complete,
2655 * so represents the appropriate moment to deallocate any resources
2656 * that were allocated to the request in the prep_rq_fn. The queue
2657 * lock is held when calling this.
2659 void blk_unprep_request(struct request
*req
)
2661 struct request_queue
*q
= req
->q
;
2663 req
->rq_flags
&= ~RQF_DONTPREP
;
2664 if (q
->unprep_rq_fn
)
2665 q
->unprep_rq_fn(q
, req
);
2667 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2670 * queue lock must be held
2672 void blk_finish_request(struct request
*req
, int error
)
2674 struct request_queue
*q
= req
->q
;
2676 if (req
->rq_flags
& RQF_STATS
)
2677 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2679 if (req
->rq_flags
& RQF_QUEUED
)
2680 blk_queue_end_tag(q
, req
);
2682 BUG_ON(blk_queued_rq(req
));
2684 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2685 laptop_io_completion(&req
->q
->backing_dev_info
);
2687 blk_delete_timer(req
);
2689 if (req
->rq_flags
& RQF_DONTPREP
)
2690 blk_unprep_request(req
);
2692 blk_account_io_done(req
);
2695 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2696 req
->end_io(req
, error
);
2698 if (blk_bidi_rq(req
))
2699 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2701 __blk_put_request(q
, req
);
2704 EXPORT_SYMBOL(blk_finish_request
);
2707 * blk_end_bidi_request - Complete a bidi request
2708 * @rq: the request to complete
2709 * @error: %0 for success, < %0 for error
2710 * @nr_bytes: number of bytes to complete @rq
2711 * @bidi_bytes: number of bytes to complete @rq->next_rq
2714 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2715 * Drivers that supports bidi can safely call this member for any
2716 * type of request, bidi or uni. In the later case @bidi_bytes is
2720 * %false - we are done with this request
2721 * %true - still buffers pending for this request
2723 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2724 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2726 struct request_queue
*q
= rq
->q
;
2727 unsigned long flags
;
2729 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2732 spin_lock_irqsave(q
->queue_lock
, flags
);
2733 blk_finish_request(rq
, error
);
2734 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2740 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2741 * @rq: the request to complete
2742 * @error: %0 for success, < %0 for error
2743 * @nr_bytes: number of bytes to complete @rq
2744 * @bidi_bytes: number of bytes to complete @rq->next_rq
2747 * Identical to blk_end_bidi_request() except that queue lock is
2748 * assumed to be locked on entry and remains so on return.
2751 * %false - we are done with this request
2752 * %true - still buffers pending for this request
2754 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2755 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2757 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2760 blk_finish_request(rq
, error
);
2766 * blk_end_request - Helper function for drivers to complete the request.
2767 * @rq: the request being processed
2768 * @error: %0 for success, < %0 for error
2769 * @nr_bytes: number of bytes to complete
2772 * Ends I/O on a number of bytes attached to @rq.
2773 * If @rq has leftover, sets it up for the next range of segments.
2776 * %false - we are done with this request
2777 * %true - still buffers pending for this request
2779 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2781 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2783 EXPORT_SYMBOL(blk_end_request
);
2786 * blk_end_request_all - Helper function for drives to finish the request.
2787 * @rq: the request to finish
2788 * @error: %0 for success, < %0 for error
2791 * Completely finish @rq.
2793 void blk_end_request_all(struct request
*rq
, int error
)
2796 unsigned int bidi_bytes
= 0;
2798 if (unlikely(blk_bidi_rq(rq
)))
2799 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2801 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2804 EXPORT_SYMBOL(blk_end_request_all
);
2807 * blk_end_request_cur - Helper function to finish the current request chunk.
2808 * @rq: the request to finish the current chunk for
2809 * @error: %0 for success, < %0 for error
2812 * Complete the current consecutively mapped chunk from @rq.
2815 * %false - we are done with this request
2816 * %true - still buffers pending for this request
2818 bool blk_end_request_cur(struct request
*rq
, int error
)
2820 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2822 EXPORT_SYMBOL(blk_end_request_cur
);
2825 * blk_end_request_err - Finish a request till the next failure boundary.
2826 * @rq: the request to finish till the next failure boundary for
2827 * @error: must be negative errno
2830 * Complete @rq till the next failure boundary.
2833 * %false - we are done with this request
2834 * %true - still buffers pending for this request
2836 bool blk_end_request_err(struct request
*rq
, int error
)
2838 WARN_ON(error
>= 0);
2839 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2841 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2844 * __blk_end_request - Helper function for drivers to complete the request.
2845 * @rq: the request being processed
2846 * @error: %0 for success, < %0 for error
2847 * @nr_bytes: number of bytes to complete
2850 * Must be called with queue lock held unlike blk_end_request().
2853 * %false - we are done with this request
2854 * %true - still buffers pending for this request
2856 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2858 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2860 EXPORT_SYMBOL(__blk_end_request
);
2863 * __blk_end_request_all - Helper function for drives to finish the request.
2864 * @rq: the request to finish
2865 * @error: %0 for success, < %0 for error
2868 * Completely finish @rq. Must be called with queue lock held.
2870 void __blk_end_request_all(struct request
*rq
, int error
)
2873 unsigned int bidi_bytes
= 0;
2875 if (unlikely(blk_bidi_rq(rq
)))
2876 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2878 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2881 EXPORT_SYMBOL(__blk_end_request_all
);
2884 * __blk_end_request_cur - Helper function to finish the current request chunk.
2885 * @rq: the request to finish the current chunk for
2886 * @error: %0 for success, < %0 for error
2889 * Complete the current consecutively mapped chunk from @rq. Must
2890 * be called with queue lock held.
2893 * %false - we are done with this request
2894 * %true - still buffers pending for this request
2896 bool __blk_end_request_cur(struct request
*rq
, int error
)
2898 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2900 EXPORT_SYMBOL(__blk_end_request_cur
);
2903 * __blk_end_request_err - Finish a request till the next failure boundary.
2904 * @rq: the request to finish till the next failure boundary for
2905 * @error: must be negative errno
2908 * Complete @rq till the next failure boundary. Must be called
2909 * with queue lock held.
2912 * %false - we are done with this request
2913 * %true - still buffers pending for this request
2915 bool __blk_end_request_err(struct request
*rq
, int error
)
2917 WARN_ON(error
>= 0);
2918 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2920 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2922 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2925 if (bio_has_data(bio
))
2926 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2928 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2929 rq
->bio
= rq
->biotail
= bio
;
2932 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2935 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2937 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2938 * @rq: the request to be flushed
2941 * Flush all pages in @rq.
2943 void rq_flush_dcache_pages(struct request
*rq
)
2945 struct req_iterator iter
;
2946 struct bio_vec bvec
;
2948 rq_for_each_segment(bvec
, rq
, iter
)
2949 flush_dcache_page(bvec
.bv_page
);
2951 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2955 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2956 * @q : the queue of the device being checked
2959 * Check if underlying low-level drivers of a device are busy.
2960 * If the drivers want to export their busy state, they must set own
2961 * exporting function using blk_queue_lld_busy() first.
2963 * Basically, this function is used only by request stacking drivers
2964 * to stop dispatching requests to underlying devices when underlying
2965 * devices are busy. This behavior helps more I/O merging on the queue
2966 * of the request stacking driver and prevents I/O throughput regression
2967 * on burst I/O load.
2970 * 0 - Not busy (The request stacking driver should dispatch request)
2971 * 1 - Busy (The request stacking driver should stop dispatching request)
2973 int blk_lld_busy(struct request_queue
*q
)
2976 return q
->lld_busy_fn(q
);
2980 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2983 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2984 * @rq: the clone request to be cleaned up
2987 * Free all bios in @rq for a cloned request.
2989 void blk_rq_unprep_clone(struct request
*rq
)
2993 while ((bio
= rq
->bio
) != NULL
) {
2994 rq
->bio
= bio
->bi_next
;
2999 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3002 * Copy attributes of the original request to the clone request.
3003 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3005 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3007 dst
->cpu
= src
->cpu
;
3008 dst
->cmd_flags
= src
->cmd_flags
| REQ_NOMERGE
;
3009 dst
->cmd_type
= src
->cmd_type
;
3010 dst
->__sector
= blk_rq_pos(src
);
3011 dst
->__data_len
= blk_rq_bytes(src
);
3012 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3013 dst
->ioprio
= src
->ioprio
;
3014 dst
->extra_len
= src
->extra_len
;
3018 * blk_rq_prep_clone - Helper function to setup clone request
3019 * @rq: the request to be setup
3020 * @rq_src: original request to be cloned
3021 * @bs: bio_set that bios for clone are allocated from
3022 * @gfp_mask: memory allocation mask for bio
3023 * @bio_ctr: setup function to be called for each clone bio.
3024 * Returns %0 for success, non %0 for failure.
3025 * @data: private data to be passed to @bio_ctr
3028 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3029 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3030 * are not copied, and copying such parts is the caller's responsibility.
3031 * Also, pages which the original bios are pointing to are not copied
3032 * and the cloned bios just point same pages.
3033 * So cloned bios must be completed before original bios, which means
3034 * the caller must complete @rq before @rq_src.
3036 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3037 struct bio_set
*bs
, gfp_t gfp_mask
,
3038 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3041 struct bio
*bio
, *bio_src
;
3046 __rq_for_each_bio(bio_src
, rq_src
) {
3047 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3051 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3055 rq
->biotail
->bi_next
= bio
;
3058 rq
->bio
= rq
->biotail
= bio
;
3061 __blk_rq_prep_clone(rq
, rq_src
);
3068 blk_rq_unprep_clone(rq
);
3072 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3074 int kblockd_schedule_work(struct work_struct
*work
)
3076 return queue_work(kblockd_workqueue
, work
);
3078 EXPORT_SYMBOL(kblockd_schedule_work
);
3080 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3082 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3084 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3086 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3087 unsigned long delay
)
3089 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3091 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3093 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3094 unsigned long delay
)
3096 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3098 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3101 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3102 * @plug: The &struct blk_plug that needs to be initialized
3105 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3106 * pending I/O should the task end up blocking between blk_start_plug() and
3107 * blk_finish_plug(). This is important from a performance perspective, but
3108 * also ensures that we don't deadlock. For instance, if the task is blocking
3109 * for a memory allocation, memory reclaim could end up wanting to free a
3110 * page belonging to that request that is currently residing in our private
3111 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3112 * this kind of deadlock.
3114 void blk_start_plug(struct blk_plug
*plug
)
3116 struct task_struct
*tsk
= current
;
3119 * If this is a nested plug, don't actually assign it.
3124 INIT_LIST_HEAD(&plug
->list
);
3125 INIT_LIST_HEAD(&plug
->mq_list
);
3126 INIT_LIST_HEAD(&plug
->cb_list
);
3128 * Store ordering should not be needed here, since a potential
3129 * preempt will imply a full memory barrier
3133 EXPORT_SYMBOL(blk_start_plug
);
3135 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3137 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3138 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3140 return !(rqa
->q
< rqb
->q
||
3141 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3145 * If 'from_schedule' is true, then postpone the dispatch of requests
3146 * until a safe kblockd context. We due this to avoid accidental big
3147 * additional stack usage in driver dispatch, in places where the originally
3148 * plugger did not intend it.
3150 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3152 __releases(q
->queue_lock
)
3154 trace_block_unplug(q
, depth
, !from_schedule
);
3157 blk_run_queue_async(q
);
3160 spin_unlock(q
->queue_lock
);
3163 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3165 LIST_HEAD(callbacks
);
3167 while (!list_empty(&plug
->cb_list
)) {
3168 list_splice_init(&plug
->cb_list
, &callbacks
);
3170 while (!list_empty(&callbacks
)) {
3171 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3174 list_del(&cb
->list
);
3175 cb
->callback(cb
, from_schedule
);
3180 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3183 struct blk_plug
*plug
= current
->plug
;
3184 struct blk_plug_cb
*cb
;
3189 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3190 if (cb
->callback
== unplug
&& cb
->data
== data
)
3193 /* Not currently on the callback list */
3194 BUG_ON(size
< sizeof(*cb
));
3195 cb
= kzalloc(size
, GFP_ATOMIC
);
3198 cb
->callback
= unplug
;
3199 list_add(&cb
->list
, &plug
->cb_list
);
3203 EXPORT_SYMBOL(blk_check_plugged
);
3205 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3207 struct request_queue
*q
;
3208 unsigned long flags
;
3213 flush_plug_callbacks(plug
, from_schedule
);
3215 if (!list_empty(&plug
->mq_list
))
3216 blk_mq_flush_plug_list(plug
, from_schedule
);
3218 if (list_empty(&plug
->list
))
3221 list_splice_init(&plug
->list
, &list
);
3223 list_sort(NULL
, &list
, plug_rq_cmp
);
3229 * Save and disable interrupts here, to avoid doing it for every
3230 * queue lock we have to take.
3232 local_irq_save(flags
);
3233 while (!list_empty(&list
)) {
3234 rq
= list_entry_rq(list
.next
);
3235 list_del_init(&rq
->queuelist
);
3239 * This drops the queue lock
3242 queue_unplugged(q
, depth
, from_schedule
);
3245 spin_lock(q
->queue_lock
);
3249 * Short-circuit if @q is dead
3251 if (unlikely(blk_queue_dying(q
))) {
3252 __blk_end_request_all(rq
, -ENODEV
);
3257 * rq is already accounted, so use raw insert
3259 if (op_is_flush(rq
->cmd_flags
))
3260 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3262 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3268 * This drops the queue lock
3271 queue_unplugged(q
, depth
, from_schedule
);
3273 local_irq_restore(flags
);
3276 void blk_finish_plug(struct blk_plug
*plug
)
3278 if (plug
!= current
->plug
)
3280 blk_flush_plug_list(plug
, false);
3282 current
->plug
= NULL
;
3284 EXPORT_SYMBOL(blk_finish_plug
);
3288 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3289 * @q: the queue of the device
3290 * @dev: the device the queue belongs to
3293 * Initialize runtime-PM-related fields for @q and start auto suspend for
3294 * @dev. Drivers that want to take advantage of request-based runtime PM
3295 * should call this function after @dev has been initialized, and its
3296 * request queue @q has been allocated, and runtime PM for it can not happen
3297 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3298 * cases, driver should call this function before any I/O has taken place.
3300 * This function takes care of setting up using auto suspend for the device,
3301 * the autosuspend delay is set to -1 to make runtime suspend impossible
3302 * until an updated value is either set by user or by driver. Drivers do
3303 * not need to touch other autosuspend settings.
3305 * The block layer runtime PM is request based, so only works for drivers
3306 * that use request as their IO unit instead of those directly use bio's.
3308 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3311 q
->rpm_status
= RPM_ACTIVE
;
3312 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3313 pm_runtime_use_autosuspend(q
->dev
);
3315 EXPORT_SYMBOL(blk_pm_runtime_init
);
3318 * blk_pre_runtime_suspend - Pre runtime suspend check
3319 * @q: the queue of the device
3322 * This function will check if runtime suspend is allowed for the device
3323 * by examining if there are any requests pending in the queue. If there
3324 * are requests pending, the device can not be runtime suspended; otherwise,
3325 * the queue's status will be updated to SUSPENDING and the driver can
3326 * proceed to suspend the device.
3328 * For the not allowed case, we mark last busy for the device so that
3329 * runtime PM core will try to autosuspend it some time later.
3331 * This function should be called near the start of the device's
3332 * runtime_suspend callback.
3335 * 0 - OK to runtime suspend the device
3336 * -EBUSY - Device should not be runtime suspended
3338 int blk_pre_runtime_suspend(struct request_queue
*q
)
3345 spin_lock_irq(q
->queue_lock
);
3346 if (q
->nr_pending
) {
3348 pm_runtime_mark_last_busy(q
->dev
);
3350 q
->rpm_status
= RPM_SUSPENDING
;
3352 spin_unlock_irq(q
->queue_lock
);
3355 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3358 * blk_post_runtime_suspend - Post runtime suspend processing
3359 * @q: the queue of the device
3360 * @err: return value of the device's runtime_suspend function
3363 * Update the queue's runtime status according to the return value of the
3364 * device's runtime suspend function and mark last busy for the device so
3365 * that PM core will try to auto suspend the device at a later time.
3367 * This function should be called near the end of the device's
3368 * runtime_suspend callback.
3370 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3375 spin_lock_irq(q
->queue_lock
);
3377 q
->rpm_status
= RPM_SUSPENDED
;
3379 q
->rpm_status
= RPM_ACTIVE
;
3380 pm_runtime_mark_last_busy(q
->dev
);
3382 spin_unlock_irq(q
->queue_lock
);
3384 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3387 * blk_pre_runtime_resume - Pre runtime resume processing
3388 * @q: the queue of the device
3391 * Update the queue's runtime status to RESUMING in preparation for the
3392 * runtime resume of the device.
3394 * This function should be called near the start of the device's
3395 * runtime_resume callback.
3397 void blk_pre_runtime_resume(struct request_queue
*q
)
3402 spin_lock_irq(q
->queue_lock
);
3403 q
->rpm_status
= RPM_RESUMING
;
3404 spin_unlock_irq(q
->queue_lock
);
3406 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3409 * blk_post_runtime_resume - Post runtime resume processing
3410 * @q: the queue of the device
3411 * @err: return value of the device's runtime_resume function
3414 * Update the queue's runtime status according to the return value of the
3415 * device's runtime_resume function. If it is successfully resumed, process
3416 * the requests that are queued into the device's queue when it is resuming
3417 * and then mark last busy and initiate autosuspend for it.
3419 * This function should be called near the end of the device's
3420 * runtime_resume callback.
3422 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3427 spin_lock_irq(q
->queue_lock
);
3429 q
->rpm_status
= RPM_ACTIVE
;
3431 pm_runtime_mark_last_busy(q
->dev
);
3432 pm_request_autosuspend(q
->dev
);
3434 q
->rpm_status
= RPM_SUSPENDED
;
3436 spin_unlock_irq(q
->queue_lock
);
3438 EXPORT_SYMBOL(blk_post_runtime_resume
);
3441 * blk_set_runtime_active - Force runtime status of the queue to be active
3442 * @q: the queue of the device
3444 * If the device is left runtime suspended during system suspend the resume
3445 * hook typically resumes the device and corrects runtime status
3446 * accordingly. However, that does not affect the queue runtime PM status
3447 * which is still "suspended". This prevents processing requests from the
3450 * This function can be used in driver's resume hook to correct queue
3451 * runtime PM status and re-enable peeking requests from the queue. It
3452 * should be called before first request is added to the queue.
3454 void blk_set_runtime_active(struct request_queue
*q
)
3456 spin_lock_irq(q
->queue_lock
);
3457 q
->rpm_status
= RPM_ACTIVE
;
3458 pm_runtime_mark_last_busy(q
->dev
);
3459 pm_request_autosuspend(q
->dev
);
3460 spin_unlock_irq(q
->queue_lock
);
3462 EXPORT_SYMBOL(blk_set_runtime_active
);
3465 int __init
blk_dev_init(void)
3467 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3468 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3469 FIELD_SIZEOF(struct request
, cmd_flags
));
3470 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3471 FIELD_SIZEOF(struct bio
, bi_opf
));
3473 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3474 kblockd_workqueue
= alloc_workqueue("kblockd",
3475 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3476 if (!kblockd_workqueue
)
3477 panic("Failed to create kblockd\n");
3479 request_cachep
= kmem_cache_create("blkdev_requests",
3480 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3482 blk_requestq_cachep
= kmem_cache_create("request_queue",
3483 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);