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 * When queue DYING flag is set, we need to block new req
505 * entering queue, so we call blk_freeze_queue_start() to
506 * prevent I/O from crossing blk_queue_enter().
508 blk_freeze_queue_start(q
);
511 blk_mq_wake_waiters(q
);
513 struct request_list
*rl
;
515 spin_lock_irq(q
->queue_lock
);
516 blk_queue_for_each_rl(rl
, q
) {
518 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
519 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
522 spin_unlock_irq(q
->queue_lock
);
525 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
528 * blk_cleanup_queue - shutdown a request queue
529 * @q: request queue to shutdown
531 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
532 * put it. All future requests will be failed immediately with -ENODEV.
534 void blk_cleanup_queue(struct request_queue
*q
)
536 spinlock_t
*lock
= q
->queue_lock
;
538 /* mark @q DYING, no new request or merges will be allowed afterwards */
539 mutex_lock(&q
->sysfs_lock
);
540 blk_set_queue_dying(q
);
544 * A dying queue is permanently in bypass mode till released. Note
545 * that, unlike blk_queue_bypass_start(), we aren't performing
546 * synchronize_rcu() after entering bypass mode to avoid the delay
547 * as some drivers create and destroy a lot of queues while
548 * probing. This is still safe because blk_release_queue() will be
549 * called only after the queue refcnt drops to zero and nothing,
550 * RCU or not, would be traversing the queue by then.
553 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
555 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
556 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
557 queue_flag_set(QUEUE_FLAG_DYING
, q
);
558 spin_unlock_irq(lock
);
559 mutex_unlock(&q
->sysfs_lock
);
562 * Drain all requests queued before DYING marking. Set DEAD flag to
563 * prevent that q->request_fn() gets invoked after draining finished.
568 __blk_drain_queue(q
, true);
569 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
570 spin_unlock_irq(lock
);
572 /* for synchronous bio-based driver finish in-flight integrity i/o */
573 blk_flush_integrity();
575 /* @q won't process any more request, flush async actions */
576 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
580 blk_mq_free_queue(q
);
581 percpu_ref_exit(&q
->q_usage_counter
);
584 if (q
->queue_lock
!= &q
->__queue_lock
)
585 q
->queue_lock
= &q
->__queue_lock
;
586 spin_unlock_irq(lock
);
588 /* @q is and will stay empty, shutdown and put */
591 EXPORT_SYMBOL(blk_cleanup_queue
);
593 /* Allocate memory local to the request queue */
594 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
596 struct request_queue
*q
= data
;
598 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
601 static void free_request_simple(void *element
, void *data
)
603 kmem_cache_free(request_cachep
, element
);
606 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
608 struct request_queue
*q
= data
;
611 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
613 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
620 static void free_request_size(void *element
, void *data
)
622 struct request_queue
*q
= data
;
625 q
->exit_rq_fn(q
, element
);
629 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
632 if (unlikely(rl
->rq_pool
))
636 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
637 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
638 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
639 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
642 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
643 alloc_request_size
, free_request_size
,
644 q
, gfp_mask
, q
->node
);
646 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
647 alloc_request_simple
, free_request_simple
,
648 q
, gfp_mask
, q
->node
);
656 void blk_exit_rl(struct request_list
*rl
)
659 mempool_destroy(rl
->rq_pool
);
662 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
664 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
666 EXPORT_SYMBOL(blk_alloc_queue
);
668 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
673 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
680 * read pair of barrier in blk_freeze_queue_start(),
681 * we need to order reading __PERCPU_REF_DEAD flag of
682 * .q_usage_counter and reading .mq_freeze_depth or
683 * queue dying flag, otherwise the following wait may
684 * never return if the two reads are reordered.
688 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
689 !atomic_read(&q
->mq_freeze_depth
) ||
691 if (blk_queue_dying(q
))
698 void blk_queue_exit(struct request_queue
*q
)
700 percpu_ref_put(&q
->q_usage_counter
);
703 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
705 struct request_queue
*q
=
706 container_of(ref
, struct request_queue
, q_usage_counter
);
708 wake_up_all(&q
->mq_freeze_wq
);
711 static void blk_rq_timed_out_timer(unsigned long data
)
713 struct request_queue
*q
= (struct request_queue
*)data
;
715 kblockd_schedule_work(&q
->timeout_work
);
718 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
720 struct request_queue
*q
;
722 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
723 gfp_mask
| __GFP_ZERO
, node_id
);
727 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
731 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
735 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
736 if (!q
->backing_dev_info
)
739 q
->stats
= blk_alloc_queue_stats();
743 q
->backing_dev_info
->ra_pages
=
744 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
745 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
746 q
->backing_dev_info
->name
= "block";
749 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
750 laptop_mode_timer_fn
, (unsigned long) q
);
751 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
752 INIT_LIST_HEAD(&q
->queue_head
);
753 INIT_LIST_HEAD(&q
->timeout_list
);
754 INIT_LIST_HEAD(&q
->icq_list
);
755 #ifdef CONFIG_BLK_CGROUP
756 INIT_LIST_HEAD(&q
->blkg_list
);
758 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
760 kobject_init(&q
->kobj
, &blk_queue_ktype
);
762 mutex_init(&q
->sysfs_lock
);
763 spin_lock_init(&q
->__queue_lock
);
766 * By default initialize queue_lock to internal lock and driver can
767 * override it later if need be.
769 q
->queue_lock
= &q
->__queue_lock
;
772 * A queue starts its life with bypass turned on to avoid
773 * unnecessary bypass on/off overhead and nasty surprises during
774 * init. The initial bypass will be finished when the queue is
775 * registered by blk_register_queue().
778 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
780 init_waitqueue_head(&q
->mq_freeze_wq
);
783 * Init percpu_ref in atomic mode so that it's faster to shutdown.
784 * See blk_register_queue() for details.
786 if (percpu_ref_init(&q
->q_usage_counter
,
787 blk_queue_usage_counter_release
,
788 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
791 if (blkcg_init_queue(q
))
797 percpu_ref_exit(&q
->q_usage_counter
);
799 blk_free_queue_stats(q
->stats
);
801 bdi_put(q
->backing_dev_info
);
803 bioset_free(q
->bio_split
);
805 ida_simple_remove(&blk_queue_ida
, q
->id
);
807 kmem_cache_free(blk_requestq_cachep
, q
);
810 EXPORT_SYMBOL(blk_alloc_queue_node
);
813 * blk_init_queue - prepare a request queue for use with a block device
814 * @rfn: The function to be called to process requests that have been
815 * placed on the queue.
816 * @lock: Request queue spin lock
819 * If a block device wishes to use the standard request handling procedures,
820 * which sorts requests and coalesces adjacent requests, then it must
821 * call blk_init_queue(). The function @rfn will be called when there
822 * are requests on the queue that need to be processed. If the device
823 * supports plugging, then @rfn may not be called immediately when requests
824 * are available on the queue, but may be called at some time later instead.
825 * Plugged queues are generally unplugged when a buffer belonging to one
826 * of the requests on the queue is needed, or due to memory pressure.
828 * @rfn is not required, or even expected, to remove all requests off the
829 * queue, but only as many as it can handle at a time. If it does leave
830 * requests on the queue, it is responsible for arranging that the requests
831 * get dealt with eventually.
833 * The queue spin lock must be held while manipulating the requests on the
834 * request queue; this lock will be taken also from interrupt context, so irq
835 * disabling is needed for it.
837 * Function returns a pointer to the initialized request queue, or %NULL if
841 * blk_init_queue() must be paired with a blk_cleanup_queue() call
842 * when the block device is deactivated (such as at module unload).
845 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
847 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
849 EXPORT_SYMBOL(blk_init_queue
);
851 struct request_queue
*
852 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
854 struct request_queue
*q
;
856 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
862 q
->queue_lock
= lock
;
863 if (blk_init_allocated_queue(q
) < 0) {
864 blk_cleanup_queue(q
);
870 EXPORT_SYMBOL(blk_init_queue_node
);
872 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
875 int blk_init_allocated_queue(struct request_queue
*q
)
877 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
881 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
882 goto out_free_flush_queue
;
884 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
885 goto out_exit_flush_rq
;
887 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
888 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
891 * This also sets hw/phys segments, boundary and size
893 blk_queue_make_request(q
, blk_queue_bio
);
895 q
->sg_reserved_size
= INT_MAX
;
897 /* Protect q->elevator from elevator_change */
898 mutex_lock(&q
->sysfs_lock
);
901 if (elevator_init(q
, NULL
)) {
902 mutex_unlock(&q
->sysfs_lock
);
903 goto out_exit_flush_rq
;
906 mutex_unlock(&q
->sysfs_lock
);
911 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
912 out_free_flush_queue
:
913 blk_free_flush_queue(q
->fq
);
916 EXPORT_SYMBOL(blk_init_allocated_queue
);
918 bool blk_get_queue(struct request_queue
*q
)
920 if (likely(!blk_queue_dying(q
))) {
927 EXPORT_SYMBOL(blk_get_queue
);
929 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
931 if (rq
->rq_flags
& RQF_ELVPRIV
) {
932 elv_put_request(rl
->q
, rq
);
934 put_io_context(rq
->elv
.icq
->ioc
);
937 mempool_free(rq
, rl
->rq_pool
);
941 * ioc_batching returns true if the ioc is a valid batching request and
942 * should be given priority access to a request.
944 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
950 * Make sure the process is able to allocate at least 1 request
951 * even if the batch times out, otherwise we could theoretically
954 return ioc
->nr_batch_requests
== q
->nr_batching
||
955 (ioc
->nr_batch_requests
> 0
956 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
960 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
961 * will cause the process to be a "batcher" on all queues in the system. This
962 * is the behaviour we want though - once it gets a wakeup it should be given
965 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
967 if (!ioc
|| ioc_batching(q
, ioc
))
970 ioc
->nr_batch_requests
= q
->nr_batching
;
971 ioc
->last_waited
= jiffies
;
974 static void __freed_request(struct request_list
*rl
, int sync
)
976 struct request_queue
*q
= rl
->q
;
978 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
979 blk_clear_congested(rl
, sync
);
981 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
982 if (waitqueue_active(&rl
->wait
[sync
]))
983 wake_up(&rl
->wait
[sync
]);
985 blk_clear_rl_full(rl
, sync
);
990 * A request has just been released. Account for it, update the full and
991 * congestion status, wake up any waiters. Called under q->queue_lock.
993 static void freed_request(struct request_list
*rl
, bool sync
,
994 req_flags_t rq_flags
)
996 struct request_queue
*q
= rl
->q
;
1000 if (rq_flags
& RQF_ELVPRIV
)
1001 q
->nr_rqs_elvpriv
--;
1003 __freed_request(rl
, sync
);
1005 if (unlikely(rl
->starved
[sync
^ 1]))
1006 __freed_request(rl
, sync
^ 1);
1009 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1011 struct request_list
*rl
;
1012 int on_thresh
, off_thresh
;
1014 spin_lock_irq(q
->queue_lock
);
1015 q
->nr_requests
= nr
;
1016 blk_queue_congestion_threshold(q
);
1017 on_thresh
= queue_congestion_on_threshold(q
);
1018 off_thresh
= queue_congestion_off_threshold(q
);
1020 blk_queue_for_each_rl(rl
, q
) {
1021 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1022 blk_set_congested(rl
, BLK_RW_SYNC
);
1023 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1024 blk_clear_congested(rl
, BLK_RW_SYNC
);
1026 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1027 blk_set_congested(rl
, BLK_RW_ASYNC
);
1028 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1029 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1031 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1032 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1034 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1035 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1038 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1039 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1041 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1042 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1046 spin_unlock_irq(q
->queue_lock
);
1051 * __get_request - get a free request
1052 * @rl: request list to allocate from
1053 * @op: operation and flags
1054 * @bio: bio to allocate request for (can be %NULL)
1055 * @gfp_mask: allocation mask
1057 * Get a free request from @q. This function may fail under memory
1058 * pressure or if @q is dead.
1060 * Must be called with @q->queue_lock held and,
1061 * Returns ERR_PTR on failure, with @q->queue_lock held.
1062 * Returns request pointer on success, with @q->queue_lock *not held*.
1064 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1065 struct bio
*bio
, gfp_t gfp_mask
)
1067 struct request_queue
*q
= rl
->q
;
1069 struct elevator_type
*et
= q
->elevator
->type
;
1070 struct io_context
*ioc
= rq_ioc(bio
);
1071 struct io_cq
*icq
= NULL
;
1072 const bool is_sync
= op_is_sync(op
);
1074 req_flags_t rq_flags
= RQF_ALLOCED
;
1076 if (unlikely(blk_queue_dying(q
)))
1077 return ERR_PTR(-ENODEV
);
1079 may_queue
= elv_may_queue(q
, op
);
1080 if (may_queue
== ELV_MQUEUE_NO
)
1083 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1084 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1086 * The queue will fill after this allocation, so set
1087 * it as full, and mark this process as "batching".
1088 * This process will be allowed to complete a batch of
1089 * requests, others will be blocked.
1091 if (!blk_rl_full(rl
, is_sync
)) {
1092 ioc_set_batching(q
, ioc
);
1093 blk_set_rl_full(rl
, is_sync
);
1095 if (may_queue
!= ELV_MQUEUE_MUST
1096 && !ioc_batching(q
, ioc
)) {
1098 * The queue is full and the allocating
1099 * process is not a "batcher", and not
1100 * exempted by the IO scheduler
1102 return ERR_PTR(-ENOMEM
);
1106 blk_set_congested(rl
, is_sync
);
1110 * Only allow batching queuers to allocate up to 50% over the defined
1111 * limit of requests, otherwise we could have thousands of requests
1112 * allocated with any setting of ->nr_requests
1114 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1115 return ERR_PTR(-ENOMEM
);
1117 q
->nr_rqs
[is_sync
]++;
1118 rl
->count
[is_sync
]++;
1119 rl
->starved
[is_sync
] = 0;
1122 * Decide whether the new request will be managed by elevator. If
1123 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1124 * prevent the current elevator from being destroyed until the new
1125 * request is freed. This guarantees icq's won't be destroyed and
1126 * makes creating new ones safe.
1128 * Flush requests do not use the elevator so skip initialization.
1129 * This allows a request to share the flush and elevator data.
1131 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1132 * it will be created after releasing queue_lock.
1134 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1135 rq_flags
|= RQF_ELVPRIV
;
1136 q
->nr_rqs_elvpriv
++;
1137 if (et
->icq_cache
&& ioc
)
1138 icq
= ioc_lookup_icq(ioc
, q
);
1141 if (blk_queue_io_stat(q
))
1142 rq_flags
|= RQF_IO_STAT
;
1143 spin_unlock_irq(q
->queue_lock
);
1145 /* allocate and init request */
1146 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1151 blk_rq_set_rl(rq
, rl
);
1153 rq
->rq_flags
= rq_flags
;
1156 if (rq_flags
& RQF_ELVPRIV
) {
1157 if (unlikely(et
->icq_cache
&& !icq
)) {
1159 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1165 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1168 /* @rq->elv.icq holds io_context until @rq is freed */
1170 get_io_context(icq
->ioc
);
1174 * ioc may be NULL here, and ioc_batching will be false. That's
1175 * OK, if the queue is under the request limit then requests need
1176 * not count toward the nr_batch_requests limit. There will always
1177 * be some limit enforced by BLK_BATCH_TIME.
1179 if (ioc_batching(q
, ioc
))
1180 ioc
->nr_batch_requests
--;
1182 trace_block_getrq(q
, bio
, op
);
1187 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1188 * and may fail indefinitely under memory pressure and thus
1189 * shouldn't stall IO. Treat this request as !elvpriv. This will
1190 * disturb iosched and blkcg but weird is bettern than dead.
1192 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1193 __func__
, dev_name(q
->backing_dev_info
->dev
));
1195 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1198 spin_lock_irq(q
->queue_lock
);
1199 q
->nr_rqs_elvpriv
--;
1200 spin_unlock_irq(q
->queue_lock
);
1205 * Allocation failed presumably due to memory. Undo anything we
1206 * might have messed up.
1208 * Allocating task should really be put onto the front of the wait
1209 * queue, but this is pretty rare.
1211 spin_lock_irq(q
->queue_lock
);
1212 freed_request(rl
, is_sync
, rq_flags
);
1215 * in the very unlikely event that allocation failed and no
1216 * requests for this direction was pending, mark us starved so that
1217 * freeing of a request in the other direction will notice
1218 * us. another possible fix would be to split the rq mempool into
1222 if (unlikely(rl
->count
[is_sync
] == 0))
1223 rl
->starved
[is_sync
] = 1;
1224 return ERR_PTR(-ENOMEM
);
1228 * get_request - get a free request
1229 * @q: request_queue to allocate request from
1230 * @op: operation and flags
1231 * @bio: bio to allocate request for (can be %NULL)
1232 * @gfp_mask: allocation mask
1234 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1235 * this function keeps retrying under memory pressure and fails iff @q is dead.
1237 * Must be called with @q->queue_lock held and,
1238 * Returns ERR_PTR on failure, with @q->queue_lock held.
1239 * Returns request pointer on success, with @q->queue_lock *not held*.
1241 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1242 struct bio
*bio
, gfp_t gfp_mask
)
1244 const bool is_sync
= op_is_sync(op
);
1246 struct request_list
*rl
;
1249 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1251 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1255 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1260 /* wait on @rl and retry */
1261 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1262 TASK_UNINTERRUPTIBLE
);
1264 trace_block_sleeprq(q
, bio
, op
);
1266 spin_unlock_irq(q
->queue_lock
);
1270 * After sleeping, we become a "batching" process and will be able
1271 * to allocate at least one request, and up to a big batch of them
1272 * for a small period time. See ioc_batching, ioc_set_batching
1274 ioc_set_batching(q
, current
->io_context
);
1276 spin_lock_irq(q
->queue_lock
);
1277 finish_wait(&rl
->wait
[is_sync
], &wait
);
1282 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1287 /* create ioc upfront */
1288 create_io_context(gfp_mask
, q
->node
);
1290 spin_lock_irq(q
->queue_lock
);
1291 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1293 spin_unlock_irq(q
->queue_lock
);
1297 /* q->queue_lock is unlocked at this point */
1299 rq
->__sector
= (sector_t
) -1;
1300 rq
->bio
= rq
->biotail
= NULL
;
1304 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1307 return blk_mq_alloc_request(q
, rw
,
1308 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1309 0 : BLK_MQ_REQ_NOWAIT
);
1311 return blk_old_get_request(q
, rw
, gfp_mask
);
1313 EXPORT_SYMBOL(blk_get_request
);
1316 * blk_requeue_request - put a request back on queue
1317 * @q: request queue where request should be inserted
1318 * @rq: request to be inserted
1321 * Drivers often keep queueing requests until the hardware cannot accept
1322 * more, when that condition happens we need to put the request back
1323 * on the queue. Must be called with queue lock held.
1325 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1327 blk_delete_timer(rq
);
1328 blk_clear_rq_complete(rq
);
1329 trace_block_rq_requeue(q
, rq
);
1330 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1332 if (rq
->rq_flags
& RQF_QUEUED
)
1333 blk_queue_end_tag(q
, rq
);
1335 BUG_ON(blk_queued_rq(rq
));
1337 elv_requeue_request(q
, rq
);
1339 EXPORT_SYMBOL(blk_requeue_request
);
1341 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1344 blk_account_io_start(rq
, true);
1345 __elv_add_request(q
, rq
, where
);
1348 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1353 if (now
== part
->stamp
)
1356 inflight
= part_in_flight(part
);
1358 __part_stat_add(cpu
, part
, time_in_queue
,
1359 inflight
* (now
- part
->stamp
));
1360 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1366 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1367 * @cpu: cpu number for stats access
1368 * @part: target partition
1370 * The average IO queue length and utilisation statistics are maintained
1371 * by observing the current state of the queue length and the amount of
1372 * time it has been in this state for.
1374 * Normally, that accounting is done on IO completion, but that can result
1375 * in more than a second's worth of IO being accounted for within any one
1376 * second, leading to >100% utilisation. To deal with that, we call this
1377 * function to do a round-off before returning the results when reading
1378 * /proc/diskstats. This accounts immediately for all queue usage up to
1379 * the current jiffies and restarts the counters again.
1381 void part_round_stats(int cpu
, struct hd_struct
*part
)
1383 unsigned long now
= jiffies
;
1386 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1387 part_round_stats_single(cpu
, part
, now
);
1389 EXPORT_SYMBOL_GPL(part_round_stats
);
1392 static void blk_pm_put_request(struct request
*rq
)
1394 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1395 pm_runtime_mark_last_busy(rq
->q
->dev
);
1398 static inline void blk_pm_put_request(struct request
*rq
) {}
1402 * queue lock must be held
1404 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1406 req_flags_t rq_flags
= req
->rq_flags
;
1412 blk_mq_free_request(req
);
1416 blk_pm_put_request(req
);
1418 elv_completed_request(q
, req
);
1420 /* this is a bio leak */
1421 WARN_ON(req
->bio
!= NULL
);
1423 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1426 * Request may not have originated from ll_rw_blk. if not,
1427 * it didn't come out of our reserved rq pools
1429 if (rq_flags
& RQF_ALLOCED
) {
1430 struct request_list
*rl
= blk_rq_rl(req
);
1431 bool sync
= op_is_sync(req
->cmd_flags
);
1433 BUG_ON(!list_empty(&req
->queuelist
));
1434 BUG_ON(ELV_ON_HASH(req
));
1436 blk_free_request(rl
, req
);
1437 freed_request(rl
, sync
, rq_flags
);
1441 EXPORT_SYMBOL_GPL(__blk_put_request
);
1443 void blk_put_request(struct request
*req
)
1445 struct request_queue
*q
= req
->q
;
1448 blk_mq_free_request(req
);
1450 unsigned long flags
;
1452 spin_lock_irqsave(q
->queue_lock
, flags
);
1453 __blk_put_request(q
, req
);
1454 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1457 EXPORT_SYMBOL(blk_put_request
);
1459 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1462 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1464 if (!ll_back_merge_fn(q
, req
, bio
))
1467 trace_block_bio_backmerge(q
, req
, bio
);
1469 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1470 blk_rq_set_mixed_merge(req
);
1472 req
->biotail
->bi_next
= bio
;
1474 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1475 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1477 blk_account_io_start(req
, false);
1481 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1484 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1486 if (!ll_front_merge_fn(q
, req
, bio
))
1489 trace_block_bio_frontmerge(q
, req
, bio
);
1491 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1492 blk_rq_set_mixed_merge(req
);
1494 bio
->bi_next
= req
->bio
;
1497 req
->__sector
= bio
->bi_iter
.bi_sector
;
1498 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1499 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1501 blk_account_io_start(req
, false);
1505 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1508 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1510 if (segments
>= queue_max_discard_segments(q
))
1512 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1513 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1516 req
->biotail
->bi_next
= bio
;
1518 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1519 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1520 req
->nr_phys_segments
= segments
+ 1;
1522 blk_account_io_start(req
, false);
1525 req_set_nomerge(q
, req
);
1530 * blk_attempt_plug_merge - try to merge with %current's plugged list
1531 * @q: request_queue new bio is being queued at
1532 * @bio: new bio being queued
1533 * @request_count: out parameter for number of traversed plugged requests
1534 * @same_queue_rq: pointer to &struct request that gets filled in when
1535 * another request associated with @q is found on the plug list
1536 * (optional, may be %NULL)
1538 * Determine whether @bio being queued on @q can be merged with a request
1539 * on %current's plugged list. Returns %true if merge was successful,
1542 * Plugging coalesces IOs from the same issuer for the same purpose without
1543 * going through @q->queue_lock. As such it's more of an issuing mechanism
1544 * than scheduling, and the request, while may have elvpriv data, is not
1545 * added on the elevator at this point. In addition, we don't have
1546 * reliable access to the elevator outside queue lock. Only check basic
1547 * merging parameters without querying the elevator.
1549 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1551 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1552 unsigned int *request_count
,
1553 struct request
**same_queue_rq
)
1555 struct blk_plug
*plug
;
1557 struct list_head
*plug_list
;
1559 plug
= current
->plug
;
1565 plug_list
= &plug
->mq_list
;
1567 plug_list
= &plug
->list
;
1569 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1570 bool merged
= false;
1575 * Only blk-mq multiple hardware queues case checks the
1576 * rq in the same queue, there should be only one such
1580 *same_queue_rq
= rq
;
1583 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1586 switch (blk_try_merge(rq
, bio
)) {
1587 case ELEVATOR_BACK_MERGE
:
1588 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1590 case ELEVATOR_FRONT_MERGE
:
1591 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1593 case ELEVATOR_DISCARD_MERGE
:
1594 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1607 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1609 struct blk_plug
*plug
;
1611 struct list_head
*plug_list
;
1612 unsigned int ret
= 0;
1614 plug
= current
->plug
;
1619 plug_list
= &plug
->mq_list
;
1621 plug_list
= &plug
->list
;
1623 list_for_each_entry(rq
, plug_list
, queuelist
) {
1631 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1633 if (bio
->bi_opf
& REQ_RAHEAD
)
1634 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1637 req
->__sector
= bio
->bi_iter
.bi_sector
;
1638 blk_rq_set_prio(req
, rq_ioc(bio
));
1639 if (ioprio_valid(bio_prio(bio
)))
1640 req
->ioprio
= bio_prio(bio
);
1641 blk_rq_bio_prep(req
->q
, req
, bio
);
1643 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1645 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1647 struct blk_plug
*plug
;
1648 int where
= ELEVATOR_INSERT_SORT
;
1649 struct request
*req
, *free
;
1650 unsigned int request_count
= 0;
1651 unsigned int wb_acct
;
1654 * low level driver can indicate that it wants pages above a
1655 * certain limit bounced to low memory (ie for highmem, or even
1656 * ISA dma in theory)
1658 blk_queue_bounce(q
, &bio
);
1660 blk_queue_split(q
, &bio
, q
->bio_split
);
1662 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1663 bio
->bi_error
= -EIO
;
1665 return BLK_QC_T_NONE
;
1668 if (op_is_flush(bio
->bi_opf
)) {
1669 spin_lock_irq(q
->queue_lock
);
1670 where
= ELEVATOR_INSERT_FLUSH
;
1675 * Check if we can merge with the plugged list before grabbing
1678 if (!blk_queue_nomerges(q
)) {
1679 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1680 return BLK_QC_T_NONE
;
1682 request_count
= blk_plug_queued_count(q
);
1684 spin_lock_irq(q
->queue_lock
);
1686 switch (elv_merge(q
, &req
, bio
)) {
1687 case ELEVATOR_BACK_MERGE
:
1688 if (!bio_attempt_back_merge(q
, req
, bio
))
1690 elv_bio_merged(q
, req
, bio
);
1691 free
= attempt_back_merge(q
, req
);
1693 __blk_put_request(q
, free
);
1695 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1697 case ELEVATOR_FRONT_MERGE
:
1698 if (!bio_attempt_front_merge(q
, req
, bio
))
1700 elv_bio_merged(q
, req
, bio
);
1701 free
= attempt_front_merge(q
, req
);
1703 __blk_put_request(q
, free
);
1705 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1712 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1715 * Grab a free request. This is might sleep but can not fail.
1716 * Returns with the queue unlocked.
1718 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1720 __wbt_done(q
->rq_wb
, wb_acct
);
1721 bio
->bi_error
= PTR_ERR(req
);
1726 wbt_track(&req
->issue_stat
, wb_acct
);
1729 * After dropping the lock and possibly sleeping here, our request
1730 * may now be mergeable after it had proven unmergeable (above).
1731 * We don't worry about that case for efficiency. It won't happen
1732 * often, and the elevators are able to handle it.
1734 blk_init_request_from_bio(req
, bio
);
1736 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1737 req
->cpu
= raw_smp_processor_id();
1739 plug
= current
->plug
;
1742 * If this is the first request added after a plug, fire
1745 * @request_count may become stale because of schedule
1746 * out, so check plug list again.
1748 if (!request_count
|| list_empty(&plug
->list
))
1749 trace_block_plug(q
);
1751 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1752 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1753 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1754 blk_flush_plug_list(plug
, false);
1755 trace_block_plug(q
);
1758 list_add_tail(&req
->queuelist
, &plug
->list
);
1759 blk_account_io_start(req
, true);
1761 spin_lock_irq(q
->queue_lock
);
1762 add_acct_request(q
, req
, where
);
1765 spin_unlock_irq(q
->queue_lock
);
1768 return BLK_QC_T_NONE
;
1772 * If bio->bi_dev is a partition, remap the location
1774 static inline void blk_partition_remap(struct bio
*bio
)
1776 struct block_device
*bdev
= bio
->bi_bdev
;
1779 * Zone reset does not include bi_size so bio_sectors() is always 0.
1780 * Include a test for the reset op code and perform the remap if needed.
1782 if (bdev
!= bdev
->bd_contains
&&
1783 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1784 struct hd_struct
*p
= bdev
->bd_part
;
1786 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1787 bio
->bi_bdev
= bdev
->bd_contains
;
1789 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1791 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1795 static void handle_bad_sector(struct bio
*bio
)
1797 char b
[BDEVNAME_SIZE
];
1799 printk(KERN_INFO
"attempt to access beyond end of device\n");
1800 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1801 bdevname(bio
->bi_bdev
, b
),
1803 (unsigned long long)bio_end_sector(bio
),
1804 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1807 #ifdef CONFIG_FAIL_MAKE_REQUEST
1809 static DECLARE_FAULT_ATTR(fail_make_request
);
1811 static int __init
setup_fail_make_request(char *str
)
1813 return setup_fault_attr(&fail_make_request
, str
);
1815 __setup("fail_make_request=", setup_fail_make_request
);
1817 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1819 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1822 static int __init
fail_make_request_debugfs(void)
1824 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1825 NULL
, &fail_make_request
);
1827 return PTR_ERR_OR_ZERO(dir
);
1830 late_initcall(fail_make_request_debugfs
);
1832 #else /* CONFIG_FAIL_MAKE_REQUEST */
1834 static inline bool should_fail_request(struct hd_struct
*part
,
1840 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1843 * Check whether this bio extends beyond the end of the device.
1845 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1852 /* Test device or partition size, when known. */
1853 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1855 sector_t sector
= bio
->bi_iter
.bi_sector
;
1857 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1859 * This may well happen - the kernel calls bread()
1860 * without checking the size of the device, e.g., when
1861 * mounting a device.
1863 handle_bad_sector(bio
);
1871 static noinline_for_stack
bool
1872 generic_make_request_checks(struct bio
*bio
)
1874 struct request_queue
*q
;
1875 int nr_sectors
= bio_sectors(bio
);
1877 char b
[BDEVNAME_SIZE
];
1878 struct hd_struct
*part
;
1882 if (bio_check_eod(bio
, nr_sectors
))
1885 q
= bdev_get_queue(bio
->bi_bdev
);
1888 "generic_make_request: Trying to access "
1889 "nonexistent block-device %s (%Lu)\n",
1890 bdevname(bio
->bi_bdev
, b
),
1891 (long long) bio
->bi_iter
.bi_sector
);
1895 part
= bio
->bi_bdev
->bd_part
;
1896 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1897 should_fail_request(&part_to_disk(part
)->part0
,
1898 bio
->bi_iter
.bi_size
))
1902 * If this device has partitions, remap block n
1903 * of partition p to block n+start(p) of the disk.
1905 blk_partition_remap(bio
);
1907 if (bio_check_eod(bio
, nr_sectors
))
1911 * Filter flush bio's early so that make_request based
1912 * drivers without flush support don't have to worry
1915 if (op_is_flush(bio
->bi_opf
) &&
1916 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1917 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1924 switch (bio_op(bio
)) {
1925 case REQ_OP_DISCARD
:
1926 if (!blk_queue_discard(q
))
1929 case REQ_OP_SECURE_ERASE
:
1930 if (!blk_queue_secure_erase(q
))
1933 case REQ_OP_WRITE_SAME
:
1934 if (!bdev_write_same(bio
->bi_bdev
))
1937 case REQ_OP_ZONE_REPORT
:
1938 case REQ_OP_ZONE_RESET
:
1939 if (!bdev_is_zoned(bio
->bi_bdev
))
1942 case REQ_OP_WRITE_ZEROES
:
1943 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
1951 * Various block parts want %current->io_context and lazy ioc
1952 * allocation ends up trading a lot of pain for a small amount of
1953 * memory. Just allocate it upfront. This may fail and block
1954 * layer knows how to live with it.
1956 create_io_context(GFP_ATOMIC
, q
->node
);
1958 if (!blkcg_bio_issue_check(q
, bio
))
1961 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1962 trace_block_bio_queue(q
, bio
);
1963 /* Now that enqueuing has been traced, we need to trace
1964 * completion as well.
1966 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
1973 bio
->bi_error
= err
;
1979 * generic_make_request - hand a buffer to its device driver for I/O
1980 * @bio: The bio describing the location in memory and on the device.
1982 * generic_make_request() is used to make I/O requests of block
1983 * devices. It is passed a &struct bio, which describes the I/O that needs
1986 * generic_make_request() does not return any status. The
1987 * success/failure status of the request, along with notification of
1988 * completion, is delivered asynchronously through the bio->bi_end_io
1989 * function described (one day) else where.
1991 * The caller of generic_make_request must make sure that bi_io_vec
1992 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1993 * set to describe the device address, and the
1994 * bi_end_io and optionally bi_private are set to describe how
1995 * completion notification should be signaled.
1997 * generic_make_request and the drivers it calls may use bi_next if this
1998 * bio happens to be merged with someone else, and may resubmit the bio to
1999 * a lower device by calling into generic_make_request recursively, which
2000 * means the bio should NOT be touched after the call to ->make_request_fn.
2002 blk_qc_t
generic_make_request(struct bio
*bio
)
2005 * bio_list_on_stack[0] contains bios submitted by the current
2007 * bio_list_on_stack[1] contains bios that were submitted before
2008 * the current make_request_fn, but that haven't been processed
2011 struct bio_list bio_list_on_stack
[2];
2012 blk_qc_t ret
= BLK_QC_T_NONE
;
2014 if (!generic_make_request_checks(bio
))
2018 * We only want one ->make_request_fn to be active at a time, else
2019 * stack usage with stacked devices could be a problem. So use
2020 * current->bio_list to keep a list of requests submited by a
2021 * make_request_fn function. current->bio_list is also used as a
2022 * flag to say if generic_make_request is currently active in this
2023 * task or not. If it is NULL, then no make_request is active. If
2024 * it is non-NULL, then a make_request is active, and new requests
2025 * should be added at the tail
2027 if (current
->bio_list
) {
2028 bio_list_add(¤t
->bio_list
[0], bio
);
2032 /* following loop may be a bit non-obvious, and so deserves some
2034 * Before entering the loop, bio->bi_next is NULL (as all callers
2035 * ensure that) so we have a list with a single bio.
2036 * We pretend that we have just taken it off a longer list, so
2037 * we assign bio_list to a pointer to the bio_list_on_stack,
2038 * thus initialising the bio_list of new bios to be
2039 * added. ->make_request() may indeed add some more bios
2040 * through a recursive call to generic_make_request. If it
2041 * did, we find a non-NULL value in bio_list and re-enter the loop
2042 * from the top. In this case we really did just take the bio
2043 * of the top of the list (no pretending) and so remove it from
2044 * bio_list, and call into ->make_request() again.
2046 BUG_ON(bio
->bi_next
);
2047 bio_list_init(&bio_list_on_stack
[0]);
2048 current
->bio_list
= bio_list_on_stack
;
2050 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2052 if (likely(blk_queue_enter(q
, false) == 0)) {
2053 struct bio_list lower
, same
;
2055 /* Create a fresh bio_list for all subordinate requests */
2056 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2057 bio_list_init(&bio_list_on_stack
[0]);
2058 ret
= q
->make_request_fn(q
, bio
);
2062 /* sort new bios into those for a lower level
2063 * and those for the same level
2065 bio_list_init(&lower
);
2066 bio_list_init(&same
);
2067 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2068 if (q
== bdev_get_queue(bio
->bi_bdev
))
2069 bio_list_add(&same
, bio
);
2071 bio_list_add(&lower
, bio
);
2072 /* now assemble so we handle the lowest level first */
2073 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2074 bio_list_merge(&bio_list_on_stack
[0], &same
);
2075 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2079 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2081 current
->bio_list
= NULL
; /* deactivate */
2086 EXPORT_SYMBOL(generic_make_request
);
2089 * submit_bio - submit a bio to the block device layer for I/O
2090 * @bio: The &struct bio which describes the I/O
2092 * submit_bio() is very similar in purpose to generic_make_request(), and
2093 * uses that function to do most of the work. Both are fairly rough
2094 * interfaces; @bio must be presetup and ready for I/O.
2097 blk_qc_t
submit_bio(struct bio
*bio
)
2100 * If it's a regular read/write or a barrier with data attached,
2101 * go through the normal accounting stuff before submission.
2103 if (bio_has_data(bio
)) {
2106 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2107 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2109 count
= bio_sectors(bio
);
2111 if (op_is_write(bio_op(bio
))) {
2112 count_vm_events(PGPGOUT
, count
);
2114 task_io_account_read(bio
->bi_iter
.bi_size
);
2115 count_vm_events(PGPGIN
, count
);
2118 if (unlikely(block_dump
)) {
2119 char b
[BDEVNAME_SIZE
];
2120 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2121 current
->comm
, task_pid_nr(current
),
2122 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2123 (unsigned long long)bio
->bi_iter
.bi_sector
,
2124 bdevname(bio
->bi_bdev
, b
),
2129 return generic_make_request(bio
);
2131 EXPORT_SYMBOL(submit_bio
);
2134 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2135 * for new the queue limits
2137 * @rq: the request being checked
2140 * @rq may have been made based on weaker limitations of upper-level queues
2141 * in request stacking drivers, and it may violate the limitation of @q.
2142 * Since the block layer and the underlying device driver trust @rq
2143 * after it is inserted to @q, it should be checked against @q before
2144 * the insertion using this generic function.
2146 * Request stacking drivers like request-based dm may change the queue
2147 * limits when retrying requests on other queues. Those requests need
2148 * to be checked against the new queue limits again during dispatch.
2150 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2153 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2154 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2159 * queue's settings related to segment counting like q->bounce_pfn
2160 * may differ from that of other stacking queues.
2161 * Recalculate it to check the request correctly on this queue's
2164 blk_recalc_rq_segments(rq
);
2165 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2166 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2174 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2175 * @q: the queue to submit the request
2176 * @rq: the request being queued
2178 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2180 unsigned long flags
;
2181 int where
= ELEVATOR_INSERT_BACK
;
2183 if (blk_cloned_rq_check_limits(q
, rq
))
2187 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2191 if (blk_queue_io_stat(q
))
2192 blk_account_io_start(rq
, true);
2193 blk_mq_sched_insert_request(rq
, false, true, false, false);
2197 spin_lock_irqsave(q
->queue_lock
, flags
);
2198 if (unlikely(blk_queue_dying(q
))) {
2199 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2204 * Submitting request must be dequeued before calling this function
2205 * because it will be linked to another request_queue
2207 BUG_ON(blk_queued_rq(rq
));
2209 if (op_is_flush(rq
->cmd_flags
))
2210 where
= ELEVATOR_INSERT_FLUSH
;
2212 add_acct_request(q
, rq
, where
);
2213 if (where
== ELEVATOR_INSERT_FLUSH
)
2215 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2219 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2222 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2223 * @rq: request to examine
2226 * A request could be merge of IOs which require different failure
2227 * handling. This function determines the number of bytes which
2228 * can be failed from the beginning of the request without
2229 * crossing into area which need to be retried further.
2232 * The number of bytes to fail.
2235 * queue_lock must be held.
2237 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2239 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2240 unsigned int bytes
= 0;
2243 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2244 return blk_rq_bytes(rq
);
2247 * Currently the only 'mixing' which can happen is between
2248 * different fastfail types. We can safely fail portions
2249 * which have all the failfast bits that the first one has -
2250 * the ones which are at least as eager to fail as the first
2253 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2254 if ((bio
->bi_opf
& ff
) != ff
)
2256 bytes
+= bio
->bi_iter
.bi_size
;
2259 /* this could lead to infinite loop */
2260 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2263 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2265 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2267 if (blk_do_io_stat(req
)) {
2268 const int rw
= rq_data_dir(req
);
2269 struct hd_struct
*part
;
2272 cpu
= part_stat_lock();
2274 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2279 void blk_account_io_done(struct request
*req
)
2282 * Account IO completion. flush_rq isn't accounted as a
2283 * normal IO on queueing nor completion. Accounting the
2284 * containing request is enough.
2286 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2287 unsigned long duration
= jiffies
- req
->start_time
;
2288 const int rw
= rq_data_dir(req
);
2289 struct hd_struct
*part
;
2292 cpu
= part_stat_lock();
2295 part_stat_inc(cpu
, part
, ios
[rw
]);
2296 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2297 part_round_stats(cpu
, part
);
2298 part_dec_in_flight(part
, rw
);
2300 hd_struct_put(part
);
2307 * Don't process normal requests when queue is suspended
2308 * or in the process of suspending/resuming
2310 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2313 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2314 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2320 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2327 void blk_account_io_start(struct request
*rq
, bool new_io
)
2329 struct hd_struct
*part
;
2330 int rw
= rq_data_dir(rq
);
2333 if (!blk_do_io_stat(rq
))
2336 cpu
= part_stat_lock();
2340 part_stat_inc(cpu
, part
, merges
[rw
]);
2342 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2343 if (!hd_struct_try_get(part
)) {
2345 * The partition is already being removed,
2346 * the request will be accounted on the disk only
2348 * We take a reference on disk->part0 although that
2349 * partition will never be deleted, so we can treat
2350 * it as any other partition.
2352 part
= &rq
->rq_disk
->part0
;
2353 hd_struct_get(part
);
2355 part_round_stats(cpu
, part
);
2356 part_inc_in_flight(part
, rw
);
2364 * blk_peek_request - peek at the top of a request queue
2365 * @q: request queue to peek at
2368 * Return the request at the top of @q. The returned request
2369 * should be started using blk_start_request() before LLD starts
2373 * Pointer to the request at the top of @q if available. Null
2377 * queue_lock must be held.
2379 struct request
*blk_peek_request(struct request_queue
*q
)
2384 while ((rq
= __elv_next_request(q
)) != NULL
) {
2386 rq
= blk_pm_peek_request(q
, rq
);
2390 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2392 * This is the first time the device driver
2393 * sees this request (possibly after
2394 * requeueing). Notify IO scheduler.
2396 if (rq
->rq_flags
& RQF_SORTED
)
2397 elv_activate_rq(q
, rq
);
2400 * just mark as started even if we don't start
2401 * it, a request that has been delayed should
2402 * not be passed by new incoming requests
2404 rq
->rq_flags
|= RQF_STARTED
;
2405 trace_block_rq_issue(q
, rq
);
2408 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2409 q
->end_sector
= rq_end_sector(rq
);
2410 q
->boundary_rq
= NULL
;
2413 if (rq
->rq_flags
& RQF_DONTPREP
)
2416 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2418 * make sure space for the drain appears we
2419 * know we can do this because max_hw_segments
2420 * has been adjusted to be one fewer than the
2423 rq
->nr_phys_segments
++;
2429 ret
= q
->prep_rq_fn(q
, rq
);
2430 if (ret
== BLKPREP_OK
) {
2432 } else if (ret
== BLKPREP_DEFER
) {
2434 * the request may have been (partially) prepped.
2435 * we need to keep this request in the front to
2436 * avoid resource deadlock. RQF_STARTED will
2437 * prevent other fs requests from passing this one.
2439 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2440 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2442 * remove the space for the drain we added
2443 * so that we don't add it again
2445 --rq
->nr_phys_segments
;
2450 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2451 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2453 rq
->rq_flags
|= RQF_QUIET
;
2455 * Mark this request as started so we don't trigger
2456 * any debug logic in the end I/O path.
2458 blk_start_request(rq
);
2459 __blk_end_request_all(rq
, err
);
2461 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2468 EXPORT_SYMBOL(blk_peek_request
);
2470 void blk_dequeue_request(struct request
*rq
)
2472 struct request_queue
*q
= rq
->q
;
2474 BUG_ON(list_empty(&rq
->queuelist
));
2475 BUG_ON(ELV_ON_HASH(rq
));
2477 list_del_init(&rq
->queuelist
);
2480 * the time frame between a request being removed from the lists
2481 * and to it is freed is accounted as io that is in progress at
2484 if (blk_account_rq(rq
)) {
2485 q
->in_flight
[rq_is_sync(rq
)]++;
2486 set_io_start_time_ns(rq
);
2491 * blk_start_request - start request processing on the driver
2492 * @req: request to dequeue
2495 * Dequeue @req and start timeout timer on it. This hands off the
2496 * request to the driver.
2498 * Block internal functions which don't want to start timer should
2499 * call blk_dequeue_request().
2502 * queue_lock must be held.
2504 void blk_start_request(struct request
*req
)
2506 blk_dequeue_request(req
);
2508 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2509 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2510 req
->rq_flags
|= RQF_STATS
;
2511 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2514 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2517 EXPORT_SYMBOL(blk_start_request
);
2520 * blk_fetch_request - fetch a request from a request queue
2521 * @q: request queue to fetch a request from
2524 * Return the request at the top of @q. The request is started on
2525 * return and LLD can start processing it immediately.
2528 * Pointer to the request at the top of @q if available. Null
2532 * queue_lock must be held.
2534 struct request
*blk_fetch_request(struct request_queue
*q
)
2538 rq
= blk_peek_request(q
);
2540 blk_start_request(rq
);
2543 EXPORT_SYMBOL(blk_fetch_request
);
2546 * blk_update_request - Special helper function for request stacking drivers
2547 * @req: the request being processed
2548 * @error: %0 for success, < %0 for error
2549 * @nr_bytes: number of bytes to complete @req
2552 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2553 * the request structure even if @req doesn't have leftover.
2554 * If @req has leftover, sets it up for the next range of segments.
2556 * This special helper function is only for request stacking drivers
2557 * (e.g. request-based dm) so that they can handle partial completion.
2558 * Actual device drivers should use blk_end_request instead.
2560 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2561 * %false return from this function.
2564 * %false - this request doesn't have any more data
2565 * %true - this request has more data
2567 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2571 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2577 * For fs requests, rq is just carrier of independent bio's
2578 * and each partial completion should be handled separately.
2579 * Reset per-request error on each partial completion.
2581 * TODO: tj: This is too subtle. It would be better to let
2582 * low level drivers do what they see fit.
2584 if (!blk_rq_is_passthrough(req
))
2587 if (error
&& !blk_rq_is_passthrough(req
) &&
2588 !(req
->rq_flags
& RQF_QUIET
)) {
2593 error_type
= "recoverable transport";
2596 error_type
= "critical target";
2599 error_type
= "critical nexus";
2602 error_type
= "timeout";
2605 error_type
= "critical space allocation";
2608 error_type
= "critical medium";
2615 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2616 __func__
, error_type
, req
->rq_disk
?
2617 req
->rq_disk
->disk_name
: "?",
2618 (unsigned long long)blk_rq_pos(req
));
2622 blk_account_io_completion(req
, nr_bytes
);
2626 struct bio
*bio
= req
->bio
;
2627 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2629 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2630 req
->bio
= bio
->bi_next
;
2632 /* Completion has already been traced */
2633 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2634 req_bio_endio(req
, bio
, bio_bytes
, error
);
2636 total_bytes
+= bio_bytes
;
2637 nr_bytes
-= bio_bytes
;
2648 * Reset counters so that the request stacking driver
2649 * can find how many bytes remain in the request
2652 req
->__data_len
= 0;
2656 WARN_ON_ONCE(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
);
2658 req
->__data_len
-= total_bytes
;
2660 /* update sector only for requests with clear definition of sector */
2661 if (!blk_rq_is_passthrough(req
))
2662 req
->__sector
+= total_bytes
>> 9;
2664 /* mixed attributes always follow the first bio */
2665 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2666 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2667 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2671 * If total number of sectors is less than the first segment
2672 * size, something has gone terribly wrong.
2674 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2675 blk_dump_rq_flags(req
, "request botched");
2676 req
->__data_len
= blk_rq_cur_bytes(req
);
2679 /* recalculate the number of segments */
2680 blk_recalc_rq_segments(req
);
2684 EXPORT_SYMBOL_GPL(blk_update_request
);
2686 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2687 unsigned int nr_bytes
,
2688 unsigned int bidi_bytes
)
2690 if (blk_update_request(rq
, error
, nr_bytes
))
2693 /* Bidi request must be completed as a whole */
2694 if (unlikely(blk_bidi_rq(rq
)) &&
2695 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2698 if (blk_queue_add_random(rq
->q
))
2699 add_disk_randomness(rq
->rq_disk
);
2705 * blk_unprep_request - unprepare a request
2708 * This function makes a request ready for complete resubmission (or
2709 * completion). It happens only after all error handling is complete,
2710 * so represents the appropriate moment to deallocate any resources
2711 * that were allocated to the request in the prep_rq_fn. The queue
2712 * lock is held when calling this.
2714 void blk_unprep_request(struct request
*req
)
2716 struct request_queue
*q
= req
->q
;
2718 req
->rq_flags
&= ~RQF_DONTPREP
;
2719 if (q
->unprep_rq_fn
)
2720 q
->unprep_rq_fn(q
, req
);
2722 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2725 * queue lock must be held
2727 void blk_finish_request(struct request
*req
, int error
)
2729 struct request_queue
*q
= req
->q
;
2731 if (req
->rq_flags
& RQF_STATS
)
2734 if (req
->rq_flags
& RQF_QUEUED
)
2735 blk_queue_end_tag(q
, req
);
2737 BUG_ON(blk_queued_rq(req
));
2739 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2740 laptop_io_completion(req
->q
->backing_dev_info
);
2742 blk_delete_timer(req
);
2744 if (req
->rq_flags
& RQF_DONTPREP
)
2745 blk_unprep_request(req
);
2747 blk_account_io_done(req
);
2750 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2751 req
->end_io(req
, error
);
2753 if (blk_bidi_rq(req
))
2754 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2756 __blk_put_request(q
, req
);
2759 EXPORT_SYMBOL(blk_finish_request
);
2762 * blk_end_bidi_request - Complete a bidi request
2763 * @rq: the request to complete
2764 * @error: %0 for success, < %0 for error
2765 * @nr_bytes: number of bytes to complete @rq
2766 * @bidi_bytes: number of bytes to complete @rq->next_rq
2769 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2770 * Drivers that supports bidi can safely call this member for any
2771 * type of request, bidi or uni. In the later case @bidi_bytes is
2775 * %false - we are done with this request
2776 * %true - still buffers pending for this request
2778 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2779 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2781 struct request_queue
*q
= rq
->q
;
2782 unsigned long flags
;
2784 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2787 spin_lock_irqsave(q
->queue_lock
, flags
);
2788 blk_finish_request(rq
, error
);
2789 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2795 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2796 * @rq: the request to complete
2797 * @error: %0 for success, < %0 for error
2798 * @nr_bytes: number of bytes to complete @rq
2799 * @bidi_bytes: number of bytes to complete @rq->next_rq
2802 * Identical to blk_end_bidi_request() except that queue lock is
2803 * assumed to be locked on entry and remains so on return.
2806 * %false - we are done with this request
2807 * %true - still buffers pending for this request
2809 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2810 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2812 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2815 blk_finish_request(rq
, error
);
2821 * blk_end_request - Helper function for drivers to complete the request.
2822 * @rq: the request being processed
2823 * @error: %0 for success, < %0 for error
2824 * @nr_bytes: number of bytes to complete
2827 * Ends I/O on a number of bytes attached to @rq.
2828 * If @rq has leftover, sets it up for the next range of segments.
2831 * %false - we are done with this request
2832 * %true - still buffers pending for this request
2834 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2836 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2838 EXPORT_SYMBOL(blk_end_request
);
2841 * blk_end_request_all - Helper function for drives to finish the request.
2842 * @rq: the request to finish
2843 * @error: %0 for success, < %0 for error
2846 * Completely finish @rq.
2848 void blk_end_request_all(struct request
*rq
, int error
)
2851 unsigned int bidi_bytes
= 0;
2853 if (unlikely(blk_bidi_rq(rq
)))
2854 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2856 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2859 EXPORT_SYMBOL(blk_end_request_all
);
2862 * __blk_end_request - Helper function for drivers to complete the request.
2863 * @rq: the request being processed
2864 * @error: %0 for success, < %0 for error
2865 * @nr_bytes: number of bytes to complete
2868 * Must be called with queue lock held unlike blk_end_request().
2871 * %false - we are done with this request
2872 * %true - still buffers pending for this request
2874 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2876 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2878 EXPORT_SYMBOL(__blk_end_request
);
2881 * __blk_end_request_all - Helper function for drives to finish the request.
2882 * @rq: the request to finish
2883 * @error: %0 for success, < %0 for error
2886 * Completely finish @rq. Must be called with queue lock held.
2888 void __blk_end_request_all(struct request
*rq
, int error
)
2891 unsigned int bidi_bytes
= 0;
2893 if (unlikely(blk_bidi_rq(rq
)))
2894 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2896 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2899 EXPORT_SYMBOL(__blk_end_request_all
);
2902 * __blk_end_request_cur - Helper function to finish the current request chunk.
2903 * @rq: the request to finish the current chunk for
2904 * @error: %0 for success, < %0 for error
2907 * Complete the current consecutively mapped chunk from @rq. Must
2908 * be called with queue lock held.
2911 * %false - we are done with this request
2912 * %true - still buffers pending for this request
2914 bool __blk_end_request_cur(struct request
*rq
, int error
)
2916 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2918 EXPORT_SYMBOL(__blk_end_request_cur
);
2920 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2923 if (bio_has_data(bio
))
2924 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2926 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2927 rq
->bio
= rq
->biotail
= bio
;
2930 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2933 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2935 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2936 * @rq: the request to be flushed
2939 * Flush all pages in @rq.
2941 void rq_flush_dcache_pages(struct request
*rq
)
2943 struct req_iterator iter
;
2944 struct bio_vec bvec
;
2946 rq_for_each_segment(bvec
, rq
, iter
)
2947 flush_dcache_page(bvec
.bv_page
);
2949 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2953 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2954 * @q : the queue of the device being checked
2957 * Check if underlying low-level drivers of a device are busy.
2958 * If the drivers want to export their busy state, they must set own
2959 * exporting function using blk_queue_lld_busy() first.
2961 * Basically, this function is used only by request stacking drivers
2962 * to stop dispatching requests to underlying devices when underlying
2963 * devices are busy. This behavior helps more I/O merging on the queue
2964 * of the request stacking driver and prevents I/O throughput regression
2965 * on burst I/O load.
2968 * 0 - Not busy (The request stacking driver should dispatch request)
2969 * 1 - Busy (The request stacking driver should stop dispatching request)
2971 int blk_lld_busy(struct request_queue
*q
)
2974 return q
->lld_busy_fn(q
);
2978 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2981 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2982 * @rq: the clone request to be cleaned up
2985 * Free all bios in @rq for a cloned request.
2987 void blk_rq_unprep_clone(struct request
*rq
)
2991 while ((bio
= rq
->bio
) != NULL
) {
2992 rq
->bio
= bio
->bi_next
;
2997 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3000 * Copy attributes of the original request to the clone request.
3001 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3003 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3005 dst
->cpu
= src
->cpu
;
3006 dst
->__sector
= blk_rq_pos(src
);
3007 dst
->__data_len
= blk_rq_bytes(src
);
3008 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3009 dst
->ioprio
= src
->ioprio
;
3010 dst
->extra_len
= src
->extra_len
;
3014 * blk_rq_prep_clone - Helper function to setup clone request
3015 * @rq: the request to be setup
3016 * @rq_src: original request to be cloned
3017 * @bs: bio_set that bios for clone are allocated from
3018 * @gfp_mask: memory allocation mask for bio
3019 * @bio_ctr: setup function to be called for each clone bio.
3020 * Returns %0 for success, non %0 for failure.
3021 * @data: private data to be passed to @bio_ctr
3024 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3025 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3026 * are not copied, and copying such parts is the caller's responsibility.
3027 * Also, pages which the original bios are pointing to are not copied
3028 * and the cloned bios just point same pages.
3029 * So cloned bios must be completed before original bios, which means
3030 * the caller must complete @rq before @rq_src.
3032 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3033 struct bio_set
*bs
, gfp_t gfp_mask
,
3034 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3037 struct bio
*bio
, *bio_src
;
3042 __rq_for_each_bio(bio_src
, rq_src
) {
3043 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3047 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3051 rq
->biotail
->bi_next
= bio
;
3054 rq
->bio
= rq
->biotail
= bio
;
3057 __blk_rq_prep_clone(rq
, rq_src
);
3064 blk_rq_unprep_clone(rq
);
3068 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3070 int kblockd_schedule_work(struct work_struct
*work
)
3072 return queue_work(kblockd_workqueue
, work
);
3074 EXPORT_SYMBOL(kblockd_schedule_work
);
3076 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3078 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3080 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3082 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3083 unsigned long delay
)
3085 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3087 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3089 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3090 unsigned long delay
)
3092 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3094 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3097 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3098 * @plug: The &struct blk_plug that needs to be initialized
3101 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3102 * pending I/O should the task end up blocking between blk_start_plug() and
3103 * blk_finish_plug(). This is important from a performance perspective, but
3104 * also ensures that we don't deadlock. For instance, if the task is blocking
3105 * for a memory allocation, memory reclaim could end up wanting to free a
3106 * page belonging to that request that is currently residing in our private
3107 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3108 * this kind of deadlock.
3110 void blk_start_plug(struct blk_plug
*plug
)
3112 struct task_struct
*tsk
= current
;
3115 * If this is a nested plug, don't actually assign it.
3120 INIT_LIST_HEAD(&plug
->list
);
3121 INIT_LIST_HEAD(&plug
->mq_list
);
3122 INIT_LIST_HEAD(&plug
->cb_list
);
3124 * Store ordering should not be needed here, since a potential
3125 * preempt will imply a full memory barrier
3129 EXPORT_SYMBOL(blk_start_plug
);
3131 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3133 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3134 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3136 return !(rqa
->q
< rqb
->q
||
3137 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3141 * If 'from_schedule' is true, then postpone the dispatch of requests
3142 * until a safe kblockd context. We due this to avoid accidental big
3143 * additional stack usage in driver dispatch, in places where the originally
3144 * plugger did not intend it.
3146 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3148 __releases(q
->queue_lock
)
3150 trace_block_unplug(q
, depth
, !from_schedule
);
3153 blk_run_queue_async(q
);
3156 spin_unlock(q
->queue_lock
);
3159 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3161 LIST_HEAD(callbacks
);
3163 while (!list_empty(&plug
->cb_list
)) {
3164 list_splice_init(&plug
->cb_list
, &callbacks
);
3166 while (!list_empty(&callbacks
)) {
3167 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3170 list_del(&cb
->list
);
3171 cb
->callback(cb
, from_schedule
);
3176 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3179 struct blk_plug
*plug
= current
->plug
;
3180 struct blk_plug_cb
*cb
;
3185 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3186 if (cb
->callback
== unplug
&& cb
->data
== data
)
3189 /* Not currently on the callback list */
3190 BUG_ON(size
< sizeof(*cb
));
3191 cb
= kzalloc(size
, GFP_ATOMIC
);
3194 cb
->callback
= unplug
;
3195 list_add(&cb
->list
, &plug
->cb_list
);
3199 EXPORT_SYMBOL(blk_check_plugged
);
3201 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3203 struct request_queue
*q
;
3204 unsigned long flags
;
3209 flush_plug_callbacks(plug
, from_schedule
);
3211 if (!list_empty(&plug
->mq_list
))
3212 blk_mq_flush_plug_list(plug
, from_schedule
);
3214 if (list_empty(&plug
->list
))
3217 list_splice_init(&plug
->list
, &list
);
3219 list_sort(NULL
, &list
, plug_rq_cmp
);
3225 * Save and disable interrupts here, to avoid doing it for every
3226 * queue lock we have to take.
3228 local_irq_save(flags
);
3229 while (!list_empty(&list
)) {
3230 rq
= list_entry_rq(list
.next
);
3231 list_del_init(&rq
->queuelist
);
3235 * This drops the queue lock
3238 queue_unplugged(q
, depth
, from_schedule
);
3241 spin_lock(q
->queue_lock
);
3245 * Short-circuit if @q is dead
3247 if (unlikely(blk_queue_dying(q
))) {
3248 __blk_end_request_all(rq
, -ENODEV
);
3253 * rq is already accounted, so use raw insert
3255 if (op_is_flush(rq
->cmd_flags
))
3256 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3258 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3264 * This drops the queue lock
3267 queue_unplugged(q
, depth
, from_schedule
);
3269 local_irq_restore(flags
);
3272 void blk_finish_plug(struct blk_plug
*plug
)
3274 if (plug
!= current
->plug
)
3276 blk_flush_plug_list(plug
, false);
3278 current
->plug
= NULL
;
3280 EXPORT_SYMBOL(blk_finish_plug
);
3284 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3285 * @q: the queue of the device
3286 * @dev: the device the queue belongs to
3289 * Initialize runtime-PM-related fields for @q and start auto suspend for
3290 * @dev. Drivers that want to take advantage of request-based runtime PM
3291 * should call this function after @dev has been initialized, and its
3292 * request queue @q has been allocated, and runtime PM for it can not happen
3293 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3294 * cases, driver should call this function before any I/O has taken place.
3296 * This function takes care of setting up using auto suspend for the device,
3297 * the autosuspend delay is set to -1 to make runtime suspend impossible
3298 * until an updated value is either set by user or by driver. Drivers do
3299 * not need to touch other autosuspend settings.
3301 * The block layer runtime PM is request based, so only works for drivers
3302 * that use request as their IO unit instead of those directly use bio's.
3304 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3307 q
->rpm_status
= RPM_ACTIVE
;
3308 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3309 pm_runtime_use_autosuspend(q
->dev
);
3311 EXPORT_SYMBOL(blk_pm_runtime_init
);
3314 * blk_pre_runtime_suspend - Pre runtime suspend check
3315 * @q: the queue of the device
3318 * This function will check if runtime suspend is allowed for the device
3319 * by examining if there are any requests pending in the queue. If there
3320 * are requests pending, the device can not be runtime suspended; otherwise,
3321 * the queue's status will be updated to SUSPENDING and the driver can
3322 * proceed to suspend the device.
3324 * For the not allowed case, we mark last busy for the device so that
3325 * runtime PM core will try to autosuspend it some time later.
3327 * This function should be called near the start of the device's
3328 * runtime_suspend callback.
3331 * 0 - OK to runtime suspend the device
3332 * -EBUSY - Device should not be runtime suspended
3334 int blk_pre_runtime_suspend(struct request_queue
*q
)
3341 spin_lock_irq(q
->queue_lock
);
3342 if (q
->nr_pending
) {
3344 pm_runtime_mark_last_busy(q
->dev
);
3346 q
->rpm_status
= RPM_SUSPENDING
;
3348 spin_unlock_irq(q
->queue_lock
);
3351 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3354 * blk_post_runtime_suspend - Post runtime suspend processing
3355 * @q: the queue of the device
3356 * @err: return value of the device's runtime_suspend function
3359 * Update the queue's runtime status according to the return value of the
3360 * device's runtime suspend function and mark last busy for the device so
3361 * that PM core will try to auto suspend the device at a later time.
3363 * This function should be called near the end of the device's
3364 * runtime_suspend callback.
3366 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3371 spin_lock_irq(q
->queue_lock
);
3373 q
->rpm_status
= RPM_SUSPENDED
;
3375 q
->rpm_status
= RPM_ACTIVE
;
3376 pm_runtime_mark_last_busy(q
->dev
);
3378 spin_unlock_irq(q
->queue_lock
);
3380 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3383 * blk_pre_runtime_resume - Pre runtime resume processing
3384 * @q: the queue of the device
3387 * Update the queue's runtime status to RESUMING in preparation for the
3388 * runtime resume of the device.
3390 * This function should be called near the start of the device's
3391 * runtime_resume callback.
3393 void blk_pre_runtime_resume(struct request_queue
*q
)
3398 spin_lock_irq(q
->queue_lock
);
3399 q
->rpm_status
= RPM_RESUMING
;
3400 spin_unlock_irq(q
->queue_lock
);
3402 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3405 * blk_post_runtime_resume - Post runtime resume processing
3406 * @q: the queue of the device
3407 * @err: return value of the device's runtime_resume function
3410 * Update the queue's runtime status according to the return value of the
3411 * device's runtime_resume function. If it is successfully resumed, process
3412 * the requests that are queued into the device's queue when it is resuming
3413 * and then mark last busy and initiate autosuspend for it.
3415 * This function should be called near the end of the device's
3416 * runtime_resume callback.
3418 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3423 spin_lock_irq(q
->queue_lock
);
3425 q
->rpm_status
= RPM_ACTIVE
;
3427 pm_runtime_mark_last_busy(q
->dev
);
3428 pm_request_autosuspend(q
->dev
);
3430 q
->rpm_status
= RPM_SUSPENDED
;
3432 spin_unlock_irq(q
->queue_lock
);
3434 EXPORT_SYMBOL(blk_post_runtime_resume
);
3437 * blk_set_runtime_active - Force runtime status of the queue to be active
3438 * @q: the queue of the device
3440 * If the device is left runtime suspended during system suspend the resume
3441 * hook typically resumes the device and corrects runtime status
3442 * accordingly. However, that does not affect the queue runtime PM status
3443 * which is still "suspended". This prevents processing requests from the
3446 * This function can be used in driver's resume hook to correct queue
3447 * runtime PM status and re-enable peeking requests from the queue. It
3448 * should be called before first request is added to the queue.
3450 void blk_set_runtime_active(struct request_queue
*q
)
3452 spin_lock_irq(q
->queue_lock
);
3453 q
->rpm_status
= RPM_ACTIVE
;
3454 pm_runtime_mark_last_busy(q
->dev
);
3455 pm_request_autosuspend(q
->dev
);
3456 spin_unlock_irq(q
->queue_lock
);
3458 EXPORT_SYMBOL(blk_set_runtime_active
);
3461 int __init
blk_dev_init(void)
3463 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3464 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3465 FIELD_SIZEOF(struct request
, cmd_flags
));
3466 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3467 FIELD_SIZEOF(struct bio
, bi_opf
));
3469 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3470 kblockd_workqueue
= alloc_workqueue("kblockd",
3471 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3472 if (!kblockd_workqueue
)
3473 panic("Failed to create kblockd\n");
3475 request_cachep
= kmem_cache_create("blkdev_requests",
3476 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3478 blk_requestq_cachep
= kmem_cache_create("request_queue",
3479 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3481 #ifdef CONFIG_DEBUG_FS
3482 blk_debugfs_root
= debugfs_create_dir("block", NULL
);